THE COMPLETE TECHNICAL PAPER PROCEEDINGS FROM:
A METHOD OF ANALYZING MPEG DATA IN ENCAPSULATED STREAMS
F. Eugene Rohling DVA Group, Inc.
Abstract tools for looking at stream usage of Motorola Broadband DigiCipher, Scientific Atlanta This paper describes a method of analyzing Power Key, and other access control encapsulated binary data streams for the streams are usually held close. This makes it purposes of performing detailed message difficult for an MSO to find problems in his analysis. This method evolved from a general local system, especially when he is purpose analysis tool used to analyze radar responsible for operating it. data. It is now being applied to the analysis of MPEG-2 content and access control data Encapsulated MPEG Data delivered both in-band and out-of-band. It is particularly useful for compartmentalizing the The National Access Control Service details of sensitive control and encryption (NAS) owned by Motorola Broadband and information within the MPEG data strea of an operated by AT&T (now Comcast) is an access control system.. excellent example of MPEG encapsulated data. Figure 1 shows the various layers of The method allows users to describe MPEG data. First, the DigiCipher OOB data encapsulated framed data, parsing a binary is encapsulated into MPEG private data data stream, and generating human readable message packets. When it arrives in the output that can be used to analyze and resolve headend, data is then sent from the satellite problems. The template files can be tailored receiving device (IRT) across Ethernet to the and customized to reveal varying levels of out of band modulator (OM). That is, the proprietary and confidential data within the OOB data is carried as an encapsulated MPEG binary stream. data stream within a HITS multiplex through the satellite system. [1]
INTRODUCTION Level 2 Digicipher AC Messages This paper identifies a solution that helps Encapsulation MPEG Messages test and field engineers analyze complex Level 1 Digicipher OOB Traffic MPEG data streams. It uses the familiar NAS MPEG Packets access control service as an example of data Encapsulation that has been encapsulated four times when it HITS Transport Stream Visible Layer is received within a headend system. Finally, MPEG Packets it discusses the need for these tools as new technologies emerge.
Figure 1 - NAS Encapsulation This paper specifically discusses access control data. Many off-the-shelf tools exist A standard MPEG recording tool such as for analyzing standard MPEG-2 video and the DSTS by Logic Innovations allows you to DOCSIS services. However, access control record the data stream as it is received by the systems are by their nature proprietary, and IRT. But if you want to recover only the data seen by the set-top, then you must remove the such as awk, sed, grep, lex, and perl. But HITS transport stream. converting a 100 MByte file from binary to readable text becomes unwieldy when the Several one-off tools have been built to result can generate many Gigabytes of data detunnel the data, but they are all considered and take significant time to sort through and proprietary by the AC provider. Sometimes filter that data. an MSO has legitimate reasons to determine if his access control system is operating properly It is significantly less time consuming for or if he is receiving all the data his contract analysts to process binary data and extract with NAS provides. only the information they need to do their task. Similar problems exist with Motorola DAC based local access controllers. In this HISTORY case, the problem becomes more urgent because the MSO is responsible for the The problem of analyzing a complex data operation of the DAC. stream that has been multiplexed into many layers is not unique to the cable or MPEG In many systems, access control data is industries. Instrumentation systems during the encapsulated on a TCP/IP network and sent to 1980 to 1995 time frame commonly mixed a modulating device. Rather than data being and multiplexed dissimilar data from many MPEG encapsulated in MPEG, it is now sources within a telemetry or tape recorded MPEG encapsulated within IP. While good data stream. Ethernet tools exist, they to not provide utilities to integrate with MPEG tools. [1] The Link to Radars
Compartmenting Data A good example was a radar instrumentation system developed for the F- To give the MSO the tools Motorola 15, F-16, and B-1 aircraft by Lockheed originally used to develop DigiCipher would Georgia under the Advanced Radar Test Bed be giving away the keys to their access control (ARTB) program. The requirements for that kingdom. But to give MSO’s tools that help system required it to visualize and record identify if code objects are spinning, or if TV traffic from up to four MIL-STD-1553 data Guide data is still online, or to identify if bus streams, up to four streams of telemetry channel maps are being provided to their data, several custom low, medium, and high facility are all reasonable requests. speed data streams at an aggregate rate of up to 12 Mbytes/sec. This was a feat for the A legitimate need exists to compartment 1989 designed system. They also required the the visibility of MPEG access control system to be versatile and instrument any of implementation so legitimate users can five radars on the three aircraft. The visualize it operationally without requirements finally required time stamping compromising the access control system. the data to +/- 10 microseconds.
Processing Binary Data High Speed Analysis Becomes Key Instrumenting the aircraft, multiplexing Many processing programs exist for data, recording data, and time tagging data processing text. Unix has a wealth of tools was straightforward. Much of it was performed in hardware. But the system Processing Frames of Data proved that reducing and analyzing the data became a significant labor intensive task. The MAcq filter input description was U.S. Air Force engineers likened the task of designed to process nested frames of variable finding a needle in a hay stack. length data in a serial data stream. Figure 3 shows the format of the filter file. Note that This system evolved into the bench top the format of the filter file allows recursion. Radar Instrumentation System (RIM-68) That is, optional filter frames can be nested developed by Flexible Engineering Resources, within a top level scope frame to create the Inc. (FER). This company developed a same data recursion effect often found with method of encapsulating the data in a common software recursion. This is the primary format and a method of parsing the data at benefit of applying MAcq filters to high speeds so a small number of parameters encapsulated data problems. could be visualized in both text and graphic format. The method was coined “MAcq” for Modular Acquisition. FRAME DESCRIPTOR
MACQ FILTERING [3] DATA DESCRIPTION The “macq_filter” program performed the Describes Fields within Frame analysis side of this task was called the “MAcq_filter”. It analyzed data for both real- time and post processing. It used “filters” that LENGTH DESCRIPTOR described the encapsulated nature of the data stream to both extract and process the stream Function = How to determine frame length into either human readable form or into derivative streams for off-the-shelf graphic programs to process as shown in Figure 2. Pass Filter
Input Output Function = How to Data macq_ Data determine whether this Stream filter Stream frame is of interest. (File) (File)
How to output data (Binary, Text, ...) Default = pass binary MAcq Filter File Optional recursive frame (.flt)
Figure 2 - MAcq Filter Process Figure 3 - Filter File Format
HITS STREAM NAS OOB NAS OOB MPEG PACKETS MPEG PACKETS MPEG MESSAGE
PID = NAS OOB CA Stream PID = Other Services MPEG Packets Access Control CA Stream Message to Digicipher PID = Other Services MPEG Packets Devices CA Stream PID = NAS OOB MPEG Packets
PID = Other Services
PID = NAS OOB
CA Stream PID = Other Services Access Control MPEG Packets Message to Digicipher CA Stream PID = Other Services Devices MPEG Packets
PID = Other Services
PID = NAS OOB PAT PID = Other Services pat_info.flt
PID = Other Services FILTER #ifdef SHOW_PAT PID = NAS OOB
FRAME = MPEG PACKET FRAME = MPEG PACKET
Figure 4 - Representing Encapsulation Data Description became very useful when analyzing F-16 radar data.
Each of the fields within a frame must be APPLYING MACQ TO MPEG DATA defined. The data description block within the filter identifies fields of data, such as the packet sync (47 Hex), the continuity counter, DVA Group began a research program in or the PID fields of an MPEG packet. 2002 known as “Crown Royal” or CR to identify whether MAcq could be used to parse MPEG data and generate text output files. Variable Length Data
Processing Frames of Data While MPEG packets are fixed format (188 bytes or 204 bytes), UDP / IP data is not. The length, however, can be readily Figure 4 shows how MAcq filter files can determined from the contents of the UDP be used to describe and process the NAS packet. Note the length clause contains a satellite transport stream and extract the function used to establish the length of the conditional access table (CAT). This shows a arbitrary frame. simple case of extracting OOB messages.
Need for Storage Selecting Data to be Processed
One or more pass filters look at frame Note that MPEG packets contain MPEG headers and establish whether data needs to be messages, and that MPEG messages can span passed. For MPEG data, the pass filter would multiple MPEG packets. When analyzing an likely select PIDS. For UDP data, it might MPEG stream in the general case, MPEG select UDP source and/or destination ports. messages on multiple PIDs may interleave themselves in the temporal sequence of the Once data is selected, it is then processed. MPEG stream. The MAcq scratchpad is The output section defines what data is to be useful for this case. output. Output can be formatted text such as: However, the MAcq implementation only PID=234 TIME=88:99 allows statically defined scratchpads. This was fine for only detunneling OOB data, but or it can be binary data. Outputting binary was not adequate for cross PID correlation data is quite useful for simple extraction of problems. As such, the general case of encapsulated data. That is, if all you want are providing a general PID storage for the MPEG packets from a NAS IP OOB detunneling MPEG messages was not stream going to an OM-1000, you simply adequate. Indexed scratchpads need to be detunnel the UDP packets to that device. added to the MAcq filter syntax.
Storing Data Compartmenting Knowledge
The MAcq filter allows “scratchpads” to be In this context, compartmentalization used to temporarily store data. This initially refers to the Department of Defense (DoD) style security compartmentalization used during the cold war. That is, everything is on a “need-to-know” basis. Access control providers have been Many new MPEG re-multiplexors are being reticent to only provide necessary information introduced that accept video streams across IP outside (and often inside) their corporate networks. control. Providing MSOs and vendors with too much detail places the acces control Using with Unix Pipes provider at risk, and makes the MSO vulnerable to attack. Visualizing the delivery of code objects, VOD content, channel maps, and other The MAcq filter provides a method of only necessary components of a cable system providing information on a “need-to-know” requires a tool that can output data in basis. That is, filters that describe MPEG graphical form. formatted information, or that simply announce the presence of a channel map, code The macq_filter has been used in the radar object, or conditional access table may be community to visualize its effectiveness. The appropriate for an MSO to obtain. However, tool filters, processes, and then streams the details of conditional access, especially selected data in both real-time and playback key exchanges can be hidden by simply instances into off-the-shelf 3-dimensional omitting the filters that are not needed. analysis tools.
PUTTING IT ALL TOGETHER The same can be applied to monitoring the OOB data within a headend. That is, MAcq Engineers in the cable industry have many can filter and process the access control tools at their disposal. Many off-the-shelf stream and stream data into commercially products will parse Ethernet and IP packets, (and sometimes free) third party software and others parse MPEG packets. Use of tools that display arbitrary bar graphs. This MAcq should take advantage of the strengths can be used to build tools that show code of existing tools. objects, channel maps, and other access control data as a percentage of bandwidth. Analyzing Local Access Control Data Work to Date Local AC data is often encapsulated on an Ethernet IP network. Off-the-shelf tools such DVA Group has successfully used the as Etherpeek and the Unix tcpdump utility original macq_filter program for simple tasks. provide historical recording of Ethernet IP The original program worked because 188 network in text or binary form. To make byte packets were long word aligned. It sense of the MPEG packets, however, requires enabled analysis of PID distribution, the content to be detunneled. continuity counts, and extraction of PIDS in binary form. It also allowed an encapsulated The MAcq_filter can be used to detunnel IP layer to be extracted from a given PID in an the MPEG packets and put them in a form that MPEG transport stream. MPEG analyzers can use. They can then be analyzed in native MPEG forms. But extracting an OOB stream encapsulated within IP data could not be The same solution addresses performed without being able to parse frames instrumentation of systems in which video is in byte word alignment. transported across an Ethernet IP network. SUMMARY REFERENCES
We have proven the underlying technology 1. Michael Adams, “OpenCable behind the macq_filter tool can help fill the Architecture”, 2000, Cisco Press. gaps in commercial MPEG analysis tools. 2. Dr. Keith Montierth Jr, Todd Jahng, Lisa DVA Group continues to evolve the filter tool Chiang, Greg Rohling, Gene Rohling, “Three so it properly supports the needs of embedded Dimensional Data Visualization System for cable systems in the future. the F-16 Program”, JAWS S3 Conference, June 10, 1997. MANY THANKS 3. Flexible Engineering Resources, Inc. The author extends his appreciation to “MAcq User Manual”, September 13, 1997. Michael Adams and all the people who helped write “OpenCable Architecture”. By showing a top level view of how Motorola Broadband and Scientific Atlanta conditional access systems work, we can discuss real world cable industry applications in a public forum.
ADVANCES IN OPTICAL FIBER TECHNOLOGY FOR ANALOG TRANSPORT-TECHNICAL ADVANTAGES AND RECENT DEPLOYMENT EXPERIENCE
Andy Woodfin, Jim Painter*, Boh Ruffin Corning Incorporated, *Comcast Corporation
Abstract now carry the full spectrum of voice, video, and data services. Undeniably, standard Performance assessments in analog video single-mode fiber has been the workhorse, and transport and distribution will be compared arguably the key element, in HFC design. and analyzed, based on existing commercial Improvements in transmission capabilities optical and electronic equipment used with a have, as a result, historically been designed variety of standardized optical fiber types. within the constraints of standard single-mode Particular emphasis is placed on comparison fiber characteristics. Appreciating the of capabilities with standard single-mode historical evolution of optical transmission fiber to improve SBS thresholds on the order over HFC architectures provides a useful of 2dB, with associated increases in CNR, as perspective on these constraints, and the well as improvements in CSO on the order of issues which consequently remain in nearly all 8-9dB. modern HFC optical transmission systems.
Experimental and simulated results will be AN ALTERNATE PERSPECTIVE ON presented, in addition to recent field data TRANSMISSION TECHNOLOGY collected from actual physical links deployed DEVELOPMENT by a major MSO. This is the first known commercial deployment of an alternate In early stages of HFC deployment, the optical fiber type (i.e., not standard single- benefits of transitioning from copper to mode) expressly for the purposes of improving optical fiber for CATV transport purposes analog video transport capability. were clear, with some of those advantages stated above. Significant development (and acceptance) was required in the optical INTRODUCTION transmission arena, however, to realize the large and powerful HFC networks of today. Among the primary telecommunications Optical transmission at 1310nm was typically network architectures in use today, modern viewed as sufficient where copper trunks were CATV designs provide unrivaled capability replaced with fiber, and the technology was and capacity afforded by the hybrid fiber-coax relatively mature and economically feasible. (HFC) architecture. The underlying The economics of system clustering and foundation of HFC networks is the optical regional interconnection drove the need to fiber deployed primarily in the trunk/transport adopt 1550nm transmission technology, where and distribution portions of the plant. By fiber loss is significantly less than at 1310nm eliminating RF trunk amplifiers, increasing and signals can be optically amplified with transmission bandwidth, enabling two-way erbium doped fiber amplifiers (EDFAs). transmission, and eliminating interference Standard single-mode fiber’s chromatic ingress, optical fiber has allowed CATV dispersion at 1550nm is significantly higher networks to transform into the pipes which than at 1310nm, however, which was a viable fibers. The issue of high chromatic significant issue when the only sufficiently dispersion at 1550nm, on the other hand, has linear analog transmitters were high frequency been addressed for some time in the long- chirp directly-modulated types[1]. The distance telecommunications market with development of linearized externally non-zero dispersion shifted fibers (NZDSF). modulated 1550nm transmitters addressed the NZDSF typically have dispersion on the order issue of source chirp and interaction with fiber of 3 to 4 times smaller than that of standard dispersion. However, fiber dispersion-induced single-mode fiber. Although designed self phase modulation (SPM) [2,3] was still an primarily around the considerations of high issue, in addition to exacerbation of the capacity long distance networks, NZDSF can power-limiting impact of stimulated Brillouin have direct benefit on CATV network designs scattering (SBS) (4,5) by the relatively narrow by significantly reducing the impact of linewidth emitted by externally modulated nonlinear and dispersion-related impairments sources. The severity of SBS was such as SPM and composite second order subsequently mitigated by integration of distortion (CSO). Arguably the most electrical pre-distortion and suppression significant limitation on analog transmission techniques[6], although it is still a limiting at 1550nm continues to be SBS, and the factor in a number of system designs. prevailing assumption has been that standard single-mode fiber best mitigates the effect. As The transmission technology development SBS is directly dependent on the fiber’s summarized above can be viewed in the effective area (equation 1)[6], and standard context of modification to optical fiber single-mode fiber has a larger effective area parameters, rather than working within the (typically 80µm2) than all NZDSF (typically constraints of a fixed set of assumptions. 45-72µm2)(7,8,9). However, it has been While being an interesting academic exercise, shown that some NZDSF are in fact superior it obviously does not address issues in the to standard single-mode fiber in terms of SBS installed cable plant, where the fiber threshold, by as much as 2-3dB [10,11]. infrastructure is fixed. However, such an Fibers with this capability, coupled with approach can indeed provide flexibility in optimally reduced chromatic dispersion, can designs for pending upgrades and rebuilds. show significant advantages over standard Concerning the transition from 1310nm to single-mode fiber to support real world analog 1550nm, significant reductions in attenuation transport network designs. at 1310nm would conceivably increase achievable transmission distances at the lower ASSESSMENT OF TECHNICAL wavelength and enable wider application of ADVANTAGE lower cost transmitters, allowing enabling a broader application base for 1310nm. As Details of technical capability illustrated in figure 1, however, Rayleigh scattering places a fundamental limitation on In simple terms, stimulated Brillouin the minimum achievable loss at a given scattering occurs in optical fiber due to a wavelength in current silica-based optical generation of acoustic waves in the optical fiber, and current fibers closely approach that waveguide, which create periodic variations in limit. While techniques exist to improve upon the fiber’s refractive index. This periodic these limits through exotic materials and/or variation effectively reflects part of the waveguide structures, they are not original transmitted optical power thus immediately adaptable into commercially diminishing the power seen at a receiver. The effect worsens with increasing launch power, in the figures, large area NZDSF has a so the signal reduction at the receiver cannot significantly higher SBS threshold than be overcome simply by increasing the standard single-mode fiber, in spite of the fact transmitter output. The power threshold at that it has a lower effective area (72µm2 and which SBS begins to quickly deteriorate a 80µm2, respectively). A relevant point to signal is given by: consider is that the relative differences in SBS 21A thresholds are nominally constant regardless (1) P ≅ eff th of electronic-based SBS suppression gL Beff techniques. In other words, a transmitter with where P is the SBS-dictated optical power th maximum SBS-limited power of 16dBm on threshold (in dBm), A is the fiber effective eff standard single-mode fiber could support area, L is the nonlinear interaction length, eff approximately 18dBm over large area and g is the peak Brillouin gain of the fiber. B NZDSF, while a 17dBm standard single-mode As stated previously, standard single-mode fiber rated transmitter could accommodate a fiber A is larger than that of typical NZDF, eff similar 2dB increase (to 19dBm) over large but significant variability in threshold among area NZDSF. Also significant is the fact that different fiber types due to variation in other NZDSF varieties can not support the Brillouin gain characteristics has been SBS threshold allowed on standard single- empirically explored. Regardless, this mode fiber. phenomenon is commonly overlooked and effective area dependence is typically the only Aside from variation in SBS suppression consideration made. With the appropriate capabilities among standard single-mode fiber combination of reasonable effective area and the NZDSF variants, an approximation (>70µm2) and Brillouin gain, some NZDSF can be made to assess the introduction of can support higher SBS thresholds than second order distortion in different fiber types. standard single-mode fiber. Second harmonic distortion for a chirp-free Figure 2 shows an SBS threshold externally modulated source, as determined by comparison between several commercially fiber dispersion and nonlinear refractive available optical fiber types. Considering index, can be expressed as: standard single-mode fiber as the presumed 1 1 2πN standard for SBS threshold, the most (2) β&& 22 −Ω β&& Ω222 Pzmzm 2 ,[1] λ commonly deployed NZDSF varieties were 4 2 Aeff evaluated in comparison. The three NZDSF where m is the modulation index, z is the fiber variants considered were: large area NZDSF, length, Ω is the modulation frequency, P is characterized by a relatively high effective launched optical power, N is the Kerr µ 2 2 area (approximately 72 m )[7] in comparison nonlinear-index coefficient, λ is the to other NZDSF; high dispersion NZDSF, transmitter center wavelength, and Aeff is the with relatively high chromatic dispersion at fiber effective area. Also, note that 1550nm (~8ps/nm*km)[8]; and reduced slope 2 && −= πλβ )2/( Dc NZDSF, characterized by relatively low chromatic dispersion slope and very small is the second-order fiber dispersion coefficient effective area at 1550nm (0.045ps/nm2*km , where D is the fiber dispersion coefficient. For standard single-mode fiber, large area and ~55µm2, respectively)[9]. All fibers under NZDSF, reduced slope NZDSF, and high test were at a nominal length of 50km, and dispersion NZDSF, we can consider the tested in the configuration illustrated in figure chromatic dispersion at 1550nm (17, 4, 5.2, 3, with backscattered signals detected through 8ps/nm*km, respectively) and effective area a self-heterodyne configuration. As indicated (80, 72, 55, 63 µm2, respectively. Assuming Taking Advantage of the Technical Benefits all other terms in (2) are constant, we can make a qualitative assessment of the relative A number of potential performance magnitude of CSO impairment in each fiber advantages can be identified considering the by scaling the ratio of fiber dispersion to combined impact of increased SBS effective area, D/Aeff. As is evident from the suppression and optimized chromatic table, all NZDSF should have a significantly dispersion. The most readily apparent benefit reduced CSO distortion relative to standard gained from an increase in SBS threshold on single mode fiber. large area NZDSF is the capability to support higher optical launch powers, and Fiber Type D/Aeff ratio consequently extend the distance over which (ps/nm*km*µm2) in-line optical amplifiers (EDFAs) would Standard single- 0.212 otherwise be required. Assuming equivalent mode loss characteristics on large area NZDSF and Large area 0.056 standard single-mode fiber, and considering NZDSF only SBS, a 2dBm increase in SBS threshold Reduced slope 0.094 would translate to approximately 8km NZDSF increased distance with equivalent end-of-line High dispersion 0.127 received power. Coupled with a reduced NZDSF chromatic dispersion and optimal effective area, however, large area NZDSF can further If only performance parity with standard increase capability by both supporting higher single-mode fiber is desired, the comparative powers and mitigating distortions. Indeed, assessment of SBS thresholds carries previous studies have demonstrated 100km significant implications in the choice of fiber transmission over large area NZDSF with no type to deploy. While large area NZDSF can repeaters or in-line EDFAs[10], and at shorter support any given single wavelength 1550nm distances (50km) with high launch power transmission scenario designed around demonstrated significant CSO and CNR standard single-mode fiber constraints, a advantage with large area NZDSF (CSO<- system design would otherwise require careful 65dBc, CNR>50dB), compared to standard consideration and possible power budget de- single-mode (CSO<-52dBc, CNR>44dB) and rating to avoid significant signal degradation reduced slope NZDSF (CSO<-37dBc, if deployed over other types of NZDSF (i.e., CNR>24dB). By extension, this capability high dispersion NZDSF and reduced slope could extend to supporting longer reaches or NZDSF). The true justification for a choice of superior signal integrity over a fixed distance fiber other than standard single-mode fiber with large area NZDSF while remaining would obviously come from a desire to within the constraints of existing design rules achieve performance benefits, as opposed to (e.g., maximum allowable number of simple parity with the effective standard cascaded in-line EDFAs). Link engineering (standard single-mode fiber). Therefore, the rules can also potentially be extended since gains derived from exploiting an increased the input power to cascaded in-line EDFAs SBS threshold, as well as the merits of can increase due to higher launch powers, thus reduced chromatic dispersion and other improving EDFA output CNR. optimal parameters, warrant exploration.
Given the broadcast nature of analog video displacement of more costly externally- transport, another beneficial application of modulated sources to address trunking increased launch power capability with large applications as opposed to simple signal area NZDSF would be the potential to insertion. The inherent SBS suppression increase the number of remote locations resulting from modulation-induced spectral supported with a single transmitter. broadening, coupled with the improved power Particularly for those locations not characteristics due to the lack of an immediately targeted for advanced services, attenuating modulator section, aids in drawing the economics of basic service distribution a significant comparison with conventional from a single transmitter become appealing. long reach externally modulated sources. This As an example, as illustrated in figure 4, an scenario is currently being experimentally additional 2dBm maximum launched power evaluated at Corning. could scale from a 1x8 passive splitter (loss=9dB/output) to accommodate the FIELD DATA FROM DEPLOYED CABLE additional 2dB loss encountered on each arm of a 1x12 split (loss=11dB/output). Available commercial transmission equipment operating over a contiguous link of standard single-mode fiber was not capable of supporting internal CSO requirements of Pin 68dB in the system link depicted in figure 5. 1x8 As suggested previously, a reduction in total link chromatic dispersion could potentially mitigate CSO brought about by direct A interaction between fiber dispersion and residual transmitter chirp, as well as CSO introduced by SPM-induced signal chirp (which also has some dependence on fiber Pin+2dBm effective area). Indeed, concatenating a 1x12 56.4km length of large area NZDSF to the B previously installed 53.2km of standard single-mode fiber enabled a significant improvement in CSO. With an initial Figure 4: Splitter configurations: A-with transmitter CSO of 76.9dB, the contiguous standard single-mode fiber, B-with large link of all standard single-mode fiber received area NZDSF 61.6dB and 59.4dB at channels 36 and 67, respectively. By introducing large area A promising possibility, again born from NZDSF into the latter portion of the total link, the coupled advantages of reduced chromatic thereby reducing the overall accumulated dispersion and increased SBS threshold in chromatic dispersion, received CSO values large area NZDSF, is the ability to with identical system parameters were 70.9dB significantly increase the usable range of and 71.4dB at the respective channels. For the directly modulated 1550nm PEG transmitters. two monitored channels, 9.3dB and 12dB Typically characterized by significant improvements in CSO were realized over the frequency chirp and thus severely limited by total link, reducing the impairment such that it dispersion-induced CSO, the introduction of was well within the internal requirement. Note large area NZDSF with reduced dispersion in addition the increased magnitude of could potentially allow for PEG transmitter improvement at the higher modulation consideration of system solutions that can frequency. Marginal improvements in CTB meet challenging performance requirements, were also realized with the heterogeneous extend the capabilities of existing standard single-mode/large area NZDSF link, transmission equipment, and provide with 0.3dB and 1.1dB improvements at the opportunities to deliver significant savings in respective monitored channels when network flexibility and equipment cost. The compared with the homogeneous standard capabilities of non-zero dispersion shifted single-mode fiber link. Note again the slight fibers to significantly mitigate signal increase in the performance delta at the higher distortions are beginning to be explored in modulation frequency. actual installations. Moreover, the large effective area subset of NZDSF allows for the CONCLUSION broadest range of performance capability improvements among alternate fiber types, Looking at the evolution of CATV and in comparison to standard single-mode networks and systems free from the technical fiber. constraints of the majority installed base of standard single-mode fiber allows for
0.7
0.6
0.5
0.4 dB/km 0.3
0.2
0.1
0.0 1280 1320 1360 1400 1440 1480 1520 1560 1600 1640 Wavelength (nm)
Figure 1: Typical attenuation curves for standard single-mode fiber (solid curve) and low water peak standard single-mode fiber (dotted curve), and fundamental Rayleigh scattering limit (dashed line)
20
10
0
-10 Scattered Power (dBm) -20
-30
-40 02468101214 Input Power (dBm) Large Area NZDSF Reduced Slope NZDSF High Dispersion NZDSF Std Single-Mode Figure 2: Comparison of input optical power and scattered optical power. Elbow of curve indicates approximate location of nonlinear onset of Brillouin scattering, values indicated by vertical markers for reduced slope NZDSF, high dispersion NZDSF, standard single-mode, and large area NZDSF, respectively.
Figure 3: Experimental configuration for evaluating SBS threshold
2 dB 2 dB 7.8 A
23.2 Km Out 56.4 Km *Channel #36 16 dBm 0.1 Standard single-mode 14.1 gle-mode C/N = 55.7 dB Standard sin CSO = 76.9 dB Channel #36 Channel #67 CTB = 75.5 dB *Transmitters Input distortions near theoretical limit of analyzer. C/N = 48.3 dB C/N = 50.7 dB CSO = 61.6 dB CSO = 59.4 dB CTB = 70.6 dB CTB = 72.8 dB
2 dB 2 dB 7.8 B
*Channel #36 23.2 Km Out 56.4 Km 16 dBm 05 C/N = 55.7 dB Standard single-mode 14.1 Large area NZDSF CSO = 76.9 dB Channel #36 Channel #67 CTB = 75.5 dB C/N = 50.0 dB C/N = 51.8 dB CSO = 70.9 dB CSO = 71.4 dB CTB = 70.9 dB CTB = 73.9 dB Figure 4: Configuration of installed links for comparison. A-Contiguous standard single-mode fiber link, B-Standard single-mode fiber extended with large area NZDSF.
REFERENCES
[1] M.R. Phillips, T.E. Darcie, D. Marcuse, [5] R.G. Smith, “Optical Power Handling G.E. Bodeep, and N.J. Frigo, “Nonlinear Capacity of Low-Loss Optical Fibers as Distortion Generated by Dispersive Determined by Stimulated Raman and Transmission of Chirped Intensity-Modulated Brillouin Scattering,”Appl. Opt., vol. 11, Signals,” IEEE Photon. Technol. Lett., vol. 3, pp.2489-2494, 1972. pp. 481-483, 1991. [6] G. P. Agrawal, Nonlinear Fiber Optics, 2nd [2] F.W. Willems, W. Muys, J.C. van der ed., (Academic Press, San Diego, 1995). Plaats, and R. Nuyts, “Experimental Verification of Self-Phase-Modulation [7] LEAF Fiber Product Information Sheet, Induced Nonlinear Distortion in Externally Corning Incorporated, September 2002. Modulated AM-VSB Lightwave Systems,” Electron. Lett., vol. 32, pp. 1310-1311, 1996. [8] TeralightTM Metro Fiber Product Information Sheet, Alcatel, February 2002. [3] R. Chraplyvy, “Limitations on lightwave communications imposed by optical-fiber [9] TrueWave RS Fiber Product Information nonlinearities,” J. Lightwave Technol., vol. 8, Sheet, OFS Optics, February 2003. pp. 1548–1557, 1990. [10] C. C. Lee and S. Chi, “Repeaterless [4] X.P. Mao, G.E. Bodeep, R.W. Tkach, A.R. Transmission of 80-Channel AM-SCM Chraplyvy, T.E. Darcie, and R.M. Signals Over 100-km Large-Effective-Area Dispersion-Shifted Fiber,” IEEE Photon. Derosier, “Brillouin Scattering in Externally Technol. Lett., vol. 12, pp. 341-343, 2000. Modulated Lightwave AM-VSB Transmission,” IEEE Photon. Technol. Lett., vol. 4, pp. 287-289, 1992. ANYTHING, ANYTIME, ANYWHERE: OPEN ADVANCED BANDWIDTH MANAGEMENT OF ON-DEMAND SERVICES
Michael Chen Concurrent Computer Corporation
Abstract customers the full promise of on-demand services – “anything, anytime, anywhere”. The convergence of new technologies in an affordable fashion has given rise to new To deliver this, the MSO is faced with a features that not only bolster customer bewildering array of challenges, from the demand, but also provide new revenue- selection and installation of compatible generating opportunities. The ability to equipment to the configuration, management, deliver the full promise of on-demand and maintenance of this new infrastructure. services -- “Anything, Anytime, Anywhere” -- is finally within reach, and customers are These issues facing both MSOs and clamoring for their providers to deliver. equipment vendors today, as well as other looming issues, are further discussed below. New features available now, and some of The new features and capabilities of these those envisioned for the future, are identified systems, both at present and in future, are and investigated, as are the issues which face also identified and investigated. Finally, providers and vendors today. Observations observations and recommendations are made and recommendations on the next generation for the design, procurement, and deployment of systems and their architectures are then of next-generation architectures and systems. offered in closing. GIGABIT ETHERNET ON-DEMAND INTRODUCTION SYSTEMS
Increased demand, competitive market Current on-demand systems are largely forces, and technology advances have placed being deployed using Gigabit Ethernet output. Gigabit Ethernet at the heart of new cable A typical video-on-demand (VOD) system architectures offering additional revenue employing Gigabit Ethernet looks like this: opportunities to the Multiple System Operator (MSO). (optional) VOD System Video GigE GigE GigE Optical Controllers Server Switch Transport The adoption of standard Internet (BMS, HERM, etc.) protocols has made the pervasive switching
and routing capabilities which power the Headend Upstream Out-of-Band Controllers RF Network RF Network Internet available to these video delivery (DAC/DNCS, etc.) systems. QAM Set-Top via RF Edge GigE Optical These capabilities provide a framework Box Coaxial Device Optical Transport copper which, combined with new techniques such as fiber network-based personal video recording (PVR), allow the MSO to deliver their Figure 1. Typical Gigabit Ethernet VOD system.
The Gigabit Ethernet output of a Since streaming servers can now saturate streaming server is sent to an edge device, Gigabit Ethernet links, a switch is no longer where it is combined and converted to a form technically needed for deployment. However, suitable for display on digital cable set-top the use of switches also provides new routing boxes. The server’s output may be connected flexibility that was either unavailable or cost- into a switch, and optical transport gear is prohibitive with prior output formats, and used when necessary to transmit the signal many of the new features which Gigabit across large distances. Ethernet enables are built upon this functionality. For this reason, using a The streaming server output is switched Gigabit Ethernet transmission encapsulated within User Datagram Protocol framework is still quite advantageous for (UDP) packets, defined as part of the Internet these on-demand services. Protocol (IP) standards to provide low- latency data delivery, while taking advantage Asymmetric Deployment and Expansion of the wide range of products and services the Internet explosion has produced. The division of labor between the streaming server and the edge device in the Note that the output transmission is often Gigabit Ethernet framework offers the MSO a implemented unidirectionally, since this new method for system deployment and allows the MSO to effectively double the expansion. Gigabit Ethernet’s switching and amount of fiber bandwidth available. This routing functionality allows streaming servers one-way connection may require additional and edge devices to be loosely rather than effort to configure systems initially, since tightly coupled. The MSO can then deploy many standards used with IP protocols and expand edge devices separately from the assume the existence of a bi-directional streaming servers, allowing an asymmetrical network link for proper operation. buildout of the system.
Gigabit Ethernet on-demand systems A typical asymmetric buildout will today are usually allocated dedicated network overprovision the radio frequency (RF) edge bandwidth for streaming. This often stems with more edge devices than necessary to from the difficulty of ensuring sufficient satisfy initial bandwidth demands. This is quality of service to protect on-demand because installing new edge devices is often streams from being damaged by other data difficult to do without impairing the RF signal traffic. Having dedicated bandwidth for to a node, and requires more truck rolls to streaming, which the streaming servers then accomplish. The available granularities of manage among themselves, greatly simplifies optical transport equipment often will favor the overall system, and has accelerated the having more optical transport capacity than availability of Gigabit Ethernet solutions. initially required, which may prompt the MSO to overprovision with edge devices at the To Switch or Not to Switch? same time.
Gigabit Ethernet switches were used in early deployments to aggregate the outputs of one or more streaming servers, when these servers were unable to generate enough traffic to fill an entire Gigabit Ethernet link. Server Edge Device QAM via RF Node However, successful trials of subscription Coaxial STBs Server GigE GigE Edge Device copper video-on-demand content indicate that MSOs may not need to supply customers with DVR Server QAM Edge Device via RF Switch Node Coaxial STBs boxes to satisfy their desire for more varied Server Edge Device copper on-demand content, as long as they can make QAM Server Edge Device via RF Node desirable content available to their Coaxial STBs Server Edge Device subscribers. copper Network-Based PVR Figure 2. Asymmetric deployment and expansion. In a network-based PVR approach, Such an asymmetric buildout will broadcast programming is recorded and generally add only as many streaming servers stored by the MSO at the headend, rather as required to meet current demand; as than inside a consumer’s set-top box, and is demand increases, more servers can be added made available to on-demand streaming at the headend and assigned to RF outputs of servers for transmission to customers upon edge devices. request. Some implementations of network- based PVR allow a customer to pause a REALITY TODAY program in real time and use standard navigation features such as fast forward and What can today’s Gigabit Ethernet on- rewind. demand solution currently provide? The advent of network-based PVR “Something, Anytime” solutions levels the playing field with home PVR boxes, and allows the MSO to provide Current solutions have limited on-demand the full range of broadcast programming on content available. This is often not a storage demand, in addition to PVR functionality, capacity issue, but rather a rights licensing without upgrading any customer premises issue. The limited availability of on-demand equipment. content may force the MSO to select content for the on-demand system that is assumed to However, existing carriage agreements be more compelling than the broadcast digital are likely to require renegotiation before cable offerings. This content typically broadcast programming will be allowed for includes movies, special events such as on-demand viewing, so MSOs must concerts, and popular sporting events. aggressively pursue content rights to achieve the full potential value of network-based Now Playing: “Anything, Anytime” PVR. Personal video recorders (PVRs) such as “Many Streams, Each To There” Tivo can provide a wider selection of on- demand content to the home, but the limited Current on-demand solutions can be availability of PVRs with integrated digital scaled to meet the MSO’s streaming capacity cable functionality curtails the overall benefit needs for their digital subscribers. However, to the customer. PVRs also remove content these solutions often suffer from inflexible storage control from the MSO at the home. routing that dates from the previous This raises content protection issues, which generation of transmission technology such as tend to ripple back into rights negotiations. DVB-ASI and integrated quadrature again be performed at the hub level, not at amplitude modulation (QAM) and the overall system level. upconversion. Since this transmission equipment had little or no switching and THE PROMISE OF TOMORROW routing capability, and the capability was often not cost-effective when available, each “Any Stream Anywhere” streaming session had a fixed route to its “Any Stream Anywhere” is a phrase used destination. This meant that only a smaller to describe a system where any stream being subset of on-demand servers could stream sent from a streaming server can be directed content to a given customer’s set-top box. to any set-top box. Looking from the other QAM direction, this also means that any streaming GigE via RF Server Edge Device Node STBs Coaxial server can satisfy a stream request from any GigE copper Server Edge Device particular set-top box.
QAM GigE via RF Server Edge Device Node STBs A system with this property has many Coaxial GigE copper clear advantages. Since all streaming servers, Server Edge Device not only a subset, can satisfy a node of set-top boxes, the total capacity provided by these Figure 3. Fixed server-to-edge device routing. servers can be sized against the demand of the overall system, instead of sizing each subset For the MSO, these constraints meant that individually. This both eliminates unnecessary systems had to be designed for and sized to equipment, and also greatly simplifies the the peak demand expected at each hub, rather processes for installation and expansion. than the peak demand expected from the MSOs can set aside reserve streaming overall system. MSOs responded by defining capacity to cover the entire system, rather system pricing in terms of cost per than separate hubs or nodes. simultaneous stream, independent of service grouping or location. This pushed the cost of A switched Gigabit Ethernet transmission additional equipment to satisfy per-hub rather framework can easily support the “Any than overall requirements back on the Stream Anywhere” model, using the equipment vendors, resulting in lower margins switching and routing functionality provided and profit from these sales. to direct traffic from any server to any edge device which transmits to a given set-top box. This architecture is workable, but clearly A conceptual diagram of “Any Stream not optimal for either MSOs or equipment Anywhere” for a system using a switched vendors. MSOs must deploy larger systems Gigabit Ethernet transmission framework is that would otherwise be necessary, which shown below. impacts operational and maintenance costs, as well as complicating the issue of failure recovery. Equipment vendors must absorb costs imposed by sizing constraints at each hub, rather than at the overall system level. Performing asymmetric expansion of an on- demand system is further complicated by this routing inflexibility, since the expansions must Server Edge Device QAM via RF Node A rudimentary level of RF resource STBs Server Edge Device Coaxial GigE GigE copper sharing is easily achieved with a static partitioning of the available RF frequencies Server QAM Edge Device via RF Switch Node Coaxial STBs between the services sharing the RF network. Server Edge Device copper This avoids most possibilities of conflict QAM Server Edge Device via RF Node between services, but is clearly not optimal Server Edge Device Coaxial STBs since resources unused by the assigned copper service are not available for reuse by other services. Figure 4. Gigabit Ethernet Any Stream Anywhere.
An incremental improvement can be A basic implementation of “Any Stream gained by changing the partitioning so that Anywhere” using switched Gigabit Ethernet program numbers and their associated RF can take advantage of the fact that these bandwidth can be assigned to services, instead systems are usually given dedicated network of entire RF frequencies. However, the lack bandwidth. As long as the streaming servers of mechanisms to guarantee quality of service can manage the available bandwidth properly, (QoS) at this level makes it possible for an ill- while taking into account the new switched behaved service to disrupt other services infrastructure, few if any significant changes which share the same RF frequency. should be required to add the ability to support “Any Stream Anywhere” in an Dynamic partitioning of these resources is existing centralized Gigabit Ethernet system. clearly more efficient, but requires a resource management system to arbitrate requests. If “Anything, Anywhere”: Sharing Resources the site in question uses the Scientific-Atlanta Between Multiple Services headend infrastructure, the Digital Network Control System (DNCS) is responsible for A digital cable transmission system using performing this function, using the DSM-CC Gigabit Ethernet has at least two distinct protocol specified in the MPEG-2 standard. networks with resources to manage: the However, if the site uses the Motorola Gigabit Ethernet network used between headend infrastructure, no such entity streaming servers and edge devices, and the manages the RF resources. In this case, VOD RF network between edge devices and set-top system vendors have typically implemented boxes. These systems also have an Ethernet their own internal management to handle network for command and control resource sharing. Requests from other information, but that network is managed services for resource sharing can be independently and falls outside the scope of accommodated by sending these requests to this discussion. the VOD system for fulfillment.
RF Resource Sharing At present, few services attempt to share RF resources with VOD systems, and the Each RF frequency available has two small number of involved parties makes separate but related resources to manage: the solutions by private arrangement feasible. program numbers which can be individually But as more potential services emerge, and tuned by set-top boxes, and the bandwidth providers begin to call for unified multiple which all programs using the same frequency vendor support, open standards should be must share. adopted to define the interactions required for Gigabit Ethernet Quality of Service these services to share common resources. There are several different methods, such Gigabit Ethernet Resource Sharing as IP precedence, IP Type of Services (ToS), and Differentiated Services Code Point Resource sharing for the Gigabit Ethernet (DSCP), which can be used to specify which network is simpler, thanks to both its inherent quality of service policy should be applied, if switching and routing functionality, and the any, to IP traffic. Some of these methods suite of Internet protocols available for use. overlap, and may conflict with one another if Like the RF network, Gigabit Ethernet not configured and used carefully. networks have at least two separate resources to be managed: the addresses used to identify Fortunately, streaming servers are each device on the network, and the relatively immune to this problem, since the bandwidth available for data traffic. switch that receives their output can be configured to tag all incoming traffic on an Ethernet devices generally have unique input port with particular QoS settings. The Media Access Control (MAC) addresses, so streaming server is therefore not required to only IP addresses generally need to be directly know how QoS will be implemented. managed. The Address Resolution Protocol (ARP), part of the standard suite of Internet Edge devices are not so lucky, and so protocols, handles the matching of IP should be capable of receiving input with QoS addresses with appropriate MAC addresses, tagging. QoS indications are not currently and the Dynamic Host Configuration Protocol used to signal the relative priority of (DHCP) is often used to assign IP addresses individual streams; therefore, vendors may to devices, whether on a static or dynamic note that it is safe for the edge device, as the basis. last device in the chain, to ignore the QoS indications it receives. Gigabit Ethernet network bandwidth can be statically allocated in a fashion similar to Note that although lost data can the RF bandwidth allocation described sometimes be tolerated by other applications, previously to provide a rudimentary level of within streaming video server output such sharing between services. Without any losses are almost always clearly visible and quality of service guarantees, an ill-behaved objectionable to the customer. In light of this or misconfigured service can once again fact, best-effort queuing policies to enforce disrupt other services sharing the same QoS are much more suitable for digital cable network. transmission than policies which result in lost traffic. The effects of this disruption can be significantly worse for Gigabit Ethernet, since Gigabit Ethernet Bandwidth Reservation the vastly increased bandwidth available encourages a correspondingly higher number The standard Internet protocol used to of sessions per link to share the network. But perform network bandwidth management is in this case, the Internet comes to the rescue, the Resource Reservation Protocol (RSVP). since mechanisms have been developed to This protocol allows a receiver to establish a ensure quality of service for IP and Ethernet bandwidth reservation between itself and a traffic. specified source. Dynamic partitioning of network bandwidth between services can be which is why unidirectional transport from the readily accomplished with this protocol. streaming server to the RF edge is often implemented. However, having bi-directional A crucial RSVP feature is its ability to connectivity at the edge would enable much accommodate portions of the network that simpler autodiscovery and autoconfiguration are not RSVP-aware. This feature enables methods, and allow standard protocols used the gradual introduction of RSVP at sites by the Internet such as ARP and RSVP to with existing equipment that predates or accomplish their tasks. otherwise does not support it. Although many existing Gigabit Ethernet switches The establishment of bi-directional edge support RSVP, and optical transport connectivity, with only unidirectional equipment is generally not required to do so, transport to the edge, requires a switch to existing streaming servers and edge devices exist between every edge device and the largely do not support RSVP. Even if direct optical transport feeding it. This can become support for RSVP is added, the encapsulation expensive, but an emerging new breed of of these messages within UDP multicast equipment, combining switching and optical packets may be required, as specified in transport capability in the same device, may Annex C of RFC 2205. This is due to various prove well-suited to this task. operating system and security issues regarding the use of raw sockets. As an alternative, devices may support methods such as the Unidirectional Link The fact that many Gigabit Ethernet Routing (UDLR) protocols specified in RFC switches provide RSVP support is again 3077 to logically create an upstream network advantageous to streaming servers, since path over a different connection, such as the these switches may act as a sender proxy and command and control network. hide the details of RSVP operation from the connected servers. In this situation, the Autodiscovery and Autoconfiguration switch maintains RSVP states, and generates required downstream “Path” messages in Automatic discovery and configuration response to received streaming input. methods are not strictly required for these systems to be deployed. However, for MSOs Edge devices are, once again, not as lucky unfamiliar with the intricacies of these new and may be required to directly support systems, any automation that can help reduce RSVP. The primary reason for this stems the probability of misconfiguration, and also from the fact that unidirectional transport simplify system expansion, will clearly be of from streaming server to edge device is often great value. employed to better utilize the available optical fiber. For RSVP, the receiver must initiate an However, to implement autodiscovery upstream request for bandwidth reservation, and autoconfiguration, bi-directional but it is unclear what upstream path will be connectivity and support for each device is available to the edge device. required. Set-top autodiscovery schemes can use the upstream communications link Bi-directional Edge Connectivity provided by the RF network to perform these functions, but this makes open standardization Downstream video traffic requires much difficult. Existing network equipment with more bandwidth than upstream control traffic, full support for autodiscovery and autoconfiguration methods may require The multiple possible paths introduced by modifications to work with unidirectional complex network topologies make routing links, as described above. and other management tasks much more difficult. However, complex topologies are Complex Network Topologies generally chosen because they can be more flexible, and also more resilient if problems Up to this point, the discussion of arise. Other possibilities which may drive “Anything, Anywhere” has been based on a MSOs to adopt more complex network simple centralized model, where streaming topologies include the regionalization of sources and switches are located at the master functions such as broadcast feed generation, headend, and their output is distributed to network-based PVR content ingestion, and hubs and nodes using optical transport. reserve streaming capacity. Although the simplicity of this model eases From S Master Headend Edge the discussion of issues which are not S EdgeEdge Master Headend Edge dependent on topology, real-world systems EdgeEdge are much more complicated.
GigE Switch
N Hub 1 1 0 2 6 0 0 0 0 0 0 0 2 0 1 4 4 4 4 4 4 2 2 2 2 2 2
W E Hub Hub N E W SW SSW SE Hub Hub Hub Hub Hub Hub
Edge Edge Edge EdgeEdge EdgeEdge EdgeEdge
S SE Edge Edge Edge Edge Edge Edge SW EdgeEdge EdgeEdge EdgeEdge Edge Edge Edge Master Hub Hub Headend Figure 6. This example is centralized... almost!
SSW From W Hub or S Master Headend Hub Edge EdgeEdge Edge W S EdgeEdge Hub Master Edge EdgeEdge Headend Edge EdgeEdge
0 4 Figure 5. A slightly more complicated topology. 2
GigE Switch Since the cost of putting optical fiber in
1 1 2 6 0 the ground is prohibitive, the topology of 0 0 0 0 0 0 0 4 4 4 4 4 4
2 2 2 2 2 2 0
2 available fiber often dictates that of the 1 services it carries. In other cases, the W available space at headend locations may N E Hub SW SSW SE Hub Hub (from S Hub Hub Hub Master constrain the amount of equipment that can Headend)
Edge Edge Edge be installed. In addition, headends often also EdgeEdge EdgeEdge EdgeEdge
Edge Edge Edge Edge Edge Edge act as hubs to serve local customers. Lastly, EdgeEdge EdgeEdge EdgeEdge Edge Edge Edge redundant equipment is often used to provide failover capabilities. The net result of all this Figure 7. A server elsewhere on the main ring is that most real-world architectures diverge makes things more complicated. significantly from the ideal centralized model. Existing network devices and approaching. In fact, HDVOD may have architectures have sturdy mechanisms already arrived! HDVOD content differs available to handle failure detection and from standard VOD content only in video recovery, as well as other issues such as resolution and bit rate, but support for these bandwidth reservation and quality of service. higher resolutions and bit rates can have However, in some cases, the Internet solution ripple effects throughout an on-demand does not quite fit the digital cable problem. system. Care must be taken in both the For example, reserving bandwidth for a underlying infrastructure and devices stream to a set-top box differs from the themselves to ensure that HDVOD content typical Internet case, due to the separation does not cause design limits to be exceeded. between the IP network and the RF network. A simpler problem thus becomes complicated RECOMMENDATIONS in the digital cable space. Considerations for selecting and IP Network RF Network purchasing equipment for deployment:
Serve Edge Device C Streaming servers should be able to fill a S O Gigabit Ethernet link so that switch ports and Serve W M I Edge B transport bandwidth are fully utilized. T Device I STB Serve C N H E Switches and/or routers should implement Edge R port queuing and QoS policy enforcement in a Serve Device fashion compliant with streaming content requirements. Also, switches, routers, and transport equipment should only minimally Figure 8. Gigabit Ethernet IP and RF networks. modify the nature and timing of streaming
media content. This is simplified by selecting The challenge here is to integrate network equipment verified by vendors to interoperate management functionality with the resource correctly with other components, including management of on-demand streaming both streaming server and edge device. systems. If this integration is performed at a high enough level, each piece can manage its Edge devices should be upgradeable to own responsibilities and ask the others when support both variable bit rate (VBR) and external resources are required. But if the high-definition (HD) streaming input. integration is performed at too low a level, Devices with better buffering and dejittering the labor required to properly configure a capabilities are generally preferred over their system with complex topology may require competitors. that working autodiscovery and autoconfiguration methods be devised and It must be decided up front whether implemented first. asymmetrical deployment and expansion now merits the increased capital expenditure that it High-Definition Video-on-Demand requires at initial rollout. Future cost (HDVOD) projections for needed equipment will clearly play a significant role in this decision, as will The advent of high-definition (HD) the bargaining power brought by higher- content for VOD systems is quickly volume and/or integrated purchases. Considerations for designing or deploying development of next-generation features to a system: drive the next wave of business.
The rollout and expansion of proven For the MSO, this is an exciting time to revenue sources such as on-demand services be in the business, due to the convergence of should not be delayed to wait for the promise several new technologies in an affordable of resource sharing with other services. The fashion. This recent development has given revenue to be gained now facilitates the rise to new features that not only bolster expansion for these services later, and is a customer demand, but also provide new valuable hedge against the chance that other revenue-generating opportunities. The wide services may not end up as viable range of services available to customers has opportunities for additional revenue. neven been more compelling. The ability to deliver “Anything, Anytime, Anywhere” is The network topology should not be finally within reach, and customers are complicated more than absolutely necessary, clamoring for the MSO to deliver this unless the benefits of doing so are tangible promise. The last remaining hurdle is to and compelling. standardize rollout procedures to make them suitable for mass deployment, and then the Component interactions should be kept at MSO can let the good times roll. a high level when possible to accommodate differing implementation at lower layers. This REFERENCES avoids unnecessary problems that can arise from conflicting decisions made in the design Chen, Michael, “Evolving the Network and implementation of individual components. Architecture of the Future: The Gigabit Ethernet Story”, SCTE Cable-Tec Expo The use of open standards should be Proceedings, May 2003. encouraged for interoperability whenever feasible, but may not be required for existing Braden, R., Ed., Zhang, L., Berson, S., or near-term deployments. This prevents Herzog, S. and S. Jamin, "Resource unnecessary and unavoidable delays for ReSerVation Protocol (RSVP) – Version 1 acceptance and integration from impacting the Functional Specification", RFC 2205, timetables for these deployment. September 1997.
CONCLUSION Duros, E., Dabbous, W., Izumiyama, H., Fujii, N. and Y. Zhang, “A Link-Layer Equipment vendors in this space hold an Tunneling Mechanism for Unidirectional enviable position; they are poised in a market Links”, RFC 3077, March 2001. ready to explode with new business, and are positioned well to capitalize on that fact. The Postel, J., “User Datagram Protocol”, new features needed by MSOs are already RFC 768, August 1980. being developed and deployed now, while open standards are being refined and Droms, R., “Dynamic Host Configuration proposed to allow smoother integration and Protocol”, RFC 2131, March 1997. interoperability for the future. The acceptance and adoption of these standards will allow vendors to focus on the Plummer, David C., “An Ethernet ISO/IEC International Standard 13818; Address Resolution Protocol“, RFC 826, "Generic coding of moving pictures and November 1982. associated audio information", November 1994. Nichols, K., Blake, S., Baker, F. and D. Black, “Definition of the Differentiated Gigabit Ethernet Alliance, “Gigabit Ethernet: Services Field (DS Field) in the IPv4 and Accelerating the Standard for Speed”, 1999. IPv6 Headers“, RFC 2474, December 1998. http://www.10gea.org/GEA-Accel1999_rev- wp.pdf ARTIFCIAL INTELLIGENCE IN CABLE TV APPLICATIONS
Louis P. Slothouber, Aaron Ye BIAP Systems, Inc.
Abstract
After many years and billions of dollars motely programmable computer in the home invested in a digital network infrastructure provides the cable TV industry an opportu- the cable TV industry finds itself unable to nity to provide subscriber services that both fully capitalize on that investment. Under- Microsoft and the digital broadcast satellite powered set-top-boxes, daunting integration (DBS) providers must envy. issues, lack of standards, and huge capital costs hinder the roll out of new subscriber Unfortunately, the evolutionary nature services at a time when competition from of the digital upgrade process has produced digital satellite providers is becoming acute. an architecture that is ill designed to support Fortunately, artificial intelligence (AI) tech- the multiple application services that are nologies developed over the last forty years currently in development or on the drawing are directly applicable to many of the diffi- board. Initially, the STB was primarily in- cult technical problems faced by today’s ca- tended to do little more than decode MPEG ble TV applications. Specifically, we de- video. But the abundant bandwidth that the scribe how AI techniques can be applied to digital upgrades provided allowed for rapid provide more personalized subscriber ser- growth in the number of video channels, far vices, alleviate information overload, reduce too many for the analog style scrolling guide backend server and human editorial costs, to be practical. The need to overcome in- and to use available bandwidth more effi- formation overload caused by too many ciently. channels in a scrolling guide drove the de- velopment of the interactive program guide (IPG), a remote-control driven application that was squeezed into the confines of the INTRODUCTION STB. Today, a new set of business needs and opportunities drives the development of The cable TV industry has invested an array of new subscriber services, includ- huge sums of capital in recent years to up- ing video-on-demand (VOD), T-Commerce, grade both their networks and millions of information-on-demand (IOD), PC-like consumer premises equipment (CPE) units messaging, and games. from analog to digital. This has not only in- creased the quantity and quality of video Clearly a STB that was originally in- that can be provided, but also placed a sys- tended to do little more than decode MPEG tem controlled computing device, the digital video is hard pressed to support all of these set-top-box (STB), in every subscriber services. Further, the software architecture home. This high-speed, two-way network of the STB, which modified to support a combined with a re single application (the IPG), typically re- quires costly integration to accommodate new applications and services. There are no EVOLUTION NOT REVOLUTION standards for new services. As a result, most new services require costly servers to be de- The evolutionary development of to- ployed at the cable headend to perform day’s cable TV infrastructures and applica- much of the work, while the subscriber’s tions is characterized by the reactionary STB acts as merely a dumb display device. loop, depicted in Fig. 1.
This current state of affairs is unfortu- nate. Because of the economics of the situa- tion, the currently deployed STBs are likely to remain in the field for many years to come. Yet the technical limitations of both STBs and IPG applications impose signifi- cant integration and development challenges Business Needs and that impede the roll out of new, high- Opportunities revenue generating services.
Surprisingly, there is an existing tech- nology that could be applied to currently deployed STBs facilitating the full realiza- tion of the revenue potential enabled by a digital cable TV infrastructure. Even more surprising, this technology is neither pro- Business Solutions prietary nor a recent development. Rather it and Products is the often misunderstood and under- utilized fruit of many decades of academic research: Artificial Intelligence (AI).
While popular understanding of AI re- volves around jerky robots and giant, chess- playing super-brains, the true foundations of the science of AI consist of a cornucopia of techniques for performing complex tasks, such as user modeling, application of expert Technical knowledge, dealing with uncertainty, etc. Difficulties using limited computing resources. Many AI techniques are ideally suited to solving some of the most vexing problems in today’s ca- ble TV applications, and can often do so in the restricted computing environment of cur- Figure 1. The Reactionary Loop. rently deployed digital STBs. Initially, a system operator or multi- Carefully applied AI technology prom- system operator (MSO) identifies a business ises to revolutionize the subscriber services need or opportunity that solves a business cable TV can offer, and to do so at a fraction problem (e.g., increases revenue, cuts costs, of the cost of conventional, client-server provides competitive advantage, etc.). The systems. MSO then designs or purchases products or technical solutions that satisfy the need or on old technology.1 To date, the majority of capitalize on the opportunity. Finally, the deployed cable TV STB software is not mid- product or solution reveals new opportunity, dleware based, but built around a single or technical difficulties, thus driving the resident application, the IPG2. In some cases next reactionary cycle. (e.g., DCT-2000) deployment of a new ap- plication requires direct integration with the Unfortunately, this mode of develop- IPG. In most systems out-of-band (OOB) ment produces ever more complex and bandwidth is a precious commodity, and lit- costly, ad hoc solutions, as each cycle must tle if any is available for use by third-party accommodate the shortsighted decisions applications. made on previous cycles. And the resulting solutions typically have little or no inter- In this environment the cable TV indus- compatibility without costly software inte- try faces a number of challenges, including gration. high rates of digital churn, slowing digital penetration, shrinking subscriber bases, and This state of affairs has led many MSOs the ever increasing competition of DBS. To to seek a “middleware” solution that pro- continue to grow and survive the industry vides a common foundation for future appli- has entered the next cycle in the reactionary cation and service development. Unfortu- loop, as illustrated in Fig. 2. First, several nately such systems can never entirely over- critical business needs have been identified, come the inadequacies of a hardware and including: software architecture that has evolved via the reactionary loop. Middleware solutions • Reduce digital churn rates, provid- isolate applications from the raw features ing sufficient services to keep digi- available on the STB, forever limiting the tal customers once they sign on. role of such code to simple display tasks, and locking solutions into a client-server • Compete with DBS by providing model. And while limitations imposed by feature parity and significant feature underpowered STB hardware can be allevi- differentiation, capitalizing on the ated via backend processing, this only trades two-way network. one problem for others, as this adds yet an- other layer of computation in an already • Create new features and services tight STB environment, and server-centric that can provide incremental reve- backend solutions are notoriously expensive, nue (e.g., pay-per-view). and do not scale well for large numbers of subscribers.
THE CURRENT CYCLE
As of January, 2003 there are only a few markets in the US providing next gen- eration services on modern STB hardware; 1 the majority of cable systems, by far, are run Primarily from the Motorola DCT- 2000 and the Scientific Atlanta Explorer 2000 families.
2 Mostly TV Guide/Gemstar or TV Gateway. Reduce DBS Parity & New/Incremental Digital Differentiation Revenue Sources Churn
IPG PVR VOD IOD Broadband Messaging Games T-Commerce/Ads
Excessive Manpower Lack of Information Backend Bandwidth Support Personal- Overload Server Costs Usage Costs ization
Smart Viewer Intelligent User Modelling Agents Interfaces
Figure 2. The current reactionary cycle, and the application of AI tools
To satisfy these business needs, a vari- • Games. Providing games and other ety of new features and services have been interactive entertainments on TV. defined and in some cases deployed. Among them are: • T-Commerce/Advertising. Provid- ing custom adverstising and ena- • Video-on-demand (VOD). Provid- bling online sales via TV. ing TV programming of the sub- scriber’s choice (initially PPV mov- • Broadband Access. Providing ies) anytime. Internet service via cable modem.
• Information on Demand (IOD). However, we believe that such services, Providing relevant news, weather even if successful, introduce a number of and other information on TV any- new technical difficulties that must be over- time. come for these services to be adopted by subscribers, and to generate the revenue that • Messaging. Providing instant mes- justifies their implementation costs. saging and e-mail services on TV. 1. Backend server costs. As stated 5. Lack of Personalization. All of above, most such services are implemented these services would be both more useable via expensive hardware and software de- and more successful if they were personal- ployments at the headend. And while such ized for each subscriber. Tailoring the in- solutions may work adequately when rolled formation presented to the subscriber based out, they seldom scale well with the number on their interests and preferences provides a of subscribers, and may fall victim to their more efficient, and therefore more profit- own success. able, user experience. Research has also shown that systems that require personaliza- 2. Excessive bandwidth usage. It is the tion or learning are more sticky, retaining nature of such client-server solutions that customers better than those without. [5] information has to flow back and forth be- tween client and server. Often this informa- Fortunately, AI can be applied to solve tion (e.g., clicks on the remote control) must all five of these problems, and at a fraction be transmitted out-of-band. But out-of-band of the cost of conventional, client-server bandwidth is a precious, contention-based systems. commodity, and is often inadequate to the requirements of client-server solutions. APPLYING AI TO CABLE TV
3. Information overload. Today’s in- Over the past forty years AI scholars teractive program guides are useable for the have researched a variety of hard problems several hundred channels available on digi- and developed a vast array of techniques, tal cable. But how can they hope to cope technologies, and tools for solving them. with hundreds or even thousands of new Serendipitously, most of this research was programming titles made available by VOD. performed in an era when computing re- The subscriber will suffer information over- sources were scarce, so even a conventional load, hindering their ability to find and pur- cable TV STB is often adequate for their chase VOD programming. Similar problems application. Table 1 lists several such tech- exist with the other new services that flood nologies. [1,3,4] the subscriber with unprecedented quantities of information and numbers of choices. Implicit in this discussion is that the ju- dicious application of efficient AI technol- 4. Manpower support costs. Many ogy allows much of the work that is cur- new services, particularly IOD, games, T- rently performed by backend servers could Commerce and advertising require a signifi- be performed in a distributed fashion, di- cant number of people to provide content rectly on subscriber STBs, thus eliminating retrieval and editorial services. the need for costly backend servers. Such systems have been realized and are in opera- tion today.[2] AI Technology Specific Tech- Description Cable TV Applications Class niques
Learning Rote Learning, In- Incremental improvement Modeling the viewer based ductive Learning, of task performance based on previous actions to predict Neural Networks, on rote knowledge or ex- programming of interest for Genetic Learning amples. PVR or smart IPG.
Intelligent Agents Information Re- Software that understands Automated content retrieval, trieval, Knowledge a complex task well reducing editorial staff for Management, Com- enough to automate it, per- IOD, T-Commerce. merce forming in a human role.
Expert Systems Rule-based, Logic- Software that can apply Encoding knowledge about based, Context- expert domain knowledge TV usage to provide smarter sensitive interfaces to a problem. user interfaces.
Statistical Reasoning Fuzzy Logic, Cer- Reasoning with uncertain, Widely applicable techniques tainty Factors, incomplete, or noisy input useful in learning, agents, Dempster-Shafer and expert systems. Theory, Baysian Networks
Distributed Comput- Intelligent Agents, Distributing pieces of a By pushing tasks down to the ing Edge-based comput- complex task among sev- STB, obviates need for ex- ing, Peer-to-peer eral distributed computers. pensive servers. networking Table 1. A sample of AI technologies applicable to cable TV applications.
It is beyond the scope of this paper to bargains, etc. Conventional solutions em- describe every possible application of AI ploy a staff of human editors who retrieve technology to the cable TV industry. raw content from various network sources, Instead, we will focus on three sample AI revise it for display on TV. All such data is tools, each of which is directly applicable to then broadcast out to the STBs for display. many of the technical difficulties identified But it has been demonstrated that such tasks above. These tools are summarized in can be performed by intelligent agents run- Table 2. ning directly on the subscribers’ STBs.
Intelligent Agents This approach has a number of advan- tages. First, it reduces or eliminates the need Intelligent agents are small, active soft- for an editorial staff. Second, it allows STBs ware components that understand a complex to retrieve exactly the content that is appro- task sufficiently to assist a human in per- priate for a given subscriber, and ignore eve- forming it, or to automate it entirely. One rything else. This represents a significant such task required of IOD and T-Commerce reduction in the bandwidth requirements and systems is the retrieval of content (text and often allows an on-demand request model to pictures) to be displayed to the viewer, such replace the broadcast model currently in use. as news, sports scores, stock quotes, current
AI Tool Applies to Info. Server Bandwidth Manpower Lack of Per- Overload Costs Usage Costs sonalization
Viewer IPG, VOD, Fewer pro- Runs on No client-server -- Customize views Modeling IOD, PVR, gramming or STB network traffic and content to T-Commerce product target subscriber choices interests
Smart All products Automat or Runs on No client-server -- Customize features User Inter- and services assist in STB network traffic & views based on faces obvious or subscriber abilities repetitive and context tasks
Intelligent IOD, Broad- Retrieve and Runs on No broadcast of Reduce or Select agents that Agents band Portals, display cus- STB generic info. Al- eliminate con- retrieve only de- T-Commerce tom info., lows on-demand tent editorial sired content. not every- requests staff thing Table 2. Three AI tools and their applicability to the current reactionary cycle
Smart User Interfaces works today. Using an IPG to navigate through hundreds of channels is difficult Current cable TV user interfaces (e.g., enough. But add thousands of VOD titles VOD, IPG, IOD) tend to be static, providing and no subscriber is going to want to navi- the same set of capabilities to all subscribers gate through any static hierarchy to find a at all times, regardless of the situation. The title worth paying for. However, by applying user interface, in this case, provides a means several statistical reasoning and learning of operating a tool. However, significant AI techniques from AI, STB software would be research has been devoted to producing able to monitor the programming viewed by smarter interfaces, that operate more like an a subscriber and construct a model of the automated assistant. Rather than displaying tastes and preferences of that subscriber. a channel grid in numerical order, a smart Armed with this model, a smart IPG or IPG interface might order the channels VOD user interface could present the user based upon frequency of use. Or, by apply- with a small number of choices tailored to ing statistical reasoning and expert systems a their preferences, and to provide a more dy- smart IPG would “understand” the normal namic navigation through the available titles activities that a subscriber performs, either based on subscriber tastes. By overcoming assisting or performing those activities the information overload problem, and mak- automatically. ing it easier to find and buy programming of interest, such systems should allow VOD to Viewer Modelling realize its true revenue potential.
Information overload is perhaps the most prevalent problem that results from the array of new subscriber services in the SUMMARY REFERENCES
The evolution of technical advances in 1. Barr, Avron, Edward Feigenbaum, the cable TV industry is the result of a reac- eds., Handbook of Artificial Intelli- tionary cycle. As a result, current limitations gence, vols. 1-3, HeurisTech Press, of STB hardware and IPG software applica- 1981. tions impose significant development chal- lenges that impede the efficient roll out of 2. BIAP Systems, Inc., “PiTV Over- new, high-revenue generating subscriber view”, services. http://www.biap.com/pdf/Personaliz ed Information Television.pdf, However, judicious application of AI 2002. technologies, developed over the last 40 years, significantly enhance the range and 3. González, Avelino, Douglas quality of services that can be implemented Dankel, The Engineering of Knowl- via STB applications, and at a fraction of the edge-Based Systems: Theory and cost of conventional client-server ap- Practice, Prentice-Hall, 1993. proaches. 4. Rich, Elaine, Kevin Knight, Artifi- nd AUTHORS cial Intelligence, 2 ed., McGraw- Hill, 1991. Louis Slothouber, Ph.D., Chief Scientist, BIAP Systems, Inc., 5. Yahoo User Study, 2001. [email protected].
Aaron Ye, Ph.D., Chief Technical Offier, BIAP Systems, [email protected]
BUILDING COMPETITIVE SYSTEMS: A PACKETCABLE PERSPECTIVE
Burcak Beser * Juniper Networks
Abstract PUBLIC SWITCHED TELEPHONY NETWORK PacketCable defines a network superstructure that overlays the two-way data- Since the aim of the cable telephony is to ready broadband cable DOCSIS 1.1 access match or exceed the service quality that is network. PacketCable specifications define offered by the landline telephony systems, it is how PacketCable elements interact with each very important to understand the landline other and the protocols that are used between telephony of today. these elements. Since PacketCable certification/qualification only includes the Each phone call is carried as 64 kb/s bit protocol compliance of these elements, the stream, with bits flowing regardless of whether certification/qualification does not suffice as the sender talks or not. The speech signal is the necessary means to provide a competitive encoded at a sampling rate of 8 kHz, with eight service as provided by today’s Public Switched bits per sample. The encoding is a simple Telephony Network. lookup that maps sample amplitudes from 0 to 8159 to a 7-bit table entry, with a roughly This paper details some of the features that logarithmic scale. There are two encoding are necessary for competitive telephony schemes, A-law and u-law, where the former is service. Some of these features are not covered found in European countries, while the latter is by PacketCable certification/qualification used in North America and Japan. The tests. encoding is often also being referred to by its ITU Recommendation name, G.711 or called PCM coding [1]. The u-law encoding offers a signal-to-noise ratio of 39.3 dB for a full-range INTRODUCTION signal and a dynamic range of 48.4 dB. This is roughly equivalent to that of FM radio, except The PacketCable project defines the that the audio bandwidth is far lower. protocols that are necessary for building competitive telephony. The 8 kHz sampling period yields the basic clock period, 125 us, that is found throughout Building a successful competitive telephony the digital telephone system, even when no services over cable infrastructure requires the voice is being transmitted. A number of these grade of service that is provided by Public digital signals are then multiplexed into a Switched Telephony Network (PSTN) landline single frame. For example, a T1 circuit consists services to be met. of 24 voice channels, with one byte per channel. This packaging of channels into a The grade of service provided by PSTN single digital stream is called time-division landline services has many dimensions that multiplexing (TDM). A frame consists of these require different aspects of services to be voice channels plus one or more engineered. Due to time and space limitations, synchronization bits. this paper mainly focuses on the issue of perceived voice quality. Due to the TDM nature of the PSTN Since the MOS is a subjective scale and network the delay that is perceived by the users requires subjective tests to be carried out which is mostly the propagation delay [2]. For North is not a good method of designing for a target. America the end-to-end delay worst case is For design purposes ITU E-Model can be used calculated using maximum national distance of [3]. 6000 Km is found as 33 msec. In the same manner the long distance submarine fiber The equation for the transmission rating factor connection between San Francisco and Hong R is: Kong can be found as 78 msec. It is important to note that even though the typical R = Ro - Is - Id - Ie propagation delays are much less the impact of the PBX equipment, compression CODECs Where, and multiplexers the given delays constitute • Ro, the basic signal-to-noise ratio based good reference points. on send and receive loudness ratings and the circuit and room noise; GRADE OF SERVICE • Is, the sum of real-time or simultaneous The Grade of Service can be divided into speech transmission impairments, e.g., two: the call connecting quality and perceived loudness levels, side tone and PCM voice quality. quantizing distortion;
The call connecting quality depends on • Id, the sum of delayed impairments many factors including but not limited to call relative to the speech signal, e.g., talker blocking, post-dial delay and accurate billing. echo, listener echo and absolute delay;
The perceived voice quality is generally • Ie, the Equipment Impairment factor more important than the call connecting such as packet loss and CODEC loss, if quality, people tend to forget sporadic call CODEC being used is different than connection problems, but when they have bad G.711. voice quality that they are paying for they tend to remember. CABLE TELEPHONY TARGETS
Perceived Call Quality Cable Telephony competing with landline services aims to have E-Model R-value of 80, The voice quality in the PSTN networks which corresponds to MOS scale of 4. The was historically measured using ‘mean opinion cellular services offer a much lower R-value score’. The mean opinion score measures the than 80 but they have an advantage factor that subjective quality of a voice call. Historically adds to the total and improves the R-value. For the telephony providers invited people and new services, like satellite phones a correction used various call types (with delay, echo etc.) value A is intoduced, to take into account the and recorded the results. advantage of using a new service and to reflect acceptance of lower quality by users for such The MOS is a scale of 1-5 where the PSTN services. It is assumed that the Advantage stands at 4.4 for local calls (perfect score). The Factor will be reduced over time as the service score of national calls is generally above 4, improves and the customers get used to the which is considered as satisfactory. Anything benefits of the new service. It is not below 4 may result with customer recommended to include a non-zero Advantage dissatisfaction with the service being received. Framing Look Ahead Coding Decoder Total Delay G.711/10 10 msec 5 msec 1 msec 1 msec 17 msec G.711/20 20 msec 5 msec 1 msec 1 msec 27 msec G.729/10 10 msec 5 msec 10 msec 10 msec 35 msec G.729/20 20 msec 5 msec 10 msec 10 msec 45 msec
Table 1 Delay introduced by various CODECs. Factor for IP telephony because it is a The absolute speech delay is contributed by replacement for existing services, rather than a many factors: Coding delay, Cable Access completely new service. Delay, Network Side Delay and Jitter Buffer Delay. CABLE TELEPHONY PERCEIVED CALL QUALITY Coding Delay
The perceived call quality of the cable The MTA introduces a certain amount of telephony will be discussed using the e-model. delay due to framing, look ahead processing, As with the e-model the contributing factors of and decoding. The delay introduced by the two absolute speech delay and packet drop will be PacketCable CODEC’s are given in table 1. discussed. Cable Access Delay Absolute Speech Delay The delay introduced by the Cable Access The absolute speech delay known as mouth to network on the upstream direction depends on ear or one-way delay is a very important factor in many assumptions. Some of the assumptions the perceived voice quality. Figure 1 shows the are listed below: drop in the voice quality in e-model with respect to absolute speech delay. To find the voice • The MTA’s coding/de-coding clocks are quality drop one has to calculate the absolute slaved to CMTS DOCSIS master clock: speech delay look up from the graph to find the delay related impairment in E-model [2]. If this assumption does not hold than the packets on the upstream-direction would experience a delay that is increasing/decreasing between 0 and 10 (the UGS interval) and then the same behavior will repeat as depicted in Figure 2.
Figure 1 Impact of End-to-End delay Figure 2 Upstream Delay when MTA’s are not using E-model synchronized
• The MTA’s framing interval will be msec with jitter as large as 50 msec or more. aligned to UGS intervals: When packet prioritization is being used the average delay remains about the same but the The first Dynamic Service Addition jitter reduces. (DSA) message from the CMTS will include the time reference of the first UGS Jitter Buffer grant [4]. The MTA should align its framing interval such that the time between The other delay factor is the jitter buffers on the end-of-framing and time to transmit the de-coding section. The predicament with upstream is minimized. If the time is not the jitter buffer is it is one of the items that is minimized or the framing is not aligned left for vendor differentiation. than there will be a constant value between 0 and 10 (UGS interval) upstream delay added. The jitter buffer implementations can either be adaptive or static. On static jitter buffer
implementations the buffer generally holds one or two packets, which spans one or two If the MTA has implemented both the clock framing interval delays. and framing synchronization than the delay on the upstream direction consists of: The goal of adaptive jitter buffer
management is to remove the jitter while Framing to transmit delay <1 msec minimizing the amount of delay or incremental Cable propagation delay <0.8 msec latency that is added to what's already been Cable receiving delay <0.2 provided by the network. The adaptive buffer CMTS internal delay <3 msec management schemes use interpacket arrival
time variations, doing statistical analysis on it, A total of 5 msec is assumed for the upstream then adapting the mean holding time of the direction. packets or the jitter buffer length.
For the downstream direction cable delay The problem with the static buffer consists of management algorithms is that they tend to be
conservative and assume the worst network CMTS internal delay <3 msec cases. The MTA buffer management should be Interleaving/transmit delay <1 msec designed for end-to-end IP transport jitter Cable Propagation delay <0.8 msec values no matter where the call is connected. Reception to buffer delay <0.2 msec That means, if the worst-case jitter is 15 msec
end-to-end then all the calls experience twice Making a downstream direction cable delay of the jitter value delay of 30 msec. 5 msec.
The problem with the adaptive buffer management is twofold: First most of the Network Side Delay adaptive buffer management statistical
analyses schemes are not designed against the The network side delay depends on many bursty nature of the network delay that is factors such as the distance, the number of experienced today. Second, the buffer routers between end-points, the traffic and the management is generally carried out during connection technology. The end-to-end delay silence periods, which most probably does not number for a national network is around 60-90 coincide with the events that require jitter The Packet Loss in IP networks can be buffer changes. attributed to several sources: queue overflow, synchronization, jitter buffer The incorrect jitter buffer assignment has a overflow/underflow, damaged packets. The two different impacts: When the jitter buffer is impact of packet loss on the e-model set to a value that is too low than the packets impairments depends directly on the CODEC that arrive later then the buffered time will be being used and whether Packet Loss dropped, and if the buffer is set to a too high Concealment (PLC) is implemented. value then the delay will be too high. The impact of packet loss can be prevented Depending on the network configuration if the packet loss concealment algorithms are and load the jitter on the VoIP packets may implemented on the receiver side. As depicted vary. In some cases some packets are so much in figure 3 the quality drops to 5 when PLC is delayed that setting the jitter buffer will result implemented, which is one fifth of the non- in an overall drop in voice quality. PLC G.711 system. Unfortunately the PacketCable specifications do not mandate the Packet Loss implementation of PLC when using G.711 CODEC1. Almost all IP networks exhibit Packet Loss. Figure 2 shows the e-model impact of packet Packet Loss due to Queue Overflow loss on G.711 CODEC, which is the only mandatory CODEC in PacketCable The queue overflow results when a certain specifications [5]. As can be seen in figure interface receives more packets than it can below, the impact of packet loss is tremendous send for a certain time duration. The solution on the voice quality, if 1% of the voice packets for queue overflow is two fold: prioritization of are lost than the speech impairment due to packets, control of the queuing available to packet loss is 25. Combining this with the each priority. starting point of the G.711 the quality of no- delay VoIP system with 1% packet loss results The prioritization of packets is now a well- with a e-model rating of 70 which is equivalent established standard in the IP world and is to cellular phone quality. called Differentiated Services (DiffServ) [6,7]. The PacketCable standards already support the DiffServ packet marking. The DiffServ in general is the scheme that when a high priority packet is received that packet is being sent before the lower priority packets. Even when DiffServ marking is being used, it is still possible that at some funneling points, the router would receive a larger amount of high priority packets then it can handle and has to queue the high priority packets. In this case the packets either have to be queued for a long time or should be dropped. The issue with funneling is that it would cause all calls to be
Figure 3 Impact of Packet Loss on E-model 1 The term used for G.711 is ‘RECOMMENDED’ which is much weaker in terms of testing/certification viewpoint than use of the term ‘MUST’. impacted. If a certain link can handle at most packet is being received with a jitter of +10 1000 calls and 1001st call is being connected milliseconds than it will be played without any all 1001 calls will be impacted not only the one jitter buffer induced delay. If a packet is early call that is added. by 10 milliseconds (-10 millisecond jitter) than the packet will be delayed by 10 milliseconds. The issue of packet dropping in DiffServ In short the jitter buffer regulates the delay environment can be prevented by over- variation on the packets and makes the designing the network with no funneling impression that there is no jitter experienced by points. the packets only a constant delay of 20 milliseconds added. Packet Loss due to Jitter Buffer Overflow/Underflow The real issue comes to play when a packet experiences a jitter that is more than that of the The jitter overflow/underflow is caused by jitter buffer. If a packet were earlier or later the jitter experienced by the packets. As stated than the jitter buffer allocation, the MTA before every MTA has a jitter buffer, which would drop the packet since it cannot handle can be static or dynamic. Since at any instant the packet with in the jitter buffer. As shown in the jitter buffer is perceived as being static, figure the jitter in a IP network can be depicted only the static jitter buffer case will be as a bell shape and the jitter buffer as a band considered. within/engulfing this bell shape as shown in figure 4.
The jitter buffer may not be an issue if it can be set to a value that would always be able to accommodate the worst-case jitter in the network.
Figure 4 Packets Dropped and Jitter Buffer
Lets assume that the MTA has a jitter buffer Figure 5 Example Core Router Jitter of 20 milliseconds, which is actually a +/-10 milliseconds jitter buffer. What this means is Unfortunately this is not an easy task. that if a typical packet is received by time t0 Figure 5 shows a core router performance on a then it will start to be played at time t0+20 two-interface situation that no queuing is milliseconds. Now assuming that the second necessary. As can be noticed the spread of the delay is very wide [8]. If the delay from this one router is to be accommodated then a jitter BUILDING AN EXAMPLE SYSTEM buffer of 214 milliseconds will be required. The better approach is to set an acceptable Lets assume that the objective is to design a packet-drop level and set the jitter buffer to this Voice over Cable system that would be value around 2 milliseconds to get a packet competing against the PSTN landlines in North drop around 1%. America.
Using similar analyses it can be shown that Form the viewpoint of TDM based PSTN the best settings for IP network jitter buffer are equipment the end-to-end performance of the around 1% packet drop. Any settings above the PSTN network is impressive to say the least: 1% packet drop will cause a larger delay introduced by the jitter buffer thereby dropping • Less than 1 in 1000 samples are dropped the voice quality faster than the voice quality • The end-to-end delay in North America is improvements coming from the decreased less than 35 msec. packet drop rate. Any value above the 1% packet drop would result with a decreased When the cellular phones are taken as a voice quality due to the fact that the voice base point these characteristics change as: quality drop induced by the increased packet drop will not be compensated by the reduced • As much as 3% packet loss. delay. • More than 200 msec of delay Packet Loss due to Errored Packets The perceived quality of the cellular phone The packet drop due to errored packets is calls suffers from these characteristics. generally due to RF impairments in the cable plant. This is due to the fact that the modern IP CODEC Decision transmission equipment provides reliable transmission with errored packets less than one Since the PSTN network will carry the in 10000. voice as G.711, it is assumed that the G.711 will be used on the VoIP portion(s) of the In the downstream direction the packet-drop -5 networks. If this assumption is not valid than rate is in the order of 10 due to the fact that the initial coding loss and transcoding loss on the downstream the transmitter is more should be taken into account. powerful and the bandwidth used for downstream transmission has better SNR Absolute Speech Delay (Signal to Noise) characteristics than the upstream direction [4]. Assuming that the aim is to be as close to PSTN landline services as possible the worst On the upstream direction a provider has case has to be considered. The worst-case many possibilities that would impact the scenario for the Cable Telephony is that the packet loss. Almost all of the countermeasures call starts on Cable hops into PSTN and ends at against the packet loss would have an impact Cable. As depicted in figure 1 the call starts in on the perceived bandwidth on the upstream a user calling via a PacketCable certified MTA side. For example using a 1% packet loss a in Sunnyvale, CA to another user that has typical CMTS would be able to use a 2Mbps -5 PacketCable certified MTA at the Providence, upstream channel whereas for a 10 packet RI. The call goes to PSTN on the Gateway in drop rate a 612 Kbps will be achieved. San Francisco, CA, and then exits from PSTN on the Boston, MA. The delay on the PSTN Termination segment between San Francisco and Boston is Coding/Decoding 17 msec 30 msec. Cable/Network Delay 15 msec Jitter Buffer Delay 10 msec Since the design is made for worst case, it +______will be assumed that the 10 msec packetization Absolute Speech Delay 114 msec interval with G.711 coding will be used. Looking from the table 1 the coding and Using figure 1 to resolve the E-Model quality decoding delays of G.711 CODEC with 10 drop the drop can be found as 4. msec framing is 16 msec for coding and 1 msec for the de-coding, a total of 17 msec Packet Loss CODEC induced delay is found. The amount of packet loss contributed can be Since the coding is carried out on the partitioned as: originating MTA and de-coding on the ingress to PSTN, and coding on the egress from PSTN Router queuing 0.1% and de-coding on the terminating MTA, there Jitter buffer 1% two occurrences of coding-decoding in the Cable Access (RF) 0.01% Voice sections making 34 msec of CODEC delay. Making 1.11% packet loss in origin and 1% packet loss in destination. Looking at the The delay on the network side between the impact of 2.22% packet loss on figure 3, the e- CMTS and the PSTN egress point is assumed model quality drop can be found as 7.66. to be 10 msec; when the cable access delay of 5 msec is added the total IP network side delay E-Model Result becomes 15 msec. When the G.711 coding (just sampling the The Jitter Buffer delay is assumed to be 10 voice with 8000 times a second) is being used, msec. The jitter buffer value of 10 msec should the base for voice quality would start from e- be sufficient for access to a local PSTN egress model score of 94.2. Calculating the drop due point but for the end-to-end VoIP call through to delay (4) and packet drop (7.66) would the backbone this value may be too low, result with an end-to-end quality of the 82.6 causing too much packet drop. Since the which is barely above the desired limit of 80. PacketCable does not have any means of setting the Jitter Buffer Size per call, the CONCLUDING REMARKS provider has to make a compromise between PSTN quality and end-to-end voice quality. Even though at first glance it looks like that the desired of score 80 can be achieved, some The total absolute speech delay consists of points are worth mentioning: many pieces: • The calculation is for the worst case Origination and for most of the geographical Coding/Decoding 17 msec locations the score would be higher. Cable/Network Delay 15 msec Jitter Buffer Delay 10 msec • The packet loss of 1% would only be PSTN end-to-end Delay 30 msec seen on cases where severe jitter is observed.
• The overall experience with respect to [2] ITU-T Recommendation G.114 (05/00), PSTN landline services would be lower General Recommendations on the transmission due to the fact that the introduced quality for an entire international telephone absolute delay is higher1. connection.
• When connected to high delay [3] ITU-T Recommendation G.107 (05/00), endpoints such as cellular phones, The E-Model, A Computational Model PBX’s or international calls with long for use in Transmission Planning. delays, the call quality may drop to an unacceptable level. [4] CableLabs SP-RFIv1.1-I09-020830 (08/02), Data-Over-Cable Service Interface • The use of CODEC’s that provide Specifications Radio Frequency Interface better bandwidth utilization will cause Specification the voice quality to drop further. [5] ITU-T Recommendation G.113 (02/96), • The use of a bigger Jitter Buffer to Transmission impairments. accommodate the connections to other end-points would cause voice quality to [6] IETF Request For Comments RFC2475 drop further. (12/98), An Architecture for Differentiated Services • Any additional packet loss would cause the voice quality to drop further. [7] IETF Request For Comments RFC2474 (12/98), Definition of the Differentiated REFERENCES Services Field in the IPv4 and IPv6 Headers
[1] CCITT Recommendation G.711 (11/88), [8] Light Reading (7/01); The Internet Core Pulse code Modulation (PCM) of voice Routing Test: Complete Results; frequencies. http://www.lightreading.com/document.asp?sit e=testing&doc_id=6411
1 Due to added delay most of the long distance calls will sound like high quality international calls. BUSINESS AND PLEASURE: MIXED TRAFFIC ISSUES DRIVE NETWORK EVOLUTION
Robert L. Howald, Ph.D. Motorola Broadband
Abstract architecture design options. Finally, voice circuits and IP voice also have a role in the The term “triple play” originally was redefining of the meaning of triple play. meant simply to convey the convergence of video, voice, and data in the network. It once This paper will analyze and characterize encompassed everything that was important the traffic dynamics of the various service to know about network functionality. components above. Aggregation of these However, over the years the lines have services in cases consistent with likely blurred and the elements of the triple play architectural scenarios will be discussed. have been further fine-tuned and splintered Architecture and bandwidth conclusions will into a variety of different items. All three be drawn that align with the service and parts of the triple play are multi-faceted, and traffic mixes currently being offered. all contain important variables necessary for Finally, offerings such as gaming, security, proper bandwidth and traffic management. and medical applications are some of the Video has multiple, possibly interrelated, ideas among many potential services that faces: analog, digital, narrowcast, on- have been mentioned recently. Their demand, IP, and HDTV. Data, which once significance is magnified by the amount of essentially meant DOCSIS 1.0, now includes highly interactive real-time voice and video the ability to support tiered data services that needed to support them. The implications of include best effort Internet traffic, like that such new service offerings will be discussed. offered via DOCSIS 1.0, as well as guaranteed and mission critical business INTRODUCTION services. Even simple residential Internet access is becoming a more complicated At last year’s NCTA show, a paper offering. With VoIP still in the wings, ideas planted a stake in the quicksand [8], such as online gaming have leap-frogged describing, as is self-described by the title, into play as part of the residential data mix. “A 10-yr Residential Bandwidth Demand With data just as with video, possibilities Forecast and Implications for Delivery abound that involve bandwidth and traffic Networks.” The presentation generated quite management in both the access, backbone, a few comments and questions, as any and interconnecting points in the network. discussion related to predictions of Successful deployment of tiered, prioritized, bandwidth consumption might. In particular, and guaranteed services include the paper did not focus on perceived needs or understanding aspects of the access network, the conventional “Field of Dreams” theory of including higher versions of DOCSIS, as well bandwidth growth – that is, “if you build it, as non-DOCSIS solutions and technologies they will come.” Instead, it was grounded in behind the HFC access network. Proper demand expectations based on market treatment of data services to and from the research, trend studies, and conversations access network is a critical component of with various individuals across the industry bandwidth management when considering whom the information was shared with. The approach was to identify current and coming is going on outside the residential portion of services that could be reasonably anticipated, the network, as well as at and behind the evaluate and predict behavior, and aggregate hubs that feed the distribution network. To the results. Obviously, the analysis was not do this, first we need to understand the done as an academic exercise, but as a tool relevant traffic engineering problems as well that can be used to plan a business towards as access bandwidth problems for effective expected growth areas, and to engage end-to-end system design. operators in discussion that help them plan their networking needs. This paper will introduce some of the concepts associated with the traffic As it stands today (now a third year into engineering side of the problem. The assessing the predictions), predicted behavior problem at hand can actually be summarized has deviated in a couple of areas, but none in quite easily. Never before has one network earth-shattering ways. Only one service at been asked to support so much content this point is ramping more slowly than variety with such a wide range of quality of anticipated (VoIP). As a quick data point to service (QoS) objectives. With this being the encourage active minds, the study currently assumed case for growing cable operators, expects that, in 2010, there will be about 10x the following questions are being explored the bit rate demand on the forward path as today: there is in 2003, and about 25x in the return over the same period. 1) What are the traffic implications of this multi-service, multiple goal situation? As valuable as this paper is to the 2) What are the resulting architectural company as a benchmark that is updated implications? regularly, its original intent is only a piece of the puzzle needed to get a complete snapshot The fact that cable systems are able to of the bandwidth and architectural evolution encounter this type of problem at all and landscape. In particular, the results presented learn the important issues makes a strong describe forecasts for the North American, statement for its competitive readiness in the residential market, and only the implications larger picture of broadband providers. This on the HFC access portion of the network. paper will discuss scenarios that can be used While there is certainly not a lot of extensive to evaluate question one. There are as many network infrastructure activity going on in answer to question two as there are opinions the current slowdown, there is significant on optimal architectures. We will discuss attention being given to services and common ones and general themes to be equipment that have equivalent, if not more understood. dynamic, impact on metro interconnect or backbone portions of the network. Examples SOME MATHMATICAL TRAFFIC include the growth in video-on- CONCEPTS demand(VOD), the emphasis on supporting commercial services, and enhanced data While widespread DOCSIS deployment aggregation and backhaul platforms. And, brought data traffic management to the clearly, the paper described above, while attention of the industry, the idea of concerning itself with evolution of the HFC understanding traffic characteristics and the plant aspects of service growth, implies effect on performance is not a new to cable. impacts beyond HFC access. To complete Archives at Motorola contain traffic studies the picture that enables the steadily predicted aimed at understanding the response time of bandwidth growth requires a peek into what early settop IPPV request traffic, which used a basic ALOHA protocol. ALOHA is testing study was performed as LAN essentially a free-for-all that allows a user to technology began to explode during that send a message whenever ready and, period of time [3]. basically, take their chances that no one else is doing so at the same time. If an Telephone Network Simplicity acknowledgment is received prior to a time- out waiting for it, the user knows the The purpose of traffic modeling is message got through. The study goal was to simple. By developing proper statistical understand how the settop loading and models for data in the network, it is possible acknowledgement scheme effected the to predict pipe size requirements, response time to a request from the user, bottlenecks, performance, and equipment determine the re-send likelihood, and requirements. Historically, there are two understand the system breaking point. paradigms – telephone networks and data Implementation details were derived from networks. The phone network is traffic this study. Through traffic modeling, the engineered to minute decimal place (the “five analysis was able to show that about 40% nines”) precision. The unit of Erlangs is more settop returns could be accommodated used to describe voice traffic volume. The at the HE equipment if the acknowledgments voice traffic arrivals are characterized sent to the settop were modified in the way statistically as a process with call arrivals that they were originally designed to be delivered. exhibit a Poisson characteristic. This Clearly, equipment cost savings were directly information is used with the well-known obtained in this simple case. Erlang formula to determine trunking capacity necessary to ensure that circuit As a second example, prior to full two- availability can be guaranteed to the high way activation of cable plants, Motorola had level described above. Traffic engineering is deployed an early SurfBoard cable modem possible with precision because of the well- with a telephony return path. Traffic studies understood nature of voice traffic with many were commissioned to understand this drastic years of historical precedent, and the single- network asymmetry, the impact of PC service nature of that system at inception. hardware, and the effect of TCP/IP implementation on the PC. The analysis Data traffic, on the other hand, has not characterized how the telephony modems of historically been heavily traffic engineered. the time (14.4 kbps and 28.8 kbps) Providing plenty of excess bandwidth has compromised downloads that had much been the protection against performance higher raw throughput capability. degradation due to congestion, and there are Performance of FTP transfer of large files still advocates of cheap bandwidth and less was compared against symmetrical 10 Mbps complexity as the way to continue. Others and 100 Mbps point-to-point Ethernet to argue that, besides the inherent cost of higher understand the user experience relative to, for performance equipment associated with example, the office environment. This data under utilizing the network, flows are likely was use to support configuration guidelines to encounter some bottleneck in an end-to- for the product. end system, particularly as the routes grow longer and more complex. Delivering In the 1980’s, Ethernet itself, repeatable QoS for high-performance standardized around a carrier-sense, collision services is not practical through pure detect multiple access (CSMA/CD MAC) bandwidth means in such cases. protocol came under traffic analysis scrutiny. We have mentioned the idea of Erlangs. A widely referenced throughput analysis and Telephone system trunking curves – how many circuits must be deployed as a function nature of the discovery, others were inspired of subscribers to assure a given blocking to look closely at their own assumptions. criteria – are available in many classic Subsequent findings included self-similar textbooks and papers. The results provide properties of ATM traffic, metro area traffic remarkably straightforward formulas for (MAN), wide area network traffic (WAN), telephone network design – a formula that and also for multimedia traffic, such as depends only on voice traffic offered (arrival compressed digital video streams and Web of calls and duration) and the statistical traffic. assumption of a Poisson process for call arrivals. What statistical characterization can The unearthing of self-similarity created be used for other services, such as data? Are a camp of network theorists that felt that the the answers as conveniently simple? book on traffic theory now had to be re- Unfortunately, this answer is no. written. Traditional models generally focused on Markovian behavior, which relies Long-Range Dependence (LRD) on limited memory of prior traffic – in other words, correlation is lost over time, and The finding that data traffic has a self- smoothing out occurs as the time scale is similar characteristic was one of the major broadened. The stock market is a good traffic modeling discoveries to date for this example of this expected smoothing, relatively young discipline. Self-similarity – although studying these curves are probably also called fractal or long-range dependent best avoided at this juncture. The impact of behavior – implies that, regardless of time this correlation lasting over broad time scale, the traffic pattern has the same basic periods has implications for policing, structure. When we say “traffic”, we are scheduling, congestion control, and statistical talking about the baseband data volume and multiplexing gain. trends observed at the output of a CMTS or switch serving MPEG VOD streams, for The surprising finding naturally led example. This was an unusual finding, in researchers to search for the reasons for it. that it indicates that there is correlation The cause of self-similar behavior was found across much wider time scales than to be associated with the fact that the previously thought, and the assumption that distribution of transmission content is smoothing occurs when observed over long heavily-tailed. That is, the tails of the periods was proven inaccurate. Another probability density function do not decay surprising way to envision this characteristic rapidly. What this means is that, rather than is to think in terms of our basic seeing the likelihood of the size of a understanding that data traffic is bursty, transmission flow occurrence decreasing which we usually associate with short time exponentially as the flow size increases, this dependence in our minds. However, self- drop-off in likelihood is not so drastic. There similar traffic indicates that long bursts is a very wide variation in the size of packet separated by long time intervals are flows that throws off traditional statistical characteristic of the traffic as well. models – large files, MP3’s, JPEGs, database Intuitively, we would have expected the activity. In fact, it has been shown that such wider time scale to smooth out the peaks and heavily-tailed characteristics are a sufficient valley around a mean. condition for self-similarity. As important is to recognize what self-similarity is not The seminal paper showing self- caused by. The breadth of examples similarity at work was based on an Ethernet indicates that self-similarity is not associated analysis, but because of the astounding with the delivery format generated to carry the information – i.e., it is not a protocol arrival rate of overlapping bursts to create a artifact. mathematical realization of the situation. More specifically, data traffic is assumed to Now, obviously, cable systems deploy be bursts with a Poisson distribution and equipment for carrying multimedia traffic. In associated arrival rate, where each burst is of particular, compressed video streams are on duration described by a Pareto distribution. today’s cable transport networks, with today’s most relevant example in terms of The M/Pareto model has several equipment growth and network design being variables associated with it, including the video-on-demand (VOD). Based on the Poisson arrival rate information. This above, the traffic characteristics will be the portion of the model has been shown to be same whether the video delivery is MPEG important to accurately curve fitting real over IP – such as GbE-based transport – or traffic to it [2]. The essential “real” traffic the other way around. And, of course, cable property captured by varying the Poisson companies are interested in moving data parameters is the amount of traffic being around in the form of Internet traffic from multiplexed onto a pipe for characterization. CMTS’s to ISP points of presence, and data This approach is a valuable step to a traffic from business services, both Internet directed model comprised of an aggregate of multiple or otherwise. sources of independent information. The aggregation of traffic is not significant Summarizing, then, understanding the enough to use mathematical assumptions of role of self-similar traffic patterns is valuable Gaussian behavior driven by the central limit for the applications above in designing the theorem. Models based on long-range HE to hub or hub-to-hub interconnects. As dependence provide network designers with a networks become more integrated, the value tool for developing architectures and of understanding traffic increases as the equipment requirements that support the aggregation pushes bit rates higher, making aggregated traffic. This is important to efficient use of resources yet more important. capture, as issues associated with self- As movement of different types of traffic similarity drive network changes in queuing becomes integrated, there is the further need and congestion control mechanisms then to ensure the QoS support for each. Markov-based assumptions would imply. Providers must therefore understand the implications of traffic characteristics and the Gaussian Behavior distribution of QoS needs of each. M/Pareto Model What is occurring on the Internet and to a similar extent in breadth on HFC networks While we have explained and described is the aggregation of more traffic and more a fundamental and surprising trait of many traffic types from independent sources. It is traffic types relevant to cable, making use of not difficult to envision the challenge this this model for statistical calculation requires growth entails; yet, this very growth and the fitting this knowledge into a distribution. evolution of integrated networks is The characteristic described has been shown potentially a blessing to the traffic modeler. to be a result of an aggregate of bursts of The central limit theorem provides a widely varying sizes. A model based on fundamental statistical underpinning for what randomly arriving bursts with a heavily- the nature of the traffic over time could tailed distribution is therefore called for. A evolve to – Gaussian behavior. Pareto distribution, commonly described in statistics texts, is combined with a Poisson The central limit theorem is the basis for attributed to simplified web browser many natural phenomenons that exhibit breakthroughs that led to mass acceptance, Gaussian behavior, and, in the case of and the subsequent changes made by online aggregated traffic, it becomes asymptotically providers to graphically rich interfaces. so when many independent contributors - Further traffic related information from trend under some minor, but important, caveats - studies indicate that access to broadband via are aggregated. The convenience of this is DSL or cable modem results in a user that Gaussian statistics are very well-studied increasing their time online by 50-100%, and and understood, and if the traffic statistics that the bytes consumed per month increase can be assumed Gaussian, then many 5x to 10x as well. simplifications can occur and probabilities of occurrence characterized. Multiplexing gain A very informative paper based on a can be predicted under Gaussian project at CableLabs and also presented at assumptions, and pipes designed efficiently last year’s conference [12] offered a first real for some pre-selected level of congestion comprehensive glimpse into DOCSIS traffic. avoidance. This can be used to support a Summarizing some of the key findings: desired set of policing, shaping, scheduling, and queuing mechanisms. While rapid traffic − Daily activity is a slow build growth makes network evolution difficult, throughout the day, with “busy hours” the bandwidth explosion, in general, is good between 8 pm – 12 am (peak), and a for business, and the handling of traffic from subsequent rapid drop-off until beginning a bandwidth boom potentially makes it more again at 5 am readily predictable. − Traffic is seasonal – following school SERVICE SET holidays and vacations
DOCSIS − Traffic asymmetry decreases from 3:1 to about 1.5:1 as familiarity and capabilities Since the wide acceptance and set in deployment of DOCSIS, all operators and vendors have interest in traffic characteristics − DOCSIS 1.1 enhancements to support of essentially this same basic system. As a voice traffic result in a 15% efficiency result, there have been many articles on improvement over DOCSIS 1.0 configuration of the CMTS and guidelines for a DOCSIS-based system setup. The The seasonal phenomenon represents the paper described previously [8] suggests a dominance of traffic by a younger generation doubling of DOCSIS return path traffic each of user. This particular phenomenon should year, a result that is a combination of take become less pronounced over time as these rates, modulation profiles, and usage of the kids become tomorrow’s adults, although a medium by subscribers. This variable in the generally heavier level of usage by the paper is dedicated to residential cable academic community may linger. modems – i.e. Internet users at home. This doubling effect is corroborated by other, DOCSIS 1.1 provides the ability to more general Internet traffic studies that support VoIP traffic. It does so by suggest a “Moore’s Law” for data traffic [7]. supporting multiple classes of service (CoS), This analysis notes that this trend has been whereas DOCSIS 1.0 supports only one – pretty reliable except for a period in 1995-96 best effort. DOCSIS 1.1 also allows packet where there was a burst of greater growth fragmentation to ensure that latency-sensitive voice traffic is not bogged down behind large downloaders from the limitations of dial-up “best effort” data packets. From a traffic speeds to deliver multi-megabit files. Peer- generating standpoint, DOCSIS 1.1 also to-peer traffic – pioneered by music file implements pre-equalization at the CM side, sharing through Napster, but subsequently a physical layer technique similar to pre- followed up by similar services (Kazaa, distortion, but for bits. This feature permits Morpheus) – raises a significant flag to more practical use of the 16-QAM mode, operators of broadband networks who which doubles the bit rate compared to a observe and/or police their traffic patterns. QPSK channel of the same symbol rate. In other words, the 160 ksps mode, which The impact of heavy P2P traffic on cable results in 320 kbps for QPSK, provides 640 modem systems offering no byte count or kbps in 16-QAM mode. The result at the hub rate limits is twofold. The raw bits-per- or HE is more bits-per-second pouring out of second load sees a “bias” around which the a CMTS spigot. web browsing peaks and valleys vary. The effect is to have a constant offset or mean DOCSIS 2.0 also speaks foremost to raw value associated with the streaming file throughput enhancements, It provides for a content much like any constant bit rate dual, selectable, medium access control, or (CBR) application. The difference in the MAC (S-CDMA or A-TDMA), and an modern case is the high bit rates associated enhanced modulation profile capable of 64- with the CBR-like traffic are rates that are QAM at twice the previous maximum considered high speed. Of course, with no symbol rate. Raw capability is now about 30 rate capping or policies in place to limit this Mbps. Built-in as well are enhanced type of traffic, a very small number of users interference mitigation techniques for can essentially dominate the throughput of narrowband and burst interference make use the link. Enough heavy P2P users or enough of the newest modes realistic, and use of sharing of single return channels among lower return channels possible, creating more users can therefore create a congestion effective bandwidth. The CMTS spigot scenario, as the demand for transmission therefore just got wider, or the pipe became time slots upstream outstrips supply. more fully utilized. Simulations [12] support this effect. A single users acting as a source of MP3’s, when In summary, with DOCSIS we can placed among a dozen or so other users, expect rapid, raw, bits-per-second growth, managed to consume up to half of the return more efficient bandwidth consumption in capacity over significant period of observed access, self-similarity in backhaul, and both time. This creates a clear forward-looking best-effort and class-of-service (CoS) argument for tiered service offering based mapping. upon either rate or volume limitations with the necessary prioritization and policing Peer-to-Peer (P2P) schemes to enforce the tiered structure.
Although the proliferation of peer-to- To summarize, we can therefore add to peer communication still generally falls our prior discussion of DOCSIS the need or under a residential data (DOCSIS) objective of supporting tiered services. discussion, this phenomenon is significant enough to warrant special mention. It is, of Enterprise Traffic course, certainly the case that broadband access has, in fact, enabled P2P traffic to The play for commercial services can be become as significant as it has, freeing attacked primarily in two ways – DOCSIS- based and fiber-based. Which solution is expected. This is somewhat intuitive, in that, determined by service needs at the business, while residential users find new and which basically boils down to the size of the bandwidth consuming ways to exchange and enterprise. The key features that DOCSIS download content, the shift in business 1.1 provides that make it a reasonable transaction content has not been this solution for a subset of the business market, dramatic. In terms of security, virtual private in addition to cost, are the support of voice, networks (VPNs) translate to both security enhancements that offer multiple service and bandwidth guarantee issues. In terms of classes, enhanced security, and, finally, a reliability, resiliency and 50 msec recovery more realistic opportunity to achieve 10 characterize common “carrier class” Mbps type of performance due to physical attributes expected by business customers. layer improvements that add pre- equalization. Of course, 10 Mbps has a nice Fiber-based systems that segment ring to it in light of comparison to 10Base-T spatially or via wavelength – the only non- LAN environments. Finally, DOCSIS 2.0 DOCSIS, HFC-centric approaches today, encompasses the key features of DOCSIS have no aggregation issues in the access 1.1, but also offers the 3x increased capacity network. As the data hits the first return due to symbol rate and modulation aggregation point in a hub, it is at this point improvements. In addition, the protocol that it may have to be managed, prioritized, advancements inherent in A-TDMA and scheduled alongside other traffic signaling and S-CDMA are designed to requiring bandwidth. If both business voice expose more return bandwidth to the operator and data exist, an architecture that guarantees for high-speed services that previously had bandwidth for voice and queues data packets been unusable for this type of traffic. may be necessary. Architecture decisions are made with these types of network crossroads Based on the above, we can summarize in mind – the service mix at these points by saying that business services over could consist of VOD content, CMTS DOCSIS 1.1 represents for the operator a activity that could include voice and tiered need for QoS tiers, and service level data, business voice and data, even broadcast guarantees, similar to previously mentioned digital video content. DOCSIS needs for peer-to-peer traffic. In summary, then, overall higher symmetrical bandwidth and quality of Fiber-based solutions can deliver higher service guarantees are important to this levels of service, corresponding to larger market, and traffic growth to plan for may be businesses or business campuses. Of course, less dynamic. the target here is to provide cost-competitive voice and data services with the same level Networked Gaming of service experience to the end customer – data rates, security, reliability. While The average age of a “gamer” is on the residential data traffic has grown rapidly as rise. The Gen-X and Gen-Y demographic previously described, business data traffic that grew up with the phenomenon of has taken a more modest trajectory, and Playstation, Nintendo, and now X-Box are business voice traffic is essentially flat. now growing up themselves – if perhaps in Thus, while the per-subscriber business age only. They are bringing their bad habits needs are higher, the growth in this data with them, including their passion for sector is more gradual, meaning the pipes gaming. At last year’s Western Show, not assigned to support a fiber-to-the-business much was well-attended. But, one of the best application may have longer legs than attended sessions dealt with the gaming Today, gaming traffic is a minor phenomenon and the impact on broadband. contributor to overall volume. The number of gamers is relatively small, although the Intuitively, what would we expect certainty of this growing is about a sure a bet gaming needs to be? Certainly it is real-time is there is in predicting network usage. interactivity, the magnitude of which needs Secondly, however, gaming transmission quantification relative to the well understood volume is small, as the games merely baseline needs of voice traffic. Latency is of transmit state information that is used by the key importance, as thumb actions must be software at the ends to actually render the translated into game states and complex video images. Migrating game communicated rapidly to other players. As activity to virtual reality based image important yet is probably jitter variation transport could change the transmission among peers, so that the game can be fair to volumes dramatically. everyone, and no one has a built-in edge. Studies have shown that delays that are not Video-on-Demand noticeable to voice traffic users are quite noticeable to gamers, and, indeed, can affect Data growth has some historical game outcome. precedent, including recent trends with cable modem users. These trends offer anecdotal Again, intuitively, the nature of gaming evidence supporting the bandwidth growth activity would not be expected to have the assumptions predicted in [8]. Video-on- statistical look of web browsing or streaming demand has not been characterized as media. Recent work has analyzed traffic carefully. However, that VOD has distribution of the popular “Quake” accelerated rapidly in the last couple of years application [4], observing both packet is not news, with total VOD revenues arrivals and sizes for clients and of the game increasing nearly 10x between 2000 and server. The results of this study suggested 2002 [14], and the expectation it will double that an Extreme distribution is a good fit for again by the end of next year. It is also not packet arrival of both servers and clients in news that the bandwidth needs of video most cases, as well as packet sizes of server content as a service to one subscriber greatly traffic, where the server is the point from exceed that of a data service to that which game states are updated and broadcast subscriber. VOD services are well down the to the clients. path of some of deploying the widest-pipe and most cost effective technologies, The parameters of the distribution that including Gigabit Ethernet (GbE) and WDM. define the exact shape of the broader family are functions of processing speed distributed The choice of technologies above are the among the servers and clients – another same as those discussed to support the effect that would seem intuitive if there is no bandwidth growth in residential data. network bottleneck. Of course, making sure Because of this, carrying them intermingled this is not the case is what our job is all with one another seems like a logical step for about. Some observations suggest that a split added efficiencies in the network. Important statistical model – part deterministic and part questions to consider are those associated exponential – is a better fit for client packet with the QoS needs of each, and, if arrival in some cases. Client packet sizes necessary, how Ethernet-based transport is were found to be deterministic. Future augmented with added robustness to ensure research is ongoing. such needs are met. For example, over 200 MPEG movies are carried on a single GbE pipe. The case for redundancy and fail-over 18 times as large between now and 2010. It becomes more compelling as pipes widen current contribution in term of bandwidth and carry more traffic. Whereas data consumption and traffic engineering is implementations today may have Layer 4 negligible, although it has potentially large mechanisms providing loss and flow control, architectural impacts if the enabling loss of unidirectional streams represent a technology to light this fuse is all-IP, all the non-trivial protection situation, not to time. This is a visionary decision or mention a bad day to be near the customer timetable – depending on your perspective – service center. every operator must make for the future growth and service providing capabilities of In terms of traffic, VOD has some their system. obvious diurnal dynamics. Of course, most network decisions rely on demand requests From a traffic standpoint, we have during peak usage hours on peak days. The discussed what P2P traffic does to a dynamic dynamic range is quite wide, and makes for set of flows of residential Internet. The some tempting unused bandwidth in off- effect is to create a steady “bias.” In other hours. Fortunately, these behaviors are words, the mean bits-per-second increases, predictable, and fortunately, as an example, and the traffic dynamics exist on top of this peak VOD hours will not coincide with, for mean. For streaming media, this same effect example, peak busy hours of enterprise data would be the expectation when it becomes traffic. By the same token, however, peak significant enough to matter, except that this VOD busy hours may roughly coincide with would be a downstream phenomenon. As peak residential Internet usage hours. such it could be more buried in the noise depending on the asymmetry experienced in In summary then, aggregated VOD the network. Obviously, if the mean is very streams present to us a rapidly growing, high large in comparison to the peaks and valleys, QoS, wide bandwidth application, the needs the impact of peaks and valleys on efficient of which today are essentially met via silo pipe usage is very minor. In other words, if networks. The technology trends, however, the streaming media content (or VOD point towards the same technology choices content for that matter) dwarfs data transport expected for data growth. VOD traffic varies content along the same conduit, the traffic in daily and weekly trends in predictable engineering pipe size problem is simplified, ways over time. since the ups and downs will be relatively small. IP Video/Audio Now, a significant difference between Streaming content has received quite a streaming media and most P2P traffic today bit of air time in the past few years, while is that the former is real-time content, during that time P2P traffic was really what deserving of QoS capabilities that do not caught a buzz about it and took off. What require the same level of sophistication as constitutes streaming traffic and P2P begins moving MP3’s and JPEG’s around. Since to blur, but, in general, the concept of the content still exhibits LRD regardless of streaming media conventionally applied to whether data or streaming media, the idea of a content providing service mechanisms at the ends of the pipe that streaming IP to a computer terminal or settop enforce policies and switch packets need box connected to a computer (or even a TV). awareness into the proper traffic models of The reference bandwidth projection predicts this LRD, so that queues can be build and that this type of traffic will grow to be about implemented to avoid dropped packets and blocking and supply the QoS expected for ARCHITECTING FOR MULTIPLE real time content. SERVICES
Summarizing, streaming multimedia QoS Parameters represents content characterized in prior work as having self-similar behavior. It is a There is no universal definition of relatively high bandwidth consumer on a Quality of Service QoS), just as there is no single user session basis with the QoS needs universal definition of “carrier class.” of other real-time media, but the total Nonetheless, QoS encompasses basically five bandwidth usage is low and growth path very parameters: dependent on architectural and service choices going forward. − Latency – End-to-end absolute delay − Jitter – End-to-end variation in delay Digital TV or HD − Loss – Dropped transmissions − Throughput – Bits-per-second or Broadcast digital TV has much the same Bandwidth characteristics as VOD. There are two main − Availability – Likelihood of the differences. First, in many cases, linear network being “up” supertrunking is used rather than digital transport as a low-cost alternative when high QoS has achieved buzzword status quite bandwidth digital backbone is not in place. recently, and its footprint is all over Second, the service group size is very large, standardization committees. What is making redundancy of path and equipment essentially going on are efforts to bring to the quite important. High Definition streams, for data world something it has always lacked – the most part, represent to the network guarantee-able QoS – but doing so on a engineer, digital TV traffic on steroids. The “connectionless” network while keeping as bandwidth hungry nature of HDTV is a much legacy frame and protocol structure promising possibility for inspiring intact as possible. The result is primarily networking bandwidth upgrades. profitable to the acronym maker (or marketing team). The effort is a logical Similar to streaming media – and in fact outgrowth of the indisputable fact that analogous except for the more standardized Ethernet dominates the LAN. As the LAN format of delivery – digital TV provides a aggregates to the MAN and WAN, the goal steady flow of packets and a nice averaging of leveraging the broad familiarity with of bandwidth behavior to a bit rate number Ethernet, its cost points, and the flexibility of that is a function of compression and features within Ethernet and IP as protocols statistical multiplexing of a large number of have driven traditional Ethernet and IP streams. If anything rides along the same network designers to innovative approaches channel, such as VOD, then VOD dynamics to solving this classic QoS shortcoming in would be superimposed. the hopes of scaling the local LAN to a broader market.
Not surprisingly, some techniques to enhance Ethernet resemble old ideas. For example, the use of the DiffServ protocol (differentiated services), when implemented over MPLS (Multi-protocol label switching) has the look and feel of ATM, but with a lot of different acronyms describing the details. Ethernet and flexibility of IP have made DiffServ provides the ability to classify a riding this wave a necessity in network packet with a forwarding class describing its design, to the extent that incrementally priority on a per-hop basis. The latter fact upgrading the technology to carry more than actually limits the overall QoS strength of best-effort is more palatable than addressing DiffServ on its own, but increases its major equipment and protocol overhauls and practicality. The role of MPLS is to expedite learning curves. the forwarding of packets through the network by creating label-switched paths via QoS – Who Needs It? tags on the Ethernet frames, directing packets at Layer 2, rather than making route Let’s list some of the services decisions through the network that create encountered or viewed as on the horizon. processing bottlenecks and subsequent How do these compare as far as who needs latency and jitter problems. Thus, these what for QoS? Let’s use a simple scale: schemes together offer prioritization of High (H), Medium (M), or Low (L) need for payload types, and create predefined and the particular QoS parameter below. A expedited paths through the network. This qualitative summary of QoS needs is shown scenario has indisputable similarities with in Table 1 below. Certainly, there is plenty ATM. of room for debate (I adjusted this chart more So, why re-invent the wheel? The than a dozen times), and some still require answer is simply because the dominance of more learning and evaluation. ______Table 1 – Services and QoS Need
Latency Jitter Loss
Residential Data (DOCSIS Internet) L L L Residential Voice (VoIP) H H L Business Data (DOCSIS 1.1 or higher) L L H Business Data (fiber-based) M L H Business Voice (VoIP or T1/T3) H H M Business Video (videoconference) H M L MPEG or IP Video or VOD M H M HDTV Broadcast M H M IP Audio (Radio AOL) M M M Interactive Gaming H H H ______
Clearly, we can recognize that some nature of the service, it may be loss tolerant applications are real-time, while others are or not. In general, if the content itself is to not. This fact primarily drives the latency be transduced for human senses, it is likely to and jitter QoS needs. Also, based on the be loss tolerant to some extent. Human senses are quite effective as filters. If the variety, and would certainly require some content is information for a computer to traffic engineering and modeling. interpret and process, it is likely to be less loss tolerant. Streaming media has the same type of architectural implications in the network, What tools exist to assure the level of with the difference being that DOCSIS QoS desired is achieved? In the DOCSIS supports the access portion of the network. world, the use of DOCSIS 1.1 or higher, and Thus, mapping of QoS mechanisms from one a next generation CMTS [13] provide this side of the CMTS to the other – again using capability. The CMTS is a key element the CMTS both the due media and QoS between the access and transport network, transducing – is needed. acting as both a media converter at layer 1 and protocol delineation point for layers 2 Architecture Technologies and 3. DOCSIS 1.1 provides the class of service capabilities on the HFC side, while Clearly, many services with many advanced layer three implementation such as different needs are set to co-exist. Sound per-flow queuing provide traffic management business practice means finding an efficient functionality on the network side. Thus, for means to handle them by judicious choice of the first three items in Table 1, DOCSIS 1.1 technologies, levels of integration, and a and next generation CMTS enable the healthy concern for operational costs and providing of QoS mapping from access scalability. The term “triple play” alone network to interconnect ports. For business sounds like an abbreviated set of services, voice and data based on fiber connectivity but the flavors within the triple play, as rather than DOCSIS 1.1, and segmented shown above, clearly make the problem more spatially or via wavelengths in the access complex. The access network itself is network, QoS schemes must reside in the constantly being re-thought for fresh ideas, aggregation equipment and supported such as data overlays, wireless interfaces, elsewhere in the architecture. and intelligent processing. At the aggregation points in the hub and Headends, On the video transport side, such as various technology choices exist, some of VOD and HD transport, QoS is assured by which are deployed already in silo networks the fact that these systems currently are as previously indicated in VOD cases. VOD essentially silo systems. Statistical represents a good reference example because multiplexing occurs as server content of where it is in the cycle – basically just at traverses switches, but the rules of or past the “knee” of one of those classic engagement are simplified by the singular marketing hockey stick charts, depending on content and rules easily developed from this which analyst you ask. It is a service simplified, application-specific architecture. experiencing significant growth, and it is Should these services become part of an using technology on the move in both the integrated triple-play transport network, the server and multiplexing arena, as well as in dynamics of the traffic situation could the transport pipe. VOD transport has been change significantly. For example, for a migrating from DVB-ASI transport to lower single IP pipe supporting video and data, cost, more-flexible, GbE links. And, again, there would be a heavy reliance on IP QoS use of GbE technology makes its way into schemes through some of the existing the data world as well. VOD also has been a standardization efforts to assure the video driving application for another key QoS needs are met. The end-to-end technological option - wave division capabilities are not of the “guaranteed” multiplexing (WDM). Gigabit Ethernet has some notable – gobs of bandwidth assuring that nothing shortcomings as an all-inclusive answer for gets held up. For the price of this network design. Ethernet as well as IP over overcapacity, if well managed, there would Ethernet – both designed for data – do not be no congested routes, and no pipe traffic inherently offer the resiliency, availability peaks requiring buffering and queuing delay. and network management attributes that There are smarter ways to solve the problem become important elements of a “carrier rather than relying on this brute force class” solution when so much content is approach. riding on the success of a single link. IP over Ethernet was developed as a “best effort” WDM itself is, in general, a brute force technology, and most of the ongoing efforts capacity enhancing tool. It allows a single today revolve around finding way to be better fiber to carry multiple data-bearing streams, than best effort. The previously mentioned such as GbE or 10 GbE, by using a different DiffServ and MPLS developments fall into wavelength for transmission for each. this category, and there are others. Thinking However, WDM does not easily offer QoS in terms of end-to-end IP as an attractive end consciousness. This is not a show-stopping game, MPLS, in fact, can be viewed as a way issue - intelligent wavelength management to skirt the routing limitations of an all-IP exists as a relatively mature technology. network by avoiding the per-hop calculation Integrating wavelength management as part of routes through the network. Not all of system resource management has been an developments aimed at traffic engineering anticipated direction of the telecom sector for and hardening data systems are completely some time, and could see relevance in cable new, however. TCP/IP itself is a kind of networks as well since the same premise QoS feature, and type-of-service (ToS) drove that thinking – bandwidth growth and header bits have been around since 1981. support for advanced services. Limited capabilities of these features – invented still in a “data world” context – The classic shortcoming of QoS is limit their power to meet the kind of diverse precisely the reason for equipment based on needs expected. Next Generation Sonet. There is no debating the QoS features of Sonet transport. The Another Ethernet issue is that it does not knock against Sonet had to do with its rigid inherently support circuit-based voice. structure that made it inefficient as a packet Technologies exist to create virtual circuits data transport system. Coarse bandwidth over Ethernet. Similarly, at layer 3, voice increments left excess unused bandwidth, over IP (VoIP) has been developed driving up the effective cost of doing technologically. Each today has cost business by decreasing fiber usage penalties. In short, however, Ethernet, even efficiency. GbE or 10 GbE, and even if we include wave division multiplexing (WDM), cannot go it However, precisely these data issues are alone. All-IP looks attractive from an addressed with Next Generation Sonet, all interoperability and flexibility standpoint, but the while holding firm on the guaranteed, the jury is out on guaranteeing that all of the proven, mature resiliency and reliability that QoS needs can be met even with the traffic has no comparison today among alternative engineering tools brought to the table with technologies. Furthermore, there is lots of MPLS and other tools designed to optimize existing Sonet-based infrastructure. What packet transport. Actually, right now, use of modern equipment does is build the data enough wavelengths and 10 GbE would be flexibility into formerly coarsely-grained all essentially the oft-practiced lazy man’s QoS TDM-only platforms through virtual concatenation (VC) and link capacity RPR can run on entrenched physical layers adjustment (LCAS) which allows dynamic – (i.e Sonet and Ethernet PHYs), which is i.e. supporting data – provisioning of VC important considering the amount of carriage. Such platforms contain port deployed infrastructure. The standard is still interfaces natural to data handling, such as an emerging one, and the discussions break 100BaseF and GbE, as well as the traditional down into two camps – Cisco and others. voice-related interfaces of a traditional Sonet platform. Other data-oriented features are WHAT NEXT? included as the platforms evolve to meet the shift in traffic demand. Support of standard What to make of all of these services, framing protocols for data is gathering technologies, and, in general, the many momentum (Generic Framing Protocol or choices that face network designers today? GFP), as well as standards-based packet Needless to say, there is no single magic classification and guarantee-able class-of- answer, and recommending what works best service mapping. for 2010 will simply create competition for someone’s infamous “who needs more than Next generation Sonet platform, then, 64k of memory” comment. The good news offer the guaranteed resilience and inherent is that cable operators are in the drivers seat QoS parameters for both TDM and Ethernet at the moment. The competition for service services – the resiliency still unique to Sonet provider of choice is arranged in their favor. – and now offer the added granularity and But it is unclear how long this will last given flexibility to efficiently support data needs. alternative solutions and the allure of residential broadband as something people The networking world never has quite will pay for, a few-and-far between uptick in enough protocols, and one of the latest is a beaten down economy. To capitalize on a aimed at providing efficiencies of packet- game that is cable’s to lose, and since there is based transport natively, but having the no one-size-fits-all solution at this juncture, resiliency characteristics of Sonet. Avoiding what we can recommend is a five steps “keys the traditional Sonet limitations was a key to success” approach: target of this development effort. A standards body has been formed, IEEE 1) Have a planned roll out of services to 802.17, that is developing the layer 2 provide or support. A comprehensive sample protocol known as resilient packet ring set is provided in this paper. This first step (RPR). Not surprisingly, the basic frame is, as always, a business exercise. structure of RPR was based on Ethernet, and 2) Develop a bandwidth forecast of your adds to it MPLS and Class-of-Service header own, and comprehend the traffic and QoS content, as well as other fields. aspects of each service. The given information and references provide guidance RPR is a ring protocol, with the towards both. This is a business and objectives of optimally supporting all technical exercise. previously described traffic types, 3) Understand the current and growth maximizing efficient use of counter-rotating capabilities and limitations of any existing ring bandwidth, and simplifying network infrastructure in the context of 2). This provisioning. The critical importance of especially includes the emerging standards packet resiliency is recognized as a key and techniques suited to the evolving service focus, as the traffic mix and amount no mix, and the flexibility available where there longer fit into a “best effort” paradigm as is infrastructure to build. The previous was once the case. As a layer 2 technology, section points the way towards much of the relevant activity in this area. This is a Columbia tragedy this past winter, the video technical reading & research exercise. phone images and stories of the researcher 4) Develop a time-phased service and her team offered a glimpse into the risks integration and network evolution plan for of pushing the envelope. Fast-forwarding to the traffic mix (not necessarily the same as the present, recent literature describes the an integrated network) aligned with 1), 2), wireless infrastructure in an Alabama and 3); the hard work of the first three steps hospital [9]. The article also describes the make this more straightforward than it network services the hospital uses to sounds. This is a business and technical transport mission-critical data. The concept exercise, and the one that decides whether of remote medicine involves supporting you grow, maintain, or flounder and become transmission of high-resolution images and exposed to competition. data in real-time to doctors to serve live 5) Know what end-to-end means to your patients anywhere, such as remote locations, network responsibilities, and modify any a third-world country, or on a battlefield. “silo” role & responsibility organizational Obviously, administrative and legal obstacles definition to smoothly evolve across the abound anywhere medicine is involved and network interfaces, features, standards, and lawyers are prowling. But the advantages mapping schemes. This is a technical possible are access to expert advice in a management exercise. Technical managers timely manner, access to trusted medical often used to be technical people and carry advice internationally, access to observation with them technical biases, so this one may of expert and advanced procedures by other be more difficult than it sounds. doctors and students, and less dependence on getting out and about to receive care when CONCLUSION seriously ill. Can we invent a higher QoS service need? The military designs and On the QoS front, one of the more implements its own private networks for compelling human stories in recent years mission critical data, because there is occurred when a researcher in Antarctica absolutely no margin for error. But, it can found herself stranded with her team during afford to as well, while applications such as an inaccessible time of year for rescue the above would rely on the kinds of QoS missions and in need of critical medical and traffic attention. Much like the space shuttle engineering techniques being developed and you buy this perception of today? Would discussed herein. this aggregation mix lend itself to a particularly convenient model? Some of the And what of 2010, based on the forecast constraints of the previously introduced previously referenced? Architecturally, will central limit theorem, aside from a variety of the triple play be implemented in a unified independent sources, is that the independent network, with today’s darling being an all-IP distributions have finite variance, and that format all the way to the home? Will some there not be a singularly dominant services continue to ride over separate distribution. Thus, while the broad traffic parallel networks optimized to the bandwidth mix implies central limit simplification, a and QoS required? According to the forecast single dominant service can disturb this referenced, 59% of the forward path digital convenience. Furthermore, a cornerstone traffic demand will be Internet access, 26% characteristics of self-similarity – heavy tails will be some form of VOD, 13% will be – can also imply infinite variance, another streaming audio or video content of the PC central limit theorem killer. The jury is once variety, and the remaining IP telephony. Do again out, as research continues to classify traffic trends and distributions for network Contact Info: modeling and optimal architectural design. Similarly, networking technologies will Rob Howald simultaneously evolve, making putting a Director stake in the quicksand that much more Systems Engineering perilous. Network Infrastructure Solutions Motorola Broadband [email protected] References 8) Nash, Randy, A 10-yr Residential Bandwidth Demand Forecast and 1) Adas, Abdelnaser, Traffic Models in Implications for Delivery Networks, 2002 Broadband Networks, IEEE Communications NCTA Technical Papers, May, 2002. Magazine, July, 1997. 9) Look Doc, No Wires, Communications 2) Addie, Ronald, Moshe Zukerman, and News, March 2003. Timothy Neame, Broadband Traffic Modeling: Simple Solutions to Hard 10) Robert, Jim, Traffic Theory and the Problems, IEEE Communications Magazine, Internet, IEEE Communications Magazine, August, 1998. January, 2001.
3) Boggs, David, Jeffrey Mogul, and 11) Sahinoglu, Zafer and Sirin Tekinay, On Christopher Kent, Measured Capacity of an Multimedia Networks: Self-Similar Traffic Ethernet: Myths and Reality, Digital and Network Performance, IEEE Equipment Corp, Western Research Communications Magazine, January, 1999. Laboratory, Reseacrh Report 88/4, September 1988. 12) Shaw, Terry, Bandwidth Utilization on Cable-Based High Speed Data Networks, 4) Borella, Michael, Source Models of 2002 NCTA Technical Papers, May, 2002. Network Game Traffic, 3Com Corporation. 13) White, Gerry, Tiered Data Services to 5) Brownlee, Nevil and KC Claffy, Enhance Customer Value and Revenue: Why Understanding Internet Traffic Streams: are DOCSIS 1.1 and Advanced Queuing and Dragonflies and Tortoises, IEEE Scheduling Required, 2003 NCTA Technical Communications Magazine, October, 2002. Papers, June, 2003.
6) Chu, Tam-Anh, WAN Multiprotocol 14) MultiChannel News, Broadband Data Traffic Management: Theory and Practice, Book Communications Design Conference, September, 2002.
7) Coffman, K.G., and A.M. Odlyzko, Internet Growth: Is there a “Moore’s Law for Data Traffic?”, AT&T Labs Research, July, 2000. CABLE & CE INDUSTRY COOPERATION ON UNIDIRECTIONAL DIGITAL CABLE RECEIVERS
Brian Markwalter, David Broberg Consumer Electronics Association, Cable Television Laboratories
Abstract operators signed a memorandum of understanding covering interoperability of As consumer electronics companies unidirectional digital cable products and cable and the cable industry continue to work systems. The MOU culminated months of together to accelerate the deployment of high- work, facilitated by the Consumer Electronics definition digital television (HDTV), they Association and the National Cable have created a memorandum of Telecommunications Association, to reach understanding (MOU) that defines how cable consensus on how best to achieve the mutual systems will deliver services and essential goal of retail availability of cable ready elements needed for unidirectional digital receivers while ensuring cable services are cable-ready receivers to receive such services. delivered as intended. The MOU deals with The MOU relies on Society of Cable Telecom- four impediments that prevented television munications Engineers (SCTE) and Consumer manufacturers from being able to introduce Electronics Association (CEA) standards to cable ready TVs through the OpenCable provide the framework for interoperability. process: (1) legal concerns with the available This paper describes the December 2002 POD Host Interface License Agreement, (2) MOU and focuses on the standards referenced certainty that a large percentage of cable therein that define requirements for cable systems nationwide would follow specific systems and receivers. digital transmission standards, (3) the lack of encoding rules for copy protection, and (4) a This paper provides an overview of the test or certification regime in keeping with the agreement as a foundation for providing a way televisions are typically measured for more detailed looked at the self-certification compliance. program it requires. The paper describes the categories of tests prescribed by the MSOs rightfully sought to ensure that agreement including Critical Tests, Non- in reconciling these CE manufacturer critical Tests, and Network Harm Tests. concerns their own goals not be sacrificed. Taken together, these test make up the Test These goals being: (1) cable services are Suite, jointly developed by CEA and delivered consistently whether through a CableLabs. The Test Suite is derived from leased device or a retail device, (2) cable not existing work by CableLabs as part of the be competitively disadvantaged with respect OpenCable project. This paper describes in to other video distributors, (3) operators have greater detail the foregoing testing freedom to develop and market new services, methodology and the expected benefits to the and (4) retail cable ready devices not harm the industry. cable network or allow theft of service.
Elements of the MOU obviously deal OVERVIEW OF DECEMBER 2002 MOU with certain aspects of these goals, as evidenced by the inclusion of a new DFAST In December 2002, 14 television license agreement and encoding rules. manufacturers and eight cable system Enough ink will be spent on these mostly legal matters elsewhere. This paper instead necessary to tune and display the services focuses on the standards that both parties have offered by the operator. The term Out Of agreed to rely on for compatibility and the Band indicates that the SI tables are delivered self-certification process for the retail devices. by a possibly proprietary transport to the POD and then forwarded in a standardized fashion STANDARDS THAT APPLY to the cable ready device (Host) through the Extended Channel. The Core Standards ANSI/SCTE 54 2002, titled Digital In the MOU, cable system operators Video Service Multiplex and Transport commit that cable systems with an activated System Standard for Cable Television, is now channel capacity of 750 MHz or greater shall SCTE 54 2003 after completing its revision comply with the following SCTE standards. process. This standard builds on MPEG-2 • SCTE 40 2001, as amended by Transport Stream coding to define how cable DVS/535 systems construct multi-program Transport • ANSI/SCTE 65 2002 Streams. • ANSI/SCTE 54 2002, as amended by DVS/435r4 ANSI/SCTE 28 2001, titled HOST- POD Interface Standard, is in the final And all digital cable systems shall editorial stages after completing its ballot and comply with these standards. should publish as SCTE 28 2003. This • ANSI/SCTE 28 2001, as amended by standard defines just what its title suggests --- DVS/519r2 clearly necessary for developing • ANSI/SCTE 41 2001, as amended by unidirectional digital cable products. DVS/301r4 ANSI/SCTE 41 2001, titled POD The ‘‘as amended by’’ notation Copy Protection System, is near the end of a reflected the need to point to these standards major revision related to switching the copy that were at the time being revised in the protection system to reliance on X.509 SCTE DVS committee. A quick description certificates. This standard defines how the of each standard and its status as of this interface between the POD and HOST is writing follows. protected from having to expose video content in the clear. SCTE 40 2001, titled Digital Cable Network Interface Standard, is in the final The first three standards above are an SCTE approval stages and should publish as obligation for 750 MHz cable systems to SCTE 40 2003. SCTE 40 defines the key deliver digital video by these standards. The characteristics of what the cable system HOST-POD Interface and its associated copy delivers to the television in terms RF, protection standard are an obligation of all transport layer, and other services, such as cable systems, regardless of whether digital emergency alerts and closed captioning. transmission is used. Similarly, digital cable products marketing under this MOU are ANSI/SCTE 65 2002, titled Service obligated to tune digital channels in Information Delivered Out Of Band for accordance with SCTE 40, navigate using Digital Cable Television, is unchanged since SCTE 65, respond to emergency alerts per the MOU was signed. This standard defines SCTE 54, and include a POD interface Service Information tables providing the data compliant with SCTE 28 and SCTE 41. Other Standards EIA/CEA-818-D, Cable Compatibility Requirements, collects together requirements The MOU relies on other standards, from other standards for application to digital particularly related to certain interfaces on cable systems and compatible receivers. Part leased set top boxes and retail digital cable I states minimum requirements for receiver- products. Television manufacturers commit compatible digital cable TV systems, and Part to providing DVI or HDMI interfaces on a II states minimum requirements for cable- phase-in and resolution basis and cable compatible digital TV receivers. SCTE operators commit to providing IEEE 1394 and DVS/538r1, Uni-Directional Receiving DVI interfaces on HD set top boxes on a Device Standard for Digital Cable (Input), is a phase-in basis. Cable operators expressed an proposal for standardization of receiver interest in DVI (uncompressed video) as the requirements intended to complement the preferred interface, hence the commitment by transmission standards used by digital cable television manufacturers to support it. TV systems. Neither of these documents are Television manufacturers needed support for referenced directly by the MOU, except as a compressed video interface on set top boxes sources for harm prevention test items. for recordability, explaining the inclusion of this interface on leased boxes. PROTOCOL IMPLEMENTATION CONFORMANCE STATEMENT The IEEE 1394 interface described in the MOU is actually defined by a pair of One of the tools that often is used in standards, ANSI/SCTE 26 2001 and CEA- the process of verification of a complex 931-A. SCTE 26, Home Digital Network product that follows a number of industry Interface Specification with Copy Protection, standards is the Protocol Implementation builds on EIA-775-A and EIA-779, which in Conformance Statement (PICS). This turn build on IEEE 1394, to completely document is a detailed collection of every one define how this interface is used between a of the requirements from all the referenced cable device and another CE product. CEA- standards. This document creates a 931-A, Remote Control Command Pass- traceability matrix and serves as the basis for through Standard for Home Networking, adds any conformance statement of a manufacturer the usability feature that a display device can seeking certification. pass-through remote control commands to the video source at the other end of the 1394 Since the MOU and the proposed rules interface. for a unidirectional cable receiving device were written, a team of engineers from ADDITIONAL REQUIREMENTS several manufacturers, along with staff of CableLabs and CEA, have been working to Harm Prevention Tests, those meant to complete this critical piece of documentation. protect the cable system and its ability to In the first quarter of 2003, this team deliver services, are singled out as applying to participated in meetings and conference calls all products under the MOU. A mutually totaling more than 120 hours and spent in agreed upon set of harm prevention excess of $10,000 on conference call services requirements does not exist in the form of an to this end. This concentrated effort shows SCTE or CEA standard. The MOU how critical is the element of accurately recognizes this deficit by pointing to documenting each testable requirement. EIA/CEA-818-D and DVS/538 as sources for these requirements. The PICS document contains over 600 Test Suite (JTS). This document details each unique requirements. In many cases each line of the unique test procedures that are used to item includes a direct quotation of a verify the requirements stated in the PICS. normative statement from the applicable The ATP gives instructions to the test industry standard, along with a chapter and technician who performs the test and it details verse reference location. In some cases, a the equipment settings, connections, and other requirement was stated without any citable test conditions. The ATP also defines the industry standard to reference. In those cases range of acceptable results and how the results each new requirement is added to an appendix should be documented. at the end of the PICS. There are three basic guidelines that This process of including requirements were used in creating the tests within the without an external reference does represent a ATP: (1) All of the tests are ‘‘black-box- departure from the usual process of tests’’ meaning that the tests are performed on developing a PICS. This departure from past a closed box, using only the available input CableLabs practice was necessary since the and output interfaces; (2) The tests are not MOU relies solely on published SCTE and meant to limit the type of test or procedure CEA standards and some mutually agreed that can be used to verify compliance, but are requirements derived from other sources, simply a record of an agreed upon group of including OpenCable, EIA/CEA-818-D, and tests that are applicable; (3) The test plan is DVS/538. not static or complete, further revisions are expected as additional tests are developed and The PICS documentation also serves new test equipment becomes available. as the detailed breakdown showing which requirements relate to Critical Tests and Non- The ATP is divided to match the Critical Tests. The Critical Test items are breakdown of the PICS into Critical and Non- further divided to show which apply to ‘‘Tune critical, with the Critical tests further divided and Display’’ requirements and which remain to show Harm prevention tests, security tests as Harm to Network, Security, or other harm and tune and display tests. This breakdown is related tests. The purpose of this division is to prescribed by the terms of the MOU. show which requirements apply to the different type of products defined in the MOU Each test within the ATP may be used and proposed rules. to verify one or more of the numbered requirements of the PICS. Each test identifies The final purpose of the PICS what is being tested, the test equipment to be documentation is to list the requirements that used, and the instructions on the exact settings need to be tested in the Acceptance Test Plan. of the controls and instruments used. This completes the traceability so that every Connection diagrams and further explanations test may be traced back to one or more line of the setup are provided so that all tests are item in the PICS, each of which can be traced readily repeatable. back to a normative statement of a referenced industry standard. A variety of tests are necessary to fully determine compliance with the standards. One ACCEPTANCE TEST PLAN group of tests can confirm a portion of the requirements using a POD simulation tool that The Acceptance Test Plan (ATP) is can be programmed to provide many of the another document that is included in the Joint message types that are used on the POD interface. This tool logs the response from device verifies the authenticity of the the unidirectional receiving device and certificates held by the other device. If both analyzes the response to confirm compliance. sides agree, the interface is said to be authenticated. If this process fails, cable The cable side has proposed also to services are disabled. include interoperability tests which use a genuine POD on a live cable plant. This type ‘‘Self-Certification” is the form of this of test has been found to be important in certification process that is prescribed by the previous CableLabs testing since the MOU and that relies upon the individual simulation tools do not contain proprietary manufacturer’s own statements and circuitry needed to work on a real cable plant. documentation. While the exact details of the Without those circuits, the tool is not able to self-certification process are not fully defined receive messages from a cable headend which nor agreed upon at the time this paper was is necessary to confirm the receiver is not prepared, the following basic principles are interfering with headend communications expected to be used: according to the requirements of the Harm tests. Further, there are a variety of 1) The first prototypes of the unidirectional requirements associated with the proper digital television product will be brought to reception of the OOB signals, which vary CableLabs or an appropriately qualified third widely from plant to plant that are not testable party testing facility where the Test Suite will using the simulation tools. Television be executed. Test events will be scheduled at manufacturers believe this type of CableLabs to reasonably accommodate the interoperability testing is not part of the demand and will be coordinated to make best MOU's self-certification process and offered use of resources. instead to work with cable on interoperability events. 2) If the test results reveal any failures of the Critical Tests and the product is a The ATP also includes the forms that unidirectional digital television product, then record the results of each test. Blank space is corrections must be applied and the product provided to record the measured results right resubmitted to CableLabs for re-testing as next to the defined range of acceptable results. many times as it takes to correct all the This documentation becomes part of the first Critical Test failures. If the first prototype prototype test suite results that are recorded at submitted is not a television, and has critical CableLabs. test failures, only the corrections to the Harm Prevention Test failures need be retested. SELF CERTIFICATION PROCESS 3) Once the manufacturer has successfully passed all Critical Tests and corrected all Certification is the process of verifying other test failures as needed, the passing test compliance with the required standards results are submitted to CableLabs along with necessary to earn the right to use the digital the self certification documentation. This certificates necessary to operate on a cable additional documentation includes the plant. Without these digital security affirmative conformance statement and other certificates, the product would not be details that have not been fully defined at this recognized by the cable system. When the time. digital cable receiving product is first plugged into the Point of Deployment card (POD), a 4) Once the passing test results and the Self digital authentication process ensues. Each Certification Documentation has been submitted, CableLabs authorizes the assigned REFERENCES Certificate Authority to begin issuing the X.509 certificates to the manufacturer for the The MOU is available under the FCC Further model and range of products specified. Notice of Proposed Rulemaking, FCC 03-3, in Dockets CS No. 97-80 and PP No. 00-67. 5) Subsequent products by the same manufacturer have no obligation to be tested SCTE standards are available at at CableLabs, but need only the Self www.scte.org. Certification Documentation to be authorized for digital certificates. CEA/EIA standards are available at http://global.ihs.com/. NEXT STEPS
At the time of this writing, work remained on the PICS and ATP; they are expected to be completed by the time this reaches print. There will also be some further negotiation and documentation needed to fully define the details of the Self Certification process.
Of course, the FCC must endorse the proposed rules as submitted with the MOU in order for this process to be activated. In the mean time some manufacturers are going ahead and making products designed to meet the full OpenCable requirements under the PHILA agreement while others are waiting to take advantage of the MOU process.
There also remains some risk that the FCC may not endorse the exact proposal as submitted. If that happens the MOU says the deal is off and everyone will have to reassess how to proceed.
CARRIAGE OF MPEG-4 OVER MPEG-2 BASED SYSTEMS
Ardie Bahraini Motorola Broadband Communications Sector
Abstract The MPEG-2 system standard (ISO/IEC 13881-1) provides two The MPEG-4 specifications have alternatives to carry MPEG-4 content over provided substantial advances in many MPEG-2 Transport Stream (TS). The first areas of multimedia technology. In scheme is straightforward, and provides MPEG-1 and MPEG-2 System the capability for carriage and signaling of Specifications referred only to overall individual MPEG-4 audiovisual architecture, multiplexing, demultiplexing Elementary Streams (ES) by employing and synchronization of elementary the MPEG-2 system-layer parameters streams. The MPEG-4 specification goes such as PCR, PTS and DTS. This scheme beyond these areas to encompass content could be used in existing systems that description, interactivity, and scene already use MPEG-2 Phy and Transport description to name a few. This paper layers and want to take advantage of the only addresses the overall architecture, better compression schemes as well as the multiplexing and synchronization of synthetic video coding tools offered by MPEG-4 content when carried in a system MPEG-4 part 2 (ISO/IEC 14496-2). that already supports MPEG-2 Transport MPEG is also extending the MPEG-4 Stream. video standard in its specification ISO/IEC 14496-10 (also known as JVT) which will provide significant INTRODUCTION compression advantage over both MPEG- 2 and part 2 of MPEG-4. More details MPEG-4 is the first digital audiovisual regarding this implementation will be coding standard that expands beyond provided in the first part of this paper. defining compression algorithms to address emerging computing and The second alternative defined in the telecommunication worlds. MPEG-4 MPEG-2 system layer provides the system specifications were intentionally capability for carriage of MPEG-4 scenes developed to be transport agnostic, in addition to carriage of MPEG-4 enabling MPEG-4 content to be carried audiovisual elementary Streams. Carriage over many different transport systems of this type of content over MPEG-2 such as MPEG-2, IP, ATM, etc. In Transport Stream follows both MPEG-2 particular, MPEG has amended MPEG-2 system standards as well as MPEG-4 system standard ISO/IEC 13881-1 to systems (ISO/IEC 14496 –1) allow carriage of MPEG-4 content over SL_packetize or FlexMux tools MPEG-2 Transport and Program Streams specification. This scheme can also be and this amendment is included in the implemented within existing systems in published 2000 edition of ISO/IEC 13818- order to provide the MPEG-4 object-based 1. coding and scene composition capability in addition to the better compression that
is provided by MPEG-4. The second part audio-visual Elementary Streams and is of this paper will explore this alternative specified by references [2] and [3]. The in more detail. Sync Layer manages Elementary Streams, their presentation and synchronization The last section of this paper will information as well as fragmentation and briefly explore how these random access information. MPEG-4 implementations could be used to enhance Sync Layer syntax is specified by current cable systems by migrating from ISO/IEC 14496-1 [4]. The delivery layer dual carriage of Analog and Digital to an specifies the transparent access to other all digital network in order to address layers independent of delivery technology. future bandwidth requirements as well as As depicted by Figure-2, one could providing additional services and features further divide the delivery layer into two such as HD, VOD and home Gateway sub-layers namely, DMIF (Delivery based on MPEG-4. Multimedia Integration Framework) layer that is specified by MPEG-4 ISO/IEC A HIGH LEVEL OVERVIEW OF 14496-6 [3], and TransMux Layer that is MPEG-4 DELIVERY LAYERS not specified by MPEG-4 intentionally and is left to the transport technology such MPEG-4 predecessors, namely MPEG- as IP, ATM or MPEG-2 to just name a 1 and MPEG-2, were designed to address few. specific systems. For example, MPEG-2 As shown in Figure-1, the abstract developed Transport Stream (TS) and layer demarcation between Compression Program Stream (PS) systems were Layer and Sync Layer is referred to as ESI targeted solutions toward TV (Elementary Stream Interface) and the Broadcasting and Local Retrieval of abstract layer demarcation between Sync content respectively. Hence, the MPEG-2 Layer and Delivery Layer is referred to as system was specifically designed to DMIF. optimize the transport of targeted data and In the following sections, a brief delivery systems by integrating the Sync summary of terms related to the MPEG-2 and Link layers. As a consequence, the System that are used throughout this paper MPEG-2 System is not efficient and is offered for those readers who are not cannot easily be ported to other mediums familiar with MPEG-2 Sync and Link and delivery systems without substantial Layers terminology. Then, the carriage of overhead. MPEG-4 elementary streams over MPEG- MPEG-4, on the other hand, from the 2 Transport Stream are described as beginning was designed to be flexible and depicted in Figure-3. This independent of under-layer technology implementation takes advantage of the such as the delivery system or link layer MPEG-4 Compression Layer without in order to be adaptable to different using other features and functionality delivery systems such as TV broadcasting, provided by MPEG-4. Next, system IP and ATM systems. To address this requirements and architectures are separation and be adaptable by different presented for those systems that not only systems, MPEG-4 defined three abstract attempt to use MPEG-4 compression, but layers namely: Compression Layer, Sync also intend to use other features of Layer and Delivery Layer as depicted in MPEG-4 such as Link Layer and Figure-1. The Compression Layer TransMux. This implementation is specifies the encoding and decoding of depicted in Figure-4. Finally, in the last
section of this paper, the impact of four bytes allocated for the header that MPEG-4 in the existing Cable TV system include PID in addition to other fields. is briefly discussed; although a detailed Each PID is associated with an ES of a discussion and analysis is beyond the service; therefore one program may have scope of this paper. one video PID and several Audio PIDs. Every MPEG-2 Transport Stream A SUMMARY OF THE MPEG-2 multiplex carries a set of tables known as TRANSPORT STREAM (TS) Program Specific Information (PSI) tables. These tables contain information The MPEG-2 system specification about services, which are present in the ISO/IEC 13818-1, defines two schemes multiplex. PSI data includes the following for multiplexing Elementary Streams into tables: Program Association Table (PAT), a serial bit stream namely, Transport Program Map Table (PMT), Conditional Stream (TS) and Program Stream(PS). Access Table (CAT), and Network The Transport Stream scheme is widely Information Table (NIT). Two tables that used in transmission of Audiovisual are relevant to our discussions are PAT content in CATV today and the Program and PMT tables. In a compliant MPEG-2 Stream is used mostly for storage media. multiplex, there must be only one In this section only Transport Stream Program Association Table (PAT) that hierarchy is examined since TS is the contains the list of services associated in protocol that is applicable to CATV as the multiplex. The PAT in a multiplex noted above. associates each service number with a specific PMT PID in the same multiplex. Each Elementary Stream (ES) contains PMT, in turn contains the list of PIDs coded video, coded audio or other data associated with each service and other associated with a single program. These associated data. One field of interest to streams are separately packetized and our discussion is the stream_type that is formatted into a structure defined by used to associate each component of a MPEG-2 as Packetized Elementary service identified by a PID with the type Stream (PES) as depicted in Figure 5. As of elementary stream or payload carried shown in this figure, each ES could within that PID. expand into several PES packets. Each PES packet is identified by a stream_id in CARRIAGE OF MPEG-4 ES VIA the packet header. The Stream_id that is THE MPEG-2 TRANSPORT STREAM associated with each PES packet is defined by MPEG-2 ISO/IEC 13818-1 This section discusses the System specification and identifies the encapsulation of MPEG-4 ISO/IEC 14496 type of steam that is contained in each audio-visual elementary stream in an PES. Each PES is then sliced into MPEG-2 Transport Stream. As noted Transport Stream Packets that are 188 previously, MPEG has amended the bytes that include a 4-byte header as MPEG-2 system standard ISO/IEC shown in Figure 5. The MPEG-2 system 13881-1 [1] to allow carriage of MPEG-4 specification defines each TS Packet to be content over MPEG-2 Transport and identified by a field in the header known Program Streams. This amendment is as Packet Identifier or PID that is 13 bits. included in the published 2000 edition of Thus, the payload of each TS packet could ISO/IEC 13818-1 Ref. [1]. According to contain up to 184 bytes since there are this amendment, for the carriage of
individual MPEG-4 elementary streams, New descriptors such as MPEG- only system tools from MPEG-2 [1] are 4_video_descriptor() and MPEG- used. This topology is depicted in. As 4_audio_descriptor() are defined by the shown in this Figure, MPEG-2 Link Layer 2000 edition of ISO/IEC 13818-1 for and Sync Layer are used instead of defining coding parameters of associated MPEG-4 Link and Sync layers. Hence, elementary stream. It is worth mentioning elementary streams encoded according to that these descriptors do not apply to the MPEG-4 are carried in PES packets as MPEG-4 elementary stream if MPEG-4 PES_packet_data_types with no specific Link and Sync layers are used. These alignment. In another words, from a descriptors are carried in the second loop system point of view, encoded MPEG-4 of PMT and flag to the decoder which audio-visual elementary streams are level and profile of MPEG-4 compression treated the same as MPEG-2 elementary was used to compress associated streams. For example, elementary stream Elementary Streams. These descriptors synchronization is accomplished can be found in the 2000 edition of according to MPEG-2 through decoding ISO/IEC 13818-1, section 2.6.36 and the PCR in the adaptation layer and the 2.6.38. same time base is used to synchronize all the components of a service. This is CARRIAGE OF MPEG-4 SCENE contrary to MPEG-4 in which each VIA THE MPEG-2 TRANSPORT component of a program could be STREAM synchronized to a different time base through the OCR. This section discusses the In addition, Stream_Id values for video encapsulation of MPEG-4 content which and audio elementary stream within the may consist of but not limited to: audio- PES header have been defined by this visual, IPMP, OCI streams, Object amendment to indicate that PES payload Descriptor (OD), Scene Descriptor such contains MPEG-4 audio-visual elementary as BIFS in the MPEG-2 Transport Stream. streams. As noted in the previous section, As specified by the 2000 edition of Stream_id is encoded in the PES header to ISO/IEC 13818-1, these streams are indicate to the decoder what compression carried in the SL_Packetized stream but method is used. The new updated values use of FlexMux is optional since MPEG-2 for Stream_id could be found in table 2-18 offers multiplexing tools. MPEG has of the 2000 edition of ISO/IEC 13818-1 amended the MPEG-2 system standard Ref [1]. Furthermore, the 2000 edition of ISO/IEC 13881-1 to allow encapsulation ISO/IEC 13818-1 amendment defines of MPEG-4 both SL_Packetized stream stream types that should be encoded in the and FlexMux streams in PES packets as PMT when MPEG-4 compression is used well as specifying additional descriptors to encode MPEG-4 audio, and video and other relative fields such as elementary streams. As noted in the stream_type and stream_id to aid the previous section, stream type is used to client side in distinguishing between associate compression used for each MPEG-4 content and MPEG-2. Hence, component of a service to the PID that is one could use other functionality of pointed to by PMT table. This information MPEG-4 in the existing CATV beyond is carried in the second loop of PMT. the improved compression that is provided by MPEG-4. This topology could be
represented by Figure 4 and in more detail following discussion focuses only on in Figure 6. In the balance of this section MPEG-2 TransMux. the features provided by the latest MPEG The inputs to MPEG-2 TransMux are amendment for carriage of either SL_Packetized or FlexMux streams. SL_Packetized and FlexMux in MPEG-2 The responsibility of this layer is to map Transport Streams are presented. First, a these streams into the PES and TS brief summary of Sync Layer and structure defined by MPEG-2 with the FlexMux features and functionality following constraint: SL_Packetized offered by MPEG-4 are provided as they streams are mapped into single PES relate to the discussion. Stream such that only one SL_packet As depicted in Figure 6, the Sync constitutes the payload of one PES packet Layer is located between the compression but in the case of TransMux, an integer layer and the delivery layer. The Sync number of FlexMux packets can be Layer provides a flexible set of tools that mapped into the payload of PES packets. allow incorporating time base MPEG-2 defines different a stream_id for information, fragmentation of access PES packets with the payload of units, and continuity information into data SL_packet versus PES packets carrying packets. The resulting packetization TransMux streams. The corresponding stream from the Sync Layer is referred to stream_id should be encoded in the PES as the SL_packet stream. packet header. The layer below the Sync layer is Furthermore, if SL_packets contain called the delivery layer that includes the MPEG-4 Object Clock Reference (OCR) FlexMux. The input to the FlexMux is the then PES packets carrying these SL_packetized stream from the SL as SL_packets should have PTS in the PES shown in Figure 6. FlexMux is an header and a similar requirement applies efficient and simple multiplexing tool to PES packets carrying FlexMux. In case defined by MPEG-4 designed for low of FlexMux encoded FCR, the timebase in delay and low bit-rate streams. FlexMux the FlexMux packets should be carried in was designed with low overhead since a the corresponding PES header, if present. presentation could have a large number of Also, certain MPEG-4 streams such as elementary streams. Object descriptor and Scene Descriptor Details regarding the Sync layer and tables can be carried in the MPEG-2 FlexMux are beyond the scope of this section format. These data are normally paper and can be found in ISO/IEC static and are used for random access 14496-1 system Ref[1]. similar to the way in which PSI tables are Figure 6 illustrates the data flow from used in the MPEG-2 system. the compression layer to the Sync layer Newly defined stream_types are and then to the Delivery Layer. As established for the PID streams carrying depicted in this Figure, the Delivery Layer MPEG-4 content. These stream_types can be sub-divided into two sub-layers through PMT indicate to the client that the namely, DMIF that is specified by MPEG- bit stream identified by PID in the PMT is 4 and the TransMux layer that is not a PES packet containing either specified by MPEG-4 since different SL_packets or multiple FlexMux packets, delivery layers already specify this layer. or that the bit stream contains MPEG-2 These layers may be any of: RTP, ATM section data carrying OD or BIFS or MPEG-2 just to name few. The commands. Further, a set of descriptors for carriage of MPEG-4 [Ref] in MPEG-2
[Ref] has been defined. These descriptors existing plant bandwidth by employing provide more information about the advanced compression techniques such stream and are included in the PMT. A list those offered by MPEG-4. Typically, of these descriptors can be found in MPEG-4 can provide two to three times section 2.6 of MPEG-2 [Ref]. better compression than MPEG-2 thereby lowering the effective bit rate without WHY MPEG-4 compromising picture quality. Thus a system migration to an MPEG-4 system In the early days of cable television all can accommodate a three-fold bandwidth systems focused on delivering clear increase without a costly fiber overbuilds. analog broadcast channels to viewers in Given the current investments in MPEG- fringe areas beyond the reach of broadcast 2, it is not reasonable to retire all this transmitters. As subscriber demand for relatively new equipment from service. more selection increased, additional But through a gradual transition to channels were added to the analog line-up. MPEG-4, as discussed above, it is MSOs soon discovered new sources of possible to carry MPEG-4 over an MPEG- revenue from reservation PPV, IPPV and 2 transport stream and use current or next premium subscription channels, adding generation set top boxes that could even more channels for these services. It support this dual mode during a transition soon became evident that traditional period similar to the transition that took analog CATV system architecture could place to migrate from analog to digital not support the growing service demand. systems. MSOs were running out of bandwidth and had to upgrade their systems to modern CONCLUSION Hybrid-Fiber Coax (HFC) and migrate services to MPEG-2 compressed, digital MPEG-4 is a promising, emerging video transport in order to carry more technology that has been gaining channels and provide vastly improved momentum in the CATV industry. This picture quality. At the same time ever- paper has presented different mechanisms increasing competition from the Direct to to carry MPEG-4 over MPEG-2 Transport Home [DTH] satellite networks forced Streams in current CATV systems that use MSOs to seek service differentiation. MPEG-2 Transport Streams as the MSO’s began and continue today to offer transport layer. The first scheme enhanced features such as HD, VOD, addressed the carriage of MPEG-4 Network PVR, VOIP, Streaming Media Elementary Streams over MPEG-2 and Targeted advertising just to name a Transport Streams. This scheme enables a few Ref [9]. These advanced features CATV system to take advantage of the pose new bandwidth on both the upstream improved compression offered by MPEG- and downstream HFC plant segments as 4. The second part of this paper discussed explained in Ref [9]. Thus, once again, the carriage of MPEG-4 content over bandwidth resources in a typical HFC MPEG-2. This scheme enables an MSO to network are becoming scarce. In order to offer the advanced features enabled by support these advanced features an MSO MPEG-4 in addition to the improved can choose to increase the physical compression. capacity of the plant through a costly Other advanced features such as HD, conversion to an all Fiber network such as VOIP, and various Streaming FTTH. Or the MSO can optimize the Applications can be supported by
migrating to MPEG-4 and freeing the utilization, as well as a much-needed additional bandwidth needed by these platform for the launch of diverse, applications. It is clear that MPEG-4 is evolving interactive services and overall proving to be an industry standard feature enhancements. solution for increased efficiency in plant .
Media Aware Delivery Unaware ISO/IEC 14496-2 Visual Compression Layer ISO/IEC 14496- 3 Ausio ESI - Elementary Stream Interface Media Unware Delivery Unaware Sync Layer ISO/IEC 14496- 1 System
DAI - DMFI Application Interface
Media Unware Delivery Aware Delivery Layer ISO/IEC 14496-6 DMIF
Figure 1 ISO/IEC 14496 SystemArchitecture Comp. Layer Comp.
Elementary Stream s Sync Layer Delivery Layer Delivery Layer Sync
SL SL
SL_Packetized Stream s DMIF Layer DMIF
FlexMux
FlexMux Stream s (Not Specified by MPEG-4) by Specified (Not TransMux Layer TransMux
(R T P ) (PES) UDP ATM MPEG-2 .... IP TS
Figure 2 Detailed ISO/IEC 14496 System Architecture
Media Aware Delivery Unaware ISO/IEC 14496-2 Visual Compression Layer ISO/IEC 14496- 3 Ausio ESI - Elementary Stream Interface
Sync Layer Media Aware Delivery Aware ISO/IEC 13818- 1 System
Delivery Layer
Figure 3 Carriage of MPEG-4 Elementary Stream Via MPEG-2 TS
Media Aware Delivery Unaware ISO/IEC 14496-2 Visual Compression Layer ISO/IEC 14496- 3 Ausio ESI - Elementary Stream Interface Media Unware Delivery Unaware Sync Layer ISO/IEC 14496- 1 System
DAI - DMFI Application Interface Media Aware Delivery Aware Delivery Layer ISO/IEC 13818- 1 System
Figure 4 Carriage of MPEG-4 Content Via MPEG-2 TS
Elementary Stream (ES)
PES Pkt #1 PES Pkt #2 PES Pkt #3 . . . PES Pkt #n-1 PES Pkt #n
TS Pkt #1 TS Pkt #2 TS Pkt #3 . . . TS Pkt #m-1 PES Pkt #m
TS Header TS PayLoad (184 Bytes) (4 Bytes)
Figure 5 MPEG-2 Link Layer Hierarchy
Comp. Layer
Elem entary Streams Sync Layer Delivery Layer Delivery Layer Sync
SL SL
SL_Packetized Stream s DMIF Layer DMIF
FlexM ux
FlexM ux Stream s (Not Specified by MPEG-4) TransMux Layer
(P E S ) MPEG-2 TS
Figure 6 ISO/IEC 14496 System Architecture
REFRENCES [6] ISO/IEC [1] ISO/IEC 13818-1, JTC1/SC29/WG11 N4668, March Information technology – Generic 2002, Overview of the MPEG-4 coding of moving pictures and Standard associated audio information: Systems [7] C. Herpel, A. Eleftheriadis, MPEG-4 Systems: [2] ISO/IEC 14496-1, Elemnentary Management Information technology – Coding of audio-visual objects Part 1: System [8] G. Franceschini, The Delivery Layer in MPEG-4 [3] ISO/IEC 14496-6, Information technology – Coding of [9] Use of MPEG-4 in audio-visual objects Part 6: Delivery Broadband Streaming Applications, Multimedia Integration Framework Mukta L. Kar, Ph.D. and Nasser (DMIF) Syed, Ph.D. ,Sam Narasimhan, Ph.D. and Ajay Luthra, Ph.D.; NCTA [3] ISO/IEC 14496-1, Paper, Year Information technology – Coding of audio-visual objects Part 2: Visual Author contact: [5] ISO/IEC 14496-1, Information technology – Coding of Ardie Bahraini audio-visual objects Part 3: Audio Sr. Director, System Engineering Motorola Broadband Communications Sector (678) 957-3233 [email protected] CONTROLLING AN INFINITE NUMBER OF CHANNELS
Doug Makofka Motorola, Inc. Broadband Communications Sector
Abstract from the customer. This is in contrast to the broadcast service, which in most cases can be A consumer sizes a typical cable system represented by a fixed channel map, and does based upon the number of channels that can not require any processing from the network be accessed. Cable systems that are capable in order for the tune operation to occur. A of delivering everything ‘on demand’ must be simple ‘channel up’ button push suffices. sized differently. The size of an on-demand OnDemand systems require content caches of system is related to the amount of media some type. This is in contrast to broadcast- (movies, switched broadcast, live broadcast, oriented systems which simply act as a etc.) in the on-line library. Systems capable conduit between a service originator and the of delivering any content ‘on demand’ client devices. present the experience of having an infinite number of channels. On-demand services and the content/media associated with on-demand Channel-oriented delivery systems and services can be managed separately from the media-oriented delivery systems have delivery of the service/content. For this significant differences. Media-oriented reason, OnDemand system management delivery systems need to offer generic, high- cleaves nicely into two pieces: Service- level delivery and transport functionality to Application-Content support, and Media the service and application control Delivery support. Figure 1 shows the high- subsystems that manage the media. This is in level structure of the OnDemand System, contrast to channel-based delivery systems which reflects this division. This paper will that only need to assign a service to a give an overview of each subsystem, then channel. focus on the Delivery Network, and how it provides resources to deliver ‘infinite This paper presents an architectural and channels’. functional introduction to media-oriented delivery systems, including the ramifications Service/ Delivery Network Service/ Application/ to access control, bandwidth management, Applications Content Src Delivery Node (s) Subsystem Groups network management, and media Delivery Content Billing CSI Network Subsystem transformation subsystems. Resouce Manager
Network Management Service/ Overall Application/ OnDemand Network Content Configuration Manager Configuration Module ON-DEMAND SYSTEMS STRUCTURE Module Systems that are capable of supporting a large number of on-demand services are structured differently then systems built Figure 1 High-Level OnDemand System primarily to deliver broadcast services. On- demand services are transactional by nature. Some group of equipment in the network must actively process the request for content Service/Application/Content (SAC) 2. Provide an open, high-level interface Subsystem for requesting DN resources; The Service/Application/Content (SAC) 3. Manage and control devices from Subsystem has the following primary different vendors so that they functions: cooperate seamlessly in the delivery 1. Determining how content is presented of on-demand content; to the customer – host the service, 4. Supports the automation of access control; provisioning, configuration, and 2. Manage the content – including management of the DN. ‘pitch-catch functions, and content distribution; In current VOD environments, the VOD 3. Primary billing interface; systems are either the actual or de facto 4. Supports the automated provisioning managers of the DN subsystem. This as been and management of OnDemand driven by the non-routable networking (ASI) services. between the VOD servers and the DN equipment (modulators, upconverters). With The SAC subsystem is where the majority the advent of GIGE transport of of the added value of the current VOD content/media, the delivery of content is not systems resides. The Pegasus Interactive necessarily determined by the output of the Services Architecture (ISA) standard is one server. This is a strong motivation for definition of a SAC subsystem. removing the DN network management from the content environment. Access control is completely in the context of the SAC subsystem. This Nework Management (NM) Subsystem subsystem decides whether or not The Network Management (NM) content/media will be sent to a client. The subsystem, as it relates to On-Demand Delivery Network assumes that access subsystems has the following functions: control has already been applied. If it receives 1. Maintain a consistent view of system a request to deliver content it will. Note that topology; encryption in the OnDemand system is client- 2. Maintain IP-level device based, not content-based. This is in contrast configuration (DCHP device records, to broadcast systems where encryption is for example); applied to content, and access rights are given 3. Maintain higher-level functional to clients. configuration for the Delivery Network; Delivery Network (DN) Subsystem 4. Maintain and distribute higher-level The Delivery Network is responsible for resource to content source mappings the delivery of content to the consumer/client between the Service-Application- device as directed by the SAC subsystem. Content subsystem and the Delivery The primary functions of the DN subsystem Network subsystem. are: 1. Effectively manage DN resources The Network Management subsystem can (bandwidth, encryption, transcoding, ‘see’ across the two functional subsystems. It insertion, QoS, etc); coordinates system provisioning and management activities. Even though the NM subsystem does not play an active role in the Server NG 1 F1 F2 1 delivery of OnDemand services, it is needed Device 1 S1 SG1 to make large-scale OnDemand systems Server 2 NG F1 F2 practical. Systems capable of delivering 2 Device 2 Device 3 Server ‘everything on-demand’ will be quite large, S2 3 NG F1 F3 3 SG2 and require automatic provisioning and Device 4 Server management – to the extent that equipment 4 can be placed in a rack, automatically determine its configuration and functional F1 F3 Device 5 Device 6 responsibilities when it powers up, without
NG SG3 F2 direct operator intervention. This will be DNRM DM 4 DM Device 7 DNRM DM discussed further in the Automatic DNRM
F1 F2 Configuration and Provisioning section. Delivery Network Manager Device 8 DELIVERY NETWORK (DN) STRUCTURE Figure 2 Delivery Network Structure The OnDemand system structure is based around the User-to-Network concept used in the DSM-CC User-to-Network Session Figure 2 gives a high-level look a the protocol. It is natural to use this protocol as structure of the DN. The arrows in the the basis of requesting and delivering diagram show media paths. Control paths are resources from the DN; however, other omitted. The definitions of the elements in protocols can also be accommodated – such Figure 2 are as follows: as RTSP. Server1…n These are content/media servers. They are outside the DN, but are The Delivery Network (DN) offers shown here as the connection points to the ‘resources’ to the Service-Application- DN. The connections are likely GIGE, but Content (SAC) subsystem. The SAC ASI and other connections are also subsystem requests these resources from the accommodated; DN as part of a session setup operation. The S1,S2 These are Sources, or SRC, which session is used to carry content/media are the actual connection points into the DN. associated with a service to a specific In the case of GIGE media transports, a customer’s client device. Examples of DN Source is equivalent to an IP subnet. In the resources include bandwidth, multiplexing, case of a ubiquitously switched system, there encryption, transcoding, and rate shaping. A is only one Source; major goal of the DN subsystem is to provide Device 1...n These are the actual devices these resources is a well-defined manner so or products that are going to perform that the SAC subsystem does not need to functions in the DN; understand how these resources are mapped F1…n These are the Functions performed to specific devices in the network, or how to by the Devices (mux, modulate, encrypt, control those devices. This provides a basic etc.). Each Function is described by a decoupling of the DN and SAC subsystems’ Function Block, or F-Block for short. F- architecture. Blocks map directly to the resources offered by the DN to the SAC subsystem; NG1…n These are Node Groups – that translates the standard modulator, and defined by the structure of the combining upconverter F-block commands into device- networks on the output of the Inband RF specific commands. The following is a modulators. A Node Group represents the partial list of potential F-blocks: set of ‘outputs’ that can be seen by a 1. RF Modulator (QAM64, QAM256, particular client device. It also is a primary etc.); topological grouping of client devices; 2. Upconverter; SG1…n These are Service Groups. A 3. Encryptor; Service Group is a collection of Node 4. Inserter; Groups. The motivating idea behind Service 5. Multiplexer; Groups is that they collect Node Groups that 6. JitterBuffer/QoS shaper; have the same Resources (F-Blocks). This 7. Transcoder; helps to decouple NG topology from the SAC subsystem; Delivery Network (DN) Transactions F-Block Chain Each path from a SRC The DN subsystem provides resources to a NG is called an F-Block Chain. For used by the SAC subsystem to deliver example, there is an F-Block chain from services. There are many potential on- SRC S1 through Device1, to Node Group demand service types. Video On Demand NG1. Resources F1 and F2 are available on (VOD) and its variants, Network Personal this chain; Video Recording (NPVR), Switched DNRM The Delivery Network Broadcast, and others. The DN is unaware Resource Manager (DNRM) is the part of of these service types; however, all of these the Delivery Network Manager that handles service types are accommodated by the DN the initial requests for resources from the using the same set of resources. Service\Application\Content (SAC) subsystem, allocates the resources to specific A User-Network model is used to F-Blocks, then commands the Device structure the DN transactions. The SAC Manager (DM) associated with the F-Blocks subsystem, and the customer’s client devices to do work; are each Users. The DN is the Network. The DM The Device Manager (DM) translates primary transaction structure is one of the standard F-Block behavior requests into Users requesting sessions for media to be device-specific commands. delivered by the DN. These sessions have resources associated with them. Bandwidth is Every device or product used to process the fundamental resource, but there are others media in the DN must be described as one or – those defined by the F-Blocks supported by more F-Blocks. For example, referring to devices in the DN. The typical session setup Figure 2, if F1 is a modulator, and F2 is an trans action is shown in Figure 3. upconverter, then Device1 is a product that takes its input, modulates it, upconverts it, then outputs the processed RF signal. From the standpoint of the Domain Network Resource Manager (DNRM), it doesn’t matter what Device1 is. Either the device supports standard F-Block commands directly, or there is a Device Manager (DM)
Content DNRM Client on the F-Block allocation scheme. The App resource request is then translated to F-Block PurchaseMovie parameters which are passed to the Device OK-AskForSession(srvid) Manager (DM) associated with the device SessionRequest(srvId,clientID,SrvGrp, NodeGrp) that will provide the requested F-Block SessionReqReceived(srvId) function. SessionResourceReq(srvId, ResourceLst) SessionResourceResponse(srvd, sessionId, DestinationAddress)
SessionResourceCommit(srvID) Device F-Block Chains SessionResponse(serviceId, sessionId, tuneParams) Commands ------F-Block Devices Commands DM SAC Requests for Network Level DN Functions DNRM Status
Network Level Network Level Config/Provisioning DN F-Block Chain Config/Provisioning Network Level Config Topology Status
Figure 3 Session Setup Transaction NM
Notice that the session setup transaction is compatible with the DSM-CC U-N Session Figure 4 Management Control Flows Setup Protocol. RTSP could also be accommodated by adding a few new The Device Manager (DM) accepts the F- functions. Another protocol variation is Block function request and issues the proper having the Content/App subsystem issue the commands to the device. The DM may also SessionRequest command, or by just issuing request that the DNRM adjust the ‘available a SessionResourceRequest indicating a new resource pool’ for the F-Block in questions. session is to be established. The DNRM This allows device resource allocation to could handle both scenarios from either deviate from the standard F-Block allocation protocol without any problems. scheme.
Managing the Delivery Network Automatic Configuration and Provisioning Figure 4 shows the standard control flows Delivering on-demand services to large in support of the DN. Notice that there are subscriber populations requires networks that resource/session flows, and network must support a large number of simultaneous management flows. Each are needed to ‘channels’. These systems can be orders of support the OnDemand system. There are two magnitude bigger than broadcast-oriented levels of management in the DN: system. In addition, there is are on-going 1. F-Block or resource management control transactions between subsystems. The 2. Device management. control transactions demand that each subsystem is ‘in sync’ with at least some Requests for resources are always at the F- portion of the overall topology and state of block level. DN resources are allocated on the OnDemand system. DN topology the basis of F-Block allocation. Each F-Block representation is an important part of type as an associated allocation scheme that automatic OnDemand system configuration. allows the Delivery Network Resource Manager (DNRM) to know where unallocated resources reside in the DN. The DNRM handles the resource allocation based OnDemand System Topology NG F1 F2 1 The basis of DN topology is Network Device 1 S1 Topology. An example of Network Topology SG1 NG F1 F2 is given in Figure 5. 2 Device 2 Device 3 S2 NG F1 F3 3 SG2 Device 4
SRC1 Chain 2 NG1 SRC2 Chain 1 F1 F3
SRC3 Chain 3 Device 5 Device 6
SRC4 Chain 4 NG2 NG SG3 F2 4 SRC5 Chain 5 NG3 Device 7
F1 F2
Device 8 Figure 5 Network Topology Figure 6 Chain Topology Network Topology is the framework networking that maps the DN sources into The Service/Application/Content (SAC) Chains, that reach Node Groups. In a GIGE- subsystem may need some DN configuration based system, Network Topology is a subset information in order to efficiently distribute of the GIGE IP network configuration. This content to servers. Source Topology meets shows the close connection between the DN this need. Figure 7 shows the picture of the configuration and provisioning and the DN conveyed by Source Topology. underlying network configuration – in this case IP. Server 1 F1 F2 S1 SG1 Server Once Network Topology is defined, 2 devices must be placed in the Chains between
Server the Sources and Node Groups. This S2 3 representation of topology is called Chain F1 F3 SG2 Server Topology. An example of Chain Topology is 4 given in Figure 6. Chain Topology is the configuration information used by the F1 F3 Delivery Network Resource Manager
SG3 (DNRM) to turn requests for resources into F2 F-Block allocations.
F1 F2
Figure 7 Source Topology
The SAC subsystem can use the DN without Source Topology. The content/media used by the services supported by the SAC will require specific resources of the DN. The SAC could just request those resources DNCM and the other OnDemand System without any notion of Source Topology; entities. however, some optimizations in the SAC environment are possible with a knowledge Among the DNCM’s major operations are of Source Topology. One example is that the following: content/media requiring a resource not 1. Load/Modify/Delete Network Layer available in a specific Service Group could Topology; be hidden from customers in that service 2. Load/Modify/Delete F-Block Defs; group. This would prevent requests for 3. Load/Modify/Delete Device Defs content it is impossible for the DN to deliver. (each device participating in the DN will supply a device definition Delivery Network Configuration Module package); The Delivery Network relies on topology, 4. Load/Modify/Delete NG Defs; F-Block, and Device information. This 5. Configure Chain Topology; information must also be used to drive or 6. Add /Remove Device to/from DN; modify IP-level configuration information, 7. Enable/Disable Device in DN; such as DHCP records. This information 8. Associate Network Layer With cannot be hand-crafted. It must be formed Device Layer (effect any coordination from higher-level configuration operations. between the subsystem managing The Delivery Network Configuration Module DHCP); (DNCM) automatically generates the 9. Build Configuration Script; topological and configuration information. 10. Execute Configuration Script; Graphic interfaces are used to 'stage’ the 11. Configure DNRM and DMs (these OnDemand system. The graphic interfaces managers need their own allow manipulation of the topological configuration based on system size, diagrams shown in this paper, as well as dedunancy strategy, etc.). detailed F-Block and device information. Ideally, an MSO can have individual systems configured off line by a knowledgeable systems engineer. These SAC Devices configurations can be sent to the locations Src Function Topology where the DN exists. Technicians at the DN (subset of chain DM topology) F-Block Config sites can then ‘rack and stack’ the devices DNCM needed in the DN. Using the DNCM, the Chain Topology NG Topology DNRM technicians at the DN site can apply the (subset of chain Net Config topology) Modifications Net Config configuration to the newly-installed devices NM DN without any additional configuration operations. Ongoing configuration and device changes can be handled by the local Figure 8 Delivery Network Configuration technicians at a high level. The DNCM will Module (DNCM) Environment be able to d sanity checking on these changes. The DNCM can then make sure the changes are applied in a controlled, consistent fashion Figure 8 shows the environment in across the entire OnDemand System. which the DNCM functions. It also shows the major information flows from between the CONCLUSION bandwidth. All Delivery Network functions OnDemand systems are different the must be available to the users of the Delivery broadcast-oriented systems, hence, require a Network via high-level functions that are different structure and management strategy. well-defined, open, and allow competition Yet, it is possible to manage and control large among device vendors supplying products OnDemand systems. Managing Delivery that provide Delivery Network Functions. Networks is more than just assigning Creating a Delivery Network management subsystem, that operates independently from the Service/Application/Content subsystems, will make delivering everything on-demand a technical and practical reality.
DELIVERING EVERYTHING EVERYWHERE IN THE HOME: WHOLE HOME NETWORKING
William Garrison Thomas du Breuil Motorola, Inc., Broadband Communications Sector
Abstract and photos, to this network as well as merge it with their data applications. This paper will describe the requirements for So why not just use the existing in-home an in-home network. Specifically, it will data network for this new video application? address the data rate requirements for the Well, the reason is that requirements for an in- various services and their delivery. Detailed home data network are much different from an descriptions of QoS requirements and the in-home entertainment network. An in-home relationship between the Entertainment and entertainment network needs to support Data devices and services will be given. multiple entertainment streams (some at Various wired and wireless networking HDTV rates) with excellent QoS (Quality of technologies currently in the marketplace as Service). This network also must support well as new technologies that might serve this other types of traffic, such as music, Internet, need will be covered Finally, a view of life in data and photo transport. Once this in-home this home of the future will be explored. network is in place, Voice-over-IP, i.e., telephony, and video telephony can be easily added. Cable operators have a unique INTRODUCTION opportunity in these in-home networks because they understand delivering audio and With the growth of broadband data video best. But what is required to deliver services, many consumers have found it useful these services? to install a data network. According to a 2002 Parks Associates report, 7.2 million homes REQUIREMENTS now have a data LAN and this number will Data Rates grow to 21.2 million in 2006. The primary A whole home network should be an application for these LANs is to share the infrastructure built to serve for a long time. broadband data access with multiple PCs in Just as with AC power, you would never want the home, but it is also used for printer sharing to rewire your house just to add a new and file sharing. appliance – even if that appliance did not exist
at the time you wired your house. Therefore Consumers are now buying PVRs and both current and future needs must be quickly realizing the benefits of video access considered when defining this network. via a hard drive. An obvious extension of this will be to access to this video content Relatively few homes currently have a anywhere in the home. Wouldn’t it be nice to HDTV display. Approximately 4.5% of all view a program stored on your PVR television households have a HDTV display downstairs on a TV upstairs? And consumers now. However, 26% are expected to have at will want to add other media, such as music least one HDTV display by 2008. Currently, the average household has 2.7 TV sets and by 2008 some of those homes will have multiple a system that seems ”sluggish”. Again, this is HD displays. Table 1 outlines the typical the typical experience with streaming media services that may be expected in a fully today on the PC where it takes several seconds networked home and their bandwidth to start playing or to resume play after requirements. pausing. And even with a large buffer, video glitches are common today in the streaming
environment. Application Qty Rate Total each Rate IEEE 802.11e is currently being developed Mbps Mbps as a standard QoS mechanism for wireless HDTV stream 1 19.4 19.4 systems and promises to provide a QoS which SDTV stream 3 4.5 13.5 meets entertainment video requirements. IEEE CD Stereo 1 1.5 1.5 802.11p exists for CAT-5 wired LANs, and Audio HPNA 2.0 includes a prioritized QoS, but Multichannel 1 4.5 4.5 these schemes do not support entertainment Audio 5.1 level QoS. HomePlug 2.0 does not support DVD Audio, 6 1 10 10 entertainment level QoS. However, HomePlug channel AV (the next version of HomePlug) does plan IP Data 2 1 2 to support entertainment level QoS. IP Telephony 4 0.032 0.128 Total 51 Data and Entertainment Table 1. Home Network Bandwidth Requirements. Data delivery is focused on accuracy and video is focused on timeliness. Video Quality of Service decoding is purposely designed to conceal errors while data transfers require perfection. Video requires a much higher QoS than Can both of these coexist on the same data. Many networks provide reliable service network? What tradeoffs need to be made by retransmitting a packet until it is between these? successfully received. This is the correct approach to use for delivering data. However, HOME NETWORKING TECHNOLOGIES video has a timeliness factor measured in milliseconds (or less!). If the video data is not Existing delivered by the presentation time, it would be better to skip this packet and move on to the If a home has an existing network, it is next. likely to be either a wired 10/100 Ethernet or a wireless 802.11b Ethernet. Unfortunately, In addition, a MPEG-2 TS (Transport neither of these is suitable for a whole home Stream) has a jitter tolerance measured in entertainment network. While 100 Mbps nanoseconds. A common “solution” to the Ethernet is fast enough, it does not offer QoS. jitter problem is to use a large buffer at the Plus, as a practical matter, few homes have receiver. This is demonstrated by most current Cat 5 cable running to all of the places where PC streaming media players, where 5-10 you would like to network. 802.11b offers seconds of video is buffered prior to playing. neither the QoS nor the data rate required by However, entertainment video is often an entertainment network. So, what else might interactive, so “solving” the jitter problem be used for a whole home entertainment with a large buffer at the receiver will result in network?
Wired Home Media Raw Approx. Wired networking generally offers the Networking Data Effective highest data rates and the lowest device cost. Technology Rate, Streaming A wired network could use dedicated wires Mbps Throughput, (like Cat 5) or reuse existing wires (like phone Mbps line, power line or coax). Unfortunately, none 100 Mbps Cat 5 100 90* of the currently available wired networks offer Ethernet the bandwidth and QoS required by an HPNA 2.0 Phone 10 6* MPEG-2 Transport Stream. HomePlug AV is Line the only proposed wired standard that HomePlug Power 14 6* promises to address this need, but the standard 1.0 Line has yet to be defined and first products will 802.11b 2.4 11 5* not be available until Summer 2004. There are GHz several proposed proprietary solutions for 802.11a 5 GHz 54 20* - 34** networking-over-coax that meet whole home 802.11g 2.4 54 13.5* - 34** networking requirements for Bandwidth and GHz QoS, but none of these are adopted industry Magis 5 GHz 54 40** standards. Air5 Table 2. Existing Home Network Technologies Wireless *ExtremeTech test results Current wireless technology includes IEEE **Theoretical limit 802.11a, 802.11b and 802.11g. If an adequate QoS could be layered above it, IEEE 802.11b Also note these wireless standards do not could theoretically support a standard include any provisions for Quality of Service, definition video service with a stereo audio so in practice, they can not deliver a service. However, it certainly can not be the satisfactory media delivery experience without backbone of a home with the requirements of a large decoder buffer. 802.11e specifically Table 1. addresses QoS through a prioritization scheme and may solve much of the QoS deficiency The data rate for 802.11a and 802.11g is when it is approved. Meanwhile, proprietary adequate for most of the service set shown in solutions, such as Magis Network’s Air5™ Table 1, although they too will not handle the were designed specifically to meet the needs full service set. Why won’t 802.11a or of video and audio distribution reliably. 802.11g handle 51 Mbps when it is advertised as a 54 Mbps standard? Because the effective Wired or Wireless? payload rate is less than the advertised PHY Wireless networking is a must for portable rate. The advertised raw data rate does not devices. Every networked home will have subtract the MAC overhead and other portable devices and so every home will need inefficiencies. Table 2 shows the effective a wireless network. So does a home already data rage for common networking equipped with a wireless network also needs a technologies. wired network?
The likely answer is you will need both. Wireless is required for portable devices, but it may not reach all parts of the home, it may not be able to deliver enough throughput, and So, what does the networked home of is subject to interference. This is acceptable today offer? As shown in Figure 1 in an for a portable device, but not for the backbone Ethernet configuration, the consumer has of a home entertainment system. In addition, these capabilities: portable devices are normally battery • Shared broadband for multiple PCs powered, which limits their processing power • PC printer and file sharing and hence the bandwidth they need from their • Stand-alone PVR connection to the home network. Wired • Digital Television and HDTV devices generally have no such limitations and • VOD and Impulse Pay-Per-View their bandwidth requirements will only grow • Audio sharing – MP3 to home with time. entertainment center, digital audio to PC, etc. EVOLUTION OF THE NETWORKED HOME AN INTEGRATED DATA NETWORKED HOME Where are we today? Today, many homes have RF distributed by coax and data The next step for home networks will be to networked by Ethernet or 802.11 wireless. In add the entertainment devices to the data addition, more and more homes have an network. While this is less than ideal, it will Entertainment Gateway that uses a hard drive add value to both the entertainment and data to store various content that is received, devices at very little cost. usually referred to as a PVR in the current configuration.
Cable
PC TV Cable Cable Modem Settop PC PC TV TV Gateway Ethernet Hub Cable Modem Settop PC Figure 2 – The Integrated Data Home Network
TV Gateway Ethernet Hub What will the integrated data networked Figure 1 – The Current Home Network home of Figure 2 offer? • Low bit rate video (< 1 Mbps) between There is a loose coupling between the RF PC and Gateway (with latency and and data worlds, in that the PC connects to a some glitches) cable modem in order to connect to the • Archival storage – when your PVR Internet. However, for most purposes the two disk is full, use unused capacity on worlds of entertainment and data remain two your PC separate worlds. Their closest linkage might • Remote access – move content in be the DVD disk that can be played in either slower-than-real-time from one PVR the entertainment center’s player or the PC. to another for delayed remote viewing
• Pictures stored on the PC displayed on the TV
A FULLY NETWORKED HOME
Cable In a fully networked home as depicted in PC Figure 3, the network backbone is robust TV enough to support any in-home application. Settop Entertainment, data and voice applications are PC fully supported. Location does not matter. If TV Gateway your favorite program is recorded somewhere Figure 3 – Fully Networked Home in the house, you can watch it anywhere in the house. If your favorite music is on any device in the house, you may listen to it on any Where did the cable modem of Figure 2 device in the house. Format conversions are go? Well, the home is sharing the cable handled seamlessly. modem that was already built into the Gateway. Future devices will have multiple What will the fully networked home offer? network interfaces to make connecting as easy • Quality video to/from the Gateway and as possible for the consumer. PC, including high definition content • Multiple high quality audio streams, Whole Home PVR Scenario including home theater • Watch high definition TV on your PC even if you don’t have a high You have just returned home and need definition TV some entertainment. So you plop down in the nearest chair and pick up the remote. Let’s • Watch high definition TV on a see, what is available? You want Video, standard definition TV via Gateway Recorded Programs, News. Your home format down conversion system knows that you like to get the latest • IP telephony/video telephony news, and always records the most recent
network news show for you. You don’t know And what about your car? Why wouldn’t which device in the home recorded it (my PC? you want to be able to listen to your favorite my settop?) and there really is no reason why MP3s while on the road? Your car could you should care. automatically download the most recently played songs plus any ones you specifically After you make your selection, the news designate every time you return home. starts. Well, after the first headlines, all you
want is the sports. So, you fast-forward to the Note in Figure 3 the number of wires and sports and see how your favorite team did. devices goes down. This is because the best They blew the big play? You quickly go back network is an invisible one. Communication to the menu and access the “Everything on can be via a coax network, wireless, Power Demand” system offered by my MSO, find the Line or any combination of acceptable home game, and Fast Forward to see that play. entertainment networking . Yeah, they really blew it.
What Was Going On Behind The Scenes? However, from the network’s perspective, it does matter. For content from outside the Your home devices have been home, the home network had to negotiate with autonomously recording content, based on a video server in the MSO’s to select the your preferences. Some of your preferences content and setup the session. Playing that were specifically enumerated when you set the content now requires QoS all the way from the system up, others were inferred by monitoring MSO’s headend to your TV. Not a small how you used the system. But when you challenge, because it spans multiple network plopped down, a content manager that was domains. cognizant of every device in the network put it all together for you in one place. Did the video come over a wired or
wireless network? If the person flopped in the After you made your selection, the first chair knows, then we have failed. The thing that happened is your current display network needs to work seamlessly and device negotiated with the device that held the invisibly. content. What is the best format to use? What is the best data rate? What QoS is available. CONCLUSION As an example, presume the news was recorded in HD, but the in-home network is busy and only 5 Mbps is available with the The biggest remaining question is “When QoS that you need. So, the network reserves 5 will all this happen?” The current home Mbps for this session and source device network isolates the entertainment and data down-converts the news to a new data rate networks. However, new products (such as under 5 Mbps. Replay’s and TiVo’s latest generation devices) are starting to link the data and You start watching the news and decide to entertainment worlds. This is a start, and will Fast Forward. The local device sends a likely grow over the next few years. message to the source, which starts the Fast Forward. Because the QoS minimizes the Full whole home entertainment quality amount of buffering required at the display networks are probably 3-4 years off. The device, you see the news speed up within 200 devices required to build such a network will milliseconds. be available to early adopters at boutique prices early in 2004, but mass marketable When you decide to go look at the big play, whole home networks are probably still a few you are leaving you home network. Or are years out. Standards have to be established you? Your home system can record and production volumes must ramp up before everything, so when you want something that price and ease of use meet mass market is not available locally, you can fall back on requirements. And there is a lot of software your MSO to get the content. But the MSO development required to make the network might have known that many of their invisible and user friendly. The average customers were going to look at that game, consumer must be able to take a new device and so “pushed” the content to your home home and plug it in and find that it will simply ahead of time. From your chair, it should not work – like magic. matter.
DOCSIS 1.1 – WHERE GAMING AND QUALITY OF SERVICE (QOS) INTERSECT
Kenneth Gould Time Warner Cable
Abstract [3], where are the opportunities for Multiple Service Operators (MSOs) to provide gaming Broadband on-line gaming is poised to be services that provide added value to their a key usage demand for residential high- customers and subsequently results in new speed data customers. With the recent revenue streams? releases of hugely multiplayer games such as Ultima Online and the availability of network The most obvious possibility is to simply enabled gaming consoles such as the use gaming to attract new high-speed data Microsoft Xbox and Playstation 2, there are customers to cable modem service. Every increasing opportunities for MSOs to cater to Xbox console is manufactured with an (and profit from) the demands of the Ethernet port that is the sole interface for broadband gamer. networked-based games. Xbox Live games are typically written for network play with the assumption that the bandwidth available will be less than 64Kbps upstream and VIDEO GAMING IS NOT A GAMBLE downstream. It is relatively easy to provide a DOCSIS configuration file for a cable Given the public launch of broadband- modem that limits its bandwidth enabled gaming consoles in the last year, consumption to 64Kbps. Likewise, the such as the Microsoft Xbox and Sony physical location of the Xbox relative to the Playstation 2 consoles, considerable interest physical location of a cable modem within in cable modem service has been generated the home is not a real problem given the within the gaming community. availability of wireless Ethernet bridges and wireless-equipped cable modems. This By January 7th, 2003, Microsoft opens up our potential pool of customers announced that more than 250,000 beyond households containing PCs. subscribers had signed up for the Xbox Live service that was launched on November 15th, The question is: Can MSOs offer this 2002 – this is twice as much as initial sales product without cannibalizing its existing projections [1]. With 21.5 million Sony high-speed data customer base? One Playstation 2 consoles shipped to North stumbling block is that while it is easy to America as of January 9th, 2003 [2], one can limit a cable modem service to 64Kbps, it is expect that quite a few owners will opt to far more difficult to limit a service to only purchase network adapters allowing for on- support console gaming. While the IANA line game-play over a cable modem. To a list of well-known port numbers describes lesser extent there is still demand for network both TCP and UDP ports 3074 as being the access from Nintendo GameCube customers “Xbox port”[4], our observations have shown and the customers with the more aged Sega that Xbox Live games use a wide variety of Dreamcast. ports, of which 3074 is merely the most used. This greatly limits an MSO’s ability to create As the console gaming industry is a multi- filters on a cable modem to allow Xbox billion dollar industry within North America traffic yet disable the customer’s ability to attach his PC to a “gaming cable modem” to maintained by the individual publishers. surf the web or run peer-to-peer applications. Regardless of the model for server Likewise, attempts to filter traffic based upon maintenance, the argument can be made that the MAC address of the console are fruitless co-locating the gaming servers within an given the end-user’s ability to change the MSO’s network can be a win-win situation Xbox’s MAC address at will. Similar for the publishers and the MSO. The MSO’s behavior is seen from Sony Playstation 2s. customers experience even lower network On a practical operations note, typically latency which should make the customer and ISPs like to sign up customers with the the publishers happy and the MSO benefits minimum of paperwork. Customers are by keeping more gaming traffic on their usually instructed to accept the ISP’s “Terms network and off of the backbone. and Conditions” electronically on a web page. This proves to be challenging for a Quality of Service Opportunities new gaming-only customer to complete using What value-add could an MSO possibly only a gaming console. bring to a gaming experience for which a
gamer might actually pay? After all, console It is worth pursuing the concept of gaming over a cable modem works well attracting new customers from households today. One differentiator for MSOs is the which either only contains gaming consoles ability to offer quality of service (QoS) or which contain both consoles and PCs but guarantees. While console gaming works have not yet opted for cable modem service, well today in a purely best effort data at a service tier whose bandwidth is less than environment, MSOs will soon be offering the typical residential high-speed data tier. many new services which will constrain how The MSO’s market trials are still in their much bandwidth is available for best effort infancy, and there is not yet enough statistical services. A perfect example is the offering of data to determine whether offering lower- Voice-over-IP (VoIP) in a DOCSIS 1.1- priced gaming tiers will cannibalize higher- enabled network. Assuming that the VoIP priced PC-centric tiers, but anecdotal traffic is being transmitted using the DOCSIS observations have so far indicated that Unsolicited Grant Service (UGS) traffic downgrading very seldom occurs. flows on the same upstream and downstream
channels as the traditional Best Effort Co-Location Opportunities services, then for each phone call being made Game publication is a multi-billion dollar through a CMTS, there is obviously less revenue generator for large game publishers bandwidth available for gaming. such as Electronic Arts [5], and as a result, these publishers spend a great deal of time In a bandwidth constrained environment, and money to ensure that the servers on would gamers pay to possess a guaranteed which the games are hosted are highly amount of bandwidth and guaranteed latency available, scalable to the number of dedicated to console traffic? Probably. One customers playing, and well located within can argue that as new services are rolled out, the network to provide low-latency MSOs will also be rolling out more efficient gameplay. The Xbox gaming servers seem to equipment (higher modulation profiles, provide a consistent “feel” to the gameplay as DOCSIS 2.0, etc) that will offset any the servers for each title are managed by bandwidth constraints created by new Microsoft. The game servers for PS2 games services. The counter-argument is that this is are not maintained by Sony, but are unlikely given customers’ penchant for consuming all bandwidth available to them, a combination of DOCSIS 1.1 and DiffServ and even were it true, gamers may still be or MPLS. willing to pay a small fee just to achieve Dynamic QoS is typically used today in guaranteed low latency for their consoles. PacketCable-based voice over IP (VoIP) Gamers are constantly looking for an edge deployments. In this case bandwidth is over their on-line opponents and are reserved for voice calls “on the fly” between convinced that low latency gives them that the cable modem and the CMTS only when a edge. message is generated that indicates that a customer’s phone has gone off-hook. An QoS – Background extension to the voice-centric PacketCable specifications is a promising possibility for The DOCSIS 1.1 specifications created a future gaming services. The primary foundation upon which products with quality functions defined by the PacketCable VoIP of service requirements such as latency and specifications are QoS authorization and bandwidth can be built. There are essentially admission control, generation and capture of two mechanisms for defining quality of billing information, and security. These are service, “provisioned QoS” (pQoS) or all functions desirable in a QoS-aware “dynamic QoS” (dQoS). The parameters gaming environment. CableLabs has been dictating the pQoS settings are pre-defined in working on an extension of these the DOCSIS configuration file that the cable specifications, known as PacketCable modem receives at the time that it boots. The Multimedia [6], which expands the DOCSIS configuration file would typically residential voice-centric specifications to be a define a classifier that determines which general purpose platform for delivering many packets are affected by the defined quality of IP-based multimedia services that depend service rules. The packets that meet the upon QoS. Note that while the PacketCable classifier’s parameters make up a Multimedia framework is based upon the unidirectional stream of packets known as a VoIP PacketCable specifications, the “service flow”. For example, since the implementation of a VoIP PacketCable majority of Xbox gaming traffic is service is not a pre-requisite for PacketCable transmitted to and from port 3047, a classifier Multimedia-based gaming as gaming has no can be defined which places all UDP or TCP requirements for voice specific items such as packets transmitted to, or received on, port wiretapping, PSTN interconnects, etc. 3047 onto a particular service flow. That service flow has QoS parameters associated PacketCable Multimedia Architecture for with it, such as a scheduling type (e.g. real time polling vs. best effort) and latency Gaming requirements (e.g. sub 150ms). All other The easiest way to describe the traffic could default to a standard best effort PacketCable Multimedia Architecture is to service flow which would have a lower provide a diagram of the architecture and transmission scheduler priority. discuss the functionality and interaction of each of the components as it related to Obviously, the DOCSIS 1.1 specifications providing QoS for gaming applications. only handle reserving and allocating bandwidth within the DOCSIS domain, specifically between cable modems (CMs) and the cable modem termination server (CMTS). End to end QoS can be setup with Record Keeping Server CMTS
CMTS Policy Server
Cable Modem CMTS Policy Server
Gaming Console
Gaming Server
Figure 1 - PacketCable Multimedia CMTS Architecture for Gaming Figure 2 – A single policy server can serve Our assumption is that the gaming console multiple CMTSs. has no concept of its QoS requirements nor of the PacketCable signaling like that available to a VoIP MTA (multimedia Once instructed by the Policy Server of terminal adapter) to signal its desire for QoS the gaming console’s QoS requirements, the reservations. Instead, a gaming console CMTS creates service flows for an individual simply communicates with the gaming server cable modem’s gaming traffic with the as it does today. (e.g. Xbox’s MechAssault appropriate QoS characteristics. As an game causes the Xbox to communicate with option, the PacketCable Multimedia Xbox Live servers to set up a gaming session architecture also takes into account the desire between players). to track the actual usage of the QoS-based service flows for billing purposes. These The CMTS is the gatekeeper (referred to billing records are gathered and maintained as a Policy Enforcement Point or PEP) which on the Record Keeping Server. determines whether the resources are available to reserve bandwidth between the The gaming server and the policy server cable modem and itself. Thus, the gaming are expected to reside within the MSO server must communicate the console’s network and are considered trusted devices. bandwidth needs to the CMTS. It does so The gaming server must also take on the through an intermediary known as the Policy responsibility of authenticating the gaming Server. As there could be many different console and assuring that the consoles are applications all of which are contending for authorized to request gaming services. limited bandwidth resources, the Policy Server determines the relative priority of each request (based upon business rules) to determine which requests for QoS should actually be given to the CMTS. The Policy Server is also referred to as the Policy Decision Point (or “PDP”).
Messaging Protocols You will notice that there are a few things missing which simply fall outside of the Record Keeping Server PacketCable Multimedia domain, namely RADIUS RADIUS end-to-end network QoS setup including Policy Server to Policy Server DOCSIS COPS+ DSx communications. One can imagine that CMTS Policy Server Cable gamers desire low network latency on each Modem network segment over which their gaming Gaming Console COPS+ traffic travels. This can conceptually be Unspecified handled by DiffServ or MPLS – most MSOs Protocol Gaming Server would argue that their network backbones are over-engineered and that the DOCSIS
component is where the bandwidth is the Figure 3 – Messaging Protocols most valuable resource.
Obviously, there is also a messaging flow Most relevant gaming servers have the between the CMTS, Policy Server, Gaming ability to match gamers based upon their Server, and gaming console which indicates historical levels of quality of play (that is to the success or failure of the QoS say, based upon how good the player is at provisioning. The messaging between the performing the game), and also based upon gaming console and the gaming server is the latency of the gamer’s network outside of the PacketCable specifications. connections. Obviously, one goal of the The messaging from the gaming server to the PacketCable Multimedia framework can be Policy Server and from the Policy Server to to lower the latency of an individual gamer’s the CMTS is IETF’s COPS based. Any network connection to the CMTS. The game event messaging sent from the Policy Server servers would need to report the gamer’s or CMTS to the optional Record Keeping potential latency when matching up gamers Server is RADIUS based, and the rather than their pre-service-flow-setup CMTS/cable modem exchanges to establish latency. This could require some additional QoS-based service flows is based on communication between the gaming server DOCSIS DSx messaging. and the policy server and potentially inter- policy server or inter-gaming server This has been a greatly simplified communications. explanation of QoS allocation. Upstream and downstream service flows are handled by the The gaming consoles described above are CMTS in different manners. Upstream referred to in the PacketCable Multimedia transmissions are made on a contentious, Architecture Framework as “legacy” clients shared-access medium, where downstream as these consoles are unaware of the QoS traffic is handled by the CMTS as if it were a capabilities and signaling necessary for the traditional IP router. The specifics of the QoS negotiations within the framework. QoS parameters that are associated with upstream and downstream service flows A second type of client can have some (these parameters are different) and the PacketCable awareness built-in – when a service flow scheduling types can be found in network-based game is started, the client can the VoIP-centric PacketCable 1.0 request QoS. The console can now signal to specifications. [7] the CMTS to add, change or delete bandwidth reservations, but the CMTS will implement a PacketCable Multimedia only accept the reservations if the gaming Architecture that would enable QoS server and policy server have authorized the guarantees for gaming consoles without console’s reservation. This is very similar to modification to those console or console the behavior of a VoIP MTA. This concept games. This architecture would require of building PacketCable awareness into a enhancements to the gaming servers to make console or console game will probably not the server applications capable of interacting receive much enthusiasm for implementation with the Policy Servers. by the game developers unless there is a We have seen that there are a lot of considerable client base that could make use customers playing games on our broadband of it. For that reason, we anticipate that networks, research shows that they are support for legacy clients must be well willing to spend vast quantities of money to implemented first. do so, and they have voted with their wallets to use today’s low-latency, high speed The third type of client is one which is connections - it is up to the MSOs to cement totally PacketCable aware and does not the relationship by providing a service which depend upon a gaming server to setup its is unobtainable from other providers. QoS. Instead the console is capable of transmitting its own bandwidth QoS requests References to the CMTS along with authorization [1]http://news.com.com/2110-1040- credentials. The CMTS passes the request 979556.html - “Xbox Live sign-ups beat onto the Policy Server which authenticates forecasts”, cnet.com and authorizes the consoles request. The [2]http://news.com.com/2100-1040- request message is then sent back to the 980966.html - “PlayStation 2 shipments top CMTS which will then setup the appropriate 50 million”, cnet.com service flows for that console. [3]http://www.idsa.com/IDSABooklet.pdf -
“Essential Facts about the Computer and The details of the PacketCable Video Game Industry”, Interactive Digital Multimedia signaling message structures, Software Association (idsa.com) service flow scheduling types, service flow [4]http://www.iana.org/assignments/port- management, etc are outside of the scope of numbers - “Port Numbers”, Internet Assigned this document, but should be publicly Numbers Authority (IANA.org) available in the CableLab’s PacketCable [5]http://www.business2.com/articles/mag/0, Multimedia Technical Report and 1640,45482,00.html – “Could This Be the Specifications by the time of the publication Next Disney?”, Business 2.0 (business2.com) of this document. [6]“PacketCable Multimedia Architecture
Framework PKT-TR-MM-ARCH-V01”, Summary Cable Television Laboratories Console gamers are able to use cable (cablelabs.com) modem connections today with good results. [7]“PacketCable™ Dynamic Quality-of- As MSOs deploy new services that consume Service Specification PKT-SP-DQOS-I05- more of the limited bandwidth available 021127” Cable Television Laboratories between the cable modems and cable modem (cablelabs.com) termination servers, the gamers’ user experience could become less attractive. One method to enhance the user experience is to DOCSIS™ TOOLS FOR TIERED DATA SERVICES
Doug Jones YAS Broadband Ventures
Abstract - Speed: Usually an instantaneous number measured in kilobits or megabits per Given that tiered data service is a good second. This is how “fast” the CM is economic idea (and a growing volume of allowed to operate on the network. There data support this), how does the operator can be separate speeds for the forward and implement a solution? This paper discusses return paths. the technical tools available in DOCSIS for implementing both “Speed” and “Included - Included Bytes: Usually measured over a Bytes” tiers. period of time such as a month, this is the total amount of traffic through a CM. It is usually measured as an aggregate of both INTRODUCTION forward and return traffic, though separate Business Case tiers are possible for each direction.
Cable data system usage has been studied DOCSIS provides a set of tools to for several years now and a growing body of implement both speed and Included Bytes work is available that indicates tiering curbs tiers; however, the methods can differ extreme consumption behavior. On an between DOCSIS 1.0 and DOCSIS 1.1. untiered network, 80% of the total available Specifically the operator has more choice and bandwidth is consumed by only 12% of the arguably better options available to them subscribers. On a tiered network, 80% of the with DOCSIS 1.1. But there are ways to get bandwidth is consumed by 25% of the it done regardless of the version of DOCSIS subscribers, showing a more even deployed. distribution of consumption. Given that the majority (>70%) of High Speed Data (HSD) Overall System View subscribers consume less than 2 GB (GigaBytes, where 1 GB = 1,000 This paper discusses how tiers can be MegaBytes) of data a month (combined implemented on a cable data system. upstream and downstream), curbing the Collecting DOCSIS usage data is one part of extreme consumption of a few users will free the overall solution needed to implement up bandwidth for more “average” usage tiers. A representation of the overall system subscribers and the revenue they bring in. is shown in Figure 1.
CM Management CMTS That’s about it for business motivation, Tools the data are in and tiering makes economic CM sense. The remainder of this paper discusses CM Rating Billing CMTS technical methods to implement tiering on a Engine System CM DOCSIS network. CM Customer CMTS Tools Types of Tiers CM There are two types of tiers: Figure 1
The items in Figure 1 to the right of the Choosing a speed tier begins with the CMTS are not discussed in detail in this operators service activation system. When a paper although they are important subscriber requests HSD service, the operator considerations for the back-office. generally offers a choice of several speed tiers to choose from. The service activation Usage data can be collected in either the system communicates the speed tier CM or CMTS through methods described in information to the provisioning system where this paper. A Rating Engine processes the the corresponding configuration file is data where business decisions are made to created and assigned to that subscribers’ CM. turn the raw usage information into a line When the CM boots, it is provided that item for the billing system. There are also configuration file with the appropriate speed tools to both allow the operator to manage tier information as illustrated in Figure 2. the system and to allow customers to track Service their usage before the bill shows up at their Activation 1. interact with consumer door. System to make service choices 2. push service information to provisioning system CM Provisioning 3. create configuration file CMTS System Collecting the usage data, while there are CM with speed tier infomation several methods available, is probably the CM 4. when CM boots, it's provided the correct configuration file CMTS most straightforward step of the entire CM process. Processing that data into billing Figure 2 information will be unique for each operator. Speed Tiers: DOCSIS 1.0 SPEED TIERS In DOCSIS 1.0, the maximum speeds are Description not guarantees, rather the system will provide up to that speed if there is capacity available This type of tier defines the maximum on the system. There are several reasons speeds that a user will have over the DOCSIS why the full speed may not be available, and connection. It is possible to define maximum primary among these is having too many speeds on both the upstream and downstream users attempting to access the system at the connection. same time. All networks are shared at some Example speed tiers are a user having point and engineering enough bandwidth for speeds of 128 kbps on the return path and 1.5 peak usage can solve congestion. Mbps on the forward path. The cable operator sets these numbers and it is possible In the DOCSIS 1.0 configuration file, the to assign different speed tiers to different following two parameters are used to create groups of subscribers. speed tiers for the downstream and upstream
paths: The speeds are assigned to the Cable
Modem (CM) through the CM configuration - Maximum Downstream Rate file, which is a list of instructions created by Configuration Setting the cable operator and provided to the CM - Maximum Upstream Rate Configuration every time it boots. There are many Setting parameters in the CM configuration file that the operator uses to define the data “service” These parameters are simply set to the provided to the user, but only a couple of the desired speeds and the system enforces them parameters are needed to create the speed to ensure the CM does not transmit at speeds tier. higher than allowed by their tier.
Speed Tiers: DOCSIS 1.1 The Included Bytes tier is generally a DOCSIS 1.1 supports many Quality of combination of both the upstream and the Service (QoS) parameters, the vast majority downstream usage as shown in Figure 3 of which are not needed to implement speed below. An operator could choose to offer tiers. While DOCSIS 1.1 QoS is complex, it separate tiers for upstream and downstream is as simple as DOCSIS 1.0 to implement usage. speed tiers. Included Bytes Total upstream Bytes In the DOCSIS 1.1 configuration file, the Personal CM Total downstream Bytes Computer following two parameters are used to create CMTS speed tiers for the downstream and upstream HFC network paths: Figure 3 - Downstream Maximum Sustained Traffic Rate The amount of Bytes included in these - Upstream Maximum Sustained Traffic tiers should come from the operators own Rate investigation and business plan. Two GigaBytes is equal to 2,000 MegaBytes and The names of the parameters have is a reasonable amount of data for a changed to reflect that DOCSIS 1.1 offers a subscriber just doing email and web surfing. complete Quality of Service (QoS) package. Users that are heavy into peer-to-peer These two parameters are part of that larger applications or that include large attachments QoS package, however, they function exactly with email or do a lot of file transfer may the same and cause the same effect as the consume more that this. DOCSIS 1.0 parameters. Unlike Speed Tiers that are implemented INCLUDED BYTES TIERS using the CM configuration file through an interaction with the provisioning system, Description Included Bytes tiers are implemented by Included Bytes tiers are sometimes counting the number of Bytes of data that are referred to as consumption tiers. This type of sent and received through a particular CM. tier counts how many Bytes of data are used by the CM over a period of time. An Different methods are available for analogy is to the mobile phone industry that aggregating the Bytes of data through a CM for example offers several “Included depending if the system is DOCSIS 1.0 or Minutes” tiers that include an allowed DOCSIS 1.1. These methods are described number of minute’s usage over one month. in the following sections. Similarly a fairly standard entry-level tier for HSD is including 2 GigaBytes (GB) of usage CM Byte Counters over one month. Data shows that the While this method works with all majority of HSD users consume less than 2 DOCSIS versions, it is the only DOCSIS- GB per month. For usage beyond the tier defined method of gathering consumption amount, the operator business policy information for DOCSIS 1.0 systems. A implemented in the Rating Engine would subsequent section describes enhancements determine the appropriate billing treatment available when using a DOCSIS 1.1 CMTS. for that subscriber. All DOCSIS CMs are required to implement Management Information Base (MIB) objects that can be polled using the While polling CM Byte counters is a Simple Network Management Protocol simple and easy method supported by (SNMP). Several of the required MIB DOCSIS 1.0 to implement Included Bytes objects include counters that track the tiers, using CM counters may not be a highly number of upstream and downstream Bytes reliable method due to the unpredictability of through that CM. CMs being power cycled in the home. It will be hard to guarantee accurate counts, in fact, The operator can use an SNMP the operator can expect to undercount usage workstation, also known as a Network due to the issues listed above. Management Station (NMS), to periodically poll each cable modem to collect Another reason to carefully adjust the downstream and upstream usage information polling interval is the amount of traffic the as shown in Figure 4. Once polled for, the SNMP polling of CMs places on the usage data is passed to the Rating Engine for DOCSIS network. There can be thousands analysis per operator business rules. of CMs attached to a CMTS and polling too often can add appreciable traffic to the cable data network. Depending on the number of CMs on the network and the polling interval, the SNMP polling traffic can comprise up to CM 5% of the bandwidth of the cable data CMTS CM system. This is not a trivial number as this is SNMP Rating CM polling bandwidth that could otherwise be charged Engine station CMTS for. CM CM CMTS Byte Counters CMTS CM DOCSIS 1.1 requires the CMTS to implement MIBs that count upstream and Figure 4 – Using SNMP to Poll CM counters downstream Bytes on a per CM basis. Instead of polling all the CMs, the operator The time interval the NMS uses to poll all can now poll just the CMTS as shown in the CMs on the network is an issue to be Figure 5. Note a DOCSIS 2.0 CMTS is considered for several reasons. As required to have these same counters and this subscribers can power off their CMs, usage method is equally viable there. information may be lost from time to time. When the CMs are powered on, the MIB counters are not required to reset to zero (an CM implementation detail with MIBs, its just DOCSIS 1.1 how they work). The NMS has to poll once CM CMTS SNMP Rating just to get a baseline number from which to CM polling Engine station calculate further Byte usage. DOCSIS 1.1 CM CMTS In order to detect when a CM has been CM rebooted, there is a MIB object that contains DOCSIS 1.1 CM CMTS the date/time of when the CM last rebooted. The operator can use this information to learn Figure 5 – Using SNMP to poll CMTS if the baseline number for this particular CM counters has changed.
Rating CM This method still uses SNMP to poll the Engine CMTS MIB counters at the CMTS, but since a CM consumption information
3rd Party Byte CMTS is not supposed to be power cycled CM metro IP network... counter CMTS that often, the polling frequency can be CM greatly reduced to minimize the amount of CM SNMP traffic needed to collect the data. In CMTS CM fact the Byte counters required in the CMTS were designed to count very high specifically Figure 6 – 3rd Party Byte Counter to allow the operator to poll the CMTS only once a month. As long as the CMTS is not This solution does not depend on the power cycled, the counters will accurately version of DOCSIS deployed. In fact, this count trillions of GigaBytes and it is highly solution works with non-DOCSIS cable data unlikely a subscriber could consume that systems too and so may be a consideration amount of data over a month. Using CMTS for operators that have both DOCSIS and polling, subscribers can power cycle their proprietary data systems in the same metro CMs as often as they want and the CMTS area. will still keep accurate counts of their bandwidth consumption. The 3rd party counting system can be approached in several ways. Some Ethernet A complete rollout of DOCSIS 1.1 is not switch equipment can aggregate traffic from needed to take advantage of this easier , more several CMTSes into a single data stream as reliable, and more accurate method to shown in Figure 7. This aggregation switch aggregate Byte count information. By only also takes on the additional processing task implementing a DOCSIS 1.1 CMTS and of Byte counting. On a periodic basis, leaving the CMs at DOCSIS 1.0, the rest of consumption information is transferred from the network, e.g., the back-office, does not the switch to the rating engine. need to be modified to support DOCSIS 1.1. Said another way, if only the CMTS is intraCMTS traffic may not be counted upgraded to DOCSIS 1.1 (all the CMs are Rating CM 1.0), no changes are needed to the DOCSIS Engine CMTS backoffice for provisioning DOCSIS 1.1 CM consumption information 3rd Party CM Ethernet switch/ metro IP network... CMs. The already deployed DOCSIS 1.0 Byte counter CMTS CMs will operate on the DOCSIS 1.1 CMTS CM with all the expected features available with CM CMTS DOCSIS 1.0. CM
3rd Party Counting System Figure 7
Another option for measuring cable The configuration shown in Figure 7 is modem bandwidth consumption is to place a not capable of counting traffic that “stays at traffic counting device between the CMTS home” on a particular CMTS. That is, and the Metro IP aggregation network. This intraCMTS traffic from CM to CM on a device is capable of counting the traffic into single CMTS will not pass through the Byte and out of an operator’s DOCSIS network as counter as shown in Figure 7. shown in Figure 6.
Another approach entails placing a traffic
monitoring/traffic shaping device in a data path of already aggregated CMTS traffic as A key piece of equipment needed for the shown in Figure 8. overall tiering system is the Rating Engine. DOCSIS only provides a technical means to intraCMTS traffic may not be counted implement tiers, whereas the Rating Engine
Rating is need to turn the raw data into billing CM Engine CMTS information. The Rating Engine is not CM consumption information standardized in DOCSIS, rather, this 3rd Party Byte CM metro IP network... counter CMTS functionality will be specific to each CM operator. CM interCMTS traffic may not be counted
CMTS CM
ACKNOWLEDGEMENTS Figure 8 Jerry Bennington and Terry Shaw, both at As shown in Figure 8, both intraCMTS CableLabs®, contributed to this paper. data traffic and traffic between CMTSes may not be counted with this configuration. AUTHOR CONTACT rd Finally, using a 3 Party counting system, Doug Jones in either configuration, has the potential to YAS Broadband Ventures introduce a single-point of failure in the data 300 Brickstone Square network that could affect more than one Andover, MA 01810 CMTS worth of traffic. A system used for voice: +1 978.749.9999 x226 measurement purposes only, however, may [email protected] not have this characteristic. It depends on the product.
SUMMARY There are two types of data tiers, Speed and Included Bytes. Speed tiers are implemented through the CM configuration file. Included Bytes tiers are implemented by monitoring usage data from any of several sources, though some sources are more reliable than others. Tools exist in DOCSIS to implement both types of tiers. DOCSIS 1.0 and DOCSIS 1.1 support very similar methods to implement speed tiers. However, DOCSIS 1.0 and DOCSIS 1.1 systems provide different methods to implement Included Bytes tiers. The DOCSIS community was more aware of the need for implementing data tiers in DOCSIS 1,1, therefore, that system has a more simple method to collect consumption data from the CMTS, whereas in DOCSIS 1.0 this information has to be collected from the CMs. FAST ETHERNET IN THE LAST MILE
Oleh J. Sniezko Aurora Networks, Inc.
Abstract selection of the technology, high quality service can be ensured at lower incremental Broadband HFC network operators in cost than the cost incurred by the competitors North America are uniquely positioned to for delivering equivalent service quality. serve the increasing needs for telecommunications services among small and In the U.S., HFC network footprints medium size businesses. This paper analyzes already cover approximately 80% of all the market potential for providing bandwidth SMBs. As of 1999, 1 in 5 SMBs already and other telecommunications services to this subscribed to cable TV at their business telecommunications market segment. Near- location, primarily for customer term opportunities for HFC network entertainment.i In most of these cases, HFC operators as alternative bandwidth providers networks are already within the last few and longer-term opportunities as Layer 2 and hundred feet from potential business Layer 3 service providers are also discussed. customers. This translates into lower It describes several technology solutions that incremental fiber construction costs in fiber- can be deployed in the HFC plant to support to-the-business (FTTB) architecture. transport of bandwidth-intensive applications and services. The paper also includes a With fiber to the business, high-speed discussion on the requirements such business connectivity at 100 Mbps or 1 Gbps can be applications will place on an operator’s provided cost-effectively today. This high- network. speed access can be throttled back or aggregated depending on individual customer Finally, the paper describes how a needs. Moreover, fiber to the business offers solution based on Ethernet transport can also sustained full throughput in contrast to such be deployed to support delivery of high- alternative offerings as DSL or cable modem. bandwidth residential services to MDUs/MTUs in high-density urban Ethernet Advantages environments. Several data communications systems have been developed and implemented to serve internal and external INTRODUCTION telecommunications needs of enterprises.
Among them, Ethernet has gained the Broadband HFC Advantage broadest acceptance. The following numbers
clearly support this assessment: Broadband HFC network operators in 1. 80+% of all data packets begin and end North America have a distinct advantage over their lives as Ethernet packets. their competitors in providing 2. There are 250 million Ethernet ports telecommunications services to small and deployed worldwide. medium size businesses. With the right 3. 90% of transported business data Alternatives begins and ends as Ethernet on LANs. The enterprises use almost all technologies Ethernet is well understood by small, available today to interconnect their internal medium and large enterprises. It is organizations located at different locations manageable by small businesses without a and to secure connectivity with Internet. The dedicated IT staff. Moreover, it is supported following dataiii (Table 1 and Figure 1) by the dominant standard and represents provide the distribution of deployment of mature technology. The IEEE 802.3 standard different connectivity technologies for data on Ethernet was released in 1980. It also communication services. proved to be extremely flexible and future Medium Size Extended Medium proof as new developments (Gigabit Ethernet, Businesses Size Businesses 10 GigE) do not obsolete previous Dial Up 83% 84% implementations. T1/T3 41% 61% ISDN 36% 58% Beside the fact that Ethernet technology is Switched 56 kbps 19% 35% Frame Relay 15% 21% characterized by low maintenance cost, it is Satellite 6% 3% also relatively inexpensive to implement due ATM 2% 1% to economies of scale from large installed ADSL 1% 1% base. It has shown excellent Table 1: Type of Connectivity for Data price/performance trend: 10x the performance Communication Services of the preceding generation at 3x the cost.ii
Type of Data Communication Access 90% 80% 70%
d 60% 50% 40% 30% 20% Businesses % of Surveye % of 10% 0% Dial Up T1/T3 ISDN Switched Frame Satellite ATM ADSL 56 kbps Relay Type of Data Connection Medium Size Businesses (100-499 employees) Extended Medium Size Businesses (500-999 employees)
Figure 1: The Usage of Different Connectivity Technologies for Data Communications Services
OPPORTUNITY c. receptive to national/international offerings, Market Definition d. application and service needs: voice and data networking There are many definitions of the small (Intranet, private tandem switches, and medium size businesses. The following etc.), virtual private networking presents one of the accepted definitions with (today over frame relay and ATM some characteristics of their communication networks). needs and preferences: 1. Small size businesses: Market Statistics and Opportunity a. less than 50 employees, b. fastest growing segment, Based on the definitions presented above, c. mostly ignored by local exchange the industry reportsiv show that small and carriers (LECs), medium size businesses (SMBs) represent d. highly receptive to competitive 95% of the entire U.S. business universe. offerings, They amount to 7.6 million entities e. application and service needs: approximately and this number is growing at voice support for up to 24 POTS approximately 2% annually. In 1998 alone, lines, Internet access (from 56 LAN penetration in SMBs grew by 10% to kbps to DSL speeds today) and IP 36% while the percentage of LAN connected management services, PCs grew by 24% to 13.4 million. f. billing preferences: consolidated billing for voice and data (Internet) SMB spending on IT and telecom services services. is higher than $100 billion a year. 2. Medium size businesses: a. up to 100 employees, The costs and other characteristics of the b. multi-campus/branch offices, most alternative technologies for data connectivity often within a single metropolitan today are presented below. area, 1. Low-speed T1 connections: c. considered easiest target by many a. typical installation times 30-45 LECs, days, d. receptive to regional competitive b. installation costs range from offerings, $1,000 to $2,000, e. application and service needs: c. maximum symmetrical bandwidth voice support for digital PBX of 1.5 Mbps, systems (requires fractional and d. average monthly costs range from full T-1 line provisioning), Internet $400 to $1,200; access (high speed DSL today) and 2. Low-speed, DSL connections: IP/WAN management services, a. typical installation times 4-6 f. billing preferences: consolidated weeks, billing for voice and data (Internet) b. installation costs range from $200 services. to $300 plus CPE costs of $100- 3. Large size businesses: $400, a. in excess of 100 employees, c. 10+ DSL flavors results in b. multi-campus/branch offices, complex pricing structure, d. maximum symmetrical bandwidth d. shared bandwidth perceived as a up to 1.5 Mbps in various disadvantage for business increments, applications; e. average monthly costs of $150 to 4. Data carrier connections: $400 for symmetrical DSL service, a. Installation costs range from f. must be close to central office $3,500 to $7,500, (CO); b. Dedicated symmetrical bandwidth 3. Cable modem connections: offering: 100BaseT to 1 Gbps a. typical installation times less than Ethernet, a week, c. Average monthly costs range from b. installation costs range from $0 $1,000 to $4,000, (cost to the customer, non-zero d. Limited availability depending on cost to the service providers) to market. $150, c. average monthly costs range from The data carrier connections can be quite $40 to $80, expensive even in the access network. The following table presents pricing structure for Worldcom Ethernet service offerings.v
Customer Service Interfaces Price/month Metro and WAN Corporate MAN 50 Mbps, 150 Similar to ATM and Private Line links Mbps, 622 Mbps frame relay
Enterprise access 1 Mbps to 500 $1,200 to $200,000 Dedicated Internet to Internet Mbps per circuit LAN to LAN and Enterprise private corporate network 1 Mbps to 100 $630 to $20,000 line/VPN connections Mbps per circuit
Table 2: Worldcom Ethernet Service Profiles
Market Requirements a. Equipment to provide services can be deployed based on the service There are several basic requirements demand (number of customers and specified by most businesses and some bandwidth requirements) specific requirements dependant on the b. Cost scales with the SLA business size and type. Almost the same requirements (for example, requirements are defined by service providers. equipment and route redundancy The basis requirements can be summarized in can be implemented on a as- the following points: needed basis) 1. Service scalability 3. Deployment simplicity a. Bandwidth offering scalable from 4. Future proofing 1 Mbps to 1 Gbps a. Future protocols do not render b. Symmetrical bandwidth preferable equipment providing connectivity 2. Deployment cost scalability: obsolete b. Equipment upgrades are easy to accomplish at limited and demand There are some requirements important to driven locations service providers: 5. Acceptable service reliability and 1. ROI for equipment: in months availability 2. Clear network demarcation points 6. Low maintenance costs 3. Compatible with existing HFC architecture and headend installation Some additional requirements from the and equipment following list may be critical dependant on 4. Easy to install business size and type: a. no additional active devices in the 1. Affordable and competitively priced plant to install and manage 2. Security comparable to or better than b. capable of flexible connection provided by competitive technology topologies (point-to-point, ring, and providers nested span, etc.) 3. Transparency to layer 2 and higher 5. Easy to manage protocols: a. SNMP compliant interfaces a. Allowing service providers b. remote provisioning for new (MSOs) and customers (SMBs) for business customers leveraging the existing LAN/WAN c. upgradeable via software hardware infrastructure downloads. b. Allowing for passing through Layer 2 and higher protocols, TECHNOLGIES AVAILABLE TO overlaying datacom protocols for BROADBAND HFC NETWORK robustness and security (e.g., CoS, OPERATORSvi QoS, IPSec) 4. Accessible to SMBs without a There are several data connectivity dedicated IT staff technologies that have been deployed in the 5. Capable of supporting or transparent to past. This paper will concentrate on these that critical business applications use Ethernet based interfaces. All of these a. Transparent to voice, video and technologies can be divided into two groups: data applications 1. Technologies with network processing b. Supporting: (intelligent) equipment distributed in i. point-to-point LAN transport, the access plant. ii. multipoint LAN interconnect 2. Technologies with network processing and extension, equipment centralized in headends or iii. VLANs, main hubs. iv. VPNs, and v. leased line replacement. Both groups require customer premise 6. Low latency for latency-sensitive equipment. This equipment type and applications like VoIP and video intelligence is mostly defined by medium type streaming and internal LAN requirements. Both groups 7. Support for TDM (T1, E1, DS3) traffic include several possible deployment scenarios and interfaces and topologies. 8. Sustainable, non-degrading with distance performance (unlike various Distributed Processing Equipment favors of DSL). Distributed equipment usually comprises service disruption) if located in the access IP switches/routers deployed in the field plant. between headend (or CO) and the customer. 2. Addition of new switches and traffic Several architectures are being marketed by aggregation points may lead to service vendors: disruptions for some architectures 1. FTTB/H EPON with: presented above. The service disruptions a. Switches and aggregation points in may affect data customers or any-service nodes and optical gateways on customers served by the HFC network. customer premises; or b. Switches at optical nodes and/or Centralized Processing Equipment amplifiers and optical gateways on customer premises. Similarly to the distributed processing 2. Hybrid fiber/copper pair architecture equipment solutions, the existing solutions for with: the centralized processing equipment offer a. Switches and aggregation points at several alternatives: optical nodes, 1. Ethernet over RF b. Switches in taps, and a. DOCSIS, a standard-based c. Cat6 copper pair physical layer solution with: mesh network between switches i. Steeply decreasing equipment (Cat6 to homes). prices 3. Hybrid fiber/coax architecture with: ii. Improving performance a. Switches and aggregation points at (DOCSIS 2.0) optical nodes, iii. Limited total and per channel b. Switches in taps, bandwidth (even with DOCSIS c. Coax drops to homes, and 2.0) d. Coaxial physical layer with RF iv. High maintenance cost of modulation and demodulation and upstream HFC path up-and down- conversion at each b. Ethernet over RF based on switch location and at customer proprietary solutions: premises. i. Without bandwidth conversion or All the above technologies can be ii. With bandwidth up- and down- analyzed against the set of requirements conversion. presented previously. This detail analysis can 2. Transparent Layer 1 pipe: be performed by the readers of this paper. a. Overlay systems with dedicated Here are just few comments: fiber or wavelength 1. The history of IP and other higher layer b. Integrated optical systems protocols shows that the legacy equipment not always support and not always can be The Ethernet over RF systems are upgraded to support the new protocols and comparable in performance. Proprietary has to be replaced. This usually is not a systems may have advantage in delivering problem when the legacy equipment is higher bandwidth, especially when combined located at limited locations or on customer with up- and down-conversion. However, the premises (demand-driven replacement) but proprietary character of these solutions will may pose significant problems (cost, most likely result in high equipment cost. Moreover, the coaxial shared medium is still perceived by many businesses as less reliable than dedicated fiber and copper media. The integrated solutions can be modified to provide a full, dedicated connectivity to The transparent pipe alternative in its larger businesses in an evolutionary and overlay configuration has been available for as scalable manner. This is supported by a long as IP switches and routers exist. Several significant progress in passive component established transport and IP equipment technology (colorless and WDM), especially manufacturers and some start-up companies in their capability to perform under harsh provide equipment for point-to-point, point- outside plant conditions. to-multipoint and ring topologies. The equipment has several standard interfaces APPLICATIONS AND REVENUE ranging from T1/E1 emulated TDM circuits OPPORTUNITY through 1 GigE interfaces. This type of equipment competes with ATM, frame relay Near-Term Opportunities and (recently) SONET solutions. As described above, thanks to the Ethernet Near-term opportunities for HFC network proliferation, it has cost and other advantages operators as alternative bandwidth providers over the competing technologies. This can materialize in the following areas: technology is suitable for larger business and 1. T1 replacement with T1 interfaces for is a simple extension of metro-market both data and voice services: topology into access plant. It does not offer a a. lower cost (cost of T1: $400- significant advantage to HFC network 1200/month with $1-2K installation operators as any operator with a capability of cost for a maximum of 1.544 Mbps installing or with already installed fibers to bandwidth) POPs located in proximity of large businesses b. shorter than 30-45 day waiting period could implement this data communications for T1 installation and activation technology. Moreover, this market (large c. service to businesses) has been successfully addressed i. small/medium/large businesses by ILECs, CLECs and other data carrier ii. wireless backhaul to the PSTN for companies. cell towers iii. virtually any customer of T1 The integrated solutions are usually a services today hybrid approach (at least as long as FTTH for 2. DSL replacement for data communication residential services is not deployed). Usually and Internet access it can be integrated to the node location. a. dedicated, guaranteed bandwidth, no From the node to the business, it is delivered degradation with distance on a dedicated fiber. At least one of the b. higher capacity than DSL at similar or vendors offering this technology has also lower cost (cost of DSL: $150- capability for flexible increase in capacity to 1 400/month for symmetric DSL for 128 GigE and above. The integrated technology kbps—2 Mbps; $60-200/month for has been enabled in the last several years with asymmetric DSL for 192 Kbps—1.5 the introduction of digitized technology in Mbps) upstream HFC links supporting the legacy RF c. shorter waiting period two-way communication. The figure below d. service to small and medium shows an example of the integrated Ethernet businesses solution. 3. VoIP a. integration of voice with existing data 1. SOHOs/SMBs services for small and medium size 2. MDUs businesses 3. Multi-campus businesses b. transport of T1 over Ethernet to 4. Businesses with telecommuters support legacy PBX applications with 5. Institutions: integrated or off-the-shelves interfaces 6. Schools (T1 and E1 emulators) 7. Universities c. secondary voice lines for the 8. R&D facilities residential market (MDUs) 9. Hospitals 10. Military and Government facilities These services can be provided to the 11. Sports facilities following market segments:
Fiber Node with Ethernet Capability Headend Ethernet Interface between Access and network Sections
CPE Devices in Ring Configuration or Point-to- Point Topology
Figure 2: Integrated Ethernet over HFC Network – Example
Long-Term Opportunities 3. L2 and L3 Quality-of-Service (QoS) features As the experience and familiarity with the 4. IP network services networking technology grows, longer-term 5. Alternative high-speed ISP access opportunities can be addressed and capitalized 6. Alternative IP telephony services on. These include: 7. Future services such as IP video 1. VLAN tunneling streaming, etc. 2. Link aggregation Many MSO organizations have already of solution, besides minimizing network expertise and staff to support these services to complexity, also allows freeing up valuable any businesses. forward and return spectrum that can then be allocated to other services. Moreover, in most ETHERNET TO RESIDENCES cases, it eliminates the need for costly coaxial cable rewiring in MDUs. With the The integrated Ethernet approach can be development of lower cost CPEs with limited easily extended to serve residences. This features, acting mostly as media converters, extension may happen selectively based on symmetrical Fast Ethernet (shared among cost analysis. The residential Ethernet FTTH several residences) can be also delivered to connections can take place initially in MDUs SFU residential areas. The figure below where the cost of CPEs and switches presents the evolution from FTTB to FTTH provisioning the service to individual suites through intermediate step of deploying fiber to can be shared among many tenants. This type MDUs.
CPE
CPE
CPE
Figure 3: Evolution to FTTH Ethernet for Residential Areas
This evolution will be fueled by the Moreover, the technologies allow for trendsvii,viii reported in the industry summary competing with ILECs, CLECs and data reports: carrier companies for large business market in 1. 400,000 FTTH subscribers worldwide a scalable and evolutionary manner. implemented by the end of 2002, 2. 50,000 (or 22,500 by others) FTTH FTTB applications can be readily and subscribers in North America cost-effectively extended to the MDUs and 3. 50 FTTH served communities in the MTUs today. Future price trends in the USA optical and digital technology may allow (and 4. Trends: by some reports already allow) for cost- a. 300,000 FTTH subscribers in 2003 effective implementation of FTTH systems. 800,000 FTTH subscribers in 2004 (1,400,000 i AMI-Partners by high projections for 2004) ii Metro Ethernet Forum iii AMI-Partners SUMMARY iv AMI-Partners' 2002 U.S. Small Business The demand for data communications and Market Opportunity Assessment Report v Lightreading, May 2002 other telecommunications services from vi SMBs presents a lucrative opportunity for Information from web-sites and published increased revenue. This increase can be documents of the following vendors: • achieved by leveraging MSOs’ investment in Advent broadband HFC networks. • Aurora • Cisco Multiple choices of providing Ethernet to • Harmonic businesses are available to HFC network • Jedai operators. These technologies and • Narad architectural choices should be evaluated • SwitchPoint against business and market (competition) • Wave7Optics requirements as well as against operator’s • Xtend objectives. vii J. Baumgartner, Fiber-to-the-home blazes an evolutionary path, CED, February 2003 These solutions create an opportunity for viii R. Pease, Fiber-to-home proponents speak HFC network operators to replace CLECs and out against negative deployment myths, ILECs as telecommunication service providers Lightwave, February 2003. to SMBs. This market is dramatically underserved by the traditional telecommunication service providers. FEEDER FIBER INFRASTRUCTURE FOR THE SMALL TO MEDIUM BUSINESS DATA SERVICES
Donald Sorenson Scientific-Atlanta, Inc.
Abstract ring based access technologies which leverage fiber inventories to capitalize on this This paper investigates the business promising business development opportunity. development and engineering advantages of utilizing ring topologies in last mile fiber-to- HFC feeder fiber is the key the-business applications. The use of differentiating strategic element that is traditional star or bus-star oriented network working in the favor of the cable service topologies become less than optimal when provider. For any carrier attempting to the realities associated with business address the small business market space, the services market dynamics and geographic means for effective and efficient backhaul circumstances are considered. It will be has been demonstrated to be one of the key shown how the use of rings in the access barriers to market entry. Accordingly, network can reduce business development MSO’s must optimally leverage the use of risks, is highly synergistic with existing fiber existing dark fiber. Each fiber that passes feeder plant, simplifies engineering and pockets of businesses must be able to support operations tasks over the life of the plant and as many subscribers as possible while can improve the MSO’s service deployment providing meaningful service levels. velocity. All critical success factors for the Solutions must be able support multiple MSO in the profitable deployment of fiber subscribers per feeder fiber (fiber-gain) while based business services. controlling risk, maintaining simplicity and providing low first-in cost.
INTRODUCTION Many of the data oriented access solutions being circulated today are based The access and transport fiber upon classic bus or star local area network infrastructure upgrade investment made by (LAN) topology principles that are MSO’s in support of their residential implemented over either fiber or coax broadband initiatives has positioned them infrastructure. To achieve any degree of well to become the major data services feeder fiber-gain these approaches carriers in the growing 70B$ to 90B$ small generically require the use of either passive to medium business data services market. In or active edge aggregation elements within a the U.S., MSO’s collectively now have more few thousand feet of the subscribers being access fiber passing small to medium served. Such architectures are quite suitable businesses than any other competing for applications where subscriber densities communications entity. This paper assesses and service take rates are high and can be the advantages and potential problems accurately predicted, e.g. urban or residential associated with the use of fiber applications. Unfortunately for the MSO neither of these conditions typically exists
when extending existing spare fiber for West CPE East business building access. Link Link Passive CWDM CWDM Wavelengths There is a better way! This paper will Add-Drop Ring Mux explore the comparative strategic advantages Fiber of utilizing a fiber ring topology for West CWDM Passive accessing business buildings. Coupled with CWDM Host Mux CPE Terminal Add-Drop the emerging low cost ring based networking East CWDM Mux technologies such as Resilient Packet Ringa Mux
(RPR) and Multi-Protocol Label Switching Passive b (MPLS) Fast-Reroute the MSO’s can CWDM Add-Drop minimize deployment risks and while Mux maintaining a high of degree of service flexibility. This paper will show how access CPE fiber rings are highly adaptable to varied circumstances, reduce construction and WDM OAD Ring operations demands and ease the impact on existing HFC feeder fiber. The number of stations which can be placed on an OAD ring is governed by a RING TECHNOLOGY OVERVIEW number of factors. For example, based upon the available wavelengths the maximum Broadly, fiber ring technologies are based number of ring stations is one-half of the upon either optical add-drop (OAD) available wavelength count, e.g. a CWDM multiplexing or electronic add-drop (EAD) based ring can support a maximum of 9 approaches. Each has its own advantages, stations per ring, assuming 2 wavelengths per however it will be shown that EAD station in each direction with a total of 18 approaches based upon emerging Ethernet wavelengths available. The combination of technologies can greatly simplify deployment passive optical multiplexer and ring segment and operations demands while increasing the optical losses, e.g. water-peak attenuation, number of subscribers supported per ring. may also limit the number of ring stations based upon optical budget restrictions. The Each station in a wave division incremental insertion of new stations into a multiplexed (WDM) OAD ring ring must be carefully planned to ensure that communicates to a host headend terminal the optical performances of the new and over two pairs of dedicated optical existing ring stations are not adversely wavelengths using a standard link affected. aggregation protocol such as that found in Ethernetc. Each wavelength pair is routed in The OAD ring approach requires the opposite directions around a diverse or MSO to accurately track to whom collapsed path ring, as illustrated below. wavelengths have been allocated. If successful, hundreds or thousands of subscriber wavelengths, subscriber and headend-host switch ports and network service allocations will have to be accurately correlated and tracked on a regional basis over the life of the network. The problem is further complicated by the fact that network management automation
tools do not typically extend to a manually cost of a comparably featured EAD based provisioned optical layer. These challenges ring station. could become a serious liability threatening the MSO’s ability to consistently provide a From a business development and high degree of service reliability and management point of view these integrity. The approach also presents a circumstances present a potential condition physical port scaling challenge at the host where OAD based solutions add unneeded terminal, demanding two OAD multiplexers equipment and operations expense. Over the and optically interfaced switch ports per life of the plant, headend and drop CWDM subscriber. A key advantage of the OAD optical passives, headend host switch approach is that the links between the host physical ports, headend O/E interfaces along terminal and each CPE unit are dedicated with wavelength and switch port record and private; enabling the simple and keeping complications will greatly encumber effective physical layer security, privacy and the overall cost-performance of the solution. individual station failure immunity Fortunately, all of these costs can be associated with dedicated media minimized or eliminated through the connections while operating in a resilient alternative use of an EAD based approach. shared media environment. All stations in an EAD based ring share a The apparent ease by which the add-drop common set of fibers or wavelengths which function is performed in an OAD solution greatly increases the number of stations a leads one to anticipate a reduction in ring ring can support, for example RPR supports station and headend host switch complexity up to 255 stations per ring. Ring traffic is and expense. In a linear cascade of stations, added to, dropped from or transits through used strictly for the purpose of providing best each station in the ring. As previously effort traffic over a single physical port, mentioned, many of the data network elegance in station design will indeed prevail architectures being derived for service since stations may be composed of simple provider applications, including rings, have physical layer fiber-to-twisted pair media roots in existing enterprise LAN converters. However the headend host technologies. The most common ring based switch must still bare the full brunt of being a LAN technology is Token Ringd (IEEE carrier-class device. An additional 802.5), which is based upon a single ring complication arises from the fact that the typically utilizing twisted pair copper media. headend host terminal(s) must economically scale on a per physical port basis for each port type required (e.g. 10baseT, 100bastT, End Station 1000baseT, T1, T3 and etc.) as each subscriber is added, which can in turn drive End Station End Station up costs, headend space requirements and the need to track physical fiber or twisted pair port connections for each subscriber. End Station End Station Further, station complexity and cost increase dramatically with the inclusion of higher level functions such as support of OAD rings, multiple subscriber physical and End Station End Station logical ports, varied data types, managed services and QoS sensitive traffic; effectively End Station becoming equivalent to the complexity and
Single Ring Network link layer technology and is incapable of performing packet switching functions on its A more robust duel ring approach is used e own. SONET is the defacto standard for in Fiber Distributed Data Interface (FDDI) high reliability transport within the telecom and is implemented over either twisted pair industry, however its capabilities come at a copper or fiber media. The use of a dual ring price point that is sustainable for only the gave FDDI a resiliency advantage not most demanding high revenue subscriber available in the simpler single ring approach. requirements. The reliability of SONET
rings between fixed managed facilities has been extraordinary, virtually eliminating End Station network outages due to fiber or equipment End Station End Station failures.
EAD RING FUNCTIONALITY
End Station End Station It will be useful to briefly review terminology and functionality associated with ring networks. The focus of this section End Station End Station will be on EAD rings while highlighting the key contrasts and similarities to OAD rings. End Station A degree of commonality exists in the Dual Ring Network vocabulary used in the aforementioned ring In either case, Token Ring or FDDI, the standards, and for discussion purposes here use of rings within the enterprise has clearly the terminology adopted in the IEEE 802.17 not enjoyed the widespread success of star or RPR draft standard will suffice and is bus-star technologies such as ATM and illustrated in the diagrams below. Ethernet. This has been principally due to RING higher costs and operational reliability issues RINGLET associated with disruptions in ring continuity INCLUDES ALL INCLUDES BOTH COMMON DIRECTION RINGLETS LINKS when workstations were inadvertently WEST EAST switched off, unplanned moves or RINGLET RINGLET interconnecting cabling failuresf. However FDDI has been successful as a reliable high capacity link between core network elements such as servers and routers as well as RING SEGMENT LINK STATION (SEGMENT) (ONE UNIDIRECTIONAL CONNECTION) between facilities in campus applications. In (INCLUDES TWO these cases the devices and interconnecting OPPOSING LINKS) media are managed and stationary – thus they are far less vulnerable to inadvertent user Ring Terminology manipulation. The active elements on a ring are referred
to as either stations or end-stations based Within the telecommunications domain, upon the nature of the station’s traffic. clearly the most widespread utilization of the Stations that terminate all traffic are ring topology is in the form of Synchronous commonly referred to as end-stations. The Optical Networkg (SONET). Like FDDI, term “station” will be used here since CPE or SONET is based upon the principle of headend host devices rarely terminate all resiliency through the use of multiple rings, traffic. One station on the ring will serve as however by contrast SONET is not a data
the ring’s interface to the MSO’s backbone EAD Ring Protection network by aggregating non-local traffic into Typically fiber failures are accidental, but a few high-speed interfaces. Ring segments in the case where a new station is being exist between stations and are composed of added to an OAD or EAD ring a fiber cut is two opposing links. Most, though not all, intentional and appropriate. Whether access rings will be composed of two intentionally or accidentally cut the ring ringlets. Each ringlet-link is a half-duplex enters its protection mode preventing connection with directional signaling over an protected service failure to existing optical wavelength transported on a fiber or subscribers, for example a DS1 would not with the opposite ringlet-link in a single fiber loose frame-lock. Once a new station is WDM arrangement. Occasionally, due to added to the ring all of the ring’s stations circumstances associated with a lack of fiber, automatically update their ring topology it may be desirable to operate the ring in an information and resume normal operation. It open single-ringlet mode where stations are will be shown that the ability to insert a new simply daisy chained together, thereby station at any arbitrary location within a ring forgoing many of the key carrier-oriented without interruption to protected services is a benefits of using ring-based equipment. crucial benefit to using rings in last mile
access applications. The counter rotating ringlets combined with a station’s ability to reroute traffic in A key attribute of an EAD ring is that the event of a fiber cut facilitate the each ringlet link is constrained to a simple topology’s well known resiliency protection point-to-point optical connection with no capability. As illustrated below, stations can intervening passive devices. Each station implement protection by either wrapping regenerates the link for transmission to the traffic at the stations adjacent to a ring next station allowing the link to operate over element failure or by having all stations steer great distances with standards based optical their traffic that was transiting the failure devices. A signaling constraint is typically point away from the failure. Stations not placed on the length of a ring segment, determine the condition of the ring by either and in turn the ring circumference is only continually circulating topology status limited by the combined lengths of each ring information or by monitoring optical signal segment. In the last mile applications, levels and are thus able to detect the failure station-to-station distances are typically less and redirect traffic in less than 50ms. than 10km, virtually eliminating optical budget and dispersion constraints for both the Wrapping Steering Wrapping initial ring deployment and future bandwidth enhancements. Decoupling the fiber plant from station optical budget constraints helps to both ease deployment concerns and insure that the plant will remain
Fiber Point-To-Point transparent to future upgrades. Cut Optical Link
New A fiber ring can be implemented around Station physically diverse path or within a single Ring Circumference Only Limited by Combined Station-To-Station Optical Ranges collapsed path as illustrated below. Two signaling links are required for each diverse path segment of an EAD ring, with each segment typically being supported by two
fiber strands. Four signaling links are ACCESS RING CHALLENGES required for collapsed segments typically consuming four fiber strands. Though the use of ring topologies in Diverse Path Ring Segments access applications does address many of the difficulties encountered with alternative approaches, they can and will present challenges of their own. Under specific circumstances problems with regard to security, reliability and fiber utilization can Two Signal Links Collasped arise in access rings. This section will point Ring Segments Host Station out these circumstances and explore methods that alleviate or eliminate their impact.
Four Since a ring is a shared media point-to- Signal Links point topology, security, service theft and multi-point failures in either the ring fiber or Ring Segment Paths stations must be accommodated. OAD ring As previously discussed, a segment’s fiber applications must limit CPE access to those strand usage can be reduced through the use wavelengths specifically allocated. This is of simple two wavelength WDM techniques routinely achieved by placing the service utilizing optical combiners where existing drop’s passive optical multiplexer in the fiber inventories are insufficient to support outside plant (OSP) where it is secure and one link per fiber strand. Where WDM is under the complete control of the MSO. used, each station’s east and west optical Additionally, measures must also be taken to interface’s transmit wavelength must be ensure that the ring’s host headend appropriately coordinated, e.g. all west aggregation switch provides suitable security interfaces transmitting on 1310nm and all measures to prevent successful substitution east interfaces transmitting on 1550nm. of a foreign station and/or the hacking of switch traffic or network control plane It is important to understand that a ring’s information. It is important to understand segment’s paths can diverge and merge that, because OAD ring stations do not multiple times along a common feeder fiber process neighboring station traffic, security route and within one fiber feeder serving and service theft issues are not eliminated, area. This attribute enables a great deal of they are simply moved directly back to the planning flexibility while using existing fiber ring’s hosting switch. strand inventory still available within an HFC fiber feeder cable. Rather than EAD ring technologies will pass a focusing on the feeder’s physical end points portion or all of a ring’s traffic through each and spare fiber strands at each HFC node, station on the ring. Mechanisms must be in a ring design proceeds by focusing on the place such that stations can be assured to be feeder’s many splice locations and a trusted entities; fortunately such common set of spare fiber strands that are mechanisms are native to the operation of used in all of the feeder trunks and EAD stations. An EAD station is typically branches. composed of four functional layers as illustrated below.
Subscriber Attachment Interfaces A multiple EAD station failure condition, though rare, may isolate or island subscriber MSO traffic creating a service outage irrespective Managed MAC Client of how the ring fiber is deployed. The Media Access Control condition can be addressed simultaneously at West Physical East Physical both the local station and serving area levels. West Interface Interface East Segment Segment Multiple station outages can be caused by either a common widespread condition, such as sustained power outage, or a collection of
simultaneous isolated station failures. To mitigate the possibility of multiple EAD Station Function Block Diagram simultaneous isolated station failures in a carrier environment, experience has shown The east and west physical layer that each station should be a fixed-position interfaces provide optical linkages to managed device that is operated from a neighboring ring stations. The media access standby-power source. If a station standby- control (MAC) entity manages add, drop and power source is not available, a local transit station traffic. The functional entity automatic optical bypass switch can be that is responsible for controlling all traffic implemented at either the station or the presented to the subscriber attachment building’s service drop. Thus local station interfaces is the MAC client. The client is powering conditions are not likely to cause directly managed by the MSO’s network multi-station service outage conditions. A management system. Traffic cannot be sustained utility power outage(s) within the presented to subscriber interfaces until the ring’s serving area that interrupts the client has been authenticated as a trusted operation of multiple stations is best network element and services have been addressed on a serving area basis. setup by the management system; the station does not flood its switch ports to build its Recovery from a sustained power outage forwarding tables as found in a standard may be possible through the use of optical Ethernet environment. Station authentication bypass switches at each station. Historically and control processes are typically in LAN environments this approach has not equipment-vendor specific, for example a been successful due to the added cost and system may restrict access to the network’s limitations in the number of sequential control plane with an access control bypasses that can occur at once due to optical h authentication technology such as RADIUS . loss limitations. Given the relatively rare The conditions that can trigger service conditions under which multiple stations outages on a ring are dependent upon the simultaneously fail, a strategic manual- networking technology used and the physical bypass approach may be more appropriate. fiber path, collapsed or diverse. The behavior of fiber-cut induced service outage Collapsed rings composed of fiber feeder in collapsed path OAD or EAD ring is segments can be equipped with sets of identical to that found in bus or bus-star mechanical fiber strand splices at key feeder topology; all stations downstream of a fiber splice locations. Splice locations can then cut experience an outage. Conversely a serve as a manual bypass point. In the event diverse path ring is typically resilient in the of a sustained local power outage, the case of a fiber cut. However, multiple station effected sections of the ring can be bypassed failures can present a problem for EAD rings. at the nearest splice for the duration of the outage. Unfortunately, failed segments of
diverse path rings cannot be so readily collapsed electronic add-drop ring using a isolated, and measures must be taken at the conventional four fiber approach. In these hub/headend level. cases ring fiber counts can be reduced by using either passive WDM or CWDM The business subscriber density within a techniques, as illustrated below. headend or hub serving area is likely to result in a condition where a single EAD ring will route through a hub multiple times which f Ring 2 facilitates an alternative bypass approach, as Fiber Fibers 1 Feeder Node illustrated below. Fiber Spur 18f Ring Fibers 6f WDM Combiners Splice HFC Node
Serving Area 6f
6f Target Business Building Area
WDM Fiber Ring In this case WDM combiners are implemented at the splice cases located at each end of the node’s fiber spur.
Implementing a WDM ring segment does impose additional optical requirements on Ring Spur Bypass the segment’s stations. Each station’s optical In this case an emergency manual or interfaces must transmit using different automatic optically-sensing bypass function optical wavelengths and with sufficient is implemented at the hub for each ring-spur power to overcome optical combiner such that an effected spur is isolated for the insertion losses. For example, one station duration of the outage. This restores the ring may transmit on 1310nm while the other may to full operation for the remaining transmit on 1550nm. Other CWDM subscribers on unaffected ring spurs. wavelengths may be selected as well. In any case, stations equipped with modular When planning the deployment of pluggable optical interfaces can easily collapsed fiber rings, circumstances will be accommodate this requirement while encountered where the available spare fiber facilitating the use of higher power DFB strands in a feeder cable will be insufficient transmitters for the support of ring segments to support a four fiber EAD ring segment. up to 80km in length. This condition is most likely to occur when attempting to extend ring fibers to a business FIBER ACCESS CHALLENGES AND location from a fiber feeder segment where SOLUTIONS fewer than four spare fibers are available. For example, a feeder spur to a lone HFC Clearly the primary barrier to offering node typically contains 4 to 6 fiber strands services via fiber is the cost to extend the with 1 to 2 strands held in reserve for fiber to the subscriber’s building and advanced services; this is an insufficient subsequently to the subscriber’s demark amount of free fiber for the extension of a point. Fortunately in the majority of cases
some portion of the MSO’s fiber plant is a large number of business buildings within a within a few thousand feet of the building, target area. however just being close usually is not sufficient. Effective strategies for both the Determining how to extend fiber to targeted marketing of communications business buildings is driven by where the services and fiber build-out can greatly buildings are located with respect to the reduce risks and enhance business existing fiber inventory, how many buildings development success. are involved and how they are physically laid out. The good news is that the majority of Until recently the implementation of a the business buildings in an MSO’s fiber-based business service by an MSO has addressable market are within one or two been based upon an incremental opportunity thousand feet of the MSO’s fiber plant. brought to the MSO by a service broker or by With respect to the fiber plant, business the subscriber directly. Historically, the buildings can be broadly categorized as being service is typically implemented quickly and on or off the fiber-feeder buffer or HFC node easily using a point-to-point approach. One area as illustrated below. or more feeder fiber strands are dedicated to the service at each of the subscriber FIBER SPLICE ON - NODE attachment points to the MSO’s backbone ON - NODE network. Unfortunately, due to limitations in ON - NODE 6f HEADEND 6f available spare fiber media, this approach is /HUB 6f f 24f 2 not sustainable or scalable within a general 1 ON - FEEDER business services deployment program. 18f 6f
Alternatives which allow businesses to 6f ON - NODE securely share common strands of fiber OFF - FEEDER 6f 6f media are essential and available in both ring 6f OFF - NODE and non-ring topologies. The key attribute of a successful topology lies in its ability to flexibly adapt the diverse array of circumstances that are driven by market, Business Building Locations geographical and operational Geographically, where businesses tend to circumstances. cluster relative to the fiber is dependent upon regional circumstances as well as how the Physically, the business service market MSO chose to build the plant originally. In areas where the MSO is likely to enjoy the the majority of cases the primary fiber-feeder most success and velocity are those that are runs are located along arterial roadways near existing fiber inventory and are typically which fortunately are also where most underserved by legacy business service businesses are located. However larger providers. Commonly these areas are businesses with more advanced suburban in nature with a relatively low communications needs may be more likely to density of business buildings and potential cluster off of the feeder or node areas. subscribers per street mile. Densities of 10 Typically off-node clusters will present the to 15 buildings and 15 to 20 businesses per greatest challenge in that the fewest possible street mile can be expected. This fact spare fibers will be available to serve the coupled with a low fiber service take rate cluster. will force the use of build-out strategies that can cost effectively light enough fiber to pass The number of business buildings in a cluster and their geographical layout can
complicate how fiber will be extended, this Unfortunately, circumstances are greatly is where selecting the right network topology complicated by the fact that the sale of can reduce complexity, cost and investment business services rarely proceeds in an risk. The following example illustrates a orderly fashion. Business opportunities arise combination circumstance where a small in a nearly random sequence based upon business complex is bracketed by arterial existing subscriber service contracts and roadway on the left and a residential evolving subscriber needs and available neighborhood on the right. services. Additionally, in each case the decision to serve an opportunity must be 4 HFC 3 justified on its own return on investment Business Node Park (ROI) merits. Typically the ROI model for 1 1 2 2 the first subscriber in an area will bare the brunt of the cost of extending fiber from the 1 4 Tenant 4 2 Business nearest point on the existing plant. In those 2 Building 48f 1 cases where the physical route to the 1 2 6f 1 subscriber’s building passes a number of 2 single or multi-building clusters of business Residential Serving Area
Arterial Roadway Arterial 6 2 1 Initially subscribers, as illustrated, the MSO may Served Fiber Building Feeder choose to augment the initial service extension, building out with additional fiber such that other businesses along the route can Business Park Example be easily provisioned for service as they are signed up over time; a common sense A large distribution feeder cable runs strategy, but one that increases the ROI risks along the arterial, while a small 6 fiber stub and that has practical limits in terms of how from another distribution branch reaches a many additional fibers can be extended and lone fiber node. The business complex is to where. composed of fewer than 40 businesses Once the decision has been made to housed in a number of single and multi- extend fiber into a business building area it tenant buildings. The building represented is very difficult to predict which of the by the shaded triangle houses the initially remaining businesses could become targeted business subscriber. The simplest customers and when. It will be to the and lowest cost approach to extending fiber MSO’s advantage to pass fiber by as many to the initial target location would be the buildings as budget will allow knowing that allocation of two fiber-feeder strands that are in a competitive environment service turn-up extended directly to the target building from velocity will be a key to winning service the nearest splice point. Such an approach is contracts once they are up for renewal. Thus, certainly low cost day-one, but quickly once a service is sold service turn-up can be becomes a serious liability when the core limited to installing a drop and terminal business development objective is to sell equipment. Both of which can proceed in a services to other businesses in the area. much more timely fashion than roadway construction. Ideally the task of extending fiber to the first business subscriber would cost- Perhaps based upon prior knowledge and effectively position the MSO well for future practice in these circumstances the shared sales within neighboring buildings without fiber deployment approaches most likely to significant additional cost and without be initially explored are bus-star or star consuming additional feeder fiber. topologies such as PON and remotely
switched Ethernet. In either case engineering developments are not fully mature and new challenges quickly mount, the following are buildings are being added periodically. just a few: Fiber rings are passive and EAD rings do not require field installed optical couplers or • How many buildings in the area are combiners. Additionally, all CPE optical expected to house customers? connections are point-to-point greatly easing • How many fibers per roadway branch? optical level concerns. Thus plant design • Within budget can fiber be routed along considerations along with ranging each roadway in the complex? considerations are greatly simplified. Since • Where should fiber access points be the same fibers are being used no matter placed? where a subscriber is being added to a ring a • How and where should the fiber runs simple color based fiber tracking aggregate? methodology is all that need be followed, e.g. • Where is the fiber coming from, is the a four color coding scheme that identifies the cluster too far from the host hub or east and west ring segments irrespective of headend? location on the ring. EAD rings further simplify optical layer record-keeping by • Is there sufficient remaining optical eliminating the need to track subscriber budget to reach all of the buildings? wavelength allocations. Finally, rings offer • If an active star topology is used, where the added bonus of being able to offer is the aggregating switch to be located resiliency in the form of diverse path fiber and how will it be powered? • routing. Is there a solid record keeping system such that two, three or four years after An example of how a ring topology can bundles of fiber have been installed and be used to address every business building in wavelengths have been assigned will our business park example is illustrated operations be able to quickly and below. Note the lack of centralized fiber accurately tap into the available fiber aggregation points, outdoor active equipment strands without disrupting existing and the ability to address all of the business services? park buildings with a simple six fiber buffer • Will service resiliency be required? routed along each roadway. Fiber splice • Will existing customers tolerate the cases can be added arbitrarily for service service outages that may be required to drops as subscribers are added over time, add new subscribers? deferring costs and simplifying attachments. Technicians need only know the east and By contrast a ring based topology, west ring segment fiber pair color codes to particularly EAD approaches, offers several successfully splice into the ring. However key advantages that greatly simplify or the wavelength allocations associated with an eliminate these issues. A ring approach OAD ring must still be carefully tracked. If a eliminates all design variables associated diverse fiber path is desired for the feeder with how many fiber strands and/or portion of this ring a diverse path connection wavelengths to use and where. It is very can be made to the nearby HFC node’s fiber likely that the same common set of strands stub. will be used for the entire area. A ring also eliminates the need to know ahead of time where access to a fiber buffer will be needed. This is very important considering the fact that often business park
Diverse Path Connection cornerstone to the MSO’s business services 4 HFC 3 Business Node program. Splice Park 6f 1 1 2 2 6f 6f 6f 6f Two fundamental techniques for implementing fiber rings based upon optical 1 4 Tenant 4 2 6f Business or electrical add-drop functionality have been 2 Building 6f 48f 1 1 2 outlined and shown to offer common and 6f 1 unique properties that are largely beneficial, 2
Residential Serving Area Serving Residential but can also present unique technical and
Arterial Roadway Arterial 6 2 1 Initially Served operational challenges that must be Fiber Building Feeder effectively accommodated. Ring implementations based upon optical layer Fiber Ring Access Example add-drop (WDM) techniques offer elegant physical layer means of extending private Engineering considerations aside, dedicated links to the subscriber premise that probably the most attractive aspect of using are immune to localized CPE failures. rings for business access is the overall However these techniques encumber reduction in deployment cost and business engineering, operations and management risk. Fewer components, simpler signaling functions with difficult service extension and and easier tracking lead to lower CapEx tracking challenges with little or no promise and OpEx costs, but the fact that a modest for an offsetting cost benefit. By contrast, fixed amount of fiber strategically routed ring implementations based upon electronic that can be accessed arbitrarily over time add-drop techniques greatly simplify puts the MSO in a favorable position; one deployment and management demands by of being able to extend service quickly as supporting the automation of all service opportunities arise, irrespective of take related management and tracking functions at rates, building densities and geographical the network management system level; layouts. enabling the normalization of how physical ring connections are made irrespective of CONCLUSION where or when they are implemented.
The development of business services A combination OAD/EAD strategy may programs will present MSO’s with offer a best of both worlds compromise, as significant technical and managerial illustrated below. challenges that have little historical precedence. An integrated set of simple, clear and effective business development and technical strategies that leverage existing strengths and infrastructure will be critical to success. This requirement clearly must include fiber deployment strategies that maximize service delivery capabilities while minimizing complexity, cost and business development risks. To this end it has been shown why ring based access topologies should be seriously considered as a
EAD CPE d IEEE 802.5 – 1998, “Token Ring Access Method
Passive and Physical Layer Specifications”, CWDM http://standards.ieee.org/getieee802/portfolio.html Add-Drop Host Mux EAD e Terminal ISO 9314-2:1989 Information processing systems -- West Fibre Distributed Data Interface (FDDI) -- Part 2: CWDM Passive Token Ring Media Access Control (MAC), Mux CWDM OAD Add-Drop CPE http://www.iso.ch/iso/en/StandardsQueryFormHandler East CWDM Mux Mux .StandardsQueryFormHandler Host OAD Passive f “FDDI Technology and Applications”, Authors: Terminal CWDM Add-Drop Sonu Mirchandani and Raman Khanna, Copyright Mux 1993 by John Wile & Sons, Inc
EAD g CPE “SONET, Second Edition” Author: Walter J. Goralski, Copyright 2000 McGraw-Hill
Combination EAD/OAD CWDM Ring h Internet Engineering Task Force , RADIUS (Remote Authentication Dial-In User Service), In this case two OAD ring wavelengths are http://www.ietf.org/proceedings/96dec/charters/radius allocated for EAD ring operation and are -charter.html dropped into those business buildings where one or more small to intermediate business BIOGRAPHY subscribers are located. The EAD ring’s Donald Sorenson is Manager of Advanced Access innate ability to aggregate at the CPE and the Planning in the Transmission Network Systems sector headend is highly scalable in this of Scientific-Atlanta, Inc.. Mr. Sorenson is circumstance. Larger business subscribers responsible for the engineering evaluation and who can afford access to a dedicated optical planning of Scientific-Atlanta’s next generation access link can be hosted from the remaining OAD network technologies with particular emphasis on technology migration strategies for HFC networks. wavelengths. These subscribers are likely to Donald obtained his BSEE from South Dakota School be much fewer in number thus reducing of Mines and Technology in 1978 and has maintained scalability and plant management demands. a professional engineering focus throughout his This strategy allows the MSO to offer a career. He brings 25 years of product and technical enhanced differentiated high-end service to business development experience to S-A and the telecommunications industry, of which the past twelve larger customers while at the same time have been spent in the Cable Television and Telecom easing plant scale and management issues. market segments.
REFERENCES: email: [email protected] a IEEE 802.17 Resilient Packet Ring Working Group, http://www.ieee802.org/17/index.htm bInternet Engineering Task Force, Fast Reroute Extensions to RSVP-TE for LSP Tunnels, http://www.ietf.org/internet-drafts/draft-ietf-mpls- rsvp-lsp-fastreroute-02.txt cIEEE 802.3 - 2002, “Carrier Sense Multiple Access with Collision Detection (CSMA/CD) Access Method and Physical Layer Specifications”, Section 4, Link Aggregation, http://standards.ieee.org/getieee802/portfolio.html
FLEXIBLE WHOLE-HOME NETWORKING STRATEGIES IN A MULTI-TV ENVIRONTMENT
Carlton J. Sparrell Ucentric Systems
Abstract Video Recording (PVR) technology has become a requirement of advanced set-top Intense competition between cable and boxes as service providers seek to increase satellite is becoming the driving force for revenue and reduce churn. While many whole-home digital services such as multi-TV providers are taking delivery of single-TV PVR. As service providers deploy customer PVR units in 2003, consumers have premise equipment that enables new demonstrated their desire for a multi-TV applications to reach all corners of the PVR solution with 43% of PVR households digital home, new approaches are required to having two or more units, and 74% of them manage this hardware to ensure ease of saying they want PVR on all TVs in the installation, quality of experience, and home2. flexibility for different customers’ needs. Adding one PVR box to each television is This paper introduces a powerful not only costly, but is confusing and architecture for managing distributed frustrating. Viewers must deal with different hardware resources in the networked home, recording libraries on every TV, and also and describes how a centralized resource coordinate passwords, recording schedules manager and a flexible QoS enabled IP LAN and settings. A less costly and more can be used with advanced set-top boxes, convenient solution is to use in-home low-cost clients, and other consumer networking with an advanced set-top box electronic devices to deliver advanced whole- providing recording and storage. Additional home digital services today while allowing televisions can access that same content service providers to maintain control of the through the use of low-cost media clients consumer experience. delivering content on demand over the home network. Instead of a PVR device on each TV at a cost of $350 to $500, an add-on media client can bring PVR to an additional THE NEW DIGITAL HOME TV at less than $100 retail.
Connected devices targeting the While today’s advanced set tops are one consumer are gaining momentum. Personal candidate to provide this functionality, CES computer networking has taken off as 2003 saw the commitment of the consumer broadband connections have driven data electronics world to the concept of the media gateway sales and Wi-Fi has simplified center and media gateway. Panasonic, installs. At present, some 8 million U.S. Philips, Pioneer, Samsung, Sharp, Sony, households have a network to enable Internet Thomson, and Yamaha all introduced sharing1. connected entertainment devices adding
networking and hard drives to devices such At the same time, new TV-focused as PVR, DVD, and CD appliances. consumer applications are driving the need for whole-home connectivity. Personal COMPETITIVE FORCES applications graphics into a single NTSC signal, and modulating that signal onto an The trend of network-enabled products is unused or notched-out channel while an IR the result of three powerful industry forces receiver in the family room communicates converging on the home: received key-presses back to the set-top on a separate channel. • Cable operator competition with satellite is pushing both to look for Analog distribution allows a set-top or ways of extending digital tier services media gateway solution to distribute content to every TV in the house; throughout the house without a digital • Cable has recognized the importance network. Unfortunately, this solution of the CE channel with agreements degrades digital cable to analog quality while that allow CE manufacturers to build creating privacy concerns by broadcasting digital cable ready devices; and every session “in the clear” to every TV in • The PC, the set-top and the CE media the house. The analog approach also does center suppliers are all trying to be the not scale well as hardware for every control center of the home. television needs to be added to the advanced set top. For cable operators to prevent disintermediation of their services, they need Legacy MPEG Distribution - Another to embrace an in-home networking platform approach to distributing media around the that installs easily, supplies the quality of home is to adapt the advanced set-top box to service their subscribers are accustomed to, broadcast PVR sessions digitally using HFC provides extensibility to match evolving modulation standards such as in-band MPEG customer needs, and delivers a predictable, encapsulation with out-of-band backchannels. intuitive user experience. The one advantage of applying the HFC NETWORKING OPTIONS solution to home media distribution is that it allows existing low-end set-top boxes to be As service providers move to embrace repurposed as terminals for additional sets whole-home strategies, there are a number of within the home. The repurposed equipment options for the delivery of media throughout can then utilize the in-band and out-of-band the home. The most common options are: interfaces for communicating with an advanced set-top box media server which • Analog distribution; becomes a mini-head end. • Legacy MPEG distribution; or • IP networking Drawbacks of this approach include a) the high cost of head-end networking Analog Distribution - One approach to components, and b) the asymmetric distributing media around the home is communications channel. While the overall modulating video and audio onto the existing cost of this approach is partially mitigated by coax network in the home. For example, a the reusability of legacy boxes, the high cost single advanced set-top box in the living of the low-volume, HFC networking chips room might provide a PVR session to a TV in means that new equipment cannot follow the the family room by blending the video and same low-cost volume curves of designs THE DISTRIBUTED HOME using commodity off-the-shelf silicon. The HFC networking chips also present a As service providers adopt whole-home problem in that while they allow the home networking approaches to digital content, the server to act as a mini head-end they recreate challenges include ease of installation, the inherently asymmetric communications quality of service, extensibility, and channel supported by the legacy equipment. interoperability of equipment from different The lack of a high-speed back-channel in the vendors. Add to this the desire to have a legacy equipment eliminates the possibility of flexible platform adaptable to different pooling networked resources such as tuners subscribers’ needs and competitive market and storage. This approach is also not very requirements driving time-to-market. suitable for IP traffic and hence cuts off an attractive and important array of emerging The solution is to deploy network-ready consumer services and product. equipment with software components capable of auto-detection and auto-configuration of IP Networking - The IP protocol is the all devices and resources distributed on a most broadly deployed networking solution quality of service enabled home network. worldwide and is available in several forms appropriate for whole-home media Distributed Resources distribution including wireless and over-coax. The Ucentric approach is to employ a The significant advantage of standards- centralized resource manager capable of based IP networking, regardless of the discovering, allocating, and controlling physical layer, is that it leverages commodity distributed resources. Consider the hardware hardware available at low-cost. IP resources common to the whole-home networking provides the flexibility of experience: supporting both bursty and streaming high- speed connections over a symmetric channel. Video Tuners – Not long ago, video tuner With appropriate quality of service and requirements were straightforward: One TV, resource management, IP networks support one tuner. With PVR, and now whole-home the widest range of applications and PVR, the right mix of tuners changes efficiently handle video (including trick- depending on the household. Different play), digital audio, and data networking. households will require a different number of tuners depending on the number of For service providers wishing to pursue simultaneous live-pause and pre-recording the consumer applications and associated sessions needed for their viewing habits. revenue opportunities enabled by IP networking, while supporting legacy Service providers must recognize that a equipment during the migration, a hybrid whole-home CPE network is not one size fits approach is possible with intelligent resource all. What is required is a means for adding management of both legacy and IP-based tuners incrementally based on the customer, network resources. and allowing the number to change as the customers’ needs evolve.
Conditional Access Modules - As with tuners, the conditional access equation of the past was on a per-TV basis. With a whole- solution must be able to recognize their home solution, the number of conditional capabilities. access modules required is more a function of a family’s premium content tier and viewing Bandwidth - Control of bandwidth is habits. While the ideal is to have every tuner critical to maintaining quality of service in IP CA-enabled, the economics of CA-licensing networks. A whole-home network may or POD cost suggest that in the short term contain a single subnet, such as an 802.11g- there may be premium content homes where over-coax backbone, or it may contain not all tuners are CA-equipped. multiple subnets including a wireless network and local IEEE1394 buses. Disk bandwidth Persistent Storage - The central management is also critical for time shifting component to time-shifted media is the hard multiple high-def and standard-def streams. drive. The problem with storage economics A successful resource management solution is that it is expensive for a service provider to must be able to reserve and maintain deploy, and consumers will always want bandwidth throughout the system for all more. The solution to this problem is to viewing sessions even during trickplay. leverage traditional and CE channels to allow a consumer to add more storage to their Outputs - Not all output interfaces are the network as their needs evolve. Examples of same. A resource management solution must add-on storage include 1394 drives connected be able to distinguish between active high-def to the primary ASTB or drives located in and standard-def ports, and track which other CPE on the IP network. A resource interfaces are configured to support DRM management solution needs to be able to protected content, and which are not. discover, configure and control persistent storage located anywhere on the network. Other Resources - In addition to these common AV resources, other devices are AV Encoders - MPEG-2 is the default AV converging on the home network. POTS encoding technology used today for time- lines provide necessary connections for shifting and storing analog content. As voicemail and caller-ID information. consumer oriented products evolve, this will Broadband connections provide a data link to migrate to support for more efficient the Internet. PDAs and cell phones offer encoding techniques such as MPEG-4 and alternative GUI opportunities. A resource H.26x. Whole-home PVR solutions today management solution must remain flexible require an MPEG-2 encoder per analog tuner, enough to support new devices and remain but in this evolving world, it is important to extensible to future unforeseen consumer plan resource management solutions to be applications. advanced codec-ready, and be able to distinguish different encoder capabilities on Centralized Resource Management mixed resource networks. Each of the above resources shares the AV Decoders - As with the AV encoders, common element that they all operate on there is a migration underway towards multi- streams of media or data content. Some codec capable decoders. As these products elements, such as tuners, are stream are deployed, a resource management producers. Other elements, such as decoders and their associated display outputs, are stream consumers. There are also elements that connect producers and consumers. Any their interfaces using Simple Service of these resources can be chained to form a Discovery Protocol. Applications media pipeline to provide a service, such as communicate to each other and the resource live-pause TV. manager using a networked XML interface. Standards such as UPnP and HAVi have Each network is required to have at least provided mechanisms for networked devices one resource manager-capable device. In the to broadcast their availability and allow typical home this is a single advanced set-top negotiation between devices for control of box, but in homes that contain more than one each other’s resources. While this ad hoc ASTB, a negotiation protocol is used to negotiating technique works for some lean- determine which resource manager is active. forward activities like downloading a video The active resource manager maintains a from a camcorder for digital editing, it lacks table of all available devices, their stream the level of sophistication needed to provide resources and resource interfaces. The a single point of control to reserve resources resource manager also maintains reservation in the future, string together chains of information allowing resources to be distributed resources, and resolve conflicts in assigned in the present or future. a manner essential for the lean-back experience of whole-home media on demand. Applications requiring resources communicate these requests to the resource The Ucentric centralized resource manager. These requests take the form of a manager is capable of discovering resources pipeline graph connecting resource as they are added to the home network and requirements. For example, when a new providing a single point of arbitration for the video output is activated on the system, a reservation of these resources among new session application is instantiated and authorized applications. Around each associated with that display. This session resource is implemented a unified API reserves the resources necessary to provide a allowing applications to request and control graphical user interface and pre-recorded the resources they need. MPEG-2 content to that display. If that session later needs additional resources, such HOW IT WORKS as tuner for watching live-pause content, an additional request will be made to the Devices deployed within the Ucentric resource manager at that time. environment provide a service wrapper around each resource. This service wrapper Other applications, such as the EPG, may provides a standard API for stream operations request resources for a future event. For associated with that device. For example, an example, when a user requests that a show be MPEG-2 decoder provides a streaming media recorded at a future time and date, the interface supporting the Ucentric Media application requests the associated resources Protocol (UMP). This API allows other (e.g. tuner, MPEG-2 encoder, disk capacity) components, such as a disk-stream media for the time window required. server, to deliver trick-play enabled MPEG-2 content to the decoder regardless of whether Example the two resources are in the same box or in different rooms of the house. When a new Consider a typical evening at home with the device is connected to the network, the whole-home enabled network described in device reports its available resources and Figure 1.
ASTB SERVER LIVINGROOM 2 x TUNER DISK 2 x MPEG2 ENC DCATV MPEG2 DEC TUNER DISK
CLIENT FAMILYROOM Figure 2 – Digital record pipeline TUNER MPEG2 ENC The requested pipeline includes a digital MPEG2 DEC tuner. If channel 150 had been a premium station, the scheduling application would KITCHEN have made additional requirements on this CLIENT tuner, such as associated Conditional Access CATV & IP-OVER-COAX MPEG2 DEC or POD module. The requested pipeline also indicates the required disk bandwidth and storage capacity needed to record the CATV program.
Figure 1 –Typical whole-home network The resource manager searches the resource database for resources that match This diagram illustrates a three-TV the request. The network contains one disk household. The living room TV is connected and three tuners. All three tuners have the to an advanced set-top box configured to be a same capabilities, differing only by their media server. The media server contains a location. The resource manager uses a least- number of resources including two cost algorithm to construct a pipeline analog/digital tuners, two MPEG-2 encoders, choosing a tuner in the server to avoid using and one MPEG-2 decoder. The media server network resources. also includes a hard disk with associated media services capable of sending and The resource manager checks for disk receiving media streams. The family room space both when the user schedules the TV is connected to a media client with a recording and shortly before the recording tuner, MPEG-2 encoder, and MPEG-2 begins. If no disk space is available when the decoder. A third TV in the kitchen is user schedules the event, the resource attached to a client with an MPEG-2 decoder. manager checks to see if any “delete-able” The advanced set-top box and the media files are available for deletion. If all the files clients communicate over a 2.4GHz IP on a full disk are marked as “do not delete”, network sharing the same coax as the in- the user will be alerted. house CATV. Upon successful reservation of the One morning, Dad programs the system required resources, the reservation is stored to record a hockey game at 8:00pm on digital in the resource manager reservation table for channel 150 using the EPG on the TV in the use when considering future reservation kitchen. The EPG scheduling application requests. A successful reservation is requests a reservation of an audio-video pipeline with the pipeline in Figure 2: communicated back to the application with a has added one component to the pipeline reservation identification. (Figure 4). At 7:30 that evening, the kids want to watch a show in the family room. The show SERVER they want to watch is on analog channel 32. ACATV MPEG2 DISK When they select this program from the EPG, TUNER ENCODER the application calls the resource manager to request resources. The pipeline in Figure 3 is requested by the application: CLIENT KITCHEN LAN MPEG2 DECODER ACATV MPEG2 DISK TUNER ENCODER FAMILY ROOM
Figure 4 – Granted live-pause pipeline #1
KITCHEN The LAN connection is required to connect the components in the server to the MPEG2 DECODER components in the client. The LAN is a managed resource with guaranteed quality of service. Bandwidth allocation is controlled by the resource manager. The resource Figure 3 – Requested live-pause pipeline manager assigns the bandwidth requested to send one MPEG-2 stream. There are two unassigned tuners on the network, one in the server in the living room At 7:45pm, Mom wants to watch a and one in the client in the family room. program in the kitchen. The resource While the tuner in the family room is local to manager asks for a pipeline identical to that the TV session requesting the tuner, the in Figure 3. In this case, the only tuner stream will be written to the disk in the living remaining on the network is the tuner in the room for live-pause. The least-cost algorithm family room. The resource manager leads the resource manager to assign the completes the graph in Figure 5: tuner/encoder pair in the living room to the pipeline, saving a transfer of the encoded data CLIENT SERVER twice across the network. This method LAN preserves more network bandwidth for other ACATV MPEG2 DISK uses including best-effort data transfers TUNER ENCODER between PCs sharing the network. (FAMILY ROOM)
Once the resource manager has CLIENT KITCHEN successfully mapped the requested pipeline to LAN MPEG2 actual network resources, the instantiated DECODER pipeline is returned to the application and the (KITCHEN) resources are marked as reserved (in this case indefinitely). Note that the resource manager Figure 5 – Granted live-pause pipeline #2 Two network components need to be FlexMedia LANTM software solution. The added to the graph, and twice the bandwidth Ucentric resource manager is capable of reserved on the network. managing multiple LAN technologies simultaneously, including hybrid topologies At 7:50pm, the system prepares to record such as wired/wireless or legacy MPEG and the hockey game, verifying that disk space is IP over coax. available, and removing “delete-able” content or alerting the viewers as needed. FlexMedia LANTM also provides control of bandwidth allocation for multiple streams At 8:00pm the recording of the hockey and applications through the Ucentric game commences. bandwidth broker. This bandwidth broker provides QoS mechanisms including priority At 8:05pm, Dad sits down in the living services, dynamic stream management and room to watch a program. He chooses not to bandwidth allocation. watch the game, but to look through the video library, selecting a James Bond movie The strength of FlexMedia LANTM stream recorded earlier that week. The system now management over the network is combined makes an updated request for resources with the Ucentric distributed resource (Figure 6): manager to pool all tuner, storage and network resources - allowing the number of LIVINGROOM tuners to be provisioned flexibly in DISK MPEG2 relationship to actual household need. DECODER Additional supported services include expandable storage located anywhere on the network providing convenient installation
options. Figure 6 – Play from disk pipeline CONCLUSIONS The resource manager is able to construct this graph out of resources available in the Cable operator competition with satellite server in the living room. is pushing operators to look for ways of extending digital tier services to every TV in Pipelines are torn down when they are no the house. This competition has in part lead longer needed. For example, the recording cable operators to recognize the importance resources in Figure 2 are freed when the of the CE channel as a means for making scheduled recording of the hockey game every corner of the home “cable ready”. completes. Typically, a minimal video With consumer applications evolving, and playback pipeline (Figure 6) is maintained to PC, set-top and CE media centers all trying to allow every television to have instant access be the control center of the home, cable to at least pre-recorded content. operators can leverage their quality experience with new CE relationships to FlexMedia LANTM deliver a compelling whole-home experience. To accomplish this, service providers need to The distributed resource model and adopt a flexible network approach to centralized resource manager discussed here maintain control of the consumer experience. has been implemented in the Ucentric Successful deployment of new consumer REFERENCES applications requires the flexibility of IP networks. The heterogeneous nature of these 1. Parks Associates, “Drivers for Home networks together with the need for quality of Networks and Residential Gateways,” service guarantees requires intelligent 2002 resource management. Whole-home 2. “The PVR Monitor III” from C applications require a whole-home approach Cubed, October 2002 to deployed resources, and this requirement has led Ucentric Systems to adopt a CONTACT INFORMATION centralized resource manager in the FlexMedia LAN product offering. This Carlton J. Sparrell approach provides the most flexible, Networking Systems Architect predictable, quality user experience available Ucentric Systems today. 2 Clock Tower Place, Suite 550 Maynard, MA 01754 [email protected]
IMPLEMENTING AND VERIFYING OFF-AIR DTV CARRIAGE CONTRACTS IN CABLE HEADENDS
Nandhu Nandhakumar, Jian Shen, and Gomer Thomas Triveni Digital, Inc
Abstract audio streams to reduce the bit rate in order to maximize the usage of cable bandwidth. Cable-carriage of off-air DTV broadcast streams may involve the selection and Cable operators face two problems when transformation of different components of performing such re-multiplexing: the stream. Carriage agreements may 1. The modifications to the original signal specify constraints on these processes. This may be governed by various regulatory paper lists some of the more common requirements and industry wide technical aspects of carriage agreements agreements, as well as by carriage and describes how they can be implemented contracts between the terrestrial conveniently and with low operational cost. broadcasters and cable operators. Systems that can be used to implement the Moreover, it is possible that a cable agreement can also be used to monitor and operator may have different carriage verify compliance with such carriage agreements with different broadcasters. agreements. An integrated solution is Thus, it may be necessary to apply described to meet these needs. different policies to different signals in 1. INTRODUCTION the same cable headend multiplex. 2. There is increasing desire to retain PSIP A typical ATSC terrestrial DTV metadata in off-air broadcasts when broadcast stream contains one or more video passing them through a cable plant, in programs, audio programs, and/or data order to accommodate “cable-ready” programs. It also contains the Program and consumer DTV receivers. However, System Information Protocol (PSIP) most of the MPEG-2 multiplexers used metadata used by DTV receivers for tuning, in cable headends are not designed to electronic program guides (EPG), and other handle the PSIP metadata. functions[1],[2]. Because a terrestrial broadcast stream has lower bit rate than a This paper describes a headend system QAM-modulated cable signal, and because architecture that can be used by cable cable operators may sometimes carry only a operators for processing PSIP and other selected subset of services that appear in an metadata while simultaneously enforcing off-air broadcast stream, off-air DTV and verifying carriage agreements between broadcast streams are typically re- terrestrial broadcasters and cable operators. multiplexed in the cable headend. In the re- Such a system can work in conjunction with multiplexing, the cable operators often make one or more MPEG-2 transport stream various changes to the original signal, such multiplexers to produce output streams that as filtering out unwanted programs, are fully compliant with carriage agreements elementary streams, data packets and/or and the various metadata standards optional metadata. In addition, the cable including the ATSC PSIP standard, and the operators may transcode input video and ANSI/SCTE 65 2002 standard[3].
streams, e.g. one video, two audio (one 2. Technical Components of Carriage of which could be a secondary audio agreements program/SAP). The program may also Carriage agreements negotiated between contain one or more data streams that are the local broadcaster and the cable operator associated with the A/V program. The may contain many technical elements. Some agreement may allow only specific of the more common ones are described elementary streams to be carried, e.g., below. Later we present ways in which these only the main audio track and no data elements can be implemented and even if it is program-related. monitored. 6. The stream emitted by a local 1. Specific video and audio services in the broadcaster may contain services that broadcast stream that will be carried may have no A/V content at all and only be specified. This can range from all carry data. Carriage of data–only services being carried to only a single services may be (dis)allowed. These (primary service) being carried. The services may be specified similar to the services to be carried can be specified in manner in which A/V services are the agreement by the MPEG-2 program specified. Additionally, specific number, stream packet identifiers descriptors in MPEG-2 Program Specific (PIDs), and/or virtual channel number Information (PSI) and descriptions in used by the broadcaster. ATSC DST’s may be used to identify 2. In the case where multiple services are and refer to specific content types. to be carried, the agreement may specify 7. The agreement may disallow carriage of if the broadcaster can allow services to streams that contain encrypted or disappear and appear, or if there are proprietary content that cannot be restrictions – e.g., if the services must be received by all cable customers on a present at all times, perhaps in the form non-subscription basis. of a low bit rate “barker” channel. 8. The carriage agreement may stipulate 3. The agreement may specify whether the the amount of in-band EPG information operator can reduce the bit rate of, or (in the form of ATSC A/65 compliant modify in any way, the packetized PSIP data in the incoming stream) that elementary stream (PES) structure of the will be carried. The minimum amount is services which usually results in likely to be as specified in the February increased compression/reduced image 2000 NCTA-CEA PSIP agreement quality. This operation is usually [4],[5]. implemented by exercising the 9. However, the agreement may exceed the transcoding/rate-muxing features of a minimum requirement described in the multiplexer, and sometimes by decoding NCTA-CEA PSIP agreement, and can the signal to uncompressed baseband specify the following parameters per and then re-encoding the signal. service carried: 4. If bit rate reduction is to be allowed on o Number of hours/days of program any service, the agreement may specify event titles upper and lower limits of the incoming and outgoing bit rates. o Number of hours/days of program descriptions 5. An MPEG-2 program in the incoming stream may contain multiple elementary
o Various EPG table cycle times or bit stream and create a new, output transport rates stream with only the desired services. 10. The agreement can specify branding as Depending on their source, some incoming signaled by in-band EPG data. transport streams may contain PSIP data Specifically, whether one-part of two- while others may not. For example, transport part virtual channel numbering will be streams received from off-air terrestrial used and what the specific number will broadcasts will typically contain PSIP, while be. The short name and description of streams originating from cable networks the virtual channel can also be specified. (offered as premium, hence scrambled services) may not include PSIP. 11. The off-air signal typically has content advisory information, and closed- Available multiplexers can implement captioning signaling information carried some of the elements of carriage agreements in the EIT’s. The cable operator may listed in the previous section. However, additionally require that these be currently available multiplexers are designed signaled in the MPEG-2 PSI tables, or to handle the grooming and bit-rate may agree to copy these descriptors modifications of the audio and video (during multiplexing) in the headend services only. Metadata such as the ATSC from the EIT’s to the PSI tables – if PSIP tables are typically ignored by present- legacy digital STB’s require this day multiplexers, and are either blocked information to be in the PSI tables. from passage, or are erroneously passed through – without the needed modification 12. If the cable operator can support updates to reflect the program and A/V PID to out of band EPG data - either grooming that the multiplexer implements. standards-based or proprietary in format Also, the multiplexers are meant to be used – the agreement could specify if the in- in a relatively static configuration, and are band EPG data will feed the out-of-band neither easily easily, automatically nor EPG data. This The latter is used to dynamically reconfigurable to handle some drive EPG’s in operator provided STB’s, of the cases described in the previous and in future POD-host equipped retail, section. digital-cable-ready TV’s and STB’s. A more general approach for 3. IMPLEMENTATION OF implementing the various elements of cable AGREEMENTS carriage agreements described above is to use a separate metadata processing system DTV services arrive at the cable head-end that can process and interpret the in-band over satellite and terrestrial links, as well as metadata in the off-air stream to (1) via other means (over an ATM network, for transform and generate new metadata, and example). Multiple transport streams (2) trigger pre-programmed control actions originating from different sources are that instruct the multiplexer to implement typically multiplexed into a higher needed service selection and A/V bit rate bandwidth transport stream, and are then modification actions. Figure 1 illustrates the modulated by a QAM modulator before metadata processing system in the cable being sent out to customers via cable. head-end environment. Available multiplexers can select audio/video services from each incoming
The metadata processing system monitors play out MPEG-2 packets at table-specific each incoming streams in real-time and rates. Note that not all multiplexers can provides detailed information about the support download and playout of externally contents of the stream, including the specified tables. The table playout rates existence of various services, such as data should take into consideration the bandwidth services, that are not easily discovered by a limitations indicated in the carriage traditional multiplexer. To protect the agreement. The output PSIP data stream investments already made by the cable from the metadata system is multiplexed operators, the system should work with into the output transport streams along with existing MPEG-2 multiplexers. audio and video and other elementary streams. This forms the in-band PSIP data The metadata processing system can required mainly by cable-ready DTV either get the relevant metadata (i.e., receivers for tuning to in-the-clear streams. PSIP/SI/PSI) directly from the input transport streams, or can have them passed In addition to handling in-band PSIP, the to it from the multiplexers. The latter can be metadata processing system also generates achieved by a private or standards-based an OOB SI stream. The aggregated SI data protocol to exchange desired information contains the information for all the “in-the- between the multiplexer and the metadata clear” virtual channels (VCs) in the cable processing system. By decoding the system, as well as any scrambled services MPEG-2 PSI tables and the PSIP tables, the that the cable provider chooses to include metadata processing system can identify for the purposes of discovery by POD- every component in the transport stream and enabled cable-ready DTV receivers. For tell whether it is a video, audio, data or incoming streams that contain PSIP data, the PSIP/PSI packet. Thus, a cable operator system optionally extracts the EIT and ETT does not need to continuously and manually data and converts them to the aggregated SI monitor and analyze the transport stream format described in SCTE 65 2002 [3], content to effect a multiplexer control sometimes also referred to as SCTE DVS action. The system can automatically 234. For incoming streams that do not identify the PID streams that show up in the contain PSIP, the system allows manual or transport stream, determine the actions such programmatic input of the VC information as re-mapping, blocking or passing through so that it can be included in the OOB virtual each PID stream according to the carriage channel map. policy parameters, and send appropriate In addition, the metadata processing control commands to the multiplexers. system may also export information to the When multiple streams containing PSIP cable operator’s proprietary program guide data are multiplexed to a single transport service to perform real-time update of the stream, the system needs to process the service information. Typically, the database different incoming PSIP streams and create used by the cable operator for EPG service new in-band PSIP data for the output stream. is days or weeks old. When a program, such The system decodes the original PSIP data, as a sports game, runs over time, the EPG obtains their semantic content, translates the information following the overrun program data, and merges the metadata at the table event is out-of-date. If the incoming stream level. The tables can be played out as contains updated PSIP information, this MPEG-2 transport packets either by the information can be used to update the cable metadata processing system, or the system guide. can send the tables to a multiplexer for it to
4. OPERATIONAL ISSUES XML file may capture an agreement that may be in force at multiple locations, e.g., a The operation of the metadata processing nationwide agreement between a broadcast system consists of specifying the following network and an MSO. Instead of duplicating parameters during the initial setup phase: the manual configuration process at every
location, the metadata processing system 1. Mapping of specific input should be able to export and import the programs/PIDs to specific output PID’s, XML file (corresponding to a specific or blockage of specific input agreement) for easy implementation of the PIDs/MPEG-2 programs. This operation contract elsewhere. should be performed only once. A tight integration between the metadata 5. MONITORING & VERIFICATION processing system and the multiplexer control system will ensure consistency A low cost approach for monitoring and between PID grooming and metadata verifying compliance with various carriage grooming. agreements within a headend is of high 2. Mapping of an incoming virtual value to a local system’s operations. To channel’s major-minor channel number reduce cable headend operations burden and to an outgoing/cable channel number. minimize the equipment cost, it is desirable Mapping may specify either a two-part to have the monitoring device be the same as or single-part channel. In addition, the the metadata processing system described channel name can also be changed if above, so that all aspects related to carriage desired. contract implementation can also be verified 3. The number of hours of EPG data to be and monitored continually by a single passed through. system. Thus, the system should provide 4. The frequency of optional PSIP tables convenient user interfaces for contract being played out, mainly EITs and specification, metadata transformation & ETTs, as defined by the actual table generation, multiplexer configuration, interval time or limitation on the PSIP stream monitoring/visualization, and bandwidth. configuring error-based alarming. 5. Enable/disable the copying of various 5.1 Error Monitoring descriptors from the on-air event’s EIT to the PMT. For example, the content In addition to monitoring compliance advisory descriptor that may be only in with carriage agreements, cable operators the EIT should be copied to the PMT. may need to monitor input streams to 6. Allow/disallow transcoding or prevent errors from being propagated to bandwidth reduction by the multiplexer their customers. Typical errors that may be for a specified transport stream, present in a DTV broadcast stream include: program, or elementary stream. If allowed, then specify the 1. Missing, invalid, or infrequent PSIP and maximum/mimimum bit rate for that PSI tables. Because of these metadata program/elementary stream. errors, DTV receivers may not be able to tune channels or block unwanted programs properly. Furthermore, the on- The control information can be entered screen program guide may be missing or with the help of a simple user interface and incorrect. the data can be saved in a format that is easy to view and edit, e.g., using XML. This
2. Missing video or audio elementary in an input stream and leaves significant streams. unused bandwidth, the cable operator can 3. PCR jitter, which can cause incorrect detect and use the available extra bandwidth synchronization of the audio and video to offer other services. signals, sometimes resulting in a “lip sync” problem. 5.2 Contract Verification 4. Video and audio buffers overflow or A monitoring system at the headend that underflow, which can cause ragged, continually checks for compliance with stuttering video and audio. carriage agreements is of high value to the local systems operations. All of the technical Depending on the type of errors found, components of carriage agreements the cable operators may have different described in section 2 above can be verified options to deal with the erroneous conditions automatically as described in table 1. in the transport stream. Some errors may be fixed or filtered out by the metadata Different stream monitoring strategies processing system and multiplexers. For may be applied in cable headends. One example, PSI and PSIP tables are normally possible approach is to check the input or regenerated by the metadata processing output streams periodically and manually system. Therefore, certain syntax errors, using a stream analyzer. However, this inconsistent data, and incorrect table time approach only shows a snap shot of the intervals may be automatically corrected. stream and requires tedious routine Other type of errors may be corrected with inspections that result in high operational some level of manual intervention. In cost. An alternative approach is to monitor addition, some errors may be filtered out if multiple streams simultaneously and use the they are associated with data that are not system to verify carriage contract critical. For the errors that cannot be easily compliance automatically. This approach is fixed, the cable operators may need to the most cost-effective. The operator is only contact the original broadcasters so that the required to enter the contract items that are problems can be promptly addressed. Since to be monitored in a standard template using transport stream errors can also be a simple user interface, he can repurpose the introduced in cable headends, contract specification file used for comprehensive stream monitoring (at implementing the contract. To meet the various points in the path of a stream from needs of automatic contract verification for ingest to emission) can be used to detect the multiple off-air feeds, the monitoring system occurrence and cause of errors – such as should provide the following features: equipment failure or operator error during provisioning, reconfiguration, maintenance, 1. Allow multiple inputs and performs etc. simultaneous analysis on different
streams in real-time. Besides error monitoring, another important 2. Provide remote user interface for reason for cable operators to monitor input displaying and analyzing data. streams is to know the exact content and 3. Provide an alarming function to inform bandwidth usage of an input stream so that the cable operators or the original they can optimize bandwidth usage in their broadcasters once any errors are detected backbone and QAM feeds to the home. For in the stream. example, when the program lineup changes
4. Provide a recording function that is the ability to handle frequent and periodic automatically triggered by defined error changes to incoming stream composition, conditions in a transport stream for requires new types of stream processing, further analysis. monitoring and operational support systems. 6. SUMMARY Triveni Digital has developed and deployed the StreamBridge system to meet a Whether the cable operator implements critical need in this area. StreamBridge is must-carry type of contracts or private designed to be a low cost, highly integrated carriage agreements, a large number of system to facilitate the bridging of off-air technical parameters must be addressed in streams to digital cable systems. This system implementing the agreement. The technical is designed to analyze the composition of issues will only continue to become more incoming streams, comprehend carriage numerous and complex as penetration of contracts specified in convenient XML DTV increases and service providers and format, instruct multiplexers in the headend content providers seek to innovate and to implement appropriate grooming and expand offerings. In the near future the rate-shaping actions, and monitor/verify the deployment of retail, cable-ready DTV resulting streams for carriage contract receivers based on OpenCable and related compliance. The system is scalable and a standards, and the launch of interactive single unit can support a large number of services will create these additional multiplexers / transport streams. The technical requirements. StreamBridge system provides remote user Current digital cable headends are interfaces, and can interface with equipped to deal with only a limited number multiplexers that are not co-located. The of these technical issues. Equipment system software can be easily upgraded to deployed today can only groom and rate- evolve with changing standards, operational shape audio and video programs, and they needs, and other headend equipment also require static channel line-ups. In-band changes. PSIP metadata is ignored and cannot be correctly transformed. The ability to handle this new type of stream element, as well as
Clear QAM Terrestrial Rcvr MUX Modulator A/V/PSIP Clear Terrestrial Rcvr
PSIP, SI/PSIP To Fiber Data QPSK Modulator Nodes Metadata Processing DVS System 65 2002 Encrypted A/V Satellite Rcvr QAM Satellite Rcvr MUX Modulator Clear
Figure 1: Metadata Processing System in Cable Head-End Environment.
Table 1. Common technical contract items and related monitoring parameters
Contract Item Monitoring Parameters No bit rate reduction Monitor and compare the bit rate of specified elementary stream. No video trans-coding Periodically compare the input stream with the corresponding portion of the output stream for changes in bit rate. Meet minimum bit rate Monitor the bit rate of specified elementary stream and requirement compare it with a specified value. Carry specified program or all Discover the specified programs in the input stream. programs Identify the same programs in the output based on program number, PMT PID or virtual channel mapping information. Carry specified elementary stream Discover the specified elementary streams in the input or all streams of a program stream. Verify that the specified elementary stream is present in the output stream based on PID mapping information. Carry specified data program Monitor that the data program is present in the output stream. Carry required PSIP data • Verify the presence of the virtual channel in the virtual channel table. • Verify the number of EITs present in the transport stream is the same as specified. • Verify the presence or absence of ETTs as specified. • Perform cross table analysis to check the data among different PSIP tables and between PSIP and PSI tables are consistent after re-multiplexing. Verify channel branding Check if the correct virtual channel name, major and minor channel numbers are present in the output VCT. Verify the bit rate of PSIP data. Identify the PSIP data that are related to the particular input broadcast streams. Calculate the bit rate of all PSIP data that belong to the input stream and compare it to a specified value. Carry required descriptors Analyze the tables that carry the descriptor and verify that the specified descriptors are present and correct.
REFERENCES [1] Program and System Information [4] NCTA-CEA PSIP Agreement, February Protocol for Terrestrial Broadcast and Cable 2000 (Revision A) and Amendments 1A, 2 and 3, [5] Memorandum of Understanding Among ATSC Document A/65A, 31 May 2000. Cable MSOs and Consumer Electronics [2] ITU-T Rec. H. 222.0 | ISO/IEC 13818- Manufacturers, 12 December 2002 1:1994, Information Technology — Coding of Moving Pictures and Associated Audio CONTACT INFORMATION: — Part 1: Systems. N. Nandhakumar, [3] Service Information Delivered Out-of- Triveni Digital, Inc., Band for Digital Cable Television, 40 Washington Road, ANSI/SCTE 65 2002. Princeton Junction, NJ 08550 USA
INGEST & METADATA PARTITIONING: REQUIREMENTS FOR TELEVISION ON DEMANDTM
Robert G. Scheffler, Chief Architect Broadbus Technologies, Inc.
ABSTRACT
On demand video services, such as asset distribution, content propagation, today’s Video on Demand (VOD), network loading, metadata and rules issues Subscription Video on Demand (SVOD), and need to be addressed to make Television on the fast-approaching Television on Demand a commercial reality. DemandTM (TOD®) are enhancing the consumer television experience and creating This paper will address the issues and new, exciting revenue opportunities and requirements associated with server ingest increased cash flow for cable operators and of broadcast content and content content owners alike. However, the propagation. It will also discuss the technical requirements to support these architectural implications for the VOD services are becoming more demanding and server and propose a new class of server to complex. In VOD, cable operators are support TOD requirements. The paper will seeing solid buy-in rates, repeat purchase also discuss how TOD content is managed patterns, and concurrency rates of 3%-10% through the creation and distribution of with limited marketing and promotional enhanced metadata formats in an support. With recent trials of SVOD and an environment that is controlled by studios, increased number of popular titles, distributors, and cable operators. concurrency rates have ‘smoothed’ the peak usage rates throughout the week to numbers New video server architectures and that often approach 10%-20%. However, rules-based content control and propagation with Television on Demand (TOD) services, systems become integral contributors to the consumers will have considerably more success of future on-demand services. programming choices including movies, subscription-based content, and the most VOD/TOD CONTENT INGEST popular broadcast content. It is anticipated that concurrency rates of TOD may steadily The issue of the ingest of broadcast climb to levels that approach 30%-65% -- television content is one that will become rates that mirror the total concurrent U.S. more and more important for advanced television viewing audience as measured by video services such as Television on rating services such as Nielson. Demand to become a reality. As more content is made available and concurrency Increased service usage, additional rates increase, architectural decisions will content, and new business models are have to be made to support these increased challenging MSOs to conduct unprecedented demands on the network. A new architecture network architecture preparation and comprised of higher density VOD/TOD planning. In addition, decisions related to servers with the capability to ingest
broadcast television will be required to processing capacity to focus on the ingest support ever increasing content libraries and functions. This was very labor intensive stream counts. However it is important to with a single operator feeding tapes and look at the evolution of VOD architectures entering rules to instruct the STB guide to understand how those requirements will software about the pricing and availability of change in the future. new titles (see Figure 2-1). Keeping up with content ingest was quite manageable for the VOD in the Past operator and the conventional VOD server.
In the early days of VOD, movies were VOD Today distributed on tapes. These tapes were shipped to each site that required a specific As an industry, VOD has matured movie title. Using an encoding rate of 3.375 beyond the simplistic example described Mbps and an average movie length of 100 above. VOD installations now enable 1,000 minutes, the total size of each movie was to 3,000 customers to access a library of 150 roughly 2.4 GBytes. A typical installation to 300 movies. As a result, shipping tapes to might contain a library of under 100 movies VOD enabled head-ends has proved to be a and was capable of streaming to less than logistical challenge and has evolved to a 1,000 subscribers simultaneously. newer model called pitch-and-catch, where content is distributed by private broadcast to remote stations and syndication partners via satellite (see Figure 2-2). With increased library sizes, increased stream counts and more diverse suppliers sending data, the distribution and propagation of content has shown itself to be quite a challenge. Content Tape drive can still arrive on tapes and is caught by VOD Server catchers along with trailers, posters, and Content Person rules that are required to put it all together.
Figure 2-1 Content Ingest for VOD in the past
In early VOD deployments, metadata or Catcher Catcher Catcher other business rules weren’t typically VOD Server supplied with the content. The operators themselves were responsible for deciding what rules applied to particular content and Tape drive for entering the appropriate rules into the Content Person VOD server or control system. This relatively simple model meant that most of Figure 2-2 Content Ingest for VOD today the attention was focused on the billing interfaces, set top box (STB) client, and Content aggregation companies have head-end control. With low stream counts, risen to the challenge by offering services to movie titles could be loaded during off-peak edit, adjust, and compile these diverse hours when the VOD server had more formats and metadata into a nice bundled
package to be pitched and caught. However, large amounts of content propagation. To a fundamental problem is that while quite now handle the ingest of a significant adept at low-volume streaming, amount of content, a conventional VOD conventional VOD servers usually lack in server will typically lose some, or much of their ability to simultaneously ingest large its streaming performance. quantities of content. The situation multiplies itself as we add streams, services, Today’s VOD server systems adequately storage, and begin to distribute more accommodate the demands of low- hardware throughout the network. concurrency VOD/SVOD deployments. However, adding the task of ingesting Combining SVOD with VOD numerous channels of broadcast content to conventional VOD servers creates a massive Subscription VOD (SVOD) increases the hardware and software infrastructure that existing VOD content library by adding 50- takes up a lot of space, consumes a lot of 100 movies and other content and making power, and is inherently less reliable. them available to an increased number of subscribers. Even with a limited amount of content offered, trials of SVOD to date have resulted in increased concurrency rates that may be as high as 10%-20% or 3,000 to Catcher Catcher Catcher 5,000 streams in a typical system. Ingest Server These concurrency rates place tremendous demands on the streaming capacity of the network. Also as stream Tape drives counts increase, so does the problem of Content Person VOD Servers content ingest. To increase the stream count, additional streaming servers are required. These additional servers need access to the Figure 2-3 Content Ingest for VOD in the library of ingested content. If a given piece future of content is to be made available to every customer on the network, the content needs Television on Demand using Conventional to be either locally stored or remotely VOD Servers accessible. One way to make the content accessible is to add an ingest server or Now let’s look at an example where we propagation server at the point where the expand the VOD/SVOD service offerings to content is caught or loaded from tape. This include Television on Demand (TOD). TOD ingest server could then locally store the enables cable operators to provide on content, making it available to the rest of the demand delivery of live or pre-recorded servers. Alternatively the ingest server could broadcast television services as well as the be used to propagate or distribute the movie and subscription-based content that content to the streaming servers, whether VOD/SVOD offers. TOD is especially local or remote (see Figure 2-3). attractive to television content owners Remembering that the streaming servers are because it allows the viewing and sale of primarily intended for streaming, there is a older programming that is out of fixed amount of bandwidth available for syndication. TOD enables the consumer to have PVR functionality during broadcast
television viewing without requiring a hard- offerings. When VOD, SVOD, and TOD are drive in the STB. At a minimum a TOD combined a typical system may require system should be capable of storing 1,000 20,000 to 40,000 simultaneous streams. For movies for VOD/SVOD customers, plus example, using conventional VOD servers 10,000 hours of captured broadcast capable of 500 streams each would require television. 80 servers to satisfy the stream requirement. However, as more conventional VOD With Television on Demand, ingestion, servers are added, the problem of propagation, and streaming of content needs propagating the content to all the servers to occur such that the customer still feels increases exponentially and creates the need like they are watching broadcast television. for more ingest servers to propagate the In addition to the plethora of content, content so that eventually there is a trailers, posters, and rules that VOD/SVOD hierarchy of ingest servers to streaming requires, there is now a real-time servers. A conventional VOD server is requirement for low latency content designed for streaming to customers, not for ingestion. Current VOD/SVOD systems, moving, propagating and ingesting complete with catchers, tape drives, and television content. Therefore today’s VOD content ingest propagation now have to severs are not the optimum solution for this support the ingestion of broadcast television compelling, new application. feeds (see Figure 2-4). The path of the broadcast feed to the broadcast ingest server Deploying TOD with TOD servers to the ingest and propagation server to the VOD server and then to the customers is an The critical issues that must be addressed operation that will take many seconds and to adequately support TOD are content must occur at the same time as the ingest and stream count. A new class of propagation of VOD content to the VOD TOD server is required that can ingest servers. dozens of channels of broadcast television while simultaneously redistributing thousands of streams with zero-latency. The Broadcast Ingest Server associated delays can be removed by running the broadcast feeds for ingest Broadcast Feed Broadcast Feed directly into a TOD server where they can be directly streamed to customers without requiring an external hierarchy of propagation servers. This solves the content Catcher Catcher Catcher ingest and propagation problem presented by TOD. However, a hierarchical approach Ingest and to storage is also required for off-line Tape drives Propagation Servers VOD/SVOD/TOD content access. What is Content Person needed is a distributed storage strategy with VOD Servers shared local storage as well as shared remote Figure 2-4 Content Ingest for TOD using storage that decouples the streaming VOD servers functions from the storage functions. By decoupling these functions, stream-count Consumer concurrency rates for TOD and storage-size can be scaled independently will require a much higher stream count than while storage can be placed in the network the current growth projections for VOD
where it can be used in the most cost- specific applications with unique effective way. A master head-end containing requirements. a pooled storage library would allow a group of servers to access lesser-used programs Summary of Content, Streams and Ingest without requiring local copies. By using this distributed storage architecture, each type of As needs grow and new business models content can actually be moved and are introduced, the capacity and scaling of positioned in the network for the perfect VOD streaming servers are being tested. balance between hardware and transport Content libraries are increasing and greater costs. As the needs of the network change, concurrency is leading to higher and higher the placement of system components can stream counts (see Figure 2-6). With the change as well. introduction of TOD, the added ability to ingest broadcast television with low-latency is transitioning from an interesting feature into an absolute requirement.
Broadcast Feed Broadcast Feed Conventional VOD servers are being taxed to the limit with only a modest library change rate per month. As content libraries TOD Server grow, to prevent libraries from becoming Catcher Catcher Catcher ‘stale’ with old content, an increased demand is being placed on off-line ingest.
Content Storage Even now, conventional VOD servers are
Tape drives reaching their limits in being able to keep up TOD Server with SVOD and VOD applications. Content Person Regardless of how much streaming Figure 2-5 Content Ingest for TOD using requirements increase as TOD begins to TOD servers proliferate, the cable operator will be forced to add additional servers just to handle A flexible architecture that can handle ingest tasks. Even then, the resulting system low-latency live-ingest as well as pitcher- will not adequately address the problem of catcher and tape based distribution models broadcast ingest to streaming latency. The would be ideal for cost-effectively clearly superior solution is to use a new supporting TOD applications. The capability class of specialized TOD server capable of to decouple streaming from storage, while ingesting and directly streaming with no being able to distribute the storage anywhere perceivable delay. in the network, would also significantly improve the economics of TOD. With streaming positioned in one place and storage distributed throughout the network the new architecture will scale to support even the most demanding TOD applications (see Figure 2-5). The future of VOD, SVOD, and TOD are dependent on a new architecture where scale can be controlled and each environment can be tailored for
Application Movie Library Real-Time Concurrency Stream Library Change TV Ingest Rate Count VOD 150-300 15/month 0 streams 5%-10% 1,000-3,000 SVOD/VOD 200-400 40/month 0 streams 10%-20% 3,000-6,000 TOD/SVOD 1,000 100/month 100 streams 30%-65% 20,000-40,000 VOD
Figure 2-6 System Capacities for VOD, SVOD, and TOD
METADATA AND CONTROL was over, the movie would be deleted from the server. A set of rules, or metadata, Rules are needed capturing the pre-negotiated License Window Start and End Times would be read The business of broadcast television and enforced by the VOD server. today is very complex. The participants are numerous -- content owners, content As the industry moves towards SVOD aggregators, content distributors, broadcast and ultimately TOD, the same set of and cable networks, MSOs -- and the complex rules and attributes must be applied relationships between the players are to each piece of content. Examples of dynamic. What keeps content flowing from additional rules for handling television creators to consumers is the execution and content could include: enforcement of detailed contracts. These • Specific days of week when content contracts determine the rules of “how”, is available “when”, and “by whom” content may be • One or more timeslots during the day viewed. Whether it’s a re-run episode of • Time range that the program is “Friends” that airs in syndication on TBS or available on a particular day a live broadcast of the New York Knicks on • Specific commercials that must be ESPN 2, there are specific contract-based carried with the program rules that govern the manner in which • Trick-mode rules and attributes content is handled. Therefore, it should be (specific speeds, enabled/disabled no surprise that a system of contract-based functions) rules will continue to govern (and perhaps • Specific customer groups by with greater emphasis) in a business that demographic or geographic regions combines broadcast television content with on-demand content. Rules should be entered and applied as early in the process as possible. There are When VOD was initially deployed, the rules from many levels. Examples include: rules were relatively simple. MSOs would • Content owner or studio license a window of time when a movie • Studio distribution arm would be made available to its subscribers. • Content aggregator During the licensing window, the movie • would be placed on the VOD Server and be Television network • available to subscribers. After the window Local television station
• Cable MSO title, rating, description, time, actors, • Cable local unit directors and crew, category, trailer file Some of the rules apply to VOD, some to names, poster file names, etc. This type of SVOD, and some only to TOD. The key is metadata does not change, no matter who, that there are many rules that can come from what, when, or where it is distributed. This any number of places. While it can seem metadata could clearly be embedded in the daunting, it is quite easy to create and actual content file and would stay with the manage these rules. file no matter where it goes.
Partitioning Metadata 2. Rules-specific Metadata
The Video-on-Demand Content The rules-specific metadata starts at the Specification as published by CableLabs has content creation studio. The studio decides if become the de-facto standard of how there are any specific restrictions on the metadata is created and how it can distribution and sale of this content and incorporate many of the rules necessary to passes those rules along to the content describe how on-demand content is to be distributors. For example, there may be a handled. Initially written to support VOD requirement to restrict a specific category of (movies), it has been expanded to support commercial - A “Friends” episode may SVOD. Moving forward, it is likely that the require Coke commercials, but not Pepsi. specification will need to be expanded to From there, the studio distribution arm may support all forms of on-demand content, require more specific rules. “Friends” may including broadcast television. be allowed from Monday through Friday anytime, but not Thursday from 8-9 pm, to Some metadata rules pertain to the prevent intruding on first-run episodes. specific content itself, while others apply to Further downstream, the television network how that content is distributed and sold. One may decide to allow viewing anytime on piece of content from a studio can be sent to Tuesday and Wednesday because those are many cable systems across the country. If non-peak days. The local television station the studio had to regenerate the content may want to restrict viewing from 10-11pm metadata each time, it would become a during the local news hour. painful process that nobody would want to use. However, if the content specific Content Owner Distributor Cable Operator metadata were attached or imbedded in the content itself, and the distribution specific metadata was separate, then the same content with metadata attached could be sent to many locations, with a different version of the distribution metadata. Thus, the content metadata and the rules-specific Content Rules Rules metadata has been partitioned. Rules 1. Content Metadata
Content metadata includes program Figure 3-1 Rules-specific Metadata Flow specific things such as a unique identifier,
At each step along the way, the rules can it was sold. An improved solution would be become more restricted, but cannot be less to ship the exact same content to each restricted. In this manner, the content rules downstream customer, but each would be become more and more defined as they supplied with unique rules-specific metadata propagate downstream to the network which can be changed or updated at any operator and eventually the consumer (see time without requiring the entire piece of Figure 3-1). Each system along the path is content to be resent. responsible for obeying the rules imposed upstream, and can expect each system Creating Metadata downstream to obey the rules it passes on. When they reach the cable system, the TOD With the two distinct types of metadata, menu or EPG is built using these rules for appropriate software will be required to the content received. By using this approach, author and control its creation. A key the menus for the STB can be automatically ingredient is a unique identifier used to tie and dynamically constructed. the asset together with both forms of metadata. Content Owner Distribution Cable Operator 1. Content Metadata System 1 Distributor #1 System 2 The content specific metadata is created at the earliest possible point in the production and distribution chain. The best Distributor System 3 place for this is at the studio or encoding #2 System 4 Content provider. In cases where the content is broadcast television, the content metadata Distributor System 5 could originate from the television network, Rules #3 System 6 or other production company supplying the network feed. Meta-Data Content 2. Rules-specific metadata Figure 3-2 Metadata Flow to Multiple Downstream Paths The rules-specific metadata can be created and adjusted at any point in the At each step in the process, there can be production and distribution chain, but would multiple downstream paths (see Figure 3-2) typically be originated at the same point the to both multiple distributors and cable content is generated. For live television systems. For example, the studio could sell a events, the rules could and should precede “Seinfeld” episode to the WB for certain the actual content transmission. By sending nights in a specific week, and TBS on other the rules ahead first, the STB EPG can be nights. From each step facing down, the populated, or other similar guide related metadata can fragment, meaning there is a decisions can be made. one-to-many relationship at each step of the way. This is important because at each level, Propagating Metadata a seller can sell to multiple customers. However, it would be inconvenient to have Both forms of metadata need to be sent to re-record and re-master content each time along the same path as the actual content.
When any piece of content is sold or has a limited availability window. When this distributed downstream, the content type of rule is implemented, it is important metadata is included with the actual content that the system remove such content and along with an edited copy of the rules- make the storage and streaming space specific metadata. Every copy of available as quickly as possible. Another downstream content could have a unique set situation where the propagation of content is of rules-specific metadata, but the content important is when a known high- metadata would stay the same. This allows concurrency program arrives and needs to be each downstream provider to receive propagated to many places in a large different rules, and allows them to be network to facilitate the expected high changed at a later time. When the rules demand. change only the rules-specific metadata need be resent, not the content metadata or the CONCLUSION entire program content. With this approach, any distributor in the chain can revise and In this paper, we have examined how update their rules-specific metadata as conventional VOD servers are limited in necessary. their ability to ingest content and support the increasing stream requirements of TOD. Enforcing Rules-specific metadata There is a considerable impact in the output stream count as a VOD server is asked to 1. Asset Distribution ingest more content. With most existing systems, there is a non-linear loss of To make this system viable, each video streaming capability while ingesting content. server or file server along the asset Specifically, many output streams may be distribution path must receive and obey rules lost for each single stream ingested. As the encoded in the metadata. Typically in the number of titles increases in VOD libraries role of asset distribution, all that is required the problem becomes more and more is to pass-on the rules given to us. At any apparent. To reduce the impact on a VOD point in the path, the rules can be edited to server, ingest of new content can occur become more restricted, but never less after-hours. However this is just a temporary restricted. As assets are moved downstream solution and won’t scale as ingestion to the cable plant, appropriate TOD software requirements continue to increase. With the will pick-up the rules-specific metadata. The upcoming everything on demand revolution, TOD software will use this rules-based data including Television on Demand, the ingest to build the availability matrix of programs, limitation of existing VOD server and associate a local time-slot for the architectures becomes catastrophic. The consumer. The TOD server software is then more bandwidth consumed by ingest, the responsible for ensuring that the less bandwidth is available for streaming studio/distribution/network rules and functions. Therefore more servers are permissions are obeyed. required to keep the same stream count. As more servers are added, ingest and 2. Content Propagation propagation becomes more and more complex. Elaborate ingest servers with When propagating content throughout the content propagation services are a short- cable system, there can exist specific rules term solution but problematic longer term as related to perishable content, or content that
unacceptable latencies are introduced to the and carried separately from the actual distribution of broadcast television. content to allow updating as well as customization depending on the MSO and A new breed of servers designed region that the content is destined for. specifically for Television on Demand is required. These servers need to handle over A new breed of specialized, high 100 streams of live ingest while performance TOD server with low-latency simultaneously redistributing the ingested and live content ingest capabilities, plus a content to over 20,000 output streams. The new metadata methodology, is a requirement server must not suffer any performance to realize the potential of Television on degradation in output streams while Demand for cable operators. ingesting live or non-live content. The latency through such systems must be low About the Author enough to enable live television with trick- mode functionality similar to that of DVD. Robert Scheffler is Chief Architect at The streaming elements and the storage Broadbus Technologies, a provider of next- elements must be separately scalable and generation server systems that enable cable movable within the network. operators to effectively scale and migrate their networks from Video on Demand to With the plethora of ingested content Television on DemandTM – (TOD®) from VOD, SVOD, and TOD, new means for authoring and propagating metadata Robert can be reached at: must be implemented. In addition to content metadata, a new class of rules-based (978) 264.7900 metadata will be required to protect revenue [email protected] streams by allowing a rules-based http://www.broadbus.com distribution and STB presentation of content. The metadata must be partitioned
TOD and the Broadbus logo are registered trademarks of Broadbus Technologies, Inc. All rights reserved. Other trademarks used herein are property of the respected companies.
LEGAL ISSUES IN A TRUSTED DOMAINa1
by David St. John-Larkina2 edited by Jud Carya3 Cable Television Laboratories, Inc.
Abstract makes sense.3 In other words, the laws governing both worlds are entirely self- A Trusted Domain generally provides for imposed. In Alice’s world the Queen of the delivery, retention and utilization of Hearts (presumably) sets forth the law, in the copyrighted content within a secure digital world “code is law.”4 residential network. This paper attempts to identify and to address some of the key legal One key attribute of “code,” including issues of copyright law that are presented in copy control software and digital rights a Trusted Domain in the abstract sense. An management systems (DRMs), that underlies in-depth discussion of technical and business our digital world is that it is mutable. Code is issues raised by Trusted Domains is beyond not bound to follow rigid structural or the scope of this paper. architectural guidelines; rather, code is flexible and can adapt to new or changing Contrary to those commentators who circumstances. A second important attribute criticize trusted systems as parochial or of code is that it can facilitate fast limiting,1 the thesis of this paper is that the distribution of perfectly replicated Trusted Domain can (1) preserve, if not information. These attributes have led some increase, current copyright law privileges to proclaim the vision of a technological enjoyed by consumers, (2) assure content utopia, a modern Enlightenment where owners of a secure network and (3) provide individuals share information, knowledge, distributors a new product offering. and culture at the press of a button or pulse of Ultimately, the Trusted Domain may serve as light.5 Some commentators, however, offer a model for the next generation of content- that code leads to a “dystopia [where] digital related services that preserves the technology is the handmaiden of copyright expectations of consumers, and protects the infringement” and the death of copyright rights of copyright owners alike. law.6 The fear expressed by these commentators is that digital technology will I. INTRODUCTION supplant copyright law, and that owners of digital content will use code to “undermin[e] The digital world of the 21st Century is no the utilitarian balance of copyright [law] and different than Alice’s Wonderland.2 In both threaten free expression.”7 While not entirely cyberspace and the fictional land at the end of unfounded, these fears are reactionary. It is a rabbit’s hole, there exist communities that certainly true that code or digital technology do not rely upon the scientific laws of nature could be used to usurp the general provisions or real-world social and legal norms. Just as of copyright law. Conversely, code could an invisible cat makes sense in a world of strictly enforce copyright law and restrict talking playing cards, a “worm” that traditional fair use privileges that most unobtrusively embeds itself within vulnerable consumers in the digital world now assume computers to monitor suspicious activity as a right. As originally suggested by Mark Stefik, consisting of a plurality of multimedia the concept of trusted systems offers a model components (see Illustration A., below). Part for code to exercise complete control of II sets forth the general architecture of the digital content.8 The protection of digital Trusted Domain, and describes the possible content from unlawful distribution, especially range of specific characteristics a Trusted in the post-Napster age of peer-to-peer Domain may implement. Part III explains networking (e.g., KaZaA), is an important how the Trusted Domain affirms, rather than reason to implement trusted systems. annihilates, certain copyright law principles, However, the existence of a trusted system including the first sale doctrine and fair use, does not of itself eradicate the privileges and may even be used to preserve or enhance bestowed by copyright law. This paper certain privacy protections. This paper discusses a type of trusted system, called the concludes by submitting that the libertarian Trusted Domain, that can preserve, if not Trusted Domain protects digital content and increase, copyright law privileges enjoyed by ensures the continuation of copyright consumers while concurrently assuring privileges that are consistent with the content owners of a secure network. Because expectations of both content owners and trusted systems rely upon code, the Trusted consumers alike. Domain can flexibly incorporate and closely model copyright law, as well as appurtenant II. ARCHITECTURE OF THE TRUSTED copyright privileges such as fair use. DOMAIN Moreover, there is ample reason why the Trusted Domain should be crafted to model Conceptually, trusted systems consist of a real world copyright law. Simply stated, set of protocols or rules10 that govern the use, Americans love fair use–fair use privileges management and protection of copyrighted are marketable goods that increase the value material. Physically and logically the Trusted of content to the consumer.9 Domain is embodied in a network architecture11 that can include various rules This paper discusses the Trusted Domain or functions related to the use of content in as applied in the context of a home network the Trusted Domain, including the backup,
P.C.
Non-Compliant
Trusted Domain Home Writable DVD Network Head Set End MPEG RG PVR IP STB DVD Digital TV Mobile Approved Interfaces
Illustration A. conversion, distribution, playback, recording, are customizable and may be as restrictive or storage and transport of copyrighted material. unrestrictive as necessary. As discussed in The most important attribute of a trusted the following subsections, there are two system, implied by its name, is that it must fundamental paradigms for asserting content identify Trusted and Non-Trusted Devices.12 usage rules within a Trusted Domain: the A Trusted Device is an application or libertarian Trusted Domain and a rule-based electronic device capable of identifying itself Trusted Domain. and implementing the rules of the Trusted Domain. Likewise, a Non-Trusted Device is B. The Libertarian Paradigm any application or electronic device that does not identify itself or cannot implement the One embodiment of the Trusted Domain rules of the Trusted Domain. implements only a single, simple, content usage rule: within a Trusted Domain there are Any set of Trusted Devices or Non- no content or distribution rules, apart from Trusted Devices may be combined into a the requirement that content only be Trusted Domain.13 The purpose of a Trusted distributed to other Trusted Devices. This Domain is to enforce rules applicable to open or libertarian paradigm builds upon the individual Trusted and Non-Trusted Devices. assumption (set forth above) that a Trusted Importantly, a Trusted Domain must Domain is able to identify and regulate the establish, manage and enforce rules for each connection of Trusted and Non-Trusted device connected within the domain. In other Devices to the network. Subject to initial words, the set of Trusted Devices that access and authentication of Trusted Devices, comprise the Trusted Domain must establish in this simplified form, a Trusted Domain trust within the network and maintain a would eliminate the need for complicated secure means of managing the input and copy control and content encoding rules. output of content within the Trusted Within the secure network of a Trusted Domain.14 Domain, a person would then be free to use and distribute content without restriction: any A. Usage Rules content, any time, anywhere within the Trusted Domain. Content usage rules imposed upon the Trusted Domain can generally be divided into For example, a person could distribute a two categories: distribution (or transport) movie purchased for the Trusted Domain to rules and content rules. Distribution rules all Trusted Devices capable of video- enable the Trusted Domain to verify that playback that are within the Trusted content is transferred only to devices Domain.15 Instead of restricting playback of implementing the requisite security the movie to a single DVD player, a person safeguards, e.g., transfer to other Trusted could simultaneously transfer or play the Devices. Content rules enable the Trusted movie on other Trusted Devices such as a Domain to implement requisite control over personal video recorder (PVR), LCD the content that is utilized by a Trusted projector or the digital television on the front Device. The content usage rules of a Trusted of your refrigerator. Furthermore, use of a Domain are entirely self-imposed. Because movie within the libertarian Trusted Domain they generally rely upon code, the content would also be free of any copy control usage rules imposed upon Trusted Devices restrictions (e.g., copy-once, copy-never, view-only, view-once, etc.). Instead, the for example, CSS protection on a DVD, or movie could be freely consumed and used the various copy protection methods within the Trusted Domain. So long as the applicable to CDs could be enforced in the content remains within the Trusted Domain, rule-based Trusted Domain. the content can be utilized without restriction. Although existing copy protection rules For simplicity, and for the purpose of can be modeled in the Trusted Domain, it is raising and discussing general legal topics, arguable that content owners might be more this paper focuses on this “libertarian” willing to “soften” such rules within the version of a Trusted Domain–once within the Trusted Domain because they know that the Trusted Domain, content is generally Trusted Domain network is secure, and is available any time, anywhere within the limited to the Trusted Devices on the Trusted Trusted Domain. Of course, a wide variety Domain. This reasoning may especially hold of distribution and content rules, and every true in a Trusted Domain limited to the home combination thereof, could be imposed on a environment where the number of Trusted Trusted Domain system. And, different Devices is relatively small, and the audience technical or business considerations may is limited. In other words, copy protection influence a particular desired Trusted rules that apply to a particular piece of Domain. However for academic discussion, content outside the Trusted Domain may those issues are outside the scope of this differ from the copy protection rules that are paper. Section III of this paper therefore applied to the same piece of content within proceeds to address the legal significance of the Trusted Domain. The rule-based Trusted this libertarian model in greater detail. Domain paradigm offers a wide variety of options to content owners, distributors and C. The Rule-Based Paradigm consumers. As expected, the technical and business issues are also more complex. For An alternative to the libertarian Trusted simplicity, this paper focuses on the Domain is a rule-based Trusted Domain that libertarian Trusted Domain noted above. implements one or more rules to control or to However, many of the core legal issues regulate the use and distribution of content remain the same. within the Trusted Domain. In contrast to the libertarian paradigm, the rule-based paradigm III. COPYRIGHT LAW AND THE establishes a set of rules to regulate any or all TRUSTED DOMAIN of the activity within the Trusted Domain. Various functions could be made subject to The libertarian Trusted Domain paradigm such rules, including backup, conversion, (or even a rule-based Trusted Domain with distribution, playback, recording, storage and fairly lax copy protection rules) has the transport of content. Various rule-based potential to preserve in the digital domain paradigms already exist in the digital domain. two fundamental copyright law principles: For example, content “Encoding Rules” are the protection of copyrighted content, and the required when using the Digital Transmission preservation of fair use privileges. Copy Protection (DTCP) system (e.g., on a Additionally, the distribution of copyrighted 1394 digital connector). Copy protection content within this paradigm comports with schemes that exist in physical media can also the first sale doctrine by allowing the be honored or modeled in a Trusted Domain; consumer to freely distribute content to other Trusted Devices. The Trusted Domain also copyright infringement. In a world where may preserve, or even enhance, certain data can be instantaneously replicated and privacy expectations. The following transmitted, legal protection is much too subsections discuss the legal implications of slow. On the other hand, technical measures the libertarian paradigm for the protection, gain legitimacy through the law and the law use and distribution of content within the is much better equipped to sanction people Trusted Domain. who try to infringe upon copyrights. The use of technical and legal measures to protect A. The Trusted Domain as a Compliment to content therefore establishes a the Law complementary or symbiotic relationship. Some commentators downplay the The protection of copyrighted content is differences underlying this relationship and typically accomplished via ex ante or ex post suggest that code and law are substitutes in enforcement measures. Generally speaking, their protective ability.18 technical prophylactic measures protect content ex ante, whereas legal enforcement The Trusted Domain, as an ex ante measures protect content ex post.16 Technical copyright protection mechanism, is a prophylactic measures include the use of necessary and unique compliment to the legal encryption, third-party verification, device protections afforded by the Copyright Act of and user identification, self-healing software 1976 (Copyright Act) and, as amended, by and digital certificates (that may be the Digital Millennium Copyright Act embedded in silicon). Legal enforcement (DMCA).19 The Trusted Domain implements measures include the use of contract law, a flexible, but still robust and secure, copyright law, and the anti-piracy (anti- transport layer that rides on top of a network circumvention) provisions of the Digital layer (e.g., a hybrid fiber-coax cable plant).20 Millennium Copyright Act. The Trusted If history is our guide, however, it is apparent Domain, and trusted systems generally, are that technological safeguards will “probably best classified as a technical prophylactic not be 100 percent effective.”21 The Trusted measures: Domain, by implementing multiple renewable copy protection mechanisms Trusted systems . . . achieve what (enumerated above, e.g., digital certificates), copyright law achieves. But [trusted implements a corrective means of quickly systems] can achieve [copyright resolving potential security holes.22 Because protection] without the law doing the the circumvention of technological copyright restricting. [Trusted systems] present protection measures implicates the a much more fine-grained control reproduction right,23 Congress passed the over access to and use of protected DMCA as a complementary ex post legal material than law permits, and it can enforcement regime.24 Section 1201 of the do so without the aid of the law.17 DMCA prohibits the manufacture and distribution of devices (and the rendering of The general distinction between ex ante services) for the purpose of circumventing and ex post copyright protection, however, technological measures that protect against does not suggest that code and law are unauthorized access to works.25 So, Section substitutes. Ex ante enforcement must be 1201 addresses the conduct of circumventing responsive to the immediacy of potential a technological measure that protects access.26 Congress passed this ex post Domain provide yet another tool to safeguard enforcement measure because it recognized content and reinforce ex ante technical the urgency and importance of protecting content protection measures. However, as digital content: once digital content is copied, explained below, contract restrictions in the it is very easy to duplicate and distribute.27 digital world may encroach upon traditional The effect is that Section 1201 publicly first-sale concepts and thus may diminish the discourages the circumvention of copy value of content in the Trusted Domain protection measures through the threat of an without adding any more protection to the ex post application of copyright law. content than is already incurred by the use of other ex ante and ex post copy protection Another complementary ex post copyright measures. enforcement measure is provided by contract law. Generally speaking, contract provisions B. The Trusted Domain and Benefits of the governing aspects of copyrighted works are Fair Use Doctrine enforceable.28 There is, however, disagreement among courts as to the scope of The libertarian Trusted Domain may “specific contractual provisions that would preserve, if not expand, the fair use privileges otherwise be enforceable under state law.”29 enjoyed by consumers in the analog world An expansive interpretation30 of Judge and respond to the difficulty of post-sale fair Easterbrook’s opinion affirming “shrink- use valuation problems that were historically wrap” licenses in ProCD, Inc. v. Zeidenberg left unaccounted for by market pricing highlights this disagreement, and mechanisms. The Copyright Act grants distinguishes state contract rights from the copyright owners six exclusive rights, exclusive rights in the federal copyright generally enumerated as: adaptation regime: (derivative works), distribution, display, performance, reproduction and convergence Rights “equivalent to copyright” are (digital performance and transmission rights established by law–rights that rights).33 Fair use is a defense that can be restrict the options of persons who are asserted where there is infringement of one of strangers to the author. . . . A these six exclusive rights.34 The doctrine of copyright is a right against the world. fair use is highly contentious and was at one Contracts, by contrast, generally time labeled “the most troublesome doctrine affect only their parties; strangers may in the whole law of copyright.”35 At the heart do as they please, so contracts do not of the fair use doctrine is an ongoing debate create “exclusive rights.”31 about whether the doctrine is itself dependent and restricted by technology and subject to Thus, bilateral contracts, contracts that exist economic constraints imposed by the market between two parties, “may be enforced.”32 forces. This debate is stereotypically As it pertains to preventing the circumvention between copyright owners, who regard the of trusted systems, contract law thus provides fair use doctrine as an artifact of the analog or the Trusted Domain with another means of print world that should slowly recede with enforcing copyright protection measures time, and consumers, who view fair use as an beyond technical safeguards. Establishing immutable right that is necessary for contracts that define the boundaries of promulgating one of the Copyright Act’s permissible behavior within the Trusted purposes to convey copyrighted content back into the public domain.36 Copyright owners, or not.”41 As presented below, the libertarian in this generalized sense, assert that fair use Trusted Domain paradigm not only only applies where the “transactions costs recognizes these divergent positions, but associated with clearing rights sometimes presents a model much better suited to exceeded the value of the proposed use.”37 reconcile them. Consumers, alternatively, would claim that fair use is core to the principle establishing The libertarian Trusted Domain paradigm copyright laws in the first place–i.e., to fundamentally allows unrestricted use of benefit the public–and is “not merely a content within the Trusted Domain. A book matter of economics” nor of technology.38 could be paraphrased within an electronic document, a movie clip embedded within a The rule-based Trusted Domain paradigm, home-movie, or a song transformed as a means of regulating or controlling the instantaneously to play on multiple devices specific use and distribution of content, may simultaneously. Assuming the actual use perpetuate the same quandary presented in otherwise satisfies the other parameters of this fair use debate. Rule-based usage rules fair use,42 the possibilities are endless. permit the copyright owner to price the use of Another premise of the Trusted Domain, that content on a pro rata basis.39 Accordingly, all use within the Trusted Domain will not be the hypothetical copyright owners would say metered or charged, assures the consumer that the increased technological capability to that their fair uses continue unencumbered by control use piecemeal does not run contrary pro rata licensing fees and protects that to the fair use doctrine: individual’s personal content. This model, however, is also structured to protect content Fair use, [the copyright owners] by assuring copyright owners that it remains argue, defined rights in an area where solely within the network of Trusted Devices. it was not possible to meter or charge Moreover, the libertarian Trusted Domain for use. In that context, fair use set a paradigm recognizes the legal importance and default rule that parties could always the monetary value of fair use by allowing contract around. The default rule was copyright owners to set initial distribution that use was free. prices at the convenient point-of-sale entry to the Trusted Domain and thereby capture the But as the limits of what it is possible marginal costs of fair uses that were to meter and charge for changes, the previously considered a market failure (and scope of fair use changes as well. If it thus allowed free of charge).43 Certainly, becomes possible to license every Americans love fair use. Instead of aspect of use, then no aspect of use punishing or restricting fair use, the would have the protections of fair use. libertarian model markets fair use and creates Fair use, under this conception, was new business models.44 Ultimately, then, the just the space where it was too libertarian Trusted Domain paradigm allows expensive to meter use.40 consumers to continue to enjoy their fair use privileges while providing content owners a Alternatively, the hypothetical consumers convenient mechanism to set a price for fair would state that the fair use doctrine is use in a secure environment.45 “inherent in the copyright – required whether technology makes it possible to take it away C. The Trusted Domain and Preservation of The distribution of digital content, however, the First Sale Doctrine can be fundamentally different than the distribution of books or other analog media. The libertarian Trusted Domain paradigm Where distribution of the digital content itself provides copyright protection measures that necessarily requires creation of a copy prior do not preclude application of the first sale to distribution, “[S]ection 109 does not apply doctrine. The first sale doctrine relates to the to [the] digital transmission of works.”51 distribution right and is a limitation that prohibits a copyright owner from exercising The libertarian Trusted Domain paradigm control over the distribution of a tangible somewhat restores the historical and copyrighted work past the first-sale. In other distribution-specific conception of the first words, a copyright owner may attach sale doctrine, at least in spirit. Whereas a conditions on the first-sale of a copyrighted rule-based Trusted Domain may attach work (e.g., payment of a specific price) but conditions that restrict the distribution of may not thereafter condition resale or further content to certain Trusted Devices, the distribution of the tangible copyrighted work libertarian model allows distribution to all upon any criteria. The first sale doctrine is a devices within the Trusted Domain. Notably, default rule “origin[ating] in the common law to truly comply with the first sale doctrine, aversion to limiting the alienation of personal “distribution” in this sense would technically property” and policies opposing restraints of need to be a “move.” That is, in the trade.46 Codified in Section 109 of the operation of transferring content, the storage Copyright Act, the first sale doctrine heralds place of the original content would need to be back to Bobbs-Merrill Co. v. Strauss47 in deleted or rendered unusable. 52 which the U.S. Supreme Court “construed the exclusive right to [distribute] . . . as Enabling the first sale doctrine through the applicable only to the initial sale, so that libertarian Trusted Domain allows consumers absent an appropriate contractual provision, to make use of digital content no different there could be no restriction on re-sales.”48 than how analog content, or books (with respect to further distribution, not copying), In the days of the Bobbs-Merrill Co. v. are utilized in the real world. Moreover, with Strauss case (1908), the first sale doctrine the addition of a few simple content rules, was also practical to implement with respect consumers could distribute digital content to to the media of the day – books, newspapers, other Trusted Domains implementing the etc.49 Historically speaking, it was difficult same management paradigm. Thus, the or impossible to monitor further sale or entrance to the libertarian Trusted Domain distribution of such copyrighted works, or to acts as the point-of-sale to provide the collect compensation for such. However, bargained-for uses that the first sale doctrine with the advent of code, further distribution originally enabled under earlier technological of copyrighted works can be easily tracked, constraints. monitored and regulated subject to technological controls. And, in some cases, C. The Trusted Domain and Privacy such information can prevent further distribution or use of a copyrighted work, The advent of trusted systems prompted e.g., digital content marked “view-only” many commentators to reexamine the role of would also prevent any further distribution.50 privacy norms in the digital world.53 One early commentator suggested that “the established at the point of entry into the freedom to read, listen, and view selected Trusted Domain in order to enable proper materials anonymously should be considered billing and payment.60 However, once inside a right protected by the First Amendment . . the libertarian Trusted Domain, no further ..”54 The commentator also argues that the metering is required; content may be used civil and criminal enforcement provisions of anytime and anywhere within the Trusted the pre-DMCA legislation may prove Domain. This is not to say, however, that susceptible to constitutional challenge.55 consumers may not want more monitoring Trusted systems were seen as a form of within the Trusted Domain. It is foreseeable, “private legislation” that could potentially that given the option, many consumers may disrupt the balance between preservation of a wish to monitor and store information to help copyright owner’s exclusive rights and backup and restore digital works or facilitate enrichment of the public domain.56 Trusted interactive services. systems, it was argued, could potentially marginalize, if not entirely eviscerate, As such, we submit that the libertarian copyright law.57 Trusted Domain may actually preserve certain expectations of privacy, now known Contrary to these and other dire in the analog world, in the digital domain. predictions forecasting the end of copyright law, more recent commentators noted the IV. CONCLUSION practical benefits that may arise by allowing trusted systems to manage consumer As this paper sets forth, the libertarian information. For example, automated Trusted Domain paradigm protects digital information that covers the “provenance . . . content and recognizes the value of and conditions of sale or license” may preserving copyright privileges that are 58 “substantially reduce . . . transaction costs.” consistent with the expectations of both Consumers and copyright owners also may copyright owners and consumers alike. The benefit by a system that assures the apparent benefits accruing from authenticity and integrity of digital content implementation of the libertarian Trusted 59 delivered to the home. Finally, it is now Domain paradigm are numerous. technically recognized that consumer-specific information can be anonymized. Consumers receive a convenient and Anonymizing or aggregating an individual’s standardized media platform that minimizes preferences with the preferences of other confusion about how to use content. This people allows copyright owners and platform securely and transparently protects distributors to lower transaction costs, ensure content within the Trusted Domain and the authenticity and integrity digital preserves, if not expands, content usage transmissions, while also directing expectations. sufficiently targeted information to consumers (e.g., targeted advertising). Content providers may also benefit from considerably more protection and security for The libertarian Trusted Domain may the distribution of high-value digital content. preserve, or even enhance, certain expected The unrestricted nature of the libertarian privacy norms. It is recognized that some de Trusted Domain in particular increases the minimus form of metering must be value of content, and allows content providers and distributors to create flexible new business models to capture this value. Read Anonymously: A Closer Look at “Copyright Management” in Cyberspace, 28 Likewise, consumer electronics CONN. L. REV. 981, 981-82 (1996); Mark manufacturers may benefit by a network that Gimbel, Some Thoughts on the Implications offers new market opportunities for devices of Trusted Systems for Intellectual Property and standardized interfaces for compatibility. Law, 50 STAN. L. REV. 1671, 1671-72 (1998). 2 Finally, the Trusted Domain offers ROBERT LOUIS STEVENSON, ALICE’S distributors a unique competitive network ADVENTURES IN WONDERLAND (1865). In architecture for packaging and delivering some cases, there is no distinction between content into the residential home. the two worlds. See e.g., Alice in Wonderland – An Interactive Adventure!, at In summary, the libertarian Trusted http://www.ruthannzaroff.com/wonderland (last visited Mar. 9, 2003). Domain can be used to affirm copyright law 3 principles, including fair use privileges, LESSIG, supra note 1, at 17 (Lawrence establish a digital media platform that creates Lessig offers the surveillance “worm” story value to consumers, content owners, device as a means of showing how cyberspace is manufacturers, and distributors. both like and unlike real space; the forced entry of a non-invasive worm may be so unlike the real world such as not to raise a1 The opinions expressed in this paper are Constitutional eyebrows.). that of the author and editor individually, and 4 Id. at 7 (emphasis in the original). do not represent the opinions of the 5 Mark Stefik, Letting Loose the Light, in companies or industry in which they are INTERNET DREAMS: ARCHETYPES, MYTHS employed. AND METAPHORS 220-21 (Mark Stefik ed., a2 Third-Year Law Student and Production 1996). Editor, Journal on Telecommunications and 6 Gimbel, supra note 1, at 1671; John Perry High Technology Law, University of Barlow, Selling Wine Without Bottles: The Colorado School of Law. B.S., Economy of Mind on the Global Net (Dec. Environmental Studies & Social Policy, 1993) (“Copyright is dead.”), at Northwestern University, 1997; B.S., http://www.eff.org/pub/Publications/John_Pe Computer Science, magna cum laude, rry_Barlow/HTML/idea_economy_article.ht Michigan Technological University, 2002. ml. a3 Assistant General Counsel, CableLabs. 7 Gimbel, supra note 1, at 1671-72. B.S., Applied Math & Computer Science, 8 See Stefik, supra note 5, at 220. The University of Colorado, 1986; M.E. concept of trusted systems is discussed in Engineering Management, University of Part II of this paper. Colorado, 1993; J.D. cum laude, Santa Clara 9 To realize the value of maintaining fair use University, 1993. privileges, we need look no further than the 1 LAWRENCE LESSIG, CODE AND OTHER LAWS rising popularity of personal video recorders OF CYBERSPACE 136-139 (1999); Yocchai (PVRs) such as those offered by Tivo or Benkler, Free as the Air to Common Use: SonicBlue (ReplayTV) that allow television First Amendment Constraints on Enclosure shows to be digitally recorded for later- of the Public Domain, 74 N.Y.U. L. REV. viewing (“time-shifting”). 354, 418 (1999); Julie E. Cohen, A Right to
10 17 See Mark Stefik, Shifting the Possible: LESSIG, supra note 1, at 129. How Trusted Systems and Digital Property 18 For example, while Lawrence Lessig Rights Challenge Us To Rethink Digital recognizes the possibility of this symbiotic Publishing, 12 BERKELEY TECH. L.J. 137, relationship, he advocates that code currently 139 (1997). displaces law: 11 The network architecture of a trusted What copyright seeks to do using the system is comprised individually or as a threat of law and the push of norms, union of hardware and software. trusted systems do through the code. 12 See Mark Stefik, Trusted Systems, Copyright orders others to respect the SCIENTIFIC AMERICAN 78, 79 (Mar. 1997) rights of the copyright holder before (“[T]rusted computers have the capability to using his property. Trusted systems recognize another trusted system, to execute give access only if rights are respected usage rights and to render works so that they in the first place. The controls needed either cannot he [sic] copied exactly or else to regulate this access are built into carry with them a signature of their origin.”). the systems, and no users (except Discussion of the various technological tools hackers) have a choice about whether used to verify trust are beyond the scope of to obey these controls. The code this Article. Notably, the technological displaces law by codifying the rules, measures that verify trust must be renewable, making them more efficient than they revocable and robust. These characteristics were just as rules. Trusted systems in generally require the implementation of self- this scheme are an alternative for healing (automatically upgradeable) software protecting intellectual property rights and silicon or embedded encryption – a privatized alternative to law. safeguards. They need not be exclusive; there is 13 In computer science terminology, a Trusted no reason not to use both law and Domain would most likely consist of a closed trusted systems. Nevertheless, the parallel network of devices communicating code is in effect doing what the law via an encrypted challenge-response system. used to do. It implements the law’s 14 See Stefik, supra note 5, at 229. protection, through code, far more 15 As noted earlier, business models are effectively than the law did. outside the scope of this paper; but, query Id. at 130. what the appropriate price for this type of a 19 Copyright Act of 1976, 17 U.S.C. § 106 purchase for use within a Trusted Domain (1976); Digital Millennium Copyright Act, would be relative to other “traditional” Pub. L. No. 105-304, 112 Stat. 2860 (1998). purchases of the same movie, e.g., DVD 20 See Kevin Werbach, A Layered Model for movie rental, VOD. Internet Policy, 1 J. Telecomms. & High 16 Citing a former research assistant’s paper Tech. L. 37 (2002) (offering a layered-model that attempts to discern the most efficient approach to understanding vertically-related protections in cyberspace, Lawrence Lessig communications platforms); see also, suggests that the real-world analog to the ex Douglas C. Sicker & Joshua L. Mindel, ante and ex post distinction is the difference Refinements of a Layered Model for between a fence and the law. LESSIG, supra Telecommunications Policy, 1 J. Telecomms. note 1, at 122 (citing Harold Smith Reeves, & High Tech. L. 69 (2002) (suggesting that a Property in Cyberspace, 63 U. CHI. L. REV. layered model must incorporate the 761 (1996)).
technological characteristics of individual Due to the ease with which digital telecommunications platforms). works can be copied and distributed 21 U.S. COPYRIGHT OFFICE, A Report of the worldwide virtually instantaneously, Register of Copyrights Pursuant to §104 of copyright owners will hesitate to the Digital Millennium Copyright Act, 98 make their works readily available on (Aug. 2001), available at the Internet without reasonable http://www.loc.gov/copyright/reports/studies/ assurance that they will be protected dmca/dmca_study.html. against massive piracy. Legislation 22 The cost of developing sufficient copy- implementing the treaties provides control mechanisms is not to be this protection and creates the legal underestimated. Certainly the recovery of platform for launching the global high fixed development costs would be digital on-line marketplace for difficult to recover, in the short term, if copyrighted works. charged only to a single company’s S. REP. NO. 105-190, at 8 (1998). consumers. The economic rationale for 28 Selby v. New Line Cinema Corp., 96 F. development of robust security measures is Supp. 2d 1053, 1059 (C.D. Cal. 2000) (courts better realized when internalized by a have found, generally, that breach of contract standard-setting organization. claims are not preempted by § 301 of the 23 The Copyright Act provides that the Copyright Act). 29 copyright holder, or his or her agent, has the U.S. COPYRIGHT OFFICE, supra note 21 at exclusive right to “reproduce the copyrighted work 162. in copies or phonorecords. . . .” 17 U.S.C. § 30 An alternative interpretation of Judge 106 (a)(1). Easterbrook’s ProCD opinion is that it limits 24 U.S. COPYRIGHT OFFICE, supra note 21 at “shrink-wrap” or “click-through” license 98. Notably, Congress did not “prohibit the terms past the first sale, only where there is conduct of circumventing . . . [all] copy full disclosure between the contracting control measures,” as this conduct may be parties. permitted when claimed as a fair use defense, 31 86 F.3d 1447, 1454 (7th Cir. 1996) such as for library archival work. However, (citation omitted). “fair use and other copyright exceptions are 32 Id. not defenses to gaining unauthorized access 33 17 U.S.C. § 106 (1996). to a copyrighted work.” Id. at 11-12. The 34 The fair use doctrine was first articulated in analogy is clear: quoting a lawfully acquired Folsom v. Marsh, where Justice Story book may be a fair use, quoting a book stolen enumerated factors to decide issues of fair out of a safe is not. use: 25 Id. at 10. [W]e must . . . look to the nature and 26 Id. objects of the selections made, the 27 STAFF OF HOUSE COMMITTEE ON THE quantity and value of the materials JUDICIARY, 105TH CONG., SECTION-BY- used, and the degree in which the use SECTION ANALYSIS OF H.R. 2281 AS PASSED may prejudice the sale, or diminish BY THE UNITED STATES HOUSE OF the profits, or supersede the objects, REPRESENTATIVES ON AUGUST 4, 1998, 2 of the original work. . . .” (Comm. Print 1998) (Serial No. 6). The 9 F. Cas. 342 (C.C.D. Mass. 1841) (No. Senate Judiciary Committee explained that: 4901). A full examination of the history and development of the fair use doctrine is
37 complex and beyond the scope of this paper. MARSHALL LEAFFER, UNDERSTANDING It is sufficient, herein, to note that Congress COPYRIGHT LAW § 10.17[A] (1999). codified this judicial doctrine in § 107 of the 38 Id. 1976 Copyright Act, which sets forth four 39 A proponent of rule-based usage controls, factors that are determinative of fair use: Jane Ginsburg states: (1) the purpose and character As we move to an access-based world of the use, including of distribution of copyrighted works, whether such use is of a a copyright system that neglected commercial nature or is for access controls would make copyright nonprofit educational illusory, and in the long run it would purposes; disserve consumers. Access controls (2) the nature of the make it possible for authors to offer copyrighted work; end-users a variety of distinctly-priced (3) the amount and options for enjoyment of copyrighted substantiality of the portion works. Were delivery of works not used in relation to the secured, novel forms of distribution copyrighted work as a would be discouraged, and end-users whole; and would continue to be charged for all (4) the effect of the use upon uses, whatever the level in fact of the potential market for or their consumption. value of the copyrighted Jane C. Ginsburg, From Having Copies to work. Experiencing Works: the Development of an 17 U.S.C. § 107. Access Right in U.S. Copyright Law, 35 Dellar v. Samuel Goldwyn, Inc., 104 F.2d Columbia Law School, (Public Law Working 661, 662 (2d Cir. 1939) (per curiam). Paper No. 8, 2000). 36 40 U.S. CONST. art I, § 8, cl. 8 (“[T]o promote LESSIG, supra note 1, at 136 (citations the Progress of Science and useful Arts, by omitted). securing for limited Times, to Authors and 41 Id. Inventors the exclusive Right to their 42 See infra note 34. respective Writings and Discoveries.”) 43 Bell, supra note 36, at 581-84. Tom Bell (emphasis added). Another conception of the notes that technology is more effective than fair use doctrine was as a “proxy for a fair use in responding to market failure. Id. copyright owner’s implied consent.” Tom W. Where transaction costs preclude value- Bell, Fair Use v. Fared Use: The Impact of maximizing uses of copyrighted content, Automated Rights Management on “automated rights management radically Copyright’s Fair Use Doctrine, 76 N.C. L. reduces the transaction costs of licensing REV. 557, 581 (1998) (citing MELVILLE B. access to copyrighted works . . . it responds NIMMER & DAVID NIMMER, NIMMER ON to market failure.” Id. at 583. 44 COPYRIGHT § 13.05 13-151 (disapproving of Responding to Lawrence Lessig’s the notion) [hereinafter NIMMER]). Tom Bell contention that a “small [ ] number of large notes that the proxy conception of fair use companies” necessarily force consumers to fell into disfavor because it did “not explain choose restrictive usage architectures, David why fair use protects parody and other uses of G. Post poses the rhetorical question: copyrighted material that owners find [I]f there are diverse architectures of disagreeable.” Id. at 582. privacy, of identity, and of content
protection laid before the public, why 50 However, “[t]he first sale doctrine does not is it so obvious that we will end up guarantee the existence of a secondary market choosing the one(s) that deny us those [(called an aftermarket)] or a certain price for things that Lessig (and I) think are so copies of copyrighted works.” U.S. important? COPYRIGHT OFFICE, supra note 21 at 74 David G. Post, What Larry Doesn’t Get: (responding to arguments that CSS Code, Law, and Liberty in Cyberspace, 52 technology limits the resale or aftermarkets Stan. L. Rev. 1439, 1454 (2000) (citing for used DVDs). Furthermore, “[t]o the LESSIG, supra note 1, at 130.). extent that there is a concern that region 45 A “property rights [regime, allows] the coding may limit the number of purchasers producer of intellectual property [to] charge outside North America who are willing to more than marginal cost, and thus cover the buy [region-encoded] DVDs . . . that concern total cost of producing and disseminating the has nothing to do with § 1201 [of the works.” J. Frank H. Easterbrook, DMCA].” Id. at 74, n. 263. Technological Innovation & Legal Tradition: 51 Id. at 80. The U.S. Copyright Office Enduring Principles for Changing Times, 4 explains why the reproduction right takes Tex. Rev. L. & Pol. 103, 105 (1999); see precedence over the distribution right, saying: also, Guido Calabresi & A. Douglas The Supreme Court [in the Bobbs- Melamed, Property Rules, Liability Rules, Merrill decision] drew a sharp and Inalienability: One View of the distinction between the two rights, Cathedral, 85 Harv. L. Rev. 1089 (1972) creating an exception to the vending (presenting a framework for determining (i.e., distribution) right only to the whether property, liability or inalienability extent that it didn’t interfere with the rules should protect an entitlement). production right. 46 NIMMER, supra note 36 at § 8.12[A], fn 5 Id. (citation omitted) (emphasis added). In (citing Sebastian Int’l, Inc. v. Consumer fact, the Bobbs-Merrill decision explains this Contacts (PTY) Ltd., 847 F.2d 1093, 1096 purpose, noting that “a grant of control to the (3d Cir. 1988)). copyright owner over resales would not 47 210 U.S. 339 (1908). further [the] main purpose of protecting the 48 NIMMER, supra note 36, at § 8.12[B][1]. reproduction right.” Id. at 20-21 (citing 210 49 Some commentators have called for the U.S. 339 (1908) (“[T]he main purpose [of the expansion of Section 109 to include digital copyright statutes is] to secure the right of works. U.S. COPYRIGHT OFFICE, supra note multiplying copies of the work . . . .”). 21, at 40. The Office recognizes that, 52 It is noted that the DTCP encoding rules “[p]hysical copies of works in a digital allow for such “move” operations. format . . . are subject to [S]ection 109 in the 53 See e.g., Cohen, supra note 1, at 982 same way as physical copies of works in (“[T]he new copyright management analog form.” Id. at 78 (emphasis added). technologies force us to examine anew the However, this is not to say that digital sources and extent of that freedom.”); Niva distribution of the work itself—which Elkin-Koren, Copyrights in Cyberspace– generally includes copying, not just Rights Without Laws?, 73 CHI-KENT L. REV. distribution of the physical medium—is 1155 (1998). intended to be covered under the ambit of 54 Julie E. Cohen, Some Reflections On Section 109. Copyright Management Systems And Laws Designed To Protect Them, 12 BERKELEY
TECH. L.J. 161, 185 (1997) (citing Cohen, Julie E. Cohen, Copyright and the supra note 1, at 1003-30.). Jurisprudence of Self-Help, 13 BERKELEY 55 Id. TECH. L.J. 1089, 1090 (1998). 56 Id. at 163 (“[Copyright management 58 Jane C. Ginsburg, Essay–How Copyright systems] may enable both pervasive Got a Bad Name for Itself, 26 COLUM.-VLA monitoring of individual reading activity and J.L. & ARTS 61, 70 (2002) (suggesting that § comprehensive ‘private legislation’ designed 1202 of the DMCA reduces transaction costs to augment–and possibly alter beyond and increases transaction reliability). recognition–the default rules that define and 59 Id. delimit copyright owners' rights.”); Elkin- 60 Trusted Domains generally use some form Koren, supra note 53, at 1186 (The of authentication to discern Trusted and Non- “technological ability to restrict access to Trusted Devices. And, it is recognized that information . . . provide information suppliers such authentication methods need to be with an inherent advantage over users . . . .”). sensitive to First Amendment privacy 57 Echoing John Perry Barlow’s claim that concerns. General technical techniques exist “copyright is dead,” see infra note 6, Julie to ensure a base level of privacy, including Cohen suggested that: the method of anonymizing or aggregating Copyright owners . . . envision using data to ensure anonymous use. w [self-enforcing digital] contracts to secure–and redefine–their ‘informational rights.’ Within this vision of private ordering and technological self-help, contract law rather than copyright law is paramount. Limits on information ownership set by the public law of copyright are conceived as optional restrictions that can be avoided using appropriate contractual language.
MANAGING PEER-TO-PEER TRAFFIC – BEYOND DOCSIS 1.1
John R. Pickens, PhD Greg Hutterer Allot Communications, Inc.
Abstract environment is rife with subscriber-popular, bandwidth-hungry applications, each one Managing traffic rates and assuring that vying for its share of the available HFC first- networking resources are fairly offered and mile bandwidth. consumed is perhaps the penultimate requirement and opportunity for HFC cable Peer-to-Peer (P2P) file sharing in networks. Failures translate rapidly to particular has emerged as a highly popular IP customer dissatisfaction and lost revenue. technology, primarily among home users. Current mechanisms such as DOCSIS™ 1.1 The rapid rise of protocols such as KaZaA, and PacketCable™ DQOS provide strong Morpheus, and Gnutella allow virtually every traffic management functionality for select computer to become a server, freely sharing specified applications (e.g. VOIP). However enormous audio and video files at will among the rapid demand for new applications and users of an uncontrolled global community. services (e.g. peer-to-peer), coupled with the long cycle-time of specification, Community Users implementation, testing, and deployment is FastTrack 4,464,221 rapidly bringing networks to their knees. New iMesh 1,421,256 mechanisms such as in-line flow eDonkey 582,030 classification and application signature Overnet 315,592 detection enable operators to quickly DirectConnect 151,898 understand and adapt to new application Blubster 93,883 paradigms (e.g. peer-to-peer), and fulfill Gnutella 83,439 rapidly changing subscriber demand. New Figure 1 - P2P Communities tools and interfaces are needed to accelerate (www.slyck.com) service revenue and enhance customer satisfaction. The largest P2P file sharing communities have millions of users – The FastTrack BACKGROUND community has nearly 5 million users (FastTrack uses KaZaA Lite, Grokster, Few network designers anticipated the KaZaA, and iMesh clients). No ISP or MSO exponentially growing traffic levels and ever- can fail to notice the impact of Peer-to-Peer increasing, almost viral, portfolios of file sharing. applications, architectures and protocols seen in broadband networks today. In particular, P2P file sharing applications can consume the original application traffic assumptions the majority of the total bandwidth, even with driving design of HFC cable networks have only a few subscribers active, making this been vastly exceeded. The end-user cable issue a major concern for cable providers.
The following issues are increasingly APPLICATION PARADIGMS being noted by operators: Original Broadband Application Paradigm • P2P generates high traffic loads - When broadband networks were first delaying more urgent traffic of other conceived, designed, and deployed, the subscribers and negatively affecting assumed applications model was occasional customer satisfaction. web browsing and electronic mail, with infrequent activation of bandwidth hungry • P2P traffic consumes even higher applications like file transfer. This model upstream bandwidth resources. As was primarily downstream assymetrical with knowledge of each new subscriber intermittently active users. node is propogated through the network, more and more peer nodes This assumed applications model has request uploads, thus exponentially evolved to include new planned subscription increasing upstream traffic. services such as symmetrical-bandwidth cable telephony. Both deployed networks • P2P increases operator costs by and DOCSIS [1,2] and PacketCable [3,4] forcing WAN capacity upgrades and protocol standards have been carefully HFC capacity upgrades through architected to support these sorts of carefully reducing CMTS subscriber density designed applications. and splitting fiber nodes. This initial set of applications and services paradigms is server centric – web servers, An opportunity exists for MSOs to manage email servers, and VOIP soft-switch servers. the bandwidth crisis imposed by new The server model is strong in its ability to application paradigms like P2P file sharing manage the scaling of services, centralize and also reap new service revenue by offering provisioning, and support trusted strong and flexible Quality of Service authentication and authorization schemes. functions to subscribers. Not only the traditional applications (web, email) and the But, as users, and especially developers, operator applications (VOIP) can be serviced, become aware of the high-bandwidth, low- but new application paradigms such as peer latency, and always-on attributes of to peer (from KaZaA to Grid computing), broadband, new application paradigms broadcast streaming (from MPEG video to rapidly emerge. Subscribers begin adopting internet radio), and two-way voice and video the applications. Because of the bandwidth- conferencing (from PacketCable voice to intensive traffic profiles of these new Voice/Video Instant Messaging) can be applications, the existing best-effort QoS serviced as well. Also the operator can detect mechanisms fall apart. new applications as they are activated by subscribers, and can implement policies with New Broadband Application Paradigms respect to traffic prioritization or even traffic- “The media is the message” – Marshall blocking. McLuen, 1967 [5].
Client-server is not the only (or even best) model for distributed processing. As the speed of the communications link rises, works on local data without incurring latency drops, and peer-to-peer reachability any cross-network latency. The becomes pervasive, other distributed remote environment may require call- processing models become possible. outs or data-sharing facilities to access data from the invoking Peer to peer (P2P) has achieved recent environment. notoriety, especially for its seemingly unconstrained penchant for bandwidth • Remote compute cluster model – all consumption. P2P is notable in that there is file, database, and computational often no central point of control (although resources are collocated on a remote there are some hybrid P2P with super server high speed low latency network. architectures for scalability). The list of P2P What flows between the client and the applications is long, and new applications are compute cluster is keyboard/mouse coming online every day. input and screen output (bitmapped buffers or 2D/3D graphics Is the appearance of P2P a surprise? It operations). should not be. P2P is a member of a class of new distributed applications model types, • Memory mapped model – the network each of which are enabled in the new world is viewed as a logical extension of of always-on, high-bandwidth networking. paged memory, and paged-memory working set algorithms are used to There exist at least six distributed maximize locality of data. computing models, each of which is enabled by Broadband [6]: • Distributed object model – applications are structured as • Ad hoc distributed computing model communicating objects. Objects are – no specific architecture constraints. migrated by the network operating The applications developer has a system to maximize throughput and networking API at his disposal and is minimize latency. free to generate any partitioning of functions using any protocol. Some of the paradigms can be very traffic • Remote Procedure Call model – the intensive. The compute cluster model can application uses a procedure call generate an average of almost a megabit per interface. The procedure function second of downstream bandwidth for display name and all calling parameters are updates; the memory mapped model can shipped to the remote node, and the utilize the entire available bandwidth of the application waits until a procedure Broadband-enabled virtual bus. return is invoked with the return result. A recent set of initiatives called Grid computing [7] encompasses several of the • Remote Evaluation model – models outlined above. The notable shift is fragments of applications are moved that the network itself is becoming the to the remote system on which the interconnection bus of a massively parallel data is contained. The application distributed virtual computer. As the bus speed increases, and as the latency drops, • QoS on a subscriber or tiered applications are being developed that utilize service level basis. the bus bandwidth and connectivity matrix. • Usage based billing • Denial of service protection P2P is not an aberration – it is an innate reflection of the speed and latency of always- In order for packet classification services on broadband networking, and is the first of to be implemented, however, a set of filtering many bandwidth-consumptive distributed parameters – often called flows - must be applications that will be seen by MSOs. learned and populated into tables used by the in-line packet classification function. What is needed are tools and mechanisms to classify and enforce traffic in a fair Methods for Learning Flows manner, both to the MSO provider and to the There are two existing methods for revenue generating subscriber, and which can learning flows – QoS-Smart Application quickly accommodate new distributed Signaling and In-Line Flow Classification. applications and application paradigms. HFC Cable standards currently are defining interfaces for the first method. This paper identifies requirements and interfaces for the PACKET AND FLOW CLASSIFICATION second method. Both are ultimately required and both are compatible with DOCSIS in-line Packet Classification packet classification methods. The foundation of traffic enforcement is a function called Packet Classification. Packet QoS-Smart Application Signaling Classification inspects incoming packets and Judging by the existing technical work hands off the packets to the QoS enforcement within HFC cable (DOCSIS 1.1 [1], DOCSIS function (priority queueing, congestion 2.0 [2], PacketCable [3], and PacketCable management) of the packet processing DQOS [4]), there seems to be consensus that 1 engine . flow classification (learning flows) is a required element of the total QoS solution. Typically packet classification functions The current solution focus is based on the inspect source and destination addresses, port paradigm of QoS-smart applications (smart numbers, and priority fields. Some of the with respect to QoS control). In this features that operators expect from packet paradigm the application learns flows via classification include: some unspecified internal protocol mechanism. The application client and/or its • Fair and policy-driven trusted server then explicitly signals QoS oversubscription management control and authorization elements through • Monitoring and accounting standardized signaling interfaces to in-line • Class of service management networking nodes (CMTS edge router) or • QoS on an application specific basis cascaded policy servers2. • P2P awareness
2 The PacketCable Multimedia project is currently underway and is defining an expanded QoS control 1 “Packet classification” is used synonomously with model for multiple QoS-smart applications. The “QoS enforcement” for the remainder of this paper. specifications are not yet public. Figure 2 shows how the QoS-smart model windowing platform in 2002. Thus new is used within PacketCable3 VOIP telephony. applications on this platform have no formal MGCP [8] and SDP [9] protocols QoS-smart signaling interface even available. communicate QoS control information at the application layer. The VOIP application (2) In the multi-vendor application server (Call Agent) uses PacketCable DQOS developer environment a variety of signaling to communicate flow classification implementation platforms need to be parameters to the CMTS. The CMTS utilizes harmonized – versions of Windows for DOCSIS 1.1 to communicate flow clients, embedded systems with various classification parameters to the CM. RTOSs, Linux and Windows platforms for servers. Call VOIP Agent Application Application Specific (3) Even given the presence of QoS Signaling MGCP (NCS) & SDP signaling interfaces, the application LAN DQOS (COPS) developers must choose to utilize such
Cable interfaces. Modem CMTS DOCSIS IP WAN From the subscriber’s perspective new Figure 2 - PacketCable DQOS applications are appearing at an accelerating rate, and they are easy to access and install. The subscriber wishes to utilize QoS services The QoS-smart model works well for for favorite applications (a revenue those applications that are developed in opportunity for operators), but since QoS- conformance to the model. Specific targeted intelligence is lacking within the applications services such as PacketCable VOIP telephony the subscriber (and any nearby neighbor are beneficiaries of the QoS-smart style of sharing a common DOCSIS MAC domain) is learning flows. destined for an unsatisfying experience.
However, for other new or legacy In-Line Flow Classification applications the QoS-smart model incurs long The second approach for flow lead times between the specification, classification (learning flows) is in-line implementation, testing and deployment classification. In this model all control-plane phases. Other delay inducing factors include: traffic is dynamically inspected and flows are dynamically learned. (1) Today’s applications development environment consists of many independent The in-line classification engine is primed vested interests and non-collaborative with external definitions of control-plane standardization authorities. A major personal signaling mechanisms, applications, users, computer operating system vendor, for networks, and application signatures. The example, removed RSVP signaling and its engine is also primed with policy definitions associated APIs needed for application reflecting both operator and subscriber QoS development from its latest generation policy attributes which govern applications traffic management in the packet 3 PacketCable Dynamic QoS (DQOS) is simplified for classification phase. purpose of understanding. Other elements such as Record Keeping Server exist in the full architecture. Figure 3 shows the in-line flow configuration by operators and subscribers classification model. All packets received quickly implements policies for new from any port are inspected by the Flow applications. A reasonable goal is to reduce Classification function (FC) which tracks and the time for supplying QoS functionality (and maintains current state for the control flows generate revenue) for new applications from of active application instances. The Flow months (or years) to weeks (or days or Classification function creates parameters in hours). a flow description table. Typically the parameters consist of IP addresses, UDP port Flow Classification Mechanisms numbers, protocol Ids and various traffic Flow classification4 requires a number of handling policies for the flow. processing-intensive mechanisms:
1. Portless flows – the flow is simply FC defined as a combination of IP addresses, and perhaps also a protocol identifier. No real learning PC is required other than detection of PORT-1…… PORT-2…… active packet flow with timeouts.
Figure 3 - In-line flow & packet classification 2. Fixed port mapping flows – the flow is simply defined as IP The Packet Classification (PC) function addresses and port numbers. No real inspects all packets in both directions and learning is required other than categorizes each packet into a specific flow detection of active packet flow based upon the parameters stored in the flow to/from fixe addresses/ports with description table. The packets are placed into timeouts for inactivity. queues and traffic policies are implemented according to the flow definitions in the flow 3. TCP with well-known-ports – TCP description table (priority, rate-limit, packet has a standard application specific discard). method for establishing flows [ref: IANA assigned number authority]. This model is similar to the model for Inspection of TCP packets and managing virus signatures in virus checking looking for the session systems today. The difference is that the establishment commands (TCP classification functions and classification SYN packets) can identify specific definitions are maintained within the flows. The well known port number network, rather than on the end-subscriber’s identifies the application. PC or workstation. 4. Out of band control protocols – The advantage of the in-line classification many application protocols use out method is that new applications can be of band methods to communicate IP rapidly identified and given packet addresses and port numbers. All classification and QoS enforcement functionality. Rapid distribution of new 4 The term “stateful classification” is sometimes used classification parameters and subsequent and is equivalent to flow classification as used in this paper. packets of the out of band control fields, which contain application specific protocol are monitored, and IP values. There is no standard for definition addresses and port numbers are of application signatures, and in some cases extracted from the control signaling. multiple fields and Boolean expressions H.323, MGCP, and SIP are three must be computed before a specific examples of out of band control application signature is matched. protocols. An in-line flow classification function 5. Protocols that use random port must inspect all combinations of the above numbers for session setup. Many application types in control streams in order applications such as some P2P file to unambiguously differentiate application transfer applications use random flows. Once the flow is identified, then not-well-known port numbers. specific flows and policies can be defined for These are sometimes called “port use by the in-line packet classification hopping” applications. Applications function. signatures must be detected for these protocols (see below). NEW TOOLS
6. Masquerading protocols – The in-line flow classifier is a new applications masquerade as existing architectural element. Since it inspects all well-known applications protocols packets in order to classify flows, it also such as HTTP (web) and FTP (file performs packet classification functions in transfer). In some cases these are order to enforce administrator and subscriber well-intentioned methods to pass defined policies. We call this element the through firewalls without having to Classification and Enforcement Engine impose on firewall managers. For (C&EE). See [10, 11] for a specific example example, some implementations of of a C&EE and its application in managing voice/video instant messenger peer to peer traffic. services can be carried as HTTP traffic both to utilize HTTP security The C&EE is typically external to existing and to achieve firewall traversal. In network nodes (e.g. CMTS and CM), other cases they are pernicious ways although it could be implemented internal to of fooling the network into thinking a CMTS or CM if the platform has enough this is a friendly application (e.g. packet processing horsepower and is not KaZaA), and obtain preferential limited by inflexible ASIC functionality. bandwidth and access rights. Application signatures must be Application FC KaZaA… DOCSIS utilized to ferret out any of these 1.0 or PC IP IP higher sorts of flows. LAN CM CMTS C&EE WAN
For several of the application types above, application signatures are required. Figure 4 - C&EE with CMTS/CM Application signatures are defined as application-specific protocol elements The C&EE provides both flow embedded deep within packets, e.g. HTTP classification (FC) and packet classification (PC) functions. Policies can be defined and The CMTS utilizes the DOCSIS 1.1 (and enforced for all applications, including P2P above) signaling interface to communicate file transfer, even in DOCSIS 1.0 systems. relevant parameters to the packet New applications can be detected and classification function of the CM. No configured by the system administrator. application server is required in order to Database updates from the C&EE supplier support or add new applications. This can update the knowledge base of application integration is achieved using existing and protocol types with quick turnaround PacketCable DQOS signaling interfaces and times. parameter definitions.
End-to-End System Architecture Evolution Application FC KaZaA… PC DOCSIS PC Since the C&EE, CMTS, and CM are all PC 1.1 ! 2.0 IP IP C&EE capable of providing packet classification LAN CM CMTS WAN (PC) functions (both DOCSIS 1.1 and 2.0), a DQOS future opportunity exists to integrate all DQOS Proxy networking elements into a strong end-to-end Call VOIP Agent QoS management architecture. In the Application Specific integrated architecture the flow classification Signaling FC MGCP (NCS) & SDP and packet classification function of the C&EE is combined with the packet Figure 6 – C&EE with QoS-smart Server classification functions of the CMTS and CM via standard signaling interfaces. Figure 6 shows concurrent operation of Administrator and Subscriber defined traffic C&EE in-line flow classification with QoS- policies can then be concurrently applied to Smart application servers flow classification both QoS-smart applications and QoS- (e.g. PacketCable Call Agent). In this unaware applications. configuration the C&EE node proxies between the downstream CMTS nodes and Application the Policy Server (e.g. PacketCable FC KaZaA… PC PC Telephony Call Agent). In this configuration DOCSIS IP PC LAN 1.1 ! 2.0 IP both QoS-smart flow learning (e.g. for CM CMTS C&EE WAN PacketCable Telephony) and in-line flow learning can coexist. DQOS (COPS)
Some protocol and parameter extensions Figure 5 - C&EE and DQOS are likely required in order to maximize the functionality of the full end-to-end QoS Figure 5 shows a configuration for the architecture5. For example, the total integrated end-to-end system. The C&EE available bandwidth is divided between the aggregates traffic from multiple CMTS flows known by the QoS smart application platforms (and their downstream CMs and (PacketCable Telephony) and the C&EE applications). The C&EE performs flow node. Also, given the wider variety of classification (FC) and utilizes the DQOS signaling interface to communicate relevant 5 The current proposal is based upon the PacketCable parameters and policies to the packet DQOS framework. Once PacketCable Multimedia is classification (PC) function of the CMTS. published appropriate modifications can be introduced to the proposal. application types supported, some new traffic [2] CableLabs, DOCSIS™ 2.0, “Data-Over- policy types may need to be defined. Cable Service Interface Specifications, Radio Frequency Interface Specification SP-RFIv2.0-I03-021218”, 12/18/2002. CONCLUSION [3] CableLabs, “PacketCable™ 1.0 Architecture Framework Technical Report - A new approach is proposed for dealing PKT-TR-ARCH-V01-991201”, 12/1/1999. with the explosive growth of new [4] CableLabs, “PacketCable™ Dynamic applications, new application types, and Quality-of-Service Specification - PKT-SP- rising subscriber expectations for fair and DQOS-I05-021127”, 11/27/2002. configurable quality of service in DOCSIS [5] Marshall McLuhan - The Medium is the HFC networks. The approach features a new Massage. New York: Bantam Books, 1967. element which can be incrementally added to [6] Pickens, “Transport protocol revolution”, the end-to-end architecture. This element is in SNA and TCP/IP Enterprise Networking, called the Classification and Enforcement Manning Publications, 1998. Engine (C&EE) and provides in-line flow [7] Ian Foster and Adriana Iamnitchi, “On classification functions for both existing and Death, Taxes, and the Convergence of new types of application traffic. The C&EE Peer-to-Peer and Grid Computing”, is compatible with the current QoS-Smart http://iptps03.cs.berkeley.edu/final- model for flow learning in PacketCable, papers/death_taxes.pdf. DOCSIS, and emerging PacketCable [8] CableLabs, “PacketCable™ Network- Multimedia standards. Based Call Signaling Protocol Specification - PKT-SP-EC-MGCP-I06-021127”, The C&EE can be implemented in 11/27/2002. DOCSIS 1.0 systems, and, with appropriate [9] IETF, RFC 2327 “SDP: Session future extensions, can utilize the PacketCable Description Protocol”, April 1998. DQOS signaling interface to fully utilize the [10] Allot Communications Inc., packet classification functions contained “NetEnforcer Data Sheet”, Allot within the CMTS and CM for DOCSIS 1.1 Communciations, Inc., www.allot.com. (and higher) systems. [11] Allot Communications, Inc., “KaZaA v2
and other new P2P Applications Technical Using new tools like the C&EE the Note”, www.allot.com. operators and subscribers will have the ability to manage and control bandwidth utilization on an application by application basis. More importantly, operators will both retain subscribers and reap new revenue for John Pickens is a Technical Advisor to Allot managed bandwidth services. Communications, Inc. and can be contacted at [email protected].
REFERENCES Greg Hutterer is the Director of Service
Provider Sales at Allot Communications, Inc. [1] CableLabs, DOCSIS™ 1.1, “Data-Over- and can be contacted at [email protected] Cable Service Interface Specifications, Radio Frequency Interface Specification - or (952) 944-3100. SP-RFIv1.1-I09-020830”, 8/30/2002.
MATHEMATICAL MODEL OF INTERACTIVE PROGRAMMING GUIDE
Yakov Kamen Entelel Corporation
=< ∆ > Abstract , SID ,,, EETTCE DN , (1) where The classic interactive programming guide E - is an event description, (IPG) was designed over 20 years ago using a C - is a channel ID, that may include the grid data-presentation model. This design was ID channel name, number, ID, etc. perfectly suitable for a small number of T - is an event’s start time (time stamp) homogeneous video channels and a short (few - S ∆ hour-long) schedule. Today’s IPG must T - is an event’s length (in minutes) manage over 300 heterogeneous video, PPV, EN - event’s name (can be empty)
VOD, and music channels in a two week ED - event’s description (can be empty or schedule. It also has to manage time shifting can be a complex structure of (PVR) capabilities. The classic grid-based IPG different multi-resolution description was never designed to handle these tasks, and representations) has to be significantly modified to reflect this Note, that according to definition (1), when the new reality. The big question is how to modify same TV program is shown on different the IPG so that it wins consumers’ minds and channels or at different times, it is considered solves the new problems? A mathematical two different TV events. model of the IPG is necessary to make the right decision. TV Channels This article describes the first mathematical There are two types of channels in the IPG: model of the IPG based on the cognitive regular (TV) channels and special “on- information theory. Different popular IPG demand” (OD) channels. In the case of TV solutions are analyzed and compared based on channels, all events are linearly ordered by time the proposed model. and can not intersect in the time domain. In the
case of OD channels, events are not linearly
ordered by time. In this article we consider all
channels to be TV channels. IPG COMPONENTS AND STRUCTURE
Major Components of IPG TV Event Descriptions Each IPG represents schedule data differently, The Interactive programming guide (or IPG) but there are common rules that affect the allows the viewer to view and manipulate TV guide’s logical structure. For example, schedule data directly on the TV screen. The descriptions of events that have already passed schedule data can be described as a structured are never shown on the screen. The set {E} of TV event metadata or events. Each conventional IPG that follows existing rules event E consists of a channel ID that defines consists of the following components (Fig.1): the event’s channel, starting time, event length, sorting and searching control component, date event name, and event description. Formally and time, event listing (which consists of a an event E is defined as a structure: subset of TV events described by their names),
time and channel IDs, and the description of the highlighted event or dynamically updating Fourth, different viewers have different TV- help information [Kam01]. watching habits. The same person usually has different behavior patterns on working days, Highlighted Event weekends, on vacation, and on holidays. These Scaled TV Description patterns are not stable. They tend to change Event Time and Data over the years depending on health, family, and living conditions. Searching and Fifth, optimization criteria must be Sorting “computable” and verifiable. This means that a Control Events Listing criteria like “create an event listing such that all users will be happy” would not satisfy the goal Advertisement of this work. Sixth, the IPG user interface is, after all, a work of art. This means that the best and most useful Fig1. Typical Components of an IPG solution may not be the most practical or ergonomic. The Event listing is the most important With all of the considerations above, the component of the IPG, and therefore we will proposed mathematical model is based on a concentrate on its modeling and optimization. synthetic criteria C that consists of two The “surfability” of the schedule data is the separate criteria C1 and C2. The first criteria second most important component of the guide. C1, called “maximum listing information” criteria, estimates event listing information IPG OPTIMIZATION CRITERIA value. The second criteria C2, called “minimum energy surfing” criteria, estimates Criteria specification the effort a user has to exert to find a TV It is very difficult to define numerical IPG program he would like to watch or record. optimization criteria. First we define criteria C1 : First of all, there are different formidable Formally, each event name listing (event traditions of event listing presentation in listing) is a projection P of a subset of events different countries. on the screen. Each projected event in the Second, depending on subscription package listing is represented as =< ∆ > and location, different users have access to SID EPTTCEP N )(,,,)( , (2) different channel packages, and, as a result, where have different needs in an optimized IPG. For P(EN) - is a projection of the event’s name. instance a user that watches 12 public broadcast channels would be satisfied with any The criteria C1 is defined as the maximization IPG. However a user that is subscribed to 300+ criteria comparing event listings by total channels and actively uses different recording information projected on the screen: devices (PVR, DVD, VCR) is significantly PN )( = C1 EP i )(max , (3) more sensitive to the IPG’s efficiency. P ∑ Third, the description language significantly 1 where affects IPG event listing design, because a P(E ) - is a projection of the i-event to the hieroglyphic language demands a different data i screen; presentation esthetic than alphabetic N(P) - is the number of event names on the languages. screen projected by P.
Now we define criteria C2 . transmitted without interruption 24 hours per The criteria C2 is a minimization criteria that day, 7 days per week. compares different IPGs by the average energy o Channel Structure. In the current model the user has to spend to go from an event E0 to we assume that all available channels are TV an event E1. In this article we define an energy channels where the events are linearly ordered. unit as a single key press of the remote OD-channels do not exist in this model. controller. As a result the criteria C2 allows us o Channel Distribution. A real user has to find an IPG that requires a minimal average his own list of “informative” channels and number of key presses to go from one arbitrary “non-informative”, “noisy” or “garbage” event name to another. To formalize criteria C2 channels. In the model below we assume that we define a distance function R in the event all channels are equally informative. space such that R(Ei, Ej), or the number of o Event Independence. Information remote controller key presses needed to move located inside two arbitrary event descriptions the focus from event name Ei to the name Ej, is is independent, i.e. for every two events E1 and minimal. Lets assume that A(.) is an averaging E2 information I located in the pair of events is operator as it has been defined in [Kam94]. equal to the sum of information located in each The average distance between all pairs of event: I(E1 + E2 ) = I(E1) + I(E2). events we will call “the IPG surfing diameter” o Channel Independence. Information or just IPG diameter. It is a good measure of located inside two arbitrary channels is IPG surfing energy. Formally, Criteria C2 is an independent, i.e. for every two channels c1 and IPG diameter minimization criteria described c2 information I located in the pair of channels as is equal to the sum of information located in = each channel: I(c + c ) = I(c ) + I(c ). C2 EERA ji )),((min , (4) 1 2 1 2 jiR ),( o Semantic Equivalence. All descriptions where that consist of the same number of symbols A(x1,…xn) - is an average function between have equal amounts of information x1,…xn; o Event Information Equivalence. All R(x,y) - is the distance between objects x event names viewed at the same time are and y. equally important for users and consist of an In the proposed mathematical model all IPG equal amount of information solutions are measured by the criteria C1 and C2. EVENT LISTING INFORMATION
ASSUMTIONS AND CONSTRAINTS Event Listing Modeling
There are numerous event listing models. The Any mathematical model is based on a set of event listing information criteria C measures basic assumptions and constraints that allow 1 the quantity of the information not its quality. one to make non-trivial general conclusions. From C point of view all event names viewed Described below are the major assumptions 1 at the same time are equally important for a and constraints of our IPG mathematical user and consist of an equal amount of model. information I . For simplicity we set I = 1. o Homogeneous Event Value. All TV 0 0 Each event name is projected on the screen into events in the schedule have the same priority the event listing’s fixed-sized “cell”. If the cell value for all users. In the real world this is smaller than the event name, the event name assumption is not correct. is truncated and it looses some amount of o Transmission Continuity. The current information. The same video content has a model assumes that all channels are always different value to the user depending on whether it is already in progress, starting now, or will be playing in the future, because the starting time matters. Obviously, an event has the maximum value for the user if it is starting now. However it is a fairly rare case: usually events have already started (currently playing event) or will start in the future (future event). Both playing and future events have less information for the user than “starting now” events. Fig.2 Grid based listing page A major assumption of this mathematical model is that the total information value of the event listing is a sum of the information values T1 T2 T3 T4 of all projected events (2), and each projected C1 Ek(C1) Ek+1(C1) Ek+2(C1) listing event can be completely described as follows: C Ek+2(C2) 2 Ek(C2) Ek+1(C2) = II ,(s ⋅∆⋅ ITTTG ,),,()s EP CEE)( 0 c 0 ∈ ∈∆ C3 Ek+1(C3) Ek+2(C3) Ek+2(C3) I ,(s CEE 0 TTTG c ],1,0[),,(];1,0[)s (5) E (3)k(C 3) = I 0 ,1 C4 Ek(C4) where I (s , s ) - is a name value function that C E E C 5 Ek(C5) Ek+1(C5) describes the amount of information
that has been “left” in the original C6 Ek(C6) Ek+1(C6) name after the projection P;
∆ TTTG ),,( -is a time value function that C7 E (C ) Ek+1(C7) 0 c k 7 describes the information value Fig. 3 Formal model of the grid listing that the current event has compared to the information value it would have In the formal model (Fig.3) C1,…, C7 are if the event started immediately; sequential channel IDs; T1,…, T4 are standard sE -is the size of the event name; time intervals (usually 30 minutes). Ek(Ci) is the sC -is the size of the cell. name of the event that is being transmitted on 1 channel “i” during the time interval T . Ek+l(Ci) Most Popular Event Listing Models is the name of the event that will be transmitted Three examples below describe the most on channel “i” after the event with the name popular event listing organization schemes. Ek(Ci) at the “l”-step. Example1. Grid based event listing. Grid data Example2. Link-list based event listing. The representation is the most popular listing link-list listing shows the maximum number of design approach in the US. It consists of a set events of the currently highlighted channel on of time-proportional rectangular cells that are the same screen. The Link-list solution is very used to show event names. A simple example useful for digital video recording (DVR or of a grid listing is shown in Fig. 2. An abstract PVR) enabled systems. In this example we description of the grid listing page is shown on defines two link-list schemes based on wide Fig.3. and narrow cells. The first scheme (Fig. 4) is using wide cells to present event listings, the second scheme (Fig.5) is using narrow cells to C1 C 2 present event names. E (C ) t(Ek(C1)) Ek(C1) t(Ek(C2)) k 2
1 T C3 page 1 t(Ek+1(C1)) E (C ) t(Ek+1(C2)) Ek+1(C2) k+1 1 C 1 Ek(C1) Ek+1(C3) t(Ek+2(C1)) E (C ) t(Ek+2(C2)) Ek+2(C2) k+2 1 Ek+2(C3) C2 Ek(C2) t(Ek+3(C1)) E (C ) t(Ek+3(C2)) Ek+3(C2) k+3 1 C E (C ) 3 Ek(C3) k+3 3 t(Ek+4(C1)) Ek+4(C1) t(Ek+4(C2)) Ek+4(C2)
C4 E (C ) Ek(C4) k+4 3 t(Ek+5(C1)) Ek+5(C1) t(Ek+5(C2)) Ek+5(C2)
C5 Ek(C5) Ek+5(C3) t(Ek+6(C1)) Ek+6(C1) t(Ek+6(C2)) Ek+6(C2) C 6 Ek(C6) Ek+6(C3) Fig. 6 Formal model of the matrix listing C 7 Ek(C7) Ek+7(C3) In the figure above C1 and C2 are two sequential
Fig. 4 Link-list based event listing (wide cells) channel IDs; Ek(Ci) is the name of the event that is being transmitted on channel “i” during the time interval T1. E (C ) is the name of the T1 C p age T1 k+m i 9 1 event that is transmitted on channel “i” at the
C1 Ek(C1) E (C ) Ek(C1) C8 k+1 9 “m”-step; t(Ek (Ci)) is the starting time of the event “k” on the channel Ci. E (C ) Ek+2(C9) C2 k 2 Ek(C2) C9 How would one decide which name listing representation is more informative? This can C E (C ) E (C ) C 3 k 3 k+3 9 Ek(C3) 10 be done by comparing information presented on the “average” page of the listing using C4 Ek(C4) E (C ) E (C ) C11 k+4 9 k 4 formula (5) and its realization described below.
C5 Ek(C5) Ek+5(C9) Ek(C5) C12 Name Value At first glance it is beneficial to show as many C6 Ek(C6) E (C ) Ek(C6) C13 k+6 9 event names on the same listing as possible.
C7 Ek(C7) Ek+7(C9) Ek(C7) C14 However, screen space is always limited and the visible part of the name inevitably shrinks Fig.5 Link-list based event listing when new “cells” are added to the screen. (narrow cells) The name value function (name value) estimates the amount of information left in an Example 3. Event Matrix Listing. This type of event name of size sE after its projection into an data representation is popular in some event listing cell of size sC. European countries (Fig.6). Below we will define the name value as a monotonic function of two variables structure, etc. Schematically the typical s C ,1 ≥1 s function may look like the following (Fig. 8 ) E s s = C C <<<< Value )s ,(sI CEE )s f (0 ,1) a ,1 (6) s s . E E 1 s 0 C < a . s 0 E tstart -tcurrent where f(x) -is a monotonically increasing function; Fig.8 Time value a -is a threshold parameter a ∈ [0,1] that defines the average loss of For simplicity of the model we assume that information that transforms data into future events’ time value decreases with a noise. constant speed. Accepting this assumption, ∆ For simplicity we will linearly approximate f(x) 0 TTTG c ),,( can be approximated with the as: formula (8) below: 1 a s m df )( = d − d = C ,, (7) − TT − a 11 − a s c 0 E =∆ ∆T 0 TTTG c ),,( (8) As a result, formula (6) will look like the − TT following (Fig.7): k , kb = c 0 ∆T
Name Value . where 1 m -is a parameter that defines time value degradation speed for currently playing . 0 a 1 events; sC b -is a parameter that defines time value sE Fig.7 Name value function approximation degradation speed for future events. In practice, both information degradation speed According to experimental data analysis a = parameters m and b would vary per user, time 0.6 is a good approximation of the name of day, or type of equipment (PVR). Based on threshold value for English-language based US experiments we found that m belongs to the TV schedule. interval [0.2, 1] and b belongs to the interval [0.7, 0.9]. Assuming that the currently airing Time Value event’s start time is distributed uniformly we approximate (8) with: As mentioned above, the information value of an event name in the event listing depends on 5.0 m the event’s starting time. In the chosen model, TTTG ),,( =∆ −TT , (9) the time value for future events monotonically 0 c k = c 0 kb )int(, decreases over time. The time value of the ∆T currently playing event is a monotonically growing function and as a result, the longer the where event has been playing, the less value it has to int(x) -is the integer part of x. the viewer. The exact functional tie between the event Comparison of Models starting time and its time value depends on With a few additional assumptions formulas many parameters, including subjective (5)- (9) allow us to compare different models characteristics of the user, event’s genre, of event listings. In this article we compare models described in the examples above: Table3 shows that there is no event listing “Grid”, “Link-list wide”, “Link-list narrow”, model that is “the best” for all cell sizes. “Matrix wide”, and “Matrix narrow” . We will compute the average listing information values GUIDE SURFING for the four cell sizes: 8, 10, 12, and 14. For comparison we assume that the standard time Surfing Control interval T is 30 minutes, and the number of The minimum energy surfing criteria C2 would visible cells is 28 (7x4). In the model benefit IPG solutions that use a lot of special comparison process we used empirical data keys that “short cut” the most popular step collected from real US TV schedules (English sequences. But the idea of improving the language). This data includes a tabulated surfing experience by adding special keys does average name value function for each cell size, not work. First, screen space and remote called a cell power table. A fragment of the cell controller buttons are limited. Second, it is power table is presented in Table 1. impossible to convince a user to learn an “F-16 cockpit” style remote controller to surf TV in Table1. Cell power table (fragment) the dark. To make the minimal energy surfing cell size (symbols) criteria meaningful, we assume that all 8 9 10 11 12 designed models must use the same minimal Name 0.24 0.30 0.38 0.41 0.50 value set of surfing keys: up, down, left, right, select, and ten digits 0-9. It also includes a tabulated histogram of event duration distribution (see Table2). Channel and Time Distance Without limitations we would consider that the Table2. Event duration histogram distance R(Ei, Ej) between two arbitrary events Event duration (min) Percent (%) Ei and Ej, used in formula (4), is a “manhattan” metric in the channel/time coordinate space. In 1-30 56.8 31-60 20.7 other words, 61-120 10.4 C T >120 12.1 R(Ei, Ej) = R (Ei, Ej)+ R (Ei, Ej), (10)
Final comparison results are presented in Table where C 3 below R (Ei, Ej) -is the distance between the channels of events Ei and Ej ; T Table3. Models Comparison R (Ei, Ej) -is the distance between the times of Listing cell size (symbols) events Ei and Ej. type 8 10 12 16 In formula (10) RC(.) is called a channel Grid 3.98 5.07 5.84 6.83 distance and RT(.) is called a time distance. Link-list 4.09 5.64 6.40 7.58 wide Link-list 2.29 3.96 6.49 9.17 Users’ Tasks and Models narrow Users surf the IPG to solve three main tasks: Matrix 4.03 5.37 6.15 6.89 Task A. Find something to watch now. wide Task B. Find something to watch soon. Matrix 0.80 1.92 4.12 7.13 narrow Task C. Find something to record or to watch in the future. In the case of task A, time distance is equal to zero and only the channel distance has to be estimated. When the user knows the channel The channel matrix module uses screen space number, the optimal surfing solution is dialing to show the maximum number (L) channel IDs that channel number. In this case, the IPGs (in visual or textual format) on the screen diameter is equal to the average number of (L>N). The channel matrix (Fig. 7) has almost digits in a channel number. no information about the playing event names. When the desired channel number is not This means that the user has to surf inside the known but the channel name is, task A is to matrix page to check several channels before tune to the channel based on its name with the he will make his decision to switch to a minimal number of key presses. Assuming that channel. The channel diameter of an IPG that channels are uniformly distributed, and that the includes a channel matrix module can be name listing is the only surfing solution, the approximated with the formula: expected IPG diameter is approximated with C 5.0 K +≈ EERA ji )),(( L,5.0 (12) the following formula: ji ),( L = C 5.0 K +≈ where ji EERAEERA ji )),(()),(( N,5.0 (11) ji ),( ji ),( N L -is the number of channel IDs in the where channel matrix (L>N). K -is the total number of channels; N -is the number of channels visible on Note that diameter (12) is smaller than the event listing. diameter (11) only if L is closer to K than N, i.e. if N < L < 2 K- N. This means that the Function (11) achieves its minimum when matrix module would improve the surfing = KN and it is equal to N . When N is close experience only in the case of a large number to K , simple event listing scrolling is the of playing TV channels. optimal surfing method in the channel domain. Another solution is to create a new module However when the difference is large enough, that stores channel IDs alphabetically in the there are additional opportunities to minimize “notebook” style. Using the “optimal” the channel diameter by implementing multi- notebook module channel diameter can be resolution data representation modules. The decreased to: simplest idea of multi-resolution channel list C +≈ ji KEERA 2)),(( (13) representation is the idea of a “channel matrix” ji ),( (Fig.7, courtesy iSurfTV Corporation). Let us now analyze tasks B and C. Both of them require the user to surf in the time domain. Below we will compute the time diameter in both tasks B and C. As with (11) we will approximate the time diameter of the linear event listing, when surfing in the time domain, with the following formula: T 5.0 K +≈ EERA ji )),(( N,5.0 (14) ji ),( N where K -is the total length of the schedule in hours (K varies from 1 to 720 hours);
Fig.7 Channel Matrix example N -is the average period of time visible at the event listing (N varies from 0.75 to 6 hours). Formula (14) allows us to formalize the FUTURE DEVELOPMENT concept of “playing soon” events as events that would start playing in the time interval when In this article we presented the first the linear time surfing is the most optimal mathematical model of an interactive solution. Formally this time interval is defined programming guide. A lot of assumptions and 2 as the interval [TC; TC+N +N], where TC is the constrains make the usability of this model current time. limited in practice. Therefore many of these Based on the definition of “playing soon” constrains can be waved without serious events we will estimate the time dimension in complications. A model’s complication would task B based on formula (14). be compensated by its improved practicality. The optimal solution of task C is based on a The maximal event listing information and multi-resolution time representation. In this minimal energy surfing criteria also can be model we will analyze three competing generalized and improved. For instance, a implementations. channel’s probability of being watched can be The first implementation is a homogeneous added to the event listing information criteria. grid that positions days on the first dimension In the minimal energy surfing criteria we can and time intervals on the second dimension. replace the “key press”, as a measurement unit, The time diameter in this approach could be with time. approximated with the formula: Several important questions had not been T K 12++≈ discussed in this article. For example a model’s EERA ji )),(( 1 (15) ji ),( 48 N robustness to the parameter’s variation is extremely important for practical The second implementation is a set of two implementation. screens: a day listing screen, and a screen with 12 one-hour intervals. The time diameter is REFERENCES approximated by: [Kam 01] Yakov Kamen, “Components and T K +≈ EERA ji )),(( 13 (16) Advanced Functionalities of the Next ji ),( 336 Generation Interactive Programming Guide,” Proc. “The 2nd Annual Digital TV The third implementation is a set of three Application Software Environment (DASE) layers: day, part of the day (morning, day time, Symposium 2001:End-to-End Data prime time, evening, etc.), and one hour time Services, Interoperability & Applications”, intervals inside each part of the day. The time Gaithersburg, June19-20, 2001, p.307-316 diameter in this implementation is fractionally [Kam 94] Yakov Kamen and Sergei smaller than diameter (16). Ovchinnikov “Meaningful Means on Comparing formula (14), (15) and (16) we can Ordered Sets”, Proc. “Fifth International conclude that solution (14) is preferable for a Conference IPMU”, Paris, July 4-8, 1994, very short schedule (less than 2 days), solution p.1009-1013 (15) is preferable when the schedule fluctuates between 2 and 10 days and solution (16) is preferable when the schedule is longer than 10 days.
METHODS TO INCREASE BANDWIDTH UTILIZATION IN DOCSIS 2.0 SYSTEMS
Daniel Howard, Laura Hall, Keith Brawner, and Hans Hsu Broadcom Corporation Nick Hamilton-Piercy, Reynold Ramroop, and Sheng Liu Rogers Cablesystems
Abstract supported on today's cable plants, even in bands which previously could not Several methods to increase support more than QPSK. These facts bandwidth utilization in DOCSIS 2.0 are born out by the results of a field trial systems are presented, and results of a in which both 1024 QAM on the down- field trial that demonstrated several of stream and 256 QAM on the upstream the methods are reported. On the down- were demonstrated, the latter in the stream where the majority of packets presence of ingress noise on the cable transported are large packets, the use of upstream. The conclusion is that many wider bandwidth RF channels of 12 or cable plants are already capable of sup- more MHz is discussed, where statistical porting higher orders of QAM modula- multiplexing gains improve the channel tion, and thus capacity increases of up to capacity above and beyond the factor by 33% above that already provided by which the bandwidth is increased. Simi- DOCSIS 2.0 are possible. larly, going to higher order modulation on the downstream (1024 QAM) also increases the downstream capacity, and in particular can provide MSO's the INTRODUCTION ability to get more HDTV channels per 6 MHz of downstream bandwidth or 25% It is generally recognized that more data capacity in DOCSIS down- bandwidth needs of users on cable plants streams. will continue to increase as new applica- tions and higher quality versions of ex- On the upstream where small isting applications emerge. Applications data and voice packets can be the major- that increase bandwidth needs on the ity of packets transported, bandwidth downstream of cable plants include high saving techniques such as dynamic definition TV (HDTV), video-on- header suppression and synchronous demand (VOD), voice over IP (VoIP), operation reduces the packet overhead and higher speed data service. The that can account for a significant frac- growth in data rate requirements over tion of the minislots required to trans- time is still considered exponential, dou- port the packet. For medium and large bling every year at least, however the packets on the upstream, the use of trend can perhaps be better appreciated higher order modulation up to 256 QAM via applications that are known to be can be used to expand the capacity, and increasing the current data rate require- when combined with the ingress and im- ments for cable operators in particular. pulse robustness features of DOCSIS 2.0 For example, activities such as more systems, 256 QAM upstreams can be frequent and larger file transfers via email and web browsing due to imbed- schemes most vendors provide for can- ded high quality photographs, audio, and cellation of ingress. video will drive both upstream and downstream data rate requirements DOCSIS 2.0 does not address upward in the near future. In particular, capacity increases in the downstream, the transfer of video files over broad- however, and if data rate requirements band Internet connections is already part are doubling every year, then in two of current product lines for personal years the three-fold increase in upstream video recorder devices such as RePlay capacity provided by DOCSIS will be TV’s current offering [1], where the used up and additional improvements in capability to share movies over the bandwidth may be needed. Internet with friends and family is touted. Hence, there is still a need to consider techniques that increase capac- On the upstream, home based ity in both the upstream and the down- servers and peer to peer networking stream on cable plants. In this paper, applications a la Napster (and its current several techniques for increasing the look-alikes) will continue to drive capacity on cable plants are presented, bandwidth needs upward. And upstream and field tests of some of these tech- bandwidth must be provided in a more niques are also presented as proof of robust manner, since RF interference is their viability. The techniques include frequently present. When RF spectrum higher order modulation on both the below 25 MHz must be used for addi- upstream and downstream, synchronous tional upstream data channels, the modu- operation on the upstream, and dynamic lation technology must be robust to payload header suppression on the up- ingress, impulse, and higher thermal stream. noise conditions. METHODS AND BENEFITS OF IN- The upstream capacity has re- CREASING CAPACITY cently been addressed by DOCSIS 2.0 ON THE DOWNSTREAM technology, which provides significant increases in upstream capacity and Wider Channels robustness on cable plants. Compared to the DOCSIS 1.1 maximum rate of 16 On the downstream of cable QAM operation at 2.56 Mega- plants, there are two main techniques symbols/sec, DOCSIS 2.0 technology that can be used to increase the raw enables at least 64 QAM at 5.12 capacity of data services: increasing the Megasymbols/sec, an improvement of total channel width, and increasing the three times in raw capacity. DOCSIS order of modulation. It should also be 2.0 provides this improvement by pro- noted that increased video compression viding a new modulation technique, via MPEG4 is another way to get more SCDMA, which is inherently robust to channels in the same RF bandwidth, impulse noise, by increasing the robust- however in this paper we focus on the ness of TDMA via byte interleaving and media access control (MAC) and physi- increased FEC, and by the proprietary cal (PHY) layers.
Increasing the downstream chan- fic, size of the original channel, and the nel width can be done in two manners: number of channels being multiplexed. first, the symbol rate can be increased. One of the issues associated with Since cable downstreams are channel- using larger channel widths on the ized on 6 MHz spacing in North Amer- downstream is that legacy modems can- ica and 8 MHz spacing in Europe, the not use the larger channels, being limited most efficient manner to use for increas- to conventional 6 MHz channels. Thus, ing the symbol rate of DOCSIS down- the second method of increasing the sta- stream signaling would be in integer tistical multiplexing gain on DOCSIS multiples of 6 MHz (or 8 MHz in downstreams is to combine multiple Europe). Hence, the first method of downstream channels logically so that a increasing the downstream channel ca- modem can receive on multiple 6 MHz pacity is to increase the downstream channels simultaneously. This method symbol rate from about 5 Megabaud to allows future modems to access larger 10 Megabaud and the subsequent down- channels, and headend schedulers to use stream RF bandwidth to 12 MHz, or leftover capacity in the individual chan- twice the current 6 MHz RF bandwidth. nels more effectively, but at the same time permits legacy modems to continue Note that doubling the channel to use the individual 6 MHz channels. A width does not necessarily increase the version of this technique was described spectral efficiency of the transmissions in a recent NCTA/SCTE paper by ATT since the alpha factor used in symbol [3], where 40% gains in channel utiliza- shaping may remain constant. Even tion from statistical multiplexing were though the guard band in between the shown for combinations of four down- two individual channels is removed, stream channels. more guard band on the edges of the sig- nal spectrum is required in terms of Hz Higher Order Modulation for the same value of alpha when a lar- ger symbol rate is used. Since doubling Wider downstream channels will the channel width does not increase the clearly provide some level of increased spectral efficiency, the main benefit of performance on the downstream, how- this technique lies in the additional sta- ever traffic variation and the ratio of new tistical multiplexing gain that comes to legacy modems cause statistical mul- from wider channel widths. Essentially, tiplexing gains to be variable and diffi- leftover capacity in each of the individ- cult to predict for multiple and/or wider ual channels from gaps in scheduling downstream channels. Another tech- transmissions can be combined and used nique, which gives clear and predictable for additional transmissions in the single, gain in the downstream channel, is the wider channel. Further, latency can be use of higher order modulation, for reduced by exploiting earlier opportuni- example 512 QAM or even 1024 QAM. ties to transmit in the combined channel In the latter case, the spectral efficiency instead of waiting for opportunities in a is increased from 8 bits/symbol (256 single, smaller channel. Estimates of QAM) to 10 bit/symbol, an increase of statistical multiplexing gain vary from 25%. Further, by increasing the channel 10% to 40% [2], depending on the traf- capacity directly, there will also result a statistical multiplexing gain in that channel. Figure 1 depicts the received below analog levels. But if the noise constellation of a 1024 QAM down- floor of the plant was insufficient, stream signal. clearly when all analog channels are replaced by digital carriers, the resulting digital carriers can be transmitted at higher levels due to the laser power margin freed up by removing the analog channels. And if only a subset of analog channels were boosted in order to sup- port higher order modulation, the effect on overall plant balancing would be minimal.
The CTB and CSO issues can be dealt with using many of the techniques described in [4], examples of which in- clude additional interleaving and/or cod- ing, better equalization, and also offset- ting higher order QAM frequencies to Figure 1. 1024 QAM DS Signal avoid the strongest CTB and CSO ‘tones’. But higher order modulation on the downstream has been criticized in The fact that modern transmitter the past based on the difficulty in relia- and receiver technology has mitigated bly operating even 256 QAM on the many of these issues is born out by a downstream [4]. In particular, the crest- field test of higher order QAM on the ing of CTB and CSO on the downstream downstream, described in a subsequent and higher SNR requirements are often section below. Further tests are planned. quoted as factors that could limit or pre- vent the successful operation of higher Benefits of Downstream Improvements order modulation on the downstream. While 25% improvement from For the SNR issues (which also higher order QAM and 10-40% from arise in using double- and quadruple- statistical multiplexing may not sound wide downstream channels), it may be like drastic improvements in down- noted that in current cable downstreams, stream capacity, consider the benefits for the transmit power of digital channels is an application that is currently in the backed off from the level used by analog headlines for cable operators: HDTV. channels, by up to 10 dB. The reason is Currently in a 6 MHz downstream chan- that analog TV requires an SNR of up to nel using 256 QAM, cable operators can 46 dB for high quality video, while a deliver 2 High Definition (HD) channels practical digital TV receiver requires without degradation, 3 HD channels only about 30 dB of SNR. In going to using statistical multiplexing and allow- higher order modulation such as 1024 ing a slight degradation in quality. If the QAM, another 6 dB of SNR would channel width were doubled to 12 MHz, likely be required, which is still 10 dB then a 40% statistical multiplexing gain would mean that 5 channels could be The notion of HD video on demand be- delivered in 12 MHz with no degrada- comes quite viable under such scenarios. tion, and a 24 MHz wide downstream channel could deliver 11 HD channels. METHODS FOR INCREASING UP- But now consider increasing the STREAM CAPACITY order of QAM to 1024 on the down- stream. The additional 25% raw capac- Higher Order QAM ity plus an additional statistical multi- plexing gain leads to 15-16 HD channels Higher order QAM can also be in 24 MHz of RF bandwidth. This trans- used on cable upstreams, which means lates to 4 HD channels per 6 MHz of RF greater than 64 QAM TDMA or 128 bandwidth with no degradation, or dou- QAM/TCM SCDMA can be transmitted. ble the current number of channels. And If for example, 256 QAM TDMA is used this doubling would also apply roughly on the upstream, up to 33% additional to DOCSIS data downstreams, where capacity is provided by using 8 bits per users could double current download symbol instead of 6. And this additional speeds as an effective counter to compe- capacity can be provided in a completely tition, or as a means of attracting small compatible manner with existing, legacy to medium businesses to cable modem cable modems, since the burst nature of service. upstream transmissions means that higher order QAM transmissions can be Note that as mentioned earlier, mixed with lower order QAM in the additional compression technologies same manner as DOCSIS 2.0 transmis- such as MPEG AVC (also termed sions are mixed with DOCSIS 1.x MPEG4 part 10 or ITU H.264) will fur- transmissions. ther improve the bandwidth utilization of digital video. High definition video But the upstream must be robust using AVC is projected to use between 3 to ingress, impulse, and thermal noise and 7 Mbps, depending on the content conditions. As it turns out, most CMTS type, with sports programming being one vendors have included some form of of the more difficult types. By compari- proprietary ingress cancellation process- son, current MPEG-2 HD transmissions ing in their designs, and the addition of typically use approximately 18 Mbps for SCDMA and TCM to the DOCSIS 2.0 the video stream. Higher data rates can specification significantly improves the be used with AVC for even better qual- robustness to impulse noise as well as ity, and lower date rates can be used providing several dB more robustness to where some degradation is acceptable thermal noise. and for content that is relatively easy to compress. As a result, in general ap- Consequently, higher order QAM proximately 2.5 to 3 times as many AVC turns out to be quite viable even on to- HD video streams can fit into the data day’s upstreams. In the next section, rate previously occupied by MPEG-2 field tests of higher order QAM on the HD. Combining this with the previous upstream, including 256 QAM TDMA, example of 1024 QAM/quad channels, are presented, where the higher order this would translate into 10-12 HD QAM was operated reliably even in the channels per 6 MHz of RF bandwidth. presence of three ingress signals. More Efficient Small Packets But the packet payload itself can also be reduced. A technique known as The burst nature of upstream dynamic payload header suppression transmissions leads to variation in the (DPHS), which extends the current fixed benefit of higher order QAM on the payload header suppression scheme of upstream, however. Since longer pre- DOCSIS in a simple manner, can be ambles are typically required when used to reduce the header of small pack- transmitting higher order QAM on the ets to the point where the packet dura- upstream, and the preambles of small tion is about a third of the original dura- packets can be a significant portion of tion for small data packets such as TCP overall packet duration on the upstream, ACKs. This translates into three times the benefits of higher order QAM on the bandwidth utilization of small pack- small packets can be less than 33%. ets, and when synchronous operation is Note that on medium and large packets, added, small packets can be up to 4 which account for the majority of band- times more efficient. Unlike schemes width consumed on upstreams without that only address TCP ACK packets, VOIP service, the preamble is such a DPHS also applies to other data packets small fraction of the packet duration that and to voice packets. On a data-only 256 QAM provide a 33% improvement network, where only 12% of the band- in bandwidth utilization. But especially width is consumed by small TCP ACK for small packets using the conventional packets, DPHS can provide about 12% TDMA approach, the improvement can overall network bandwidth utilization be less than 10%. improvement, while a TCP ACK-only technique would only provide about 8% Since transmitting VOIP packets capacity improvement. On a network using compressed voice will signifi- with say 50% compressed voice and cantly increase the number of small 50% data traffic, the results are 16% im- packets on the upstream, methods of im- provement for DPHS and 4% for ACK- proving the efficiency of small data and only techniques, while on an all-voice voice packets will be required. Several network, DPHS can provide up to 25% methods are available as extensions to improvement while ACK-only tech- DOCSIS 2.0. First, using synchronous niques provide no improvement. SCDMA, instead of the TDMA currently in use, which is quasi-synchronous, Thus by combining higher order permits reduction of the preamble of QAM with techniques to address small small packets without degradation in ro- voice and data packets such as synchro- bustness. The synchronous mode is nous CDMA and DPHS, the overall required to maintain code orthogonality bandwidth utilization on the upstream in SCDMA mode, but has the added can be increased by up to 33% regard- benefit of reducing the preamble over- less of packet size. Synchronous mode head on small packets. For example, a of operation and DPHS have no impact 20-30% reduction in packet size can be on ingress robustness, and both can be obtained for highly compressed voice applied to SCDMA mode in order to be packets using synchronous transport, robust to impulse noise. The fact that depending on the specific burst profile higher order QAM on the upstream is that is in use. robust to ingress is born out by field tests Further tests of higher order described in the next section. downstream modulation are in process and may be reported at the NCTA Na- HIGH-ORDER QAM FIELD TESTS tional Show.
1024 QAM Downstream Test 256 QAM Upstream Test
A test of 1024 QAM was performed on a The same prototype system was used to live cable plant in Rogers Cablesystems test high order QAM on the upstream. that was well-maintained, but nonethe- In this case, a range of upstream fre- less had measurable levels of CTB and quencies was made available for testing, CSO. The test setup is shown in Figure some of which included up to 3 ingres- 2 at the end of this paper. The 1024 sors. The range was small however, QAM transmitter was located in the hence lower symbol rates had to be used headend while the cable modem 1024 in the test in order to compare ingress QAM receiver was located in a van and free operation to operation in the pres- was connected to the cable plant in a ence of ingress. First, 64 QAM opera- residential location. There were 3 active tion (the current maximum TDMA mode components (amplifiers) between the for DOCSIS 2.0) was validated in RF fiber node and the CM 1024 QAM spectrum that was free of ingress using receiver. No degradation to existing an 800 kHz wide signal. Next, the signal adjacent 256 QAM carriers resulted was moved so that three ingressors were from the test. present and using ingress cancellation processing, 64 QAM operated reliably First, a check of 256 QAM opera- with less than 0.01% packet error rate tion was made using transmit power lev- (PER). els that were identical to those used by current digital transmitters, and no errors Next, higher order QAM was in transmitted packets were observed. tested, both with and without ingress Next, since the SNR appeared to be suf- present. 128 and 256 QAM TDMA ficient for 1024 QAM, the modulation were seen to operate reliably with less was increased to 1024 QAM using the than 0.1% PER when no ingress was same power level. Errors were detected, present. The 256 QAM signal was then and thus the transmit power level was moved to a frequency where 3 ingressors increased by 6 dB, and the system rebal- were present and with the ingress cancel- anced, with a resulting SNR of about 36 lor disabled. The PER rose to 96%, dB. The majority of errors disappeared, however when the ingress cancellor was however occasional errors were seen at engaged, the PER dropped to less than random times which thus could not be 1% PER. ascribed to CTB and/or CSO cresting as described in [4] since they were not pe- CONCLUSIONS riodic. Possible causes include hardware and software issues in the prototype sys- The deployment of high defini- tem used. tion TV will challenge cable operators to find new ways to expand their down- stream bandwidth. The techniques pre- sented here have the potential to double upstream was shown to be ready for the number of HD channels, and when deployment in today’s cable plants. combined with emerging video compres- sion schemes, can provide up to three REFERENCES times the number of HDTV channels in a 6 MHz RF downstream channel. [1] See www.replaytv.com [2] “System-level capacity and QoS in Wider channels can and have DOCSIS1.1 upstream,” C. C. Lee and been implemented in current silicon for Jérôme Bertorelle, SCTE Emerging cable technology. The fact that error- Technologies Conference, January, 2002 free operation could be achieved at 36 [3] , “FastChannel: A higher-speed dB SNR on real cable plants confirms cable data service,” B. N. Desai, Z. Ji- the viability of 1024 QAM as a down- ang, N. K. Shankaranarayanan, D. Shur, stream modulation technique as well. A. Smiljanic, T. J. Totland, J. van der Merwe and S. L. Woodward, SCTE On the upstream, higher order Emerging Technologies Conference, modulation, when combined with tech- January, 2002 niques such as dynamic payload header [4] “Statistical properties of composite suppression, provides robust and reliable distortions in HFC systems and their ef- data return service to residential custom- fects on digital channels,” R. D. ers that provides up to 33% improve- Katznelson, NCTA Cable 2002, May ment in bandwidth utilization, even on 2002 plants with ingress present on the up- stream. In particular, 256 QAM on the
Test Point Amplifier Amplifier Amplifier Fiber (Outside) Node Tap Attenuator Fiber Distribu- tive Network BCM3360 BCM93138 QAM Demodula- QAM Burst ASI BCM3034 tor Modulator RF Com- BNC QAM Modulator MPEG Video (1024 QAM) (256 QAM) biner RF- (1024 QAM) Streamer Output BMC 7020 Laptop Laptop RF Splitter Test Point HD Video PC PC
BCM93138 (Inside) Decoder QAM Burst Spectrum Laptop Demodulator TV Analyzer PC (256 QAM) Monitor
Figure 2. Field Test Setup
MODELING THE SCALING PROPERTIES OF VIDEO ON DEMAND ACCESS NETWORKS: SIMULATED TRAFFIC AND WORKLOAD ANALYSIS
Junseok Hwang, Srinivasan Nallasivan
ABSTRACT to play a major role among applications for the next-generation Internet. On the other Several different approaches –Pay per View, hand, digital broadcasting and compressed Near Video on Demand (NVOD) and Video audio/video such as DVD (MPEG2) and on Demand (VOD) have been used to Video on demand are considered high- deliver video services to customers over quality services and have become cable networks, and a variety of network increasingly popular at home. Distribution architectures have been proposed for VOD. of On-demand digital content is one of the This paper will model the performance major issues that need to be solved with a characteristics of different VOD tradeoff between bandwidth and customer architectures and pay special attention to satisfaction. Cable operators have been their scaling properties. To observe constantly battling around to provide the fundamental video stream traffic best services to their subscribers by cost characteristics and the scalability of servers efficient, scalable and suitable technology to and the transmission infrastructure, we support these high bandwidth applications. propose to perform simulation experiments This paper will discuss the scalability issues for various VOD architectures to reveal of various architectures by various which bottlenecks were the most serious. simulation experiments considering load Different VOD architectures assume balancing on the head end and suitable different locations or types of bottlenecks. server caching at the distribution end. Sensitivity analysis will be conducted by changing the values of various inputs TECHNOLOGY OVERVIEW (including technical ones such as headend locations, content distribution and streaming Architecture Studies On Performance mechanisms). Simulation is done using And Scalability different load balancing scenarios such as There are three main common architectures server load, round robin and the scalability that are being used to locate the video issues are discussed by using server caching servers and edge devices. They can be a at the local hubs. Failure mode recovery fully distributed, fully centralized and analysis is also conducted as one of the partially centralized. One of the key issues scenarios to study the fault tolerance issues that cable operators are still working through in VOD networks. is where to locate the servers in their networks. They can choose to locate the INTRODUCTION bulk of their servers in the headed or in the hubs (which are closer to homes). The Internet broadcasting and streaming contents architecture must be capable of providing a has recently been attracting a great deal of good scalable, flexible model and it should attention, despite their inadequate content also support high efficient bandwidth quality. The demand for such services is applications at a low operational cost per projected to continue to increase in the near stream. future, and streaming contents are expected Fully distributed architecture involves program transport streams (MPTS). These installation of servers at the hub. This SPTS or MPTS are mapped into approach greatly helps to reduce the ASI/ATM/Gbe/packet ring. IP-based servers transportation cost, but since there is a are used as storage servers as IP takes duplication of content in the hubs, it advantage over other servers in using increases the storage cost and it becomes MPEG-2 over IP, thereby reducing the cost difficult to manage the network thereby significantly in the system. Gigabit Ethernet increasing the operational cost. Distributed interface helps in increasing the throughput architecture involves distribution of per rack unit and the no of streams that can streaming transport in QAM/RF channels be encapsulated can be calculated as which is highly bandwidth demanding [1] available bit rate to the total bit rate and requires efficient use of bandwidth at supported by Gigabit Ethernet. [3] HFC network . b.Edge Qam Nodes Fully centralized architecture involves use QAM devices convert VOD server output of a one-server farm and other edge devices (MPEG-2) to coax channels (6-8 MHz). It at the head-end. It overcomes the drawbacks initally receives the MPEG-2 video and it of the distributed architecture by providing a re-stamps the packet that were delayed by low storage and operational cost, but it the jitter and re-routes the packet to requires high bandwidth transportation appropriate destinations. Its main role is in between headend and the hubs. This fixing up the jitter introduced due to the architecture is not suitable for larger encapsulation in the network.[2] distances exceeding 25 kilometers [2]. c.Transport Network Partially centralized architecture involves using the video servers at the headend and The video content is distributed from edge devices at the hubs. This architecture headends to hubs. The transport architecture overcomes the drawback of both centralized may be ranging from ATM or IP over Giga and distributed architecture and it can be bit ethernet or IP cloud. The resilient packet effectively utilized by using a Gigabit can also be used to provide redundancy over Ethernet backbone thereby providing long the circuit in transmitting the video traffic distance transport, increased carrying incase of a ring failure or central /remote capacity and providing a flexible node headend failure. architecture. Most cable operators have not yet decided d.Setup-Top Boxes on one scheme or the other, with many using Set-top receivers at the customers premise different schemes in different markets. acts as client nodes to VOD servers and terminate QAM signals to extract incoming Components Of Video On Demand VOD streams. a.Vod Servers VOD servers host large volumes of digital content supporting MPEG compression format. VOD servers encapsulate individual MPEG streams as single program transport mechanisms (SPTS) or into multiple RELATED WORK ON DESIGN data blocks retrieved per service round for CONSIDERATION FOR streams with lower playback rate. SCALABILITY ISSUES The Placement Scheme decides the cluster Scheduling Disk Issues size and stores video files across all clusters Real-time constraints make traditional disk and verifies that a proposed placement scheduling algorithm, such as first come scheme meets the placement requirements. It is an important factor for load balancing on servers. David Du [7] suggested that the first serve, short seek time first, and scan, placement scheme has to satisfy the inappropriate for VOD. Studies on VOD following two requirements. networks [6] suggest that two scheduling algorithms can be used for real time scheduling. It is necessary to include video data from each video file in a cluster. This is because that the types of video files requested by The best-known algorithm for real-time retrieval processes are unpredictable. Sub scheduling of tasks with deadlines is the requests within a service cycle may read earliest deadline first algorithm (EDF). The video data from any video files available in media block with the earliest deadline is the server. fetched first. The disadvantage of this algorithm is excessive seeks and poor utilization of the server's resource. [6] The continuity of data block for each video file should be maintained between clusters. All data blocks should be stored within one Under round-based algorithms, a server cluster and their corresponding next data serves all streams in units of round. During blocks should also be stored in a cluster each round, the server retrieves a certain range. If this method is not followed, after number of blocks for each stream. Since serving current sub requests, the following MPEG-2 results in variable-bit-rate requests in next service cycle will read data compressed streams, the number of blocks blocks not from a cluster range. that must be retrieved for each client in each round will vary according to the compression ratio achieved for each block. Load Balancing A simple scheme that retrieves the same Whenever a load balancer receives a packet number of blocks for each stream (generally from the client machine, it must choose an referred to as a round robin algorithm) is appropriate server to handle the request. inefficient since the maximum playback rate Load balancer will use the policy to among all streams will dictate the number of determine as which server is appropriate. blocks to read. This results in streams with Popular load balancing methods include: smaller playback rates retrieving more data blocks than needed in each round. This may Random: The load balancer chooses one of overflow some clients' buffer as well as the candidate servers at random. decrease the capacity of the server. Round-Robin: Round robin policy is a Consequently, more clients can be method of managing server congestion by accommodated by reducing the number of distributing connection loads across multiple servers. The load balancer cycles through headend and the edge devices. The the list of candidate servers depending on other alternatives could be 10 the selection weight specified. It can be Gigabit Ethernet or resilient packet classified as ring [3] • Server Load: The load balancer chooses the The end-to-end delay measured in candidate server with the lowest CPU load the network must be minimum for among all the servers efficient transport of high bandwidth streaming applications No of connections: The load balancer checks • Jitter is one of the factors affecting the server for the number of connections. the quality of service.A smaller jitter When a new request is made, the load provides with a high quality picture balancer checks and makes the connection to the customer with the server with least number of • The utilization in the network is connections. proportional to the number of connection that the server SIMULATION MODEL AND supports.The utilization must be ANALYSIS initally low in order to support multiple connections in the future This section explains the various simulation therby increasing the scalability models and the assumptions done for demands of the network. conducting our study. Model Description Requirement Assumptions The goal of this simulation is to evaluate the In order to model the VOD architecture end-to end delays of a VOD network using properly,we use the following requirement load balancing and server caching assumptions in our study. techniques.We use a ATM cloud as the backbone for the network and high speed • High-speed connection exist 1000 basex links are used to connect the betweent the headend and the hubs headend servers to the headend which helps to avoid delay thereby gateway.These are the links that carry the reducing the packet loss rate and high speed streaming MPEG-2 streams from minimum latency in the network the headend to the headend gateway. OC3 links are used to connect the headend and • Connection oriented transfer is edge devices gateway. Three IP based necessary for the timely arrival of servers were used. TheMPEG-2 streams can packets and to provide high level of be encapsulated as SPTS so that each stream QoS. can be used to transport audio,video and sent • The data stored in the servers must to any desired IP destination at the lowest be efficiently managed. This aspect cost.GBE is used as a single GBE can deals with disk scheduling, data provide a througput of 1000mbps. placement schemes and cluster management. Servers with local CISCO 12016 Gigabit switch router is used buffers are necessary for transfer of as the gateway and some of the key features data without delay and jitters. of this model are • ATM backbone is used for o An IP forwarding rate of 60,000,000 simulation analysis between the packets/sec o The router model implements a Scenario 1: Comparison of traffic in the "store and forward" type of network switching methodology. Two different architectures namely the centralized and distributed architectures Multiple GBE server outputs are aggregated were simulated for heavy traffic conditions . into one GBE output and a payload of Distributed architecture is preferred over 900Mbps is recognized out of the GbE centralized architecture in terms of all switch. The atm32_cloud node model reduced congestion and delay, as the servers represents a cloud through which ATM are located near the hubs, but management traffic can be modeled using 32 input/output of media storage is a problem. This is the physical links. Bandwidth management is only disadvantage as the servers are located highly critical in ATM networks as a delay internally at different places unlike of few milli seconds in a highly congested centralized architecture where servers are network can cause the cells to be dropped located at a single point. and lost. This results in retransmission of packets and it might compound congestion Figure 2 shows the voice traffic sent for both [6]. (Although voice and video are not the architectures. It is shown that voice retransmitted,this may cause a degradation traffic sent for distributed architecture in the performance of the enitre network.) increases linearly with time and voice traffic sent for centralized architecture remains Some of the other parameters that are being steady after reaching certain time period. used for the network are: This shows that the distributed architecture o Frame interarrival time information has high throughput relatively compared to is assumed as 15 frames/sec. centralized architecture, which is evident o Frame size information is assumed from Figure 2. as 128*240 pixels. Scenario 2: Load Balancing and Server Type of service is assumed as high quality Caching streaming multimedia. Simulataneous traffic generation is there in network until the end a. No Load Balancing of the profile.Traffic is generated exponentially using a mean factor of 30. In this scenario,no load balancing is applied on the servers and performance analysis of Simulation Scenarios the network is done and the end-to end delays are measured. The end to-to-end Various simulation experiments were delays are found to be relaitvely high when conducted and the performance of the VOD compared to other scenarios.The end-to end network under different test conditions were delay for this network using no observed.There was a comparison of the loadbalancing is around 0.2 secs on an different VOD architectures and the end-to average and it is considered to be higher end-delay, traffic of the network were compared to load balanced and server analysed under two different scenarios cached scenario. namely load balancing and Server caching.There was a failure mode analysis b.Load Balancing of the network and the effects of the load balancer on the failure mode recovery are In this scenario,a load balancer is used to discussed below. control the load acting on the three servers.There are three policies that are simulation.The traffic dropped was analyzed being used . in this scenario and it was found that there The are classified as Round Robin,Server was no packet loss in the network. Load and the no of connections.The first server is loaded with a selection weigth of CONCLUSION 10 and the other two servers are chosen with This paper provides a good insight into the a selection weight of 5.The first server is scalability isssues on VOD networks.We loaded twice as that of other two servers and discuss the scalability issues in different the end –to end-delay was found to be perspectives as how disk scheduling and greatly reduced to 0.009 secs from 0.12 secs data placement can affect the performance in a scenario that doesnot use load of VOD networks. Three alogorithms balancing. namely EDF, round-based algorithm and QMPS are discussed under the scheduling c.Server Caching disk issues. Data placement decides about In this scenario,an additional server is the admission scheduling schemes and duplicated with contents of one of the head- discusses the cluster placement issues. Our end servers and is installed at the hub,and simulation study shows the overall packet- the end-to-end delays were analysed for this end-to-end delay in the VOD network under scenario.The end-to-end delay was found to different scenarios. This shows that load increase,but the traffic in the network was balancing with server load policy can decreased considerably as there was data definitely be a good suggestion for high duplication in the hubs. performance in the network. Delay is one of the major factors in high speed network and d.Failure Recovery this shows that load balancing can certainly help in reducing the end-to-end delay of the Failure recovery analysis was conducted in network and it can also help with fault VOD networks.One of the servers was set to tolerance capabilities, if there is a problem fail while the servers are handling the high of server failure during operation in the traffic in the network.The load balancer network. Server caching can be one of the helps to isolate the server and it distributes possible solutions to reduce the traffic in the the load among the other two servers in the network, if streaming media is sent to longer networks.Towards the end of the distances,but it always has a drawback on simulation,the server recovers and it again the management and storage associated with couples with the other two servers to handle the duplication of data. the load.The end-to end delays and the traffic of the network were analyzed and the Future work graphs were obtained in Figure 6 . Resilient Packet Ring (RPR) is a new When the server three fails,the end-to end transport standard and can be used for more delay increases and its 24.9 secs at the point efficient multiprotocol transport. This of failure and it gradually decreases towards standard combines the best attributes of the end of the simulation as the server three SONET,WDM,and GbE to provide with recovers.Thetraffic in the network can be high quality of streaming content with seen as decrease during the period of failure redundant links. However this standard is at and gradually increase towards the end of the initial stages of development and it could be expected to be more expensive than the other options that were discussed.
Figure 1.Components of VOD network [1]
Figure 2 : Video on demand network - Topology
Figure 2 : Comparison of voice traffic received
Figure 3: No Load Balancing Scenario
Figure 4: Load Balancing Scenario
Figure 5: Server Caching
Figure 6 :Failure Recovery Scenario
Figure 7 :Overall Scenario Summary
REFERENCE [4] Y.Doganat and A.Tantawi, Making a Cost Effective video server, IEEE [1] Fernando Amendola and Dr.Eric Multimedia 1995 Schweitzer, Scalable VOD over DWDM Gigabit Etherent –Recent Deployment [5] Opnet Modeler and User Manuals Experiences [6] Kyung Oh Lee,Heon y.Yeom,An [2] White Paper on Network and Access Effective Admission Control Mechanism for Architecture for On-Demand Cable Variable Bit-Rate Video Streams.Seoul Television . –Harmonic Inc March 2002 National University,Korea
[3] White Paper on GigabitEthernet –over – [7] Horng-Juing Lee and David Du DWDM Transport Architetures for On- Distributed Multimedia Center: Cluster Demand Services –Harmonic Inc September Placement :A Data placement Scheme for a 2002 Mass Storage System on Storage System of Video-On-Demand Servers. MULTI-ROOM DVR: A MULTI-FACETED SOLUTION FOR CABLE OPERATORS
Dave Clark Scientific Atlanta, Inc.
Abstract a DVR. In this same survey, 81% of the respondents said that “TV is more fun than it Now that your subscribers have come to was before DVR.” Clearly, DVR is a depend on a Digital Video Recorder (DVR) convenience that changes the way people for their TV watching experience, how do they watch video content. enjoy that experience on other TVs in their home? Why limit DVR -- your subscribers’ Once consumers become accustomed to a ability to watch what they want when they DVR on their primary TV, they don’t want to want -- to just one room in the house? watch TV without DVR functionality. From that perspective, the ability to, say, start The Multi-Room DVR solution currently watching a program in the living room then being developed by Scientific-Atlanta will give watch the last half of it in the bedroom makes DVR users the ability to watch content sense. That’s one of the reasons why people recorded on their DVR set-top from other have more than one TV in their homes in the rooms in the house. As this paper illustrates, first place. Multi-Room DVR moves the such a solution will give subscribers more consumer experience of DVR to this highly flexibility in their TV viewing routine as it desired, next level. offers an attractive business case for cable operators to expand on their DVR success. ENTERTAINMENT NETWORKING
While home networking is becoming a WHY MULTI-ROOM DVR? topic of intense discussion and analysis, it is important to distinguish between the two The first question that must be answered types of services that make up “home when looking at Multi-Room DVR is, networking.” “Data networking” includes “Why?” What problem does Multi-Room multiple PC’s and high speed data. address? Isn’t DVR itself just ramping up? Is “Entertainment networking” focuses on this just a solution looking for a problem? distributed video content and music. Multi- Room DVR is an important first step toward Available research indicates that the need Entertainment networking and a successful is real. In a recent survey of people who have Multi-Room DVR solution will need to be Scientific-Atlanta’s Explorer 8000 Home able to evolve as entertainment networking Entertainment Servers in their home, 67% evolves. indicated they were ‘very interested’ in being able to access a DVR from other TVs in their Key Drivers for Entertainment Networking home. The survey also showed altered TV viewing habits with 65% saying they watch Many factors must be considered and more TV programs now than before they had addressed when evaluating a solution for entertainment networking (including Multi- ONE APPROACH TO MULTI-ROOM DVR Room DVR). The first factor to consider is whether or Scientific-Atlanta’s Multi-Room DVR not average cable subscribers can understand solution, currently under development, and use an entertainment network. Will they addresses all of these issues. From the embrace it or will it become just another beginning, the main development focus has novelty device whose appeal quickly comes been to provide a simple, low-cost solution and goes? If the consumer experience doesn’t that can be ready to release to the market in reflect a simple, easy to use solution, then the time for operators to take advantage of it. The Multi-Room DVR concept itself becomes a solution also is designed to leverage the moot point. success and strength of the deployed Explorer 8000 Home Entertainment Server platform. As important as consumers are to a successful Multi-Room DVR solution, other Ease of Use, Security, Quality and Value groups must also be considered. Content owners (e.g., programmers, movie studios, The “ease of use” issue was the first issue etc.) have a very real interest in how a to tackle in developing an effective solution. solution is implemented. They want to be This was addressed by making sure that the assured that the solution fosters a safe, secure user interface experience on the client set-tops environment that keeps content in the home, mirrored that of the DVR or server set-top. not illegally copied, for example, on the So, no matter where subscribers access Internet. They need to know that the quality of recorded content, the same, familiar look and their content will not be degraded. While feel greets them. keeping the content secure and safe, however, the system must allow the content to be able The next issue addressed was that of to be used freely by consumers, within the secure content. The goal was to leverage the confines of their own homes. cable operator’s ability to deliver secure, digitally encrypted content from the headend Cost to both consumers and cable to the subscriber’s home -- in other words, to operators is another key factor to consider. turn an Explorer 8000 Home Entertainment Are there adequate revenue opportunities to Server into a mini-headend. This enables the offset costs associated with a Multi-Room Explorer 8000 DVR set-top box to digitally DVR solution? If the cable operator transmit encrypted content over the home’s establishes a Multi-Room DVR set-up, will coaxial wiring to other (client) digital set-tops revenue generating set-tops be displaced? Or in the home. By providing a safe, secure path will the operator be able to generate additional from a DVR set-top to other set-tops within revenues throughout the home? And what the home, content providers have the same about operational support costs? Does the technology that they rely upon today when DVR solution cause an increase in calls to digital content is delivered to digital cable set- your support center? Or will the solution be top boxes. easy for consumers to embrace and operators to maintain? These are all important questions The next area to address in developing a to consider when analyzing a potential Multi-Room DVR solution was content solution. quality. We had to make sure that content would not be degraded when transmitted from room to room. The Scientific-Atlanta solution Business Case assures content integrity because the home server set-top communicates with the client The current and growing success of DVR set-tops using MPEG-2 digitally encrypted products and interest in home networking signals over a fully integrated digital network provide the subscriber momentum that will in the home. So, not only can one room watch fuel Multi-Room DVR migration. Market a recorded program from the home server’s research shows that subscribers want to access hard drive, but up to three client set-tops and DVR in multiple rooms – and they are willing the home server set-top itself can all to pay for this service. The Scientific-Atlanta simultaneously watch any recorded program. solution gives cable operators the ability to The system allows all four viewers to pause, compete with upcoming Multi-Room DVR fast forward or rewind different programs – or solutions from DBS providers. It also assures the same program -- independently, without content providers that the quality of their affecting the other viewers. content will remain intact. Combine these facts with a low-cost implementation that One of the strengths of Scientific- heavily leverages existing hardware and it is Atlanta’s Multi-Room solution is enabling easy to see that the type of Multi-Room DVR operators to use existing and deployed entry service described here offers clear value to level set-tops (e.g., the Explorer 2100) as subscribers, content providers and cable client set-tops, minimizing capital costs for operators. the operator. Older digital set-tops that might be churned back to the cable operator’s SUMMARY warehouse can now be given new life as client set-tops in a Multi-Room DVR system. This Scientific-Atlanta’s digital cable Multi- solution also provides additional revenue Room DVR solution provides the ability to opportunities through other TV’s in the home share content in multiple rooms without the because the client set-tops are fully functional expense of a hard drive at each TV. This digital set-tops, not just ‘slave’ units to a solution leverages proven digital cable home entertainment server. So, they can also technology and security, while building on the handle any applications that the cable operator success of digital cable DVR set-top has deployed (e.g., Video-on-Demand, Pay- deployments. The Multi-Room DVR system Per-View, and any others). provides a low cost, safe, secure, high-quality content delivery method to both currently deployed and previously deployed digital cable set-tops from Scientific-Atlanta. Multi- Room DVR positions cable operators to get in on the ground floor of entertainment networking opportunities. Subscribers, content providers and cable operators all stand poised to benefit.
NETWORK DESIGN FOR A MULTIPLICITY OF SERVICES
Ran Oz S. V. Vasudevan BigBand Networks, Inc.
Abstract LEGACY CABLE SERVICE METHODS Cable plants were initially built as a The architecture of the cable plant traces single-purpose infrastructure for the one- directly to its broadcast legacy. The very way broadcast of analog television. During concept of six megahertz channelization in the last ten years, in order to take advantage North America exists because that’s the of emerging service opportunities and to amount of spectrum that conventionally head off competitive threats, cable multiple carries a single analog NTSC program. system operators (MSOs) have broadened Historically, relatively few programs were their offerings to an expanding range of offered to all subscribers in a cable system. services, including digital broadcast and The only service consideration was selective high-speed data. This progress continues subscription by some subscribers to with current trends towards additional premium or pay-per-view offerings, for growing services such as HDTV and various which scrambling techniques were devised. flavors of cable telephony and video on demand (VOD). Not much architectural sophistication was required for a plant built for the singular There is an increasingly holistic purpose of pushing out relatively few realization of the cable plant as an programs to all subscribers in systems that integrated multi-service and multimedia generally were no larger than a very few infrastructure. This realization holds hundred thousand homes passed. Because promise to enhance the efficiency, the cable operators initially constructing scalability, functionality and ease of rollout these plants were generally entrepreneurial of services, through activities such as the and debt-financed, there was an imperative proliferation of open standards, sharing to optimize the economics of plant optimized resources, and balancing construction for the task at hand, which offsetting characteristics of different subsequently motivated technological services and media. innovations in selected areas such as downlinks, amplifiers and taps. The key to achieving these benefits is to evolve the cable network from a collection Besides its economic streamlining for the of isolated silos of vertically integrated task at hand, a significant historical components supporting particular services, advantage of cable networks relative to other towards more inherent fitness for multiple networks connected broadly to homes (most services and media. Greater inclusion of notably telecommunications networks), is abstraction layers allows more best-of-breed the absolute transmission capacity of the providers of particular aspects of coaxial cable which can carry so many rich functionality, including the extension of media programs. However, telecom- shared functionality across multiple media munications networks historically boast and multiple services. other distinctive aspects of functionality such as line powering, two-way To its credit, and true to its transmission, and dedicated switching of entrepreneurial roots, the cable industry has content to particular subscribers. always looked at expanding multimedia and multi-service offerings as a bona fide New services beyond analog broadcast business opportunity and not merely television have emerged on the cable competitive reaction. The earliest forays into platform over the past ten years, and interactivity and on-demand consumption go continue to do so today. Each of these back several decades. But with progress services goes through its evolutionary stages made in the satellite and of proof-of-concept, economic validation, telecommunications industries, as well as widespread implementation and scaling over the greater application viability of services time. Because in part of the fact that mere like VOD, the time has emerged for cable to functional viability must be established first, realize its multimedia, multi-service leading early proponents tend to provide potential for defensive as well as offensive complete vertical integration of the reasons. technologies required to provide such services. A byproduct of such integrations is Beyond analog broadcasting, the first the complete autonomy of that service over services that the cable industry has widely a discrete count of six megahertz channels, deployed are digital broadcasting and cable sometimes with granularity to produce Internet access. These services represent content for the same channel distinctively by directly competitive spaces with DBS and node in the hybrid fiber/coaxial (HFC) DSL (which has grown as a high speed data architecture. The fixed allocation of service, but has yet to materially provide a channels and associated resources from video alternative), respectively. Because the cable’s analog broadcast legacy is thus deployment of these two digital media extended to services that exist within their services over cable required various own dedicated silos. enhancements to the cable plant, the last ten years have seen several profound cable MULTI-SERVICE OPPORTUNITY AND industry undertakings, including the COMPETITIVE IMPERATIVE development of the HFC architecture (with its node-level multicast granularity), return From time to time, the path capability in the plant, and several telecommunications industry has touted its widely embraced digital standards such as development of digital subscriber line (DSL) MPEG-2 transport. technology as a potential delivery method for video. Also during the last decade, the Competing effectively with impressive rapid subscriber growth achieved by direct subscriber growth in both digital broadcast satellite (DBS) operators has broadcasting and cable modem service, brought on another very real threat to the cable is currently embarking upon further traditional cable business franchise. With service expansion, highlighting areas where multiple industries proposing to provide it can leverage the investments made in the broadband networked connections to plant, and where it is uniquely or best subscriber homes carrying video, data and positioned for distinction. Such services voice content, cable has had to fortify its include high definition television (HDTV) positioning in the face of emerging and various packagings of VOD including competition. movies on demand (MOD), subscription VOD (SVOD), free VOD (FOD), networked personal video recording (NPVR) and long switching capabilities to assure that the right form advertising (LFA). programs are directed to the right service groups corresponding with subscriber Consumer adoption trends of HDTV and sessions, physical transport to get the VOD offerings are encouraging. At the same content to the node (or, in the case of server time, the cable industry prepares itself for distribution at hub sites, to propagate the the widespread rollout of telephony services, content to storage for later playback), and which have achieved encouraging popularity media processing such as de-jitter, QAM in their isolated current deployments. modulation and upconversion. Furthermore, the long anticipated interactive television (iTV) space also maintains its Some of the earliest VOD offerings promise as technologies are further refined, integrated all of these elements in a solution bolstered by popular indicators such as the offered by a single vendor. This severely success of direct response advertising and limited choices by MSOs in how particular audience participation in various forms of elements of the service could be configured, reality television. cost-effective service expansion could be achieved, and how flexible headend OPTIMIZING TECHNOLOGIES architectures over time could be IN MULTIMEDIA, accomplished. In the years since the early MULTI-SERVICE ENVIRONMENT VOD deployments, there has been an increasing disaggregation of functionality, to As these trends play themselves out, the the benefit of specialists in particular cable plant that was initially constructed for elements of functionality, enabling a one-way broadcast of analog video becomes decoupled system that allows different a much more complicated beast. We vendors to provide the server, transport, and increasingly see a single, yet multi-faceted modulation functions for VOD installations. network that carries combinations of video, voice and data content; broadcast and Lack of complete standardization or personalized sessions; passive and broad protocol flexibility continues to hem interactive consumption – among other in choices. For example, supported transport variations in subscribers’ engagement with protocols are based largely on the output media and services. format of a server. But with time, operators are increasingly able to migrate from legacy While voice, video, and data may seem ASI transport to Gigabit Ethernet transport like extremely dissimilar services, the between facilities, with easier abstraction composition and delivery of these services from which VOD server is selected. And share some technical similarities. The modulation equipment at the network edge general technical elements that are will be able to accommodate for the chosen integrated to deliver a service include the method of transport, be it legacy or source of the service, the switching of the contemporary. correct content towards the correct destination, the physical transport of content RESOURCE from source to destination, and the media SHARING ADVANTAGES processing of content. The rich variety of services and media For example, a VOD service requires increasingly carried on the cable plant servers with video storage and transactional presents opportunities to leverage and billing applications as service sources, technologies that can be optimized for their offsetting as well as their common MPEG-2, it is relatively easy to reap the characteristics. Yet, the one-way broadcast advantages of dynamic allocation of legacy of the cable plant by and large bandwidth and associated resources in maintains its influence, and services are response to real-time demand. Thus a six introduced in sub optimal fashions as a megahertz channel and its QAM and result. upconverter can be carrying a heavy load of data services at the moment described One clear example of this legacy is the above, and more VOD at other times, such continuing channelization of spectrum and as weekends and prime time or particular associated resources on a service-by-service spikes for such services as when popular basis. In the same manner that the local titles first become available. Because of the affiliate of a major broadcast network is more efficient utilization, fewer overall resources are required. Alternative methods allocated a six megahertz channel, so is such of accommodating scenarios of potential a swath of spectrum is provided on a 24-by- demand peaks through overprovisioning 7 basis to approximately ten digital require exorbitant capital expenditures and broadcast programs, or VOD capacity for drive horrendous bandwidth and resource the anticipated peak consumption among utilization. approximately 100 digital cable subscribers in a node or service group, or for shared Bandwidth and resource sharing across cable modem access by several hundred cable services leverages the advantages of subscribers to that service. statistical multiplexing in which various traffic loads, unlikely to all simultaneously This means that spectrum, and its peak, ride together. The result is greater associated resources such as QAM efficiency of shared resources and less modulators and upconverters, are likelihood of denial when particular services permanently allocated to particular services spike in demand. regardless of those services’ use at particular points in time. Because resources and Consider a hypothetical situation of spectrum are generally capable of fungibility services A, B, and C which, respectively, are across services under common standards, permanently allocated channels 1-3, 4, and this situation is suboptimal unless demand 5-6 in a conventional cable plant as for services is completely static or indicated in the image on the left, in Figure correlated. 1. The bandwidth available to subscribers of a service is limited by the channels allocated For example, if a major news story to that application. The left image shows all breaks, there may be a lot of demand for applications at exactly peak capacity, high speed data and live broadcasting and however, because of the periodic shifts in relatively little VOD activity. In such a application consumption patterns (ex: night scenario, the operator runs the risk of vs. day, weekday vs. weekend) and short underallocating high speed data resources, term irregularities that occur (ex: a highly which could deny transactional revenues or anticipated report is published via Internet, a frustrate customers (resulting in churn vulnerability). At the same time, capital is popular movie becomes available on VOD), being wasted on underutilized resources in reality a fixed-allocation system cannot be dedicated to VOD. managed to achieve anything close to such high utilization of bandwidth, and Because all of this digital media traffic is fulfillment of demand. based on common standards, such as Extra Demand for services Impossible to plan channel y 6: Capacity acit
matches allocated capacity allocations for demand p - never happens patterns - typical situation 5: , and free ca free and , d nsdcapacit Unused
d 4: C All demand met
3: 3: Unmet deman 3: y 2: 6: 2: 6: B 2: 1: 4: 5: 1: 4: 5: A
A B C A B C can carry any all channels allocating bandwidth, Dynamically of deman fulfillment maximize to service 1:
Figure 1: Dynamic allocation of bandwidth across services within channels.
The middle image shows a typical Recent technological advantages have situation in which the demand for even introduced broadcast television to service A is less than half its three the pool of services dynamically sharing channels’ capacity while demand for resources. As hundreds of simultaneous service C is well above the spectrum programs are made available to allotted. This results in lost revenues, subscribers, it can be assured that not all angry customers, or both due to of them would be watched at any unfulfilled demand for service C, which particular point in time in any particular occurs at a time when there is free node. Instead of filling spectrum in all bandwidth in channels 1-3. nodes with all broadcast programs all of the time, switched broadcast techniques By implementing dynamic bandwidth can be utilized to dynamically respond to allocation, an MSO can achieve the channel surfing and switch live situation in the right image in which programming only to the nodes where it assets are more flexibly managed as all is being actively watched, allowing channels, 1-6, can carry any of the broadcast programming to share services. The demand for A, B and C is spectrum and resources with other, combined and mapped onto this one big interactive services. band. This fulfills all demand with even more capacity available, facilitating When different services and media introduction of more revenue generating share the same bandwidth and resources, services. offsetting characteristics between them can be leveraged. For example, when functional components such as QAM, video, a real time medium, shares switching, or transport by particular bandwidth with data, which is time- vendors allows these vendors to hone sensitive but not necessarily real-time, best-of-breed deliverables in their areas peaks in video traffic can be of specialty. accommodated by shifting data traffic into the troughs of video traffic. Another Isolating and aggregating particular example is combining variable bit rate elements of functionality allows these HDTV traffic with standard digital elements to be designed for maximum video, in which case rate shaping can be economic scalability. And that applied to the standard video when the scalability should be able to be achieved HDTV requires more bandwidth, which in an incremental fashion. This assures the reliable and robust delivery improvement, combined with decoupling of both services while maintaining of components, allows the operator to HDTV quality. The combinations of choose exactly how much of particular various media within the same channels elements of functionality are required. has the long term potential to enable new generations of services that richly To exemplify the alternative, when combine media, such as multiplayer systems are vertically integrated, storage games that operate in conjunction with may be combined with media processing live video programming. for some services. Thus the operator’s scaling of infrastructure is based on When plant resources in general, and whichever is the bigger driver between bandwidth in particular, are dynamically total content provided (storage and openly allocated across all services determination) or total session capacity in response to real time demand, the (processing determination), and processes of service launch and scaling whichever component doesn’t drive the become greatly alleviated. Traditionally, installation is uneconomically launching programs or services requires overprovisioned as a result. discrete determinations of which existing offerings must be stopped or reduced to Decoupling functionality also enables accommodate spectrum. With open the operator to optimize architectural allocation, new services can be layered configurations and not be constrained by on, and if capacity contention arises, this vendor designs. In VOD, as an example, can be addressed through some operators express preference for implementation of intelligent centralized server consolidation, and implementation priorities by service, some prefer to distribute servers to hub session or subscriber, or further locations at the edge. In either case it is bandwidth enhancements such as rate generally agreed that media processing shaping to adapt video bit rates. such as QAM modulation should be located at the edge, to assure quality of ARCHITECTURAL OPTIMIZATION content, with the most economic transport methodologies used to get Many advantages are reaped with less content to the edge. By having vertical integration of services and more independent QAM components that are resource sharing among them. separated from the server, this media Specialization and innovation of processing can be maintained at the edge, with servers based in either element of functionality may not be location. viable as it would necessitate a complete forklift upgrade to the system. Servers could even be based both Alternatively, specialized components centrally and distributed in a hybrid can be modified or upgraded with configuration, which may be desirable relatively minimal changes, perhaps only either to leverage existing edge servers requiring modifications in the interface while moving towards greater of existing components. centralization, or to allow a form of near-line storage so that more popular In fact, new and old components can titles to be pushed towards the edge to coexist to drive innovation while streamline transport utilization, while maintaining existing plant investments. larger libraries are centralized on For example, in VOD there is a drive economically scalable server resources. towards Gigabit Ethernet as the best Content sourced from both locations transport method, and Gigabit Ethernet share the same QAM resources, compatible QAMs are emerging to optimally placed at the edge, and utilized optimally accommodate this trend. Yet, efficiently whether the temporal demand many systems have already invested, in is relatively higher for the popular titles many cases only recently, in ASI QAMs. or the broader library content. A solution that avoids any stranded capital is to use the Gigabit Ethernet Breaking vertically integrated transport for all VOD traffic, grow services towards greater specialization capacity with Gigabit Ethernet QAMs, and abstractions between elements of and use Gigabit Ethernet to ASI protocol functionality also protects the operator conversion to prolong the operational from technological obsolescence. With life of the already-purchased legacy full vertical integration, modifying one QAMs.
Headend Hub
Next generation QAM Switch RF Central Switch Transport to right plant VOD to right infrastructure spectrum server hub Protocol Legacy Distributed conversion QAM VOD server Figure 2: Flexibility for MSO to customize architectural deployment of best-of-breed components openly interfacing to each other (VOD example).
CONCLUSION With inherently high bandwidth, increasing sophistication through trends There is increasing volume and such as migration to packetized digital variety of multimedia services available content and node-level addressability, to be provided over broadband networks. and established standards such as MPEG-2, cable has an opportunity to The key to cable realizing its multi- emerge as the leading comprehensive service potential is for MSOs to have provider of all services and all media. control over determination of deployments among open, interoperable Situations in cable systems are highly components. Abstraction should be particular due to considerations such as increasingly implemented among what services are being emphasized, sources of service, switching, transport legacy investments already made in the and media processing. This allows the networks, and overall unpredictability cable operator to select best-of-breed over how scenarios will play themselves components, optimize plant efficiency, out going forward. The one entity that economically scale resources, and can best determine how to architect determine precise architectures. So infrastructure across media and services empowered, the cable operator is is the MSO. positioned to compete effectively and seize emerging opportunities.
Ran Oz S. V. Vasudevan Chief Technology Officer Chief Architect [email protected] [email protected]
BigBand Networks, Inc. 475 Broadway Redwood City, CA 94063 650 995 5000 http://www.bigbandnet.com NEW DEVELOPMENTS IN IEEE-1394 STANDARDS FOR THE CABLE SET-TOP BOX
Mark Eyer Sony Electronics
Abstract requires High Definition cable set-top boxes to: In December 2002, consumer electronics manufacturers joined with cable MSOs in an- “… comply with ANS/SCTE 26 (as of 10/29/03) with transmission of bit-mapped nouncing an historic agreement on digital ca- graphics (EIA-799) optional, and shall support ble compatibility. Among the provisions iden- the CEA-931-A PASS THROUGH control tified in the agreement was a commitment on commands: tune function, mute function, and the part of cable operators to provide, by July restore volume function. In addition these boxes shall support the POWER control com- 2005, IEEE-1394 interfaces on high-definition mands (power on, power off, and status in- digital cable set-top boxes acquired for distri- quiry) defined in A/VC Digital Interface Com- bution to consumers. Two standards related to mand Set General Specification Version 4.0 the 1394 interface were cited in the agree- (as referenced in ANSI/SCTE 26 2001).” ment, ANSI/SCTE 26 [1] and CEA-931-A [2]. We start with a system-level overview of This paper describes the agreed-upon func- the Digital Cable Set-top Box (DCSB) as it tionality of this high-speed serial bus network fits into a typical home network environment. interface, and provides an in-depth discussion We then move to a detailed review of CEA- of the new CEA-931-A protocol. 931-A [2], published this year by the Con-
sumer Electronics Association (CEA). We include along the way a discussion of a typical INTRODUCTION “IR Blaster” application to show how use of CEA-931-A can overcome the inherent limita- As envisioned by the engineers and pol- tions of that technique. Next, a brief summary icy-makers who forged the December 2002 of ANSI/SCTE 26 [1], first published in 1999 Agreement, the primary function of the 1394 and updated in 2001, is presented. We then interface is to enable consumer recording or summarize the various normative references time-shifting of compressed MPEG-2 au- cited in ANSI/SCTE 26 and in the primary dio/video content, and delivery on a peer-to- CEA standard for the display side, EIA/CEA- peer home network of that digital program- 849-A [4], and conclude with a discussion of ming. While the High-Definition (HD) digital two implementation issues that may be of in- cable set-top box outputs digital video for terest to designers. viewing via an uncompressed Digital Visual Interface (DVI) or optionally a High Defini- tion Multimedia Interface (HDMI) port, an SYSTEM OVERVIEW IEEE-1394 high speed serial bus interface must also be present. Figure 1 diagrams a Digital Cable Set-top Box (DCSB) at the upper left. The video out- The memorandum of understanding in the put flows from its DVI/HDMI port to a High- December 2002 NCTA/CEA Agreement [3] Definition Display. Audio connections are not shown in the simplified diagram. Digital Cable Set- IEEE-1394 Network Bus Cable top Box
Personal Disc or D- Add-on Media DVI or Video VHS Storage Server HDMI Recorder Recorder
High-Definition High-Definition Display #1 Display #2
Figure 1. System Block Diagram To the right of the DCSB in Figure 1, an the 1394 path offers HD video but only stan- IEEE 1394 Network Bus is shown intercon- dard-resolution graphics and a graphics frame necting the set-top with a variety of au- rate that is limited by hardware capacity in dio/video devices. A hard-disk- based Per- either the source or display. sonal Video Recorder (PVR) may have 100 GB or more of disk space available for tempo- Digital Recording Functionality rary storage of audio-video programming. The PVR or DCSB may be able to make use of How might a viewer use a setup like that additional storage devices that may be present of Figure 1 to make recordings of digital con- on the bus (represented here by the box la- tent? Several scenarios are possible. One pos- beled “Add-on Storage”). sibility is that the DCSB may discover the Add-on Storage device and may be able to use A Media Server is also shown in Figure it as a disk-based cache for audio/video pro- 1. While similar in many ways to the func- grams. In this scenario, the set-top manages tionality offered by the PVR, the Media files and offers a suitable user interface to ac- Server might include extra features such as cess them and to organize and manage disk archived musical recordings, access to Inter- resources. In another scenario, the PVR main- net-based content, ability to manage pre- tains its own file system, provides its own user recorded packaged media such as CDs and interface to stored programs, and performs DVDs, and ability to accept and catalog per- attended and unattended recording from se- sonal content such as digital photos and home lected source devices. It is with this latter sce- videos. nario in mind (among other considerations) that CEA-931-A was developed. At the far right side of Figure 1, a second display device is connected via the 1394 bus. As noted, an HD digital cable set-top box Compatibility with the DCSB is guaranteed built to comply with the NCTA/CEA Agree- for DTV displays supporting EIA-775-A [5] ment implements ANSI/SCTE 26 and certain and the “MPEG profile” of EIA/CEA-849-A provisions from CEA-931-A. Using these pro- [4]. These two protocols are complimentary to tocols, the PVR (or any other device on the ANSI/SCTE 26 [1]. Whereas the DVI/HDMI 1394 bus) can identify the set-top box as a port on the DCSB provides a high-bandwidth source of digital video, can turn it on and can pathway for graphics as well as for HD video, tune it to any given digital channel. The for- mat of data on the bus, including aspects of physical, link layer, transport, link encryption Room to DTV for copy protection, and audio/video compres- Room sion formats are all specified. The precision Interface Network (opt.) Bus and completeness of these provisions enables the development of this new category of con- sumer digital recording devices. IR pulses PVR “Record (proprietary function” CEA-931-A STANDARD format)
CEA-931-A [2], entitled Remote Control Command Pass-through Standard for Home “REC” key Networking, defines a standard method for REC RCU communication on the network of simple “user intents” such as those typically repre- Figure 2. “Record” Key Example sented by keys on a consumer Remote Control Typically, the DTV will address the key Unit (RCU). Additionally, the protocol recog- commands not corresponding to internal func- nizes and supports applications such as unat- tions to the specific network device currently tended recording that typically, in the analog selected as the current audio/video “input” or world, must rely on devices capable of emu- A/V source device. If several devices on the lating the infrared pulses emitted by a given network could respond to a “Record” com- device’s RCU. The new protocol offers a vast mand, for example, the only one receiving the improvement in reliability and ease of use as command will be the one currently selected as compared with the analog techniques it re- the A/V source. places. A key feature of RCU key pass-through is Remote Control Key Pass-through that the IR pulses corresponding to the exter- nal function are proprietary to the manufac- Figure 2 illustrates in a simplified way the turer of the RCU itself, yet the functions concept of remote-control key pass-through. themselves are mapped (in the device receiv- The RCU in the figure is the unit sold with the ing the IR pulses) into standard key com- DTV receiver. The format and carrier fre- mands by the 931-A protocol. quency of the infrared (IR) pulses it emits are recognized and accepted by that DTV re- What are the benefits of RCU key pass- ceiver. Some of the commands, such as power through? The primary benefit is apparent if on and off and picture controls (brightness, the device being controlled is not directly contrast, etc.) are directed at the DTV itself visible to the viewer who is holding the RCU. and are processed internally. Others, such as a It may be in another room, for example, or “Record” key may not correspond to functions simply inside a cabinet or behind a piece of supplied by the display. Key presses such as furniture such that the RCU’s IR pulses can- these can be “passed through” the DTV and not reach its front panel. In such cases, the placed onto the 1394 bus. The function of the DTV receiver, whose IR receiver is in full CEA-931-A [2] protocol is to define the stan- view of the user, acts as a relay agent to de- dard method whereby these RCU keys are liver the commands to the hidden unit. communicated across the network. Another benefit of RCU key pass-through application. This scenario involves a PVR us- is that the networked A/V devices (if they all ing a cable or satellite set-top box as a video support CEA-931-A) may be controlled by a source and an analog VCR for archival single RCU, thus eliminating the clutter and backup of disk-based material to videotape. confusion of several remotes on the coffee This setup is diagrammed in Figure 3. table. This statement comes with a caveat: certain devices may include functions associ- As shown, the user operates the system ated with dedicated keys on their native primarily from the PVR’s native remote con- RCUs. These may not be mappable in a trol unit. When a channel change is desired, straightforward way to CEA-931-A key codes. the PVR creates the necessary IR pulses to The “universal” RCU remains elusive, yet effect the channel change in the cable set-top CEA-931-A is clearly a step in the right direc- box. For archival recording, it emulates the tion. VCR’s RCU to start and stop recording.
Infrared “Blasters” Beyond the large number of cables needed to wire these pieces of equipment to- As mentioned, CEA-931-A [2] was de- gether, one must allow for the limited length signed with a scenario beyond simple key of the cables supporting the infrared emitters. pass-through in mind. Typically, it is the If either of these emitters slips out of position, viewer who operates his or her audio/video IR commands will not be received reliably. equipment via IR remote control IR Emitter by pushing keys on the RCU. Channel Numbers, ENTER Certain equipment, however, may take the place of a human Cable Set-top Box operator in order to control a PVR piece of A/V equipment when, Power On; Record; Analog Power Off (standby) for example, the user is not pre- Audio/ Video sent.
Until the advent of digital home network busses and proto- VCR cols such as CEA-931-A, analog RCU (of PVR) methods were the only option. Analog Video Monitor One approach employed “IR Figure 3. IR Blaster Example Blasters,” devices capable of emulating the IR pulses recognized by the Another reliability issue occurs because piece of equipment to be controlled. The con- commands such as the one used to tune the trolling device typically includes an IR emitter DCSB to a given channel involve several key attached to the end of a cable; the emitter is codes sent in sequence. If the codes are sent affixed in a position near the front panel of the too close together, the set-top box may not device to be controlled. This approach, while recognize them. If they are sent too far apart, workable, is fraught with difficulties. “channel surfing” is noticeably more sluggish.
To appreciate how CEA-931-A [2] may Setting up the system in Figure 3 is a be used to overcome the shortcomings of IR challenge by itself. The PVR must include a Blaster approaches, we take a look at a typical database covering as many models of au- dio/video equipment as possible. This data- and can determine each device’s type and base describes the modulation modes and in- function and what protocols are supported. No frared pulse formats corresponding to the special setup is required at all! needed commands for various models of vari- ous manufacturers’ equipment, so that the The functions previously handled by the PVR can control the user’s specific set-top IR blasters are now handled by the CEA-931- box and VCR. Typically, during the setup A protocol. When the PVR wishes to tune the procedure the user will need to answer a series DCSB to a given channel, it may simply issue of questions such as “Does your set-top box a “Tune Function” command providing the require an ENTER key to terminate channel channel number to be tuned. Bus commands changes?” and “Does a channel number in- are acknowledged, so operation is reliable and volve two digits or three?” With the setup predictable. procedure being time-consuming and prone to errors, it is no wonder many people are un- Whereas in the analog scenario the PVR’s willing to attempt it. own RCU was used to control it, now the DTV’s own RCU may be used to control the Figure 4 shows the same application PVR, since the DTV is able to pass on to it (PVR functionality with unattended recording key presses appropriate to its control. and control of an A/V source device), but this time the devices handle digital audio/video on When the PVR wishes to transfer a stored a 1394 bus and support standard protocols in- program to disc or tape, it can operate the re- cluding CEA-931-A. 1394 cables inter- corder using other CEA-931-A functions. connecting the pieces of equipment replace Power control is provided by commands de- the bulky audio/video wiring. fined in the A/VC Command Set General Specification [7], and the recording is initiated by a CEA-931-A “Record Func- tion.”
Tune Function
Cable Set-top Box Now, when the viewer browses through the channel line- Digital A/V PVR up to look for something entertain- Power On; ing, channel changes are fast: the Record Function; 1394 Power Off (standby) time interval between channel Bus number digits has been eliminated because the “Tune Function” in- Digital A/V D-VHS or volves a single command delivered Disc on the home network bus.
DTV Monitor Recorder CEA-931-A IN DETAIL
CEA-931-A [2] actually does RCU (of DTV) not itself define or invent anything Figure 4. Using CEA-931-A for Unattended Recording new. It is built on a portion of an existing industry standard devel- As soon as the equipment is powered on, oped by the 1394 Trade Association, called each unit discovers the others on the network Panel Subunit 1.21 [6]. Whereas the Panel Subunit document specifies both a “direct” Operation IDs mode and an “indirect” mode of operation, CEA-931-A uses only the simpler “indirect” Table 1 lists the RCU keys and Determi- mode. In this mode, whatever on-screen text nistic Functions supported in Panel Subunit or graphics constitutes the graphical user in- 1.21 and CEA-931-A. Those in the top por- terface is either embedded into video or deliv- tion of the table correspond to RCU keys, and ered via EIA-775-A bit-mapped graphics (de- have no accompanying parameters. Some of fined in EIA-799 [8]). those in the lower portion labeled “Determi- nistic Functions” have associated data; these The indirect mode of Panel Subunit was support applications like unattended recording designed to allow a controller device to emu- and simple device control. late the RCU of the device being controlled, hence its applicability to the needs identified As can be seen by inspection of the top by CEA for home networking. portion of the Table, all of the common RCU key functions are represented. Some are The Panel Subunit command that conveys clearly aimed at specific types of devices. An functions associated with RCU keys and “ba- example is the “Angle” key, used typically on sic user intents” is called the PASS THROUGH DVD players to cycle among video tracks of- control command. Each PASS THROUGH fering different viewing angles. control command is carried within a standard AV/C Function Control Protocol (FCP) Each of the Operation ID types given in packet defined in [9], and identifies the par- the top half of the table, when received by the ticular RCU key or user intent by an “Opera- target device, has exactly the same effect on tion ID” and (for some functions) one or more the device as the corresponding key on its own parameters. RCU would have. That means that, for exam- ple, if repeated pressings of the PLAY key would cause the device to toggle between playing and pausing playback, reception of the
Tab1e 1. Defined Operation ID values. Category User operations Navigation keys Digits 0-9, Select, Up, Down, Left, Right, Right-up, Right-down, Left-up, Left-down Menu selection Root menu, Setup menu, Favorite menu, Exit Media control Play, Stop, Pause, Record, Rewind, Forward, Fast forward, Eject, Backward, Angle, Subpicture Channel control Channel up, Channel down, Previous channel Miscellaneous Power, Volume up, Volume down, Mute, Sound select, Input select Deterministic Functions Name Function Parameter Play function Start (or continue) playing content Speed and direction of play Record function Start (or continue) recording - Pause-play, Pause-record Pause playback or recording - Stop function Stop playback or recording - Mute function Mute audio - Restore volume function Restore audio to previous volume level - Tune function Tune to indicated channel (or virtual channel) One- or two-part channel number Select disk function Select indicated physical media Disk number (1-65,535) Select A/V input function Select indicated A/V input A/V input (1-255) Select audio input function Select indicated audio input Audio input (1-255) “play” Operation ID would have the same plemented in that device, and if implemented, toggling effect. Both “key down” and “key whether it will be acted upon. up” events are represented. DTV 1394 INTERFACE The Deterministic Functions listed in the lower half of Table 1, on the other hand, are While the December 2002 Agreement defined such that the result of the command is does not cite specific requirements for a DTV entirely predictable. Toggling is not allowed. receiver connected to the HD digital cable set- Accordingly, reception of the “Play Function” top via 1394, the applicable standards are well in the target device must result in playback known in the industry. CEA has developed a either starting or continuing. Note that control logo program called “DTV Link” based on the of device power is not included here. CEA- “MPEG-2 profile” of EIA/CEA-849-A [4] 931-A specifies that the POWER control (visit http://www.ce.org/dtvlink for details). command specified in [7] is to be used for de- terministic control over device power state. Any DTV display device that is compli- ant with the requirements of the EIA/CEA- The benefit of Deterministic Functions is 849-A MPEG-2 profile is compatible with the that the controller device no longer needs to HD digital cable set-top box. Simply put, try to keep track of the state of the device un- “compatible” means the user will be able to der control. With IR Blaster techniques, for interconnect the DCSB to the display and then example, the Power key on the RCU might view on-screen displays and program au- toggle device power between “On” and dio/video generated by the DCSB. In technical “Standby.” If the controlling device does not terms, this compatibility guarantees: know the initial state, using the Power key might result in turning the unit off rather than • The DTV display is able to discover on. and identify the cable DCSB on the 1394 bus as a compatible source (of- Deterministic Functions supported in fering it up as a choice); Panel Subunit 1.21 and CEA-931-A include media control (Play, Record, Pause, and • The display is capable of decoding and Stop), audio control (Mute, Restore volume), displaying audio/video services includ- tuning control (Tune function), and functions ing AC-3 audio and any of the allowed to support selection of specific A/V inputs. formats for compressed MPEG-2 The Play function is particularly powerful, in video; that it includes as a parameter the speed and direction of desired playback. All the trick • The display is able to accept analog modes are included, as well as fast-forward A/V output from the DCSB, and is and rewind functions. able to switch between analog and digital DCSB outputs upon request by For reliability and error handling, the the DCSB (analog/digital source selec- Panel Subunit specification describes methods tion is discussed in more detail below); any device on the network can use to deter- mine whether a given target device supports • The display supports the MPEG-2 the protocol. It also describes how a device Transport Stream format delivered issuing a PASS THROUGH command can de- within an isochronous channel on the termine whether or not the command is im- 1394 bus in accordance with IEC OTHER PROTOCOLS AND STANDARDS 61883-4 [10]; CEA-931-A is the newest protocol appli- • The display determines which video cable to home networking, but it is only part program element to decode and display of the story. We provide a brief overview of by examining the MPEG-2 Program protocols and standards relative to digital au- Specific Information (PSI) tables: the dio/video distributed by cable to an IEEE- Program Association Table (PAT) and 1394-based home network. This discussion the Program Map Table (PMT) it ref- should not be confused with a 1394-based erences. Whenever the DCSB includes network primarily directed at A/V for the PC a single program (service) in the PAT, platform, as different video compression for- the DTV display identifies the video mats and protocols are involved there. component of that service and decodes it without need for viewer intervention. Figure 5 shows a digital cable set-top box Likewise, it will find the appropriate on the left and a DTV display on the right. As audio track (perhaps based on its indi- mentioned, the primary standard defining re- cated language) and decode it without quirements for the 1394 interface on the the need for user interaction. DCSB is ANSI/SCTE 26 [1]. The primary standard for the DTV receiver is EIA/CEA- For the same setup (DCSB connected to 849-A [4]. DTV), if CEA-931-A is supported by the DCSB, the DTV (or any other device on the ANSI/SCTE 26 network) can cause the DCSB to power on or go to standby power state, can cause audio Entitled Home Digital Network Interface outputs to be muted or un-muted, and can with Copy Protection, ANSI/SCTE 26 [1] cause the DCSB to tune to any given analog specifies requirements for the 1394 interface or digital channel. on a digital cable set-top box. It states that a compliant cable set-top box must meet re-
Cable Set-top Box DTV Display
SCTE 26 EIA/CEA-849-A Home Digital Network Application Profiles for EIA- Interface 775-A Compliant DTVs
SCTE 43 (Video) ATSC A/53 (A/V, Transport) SCTE 54 (Cable Transport) EIA-775-A DTV 1394 Interface + EIA-799 Graphics A/VC General Command Set IEC 61883-1 IEC 61883-4 DTCP Copy Protection IEEE 1394
Figure 5. Protocol Dependencies quirements for source devices given in EIA- • EIA-775-A [5], defining how the 775-A [5]. In addition, the set-top box must IEEE-1394 protocol is used for discov- implement link encryption according to the ery and connection management, deliv- “5C” method, also known as Digital Trans- ery over isochronous channels of the mission Copy Protection (DTCP). The set-top MPEG-2 Transport Stream and over box must indicate to the receiver whether to asynchronous channels of bit-mapped take its analog or digital output via the “ana- graphics defined in EIA-799 [8]; log digital source selection” method of EIA- 775-A (discussed below). • AV/C Command Set General Spec- ification [7], providing standard meth- EIA/CEA-849-A ods for device discovery and basic con- trol, this document is the foundation for Although EIA-775-A is an important ele- the family of AV/C protocols (learn ment to 1394 compatibility between the set- more at http://www.1394ta.org/); top box and the DTV, it does not specify re- quirements for the higher protocol layers. For • IEC 61883-1 [9], providing a robust example, it does not state requirements for foundation of lower-layer protocols, in- compatibility with various possible transport, cluding common methods for encap- or audio or video compression formats. sulating AV/C commands, and for for- EIA/CEA-849-A was written to address this matting isochronous packets including need, defining a number of “application pro- timestamps; files” for EIA-775-A. The one relevant to our discussion here is called the “MPEG profile.” • IEC 61883-4 [10], specifying the stan- If both a source device and a sink (display) dard method for carrying MPEG-2 device support a common EIA/CEA-849-A Transport Streams on 1394; profile, 1394 interconnectivity (including the • Digital Transmission Copy Protec- ability to decode audio and video) is assured. tion, providing a standard method for link encryption to protect against unau- Normative References thorized copying of high-value content The ANSI/SCTE 26 [1] and EIA/CEA- (learn more at http://dtcp.com/); and 849-A standards both cite a number of nor- • IEEE 1394 [14], the fundamental mative references, including: specification for the high-speed serial • ANSI/SCTE 43 [11], defining the al- bus technology. The lower layers of the lowable MPEG-2 video compression protocol stacks are defined here, formats; including physical aspects of connectors and cabling. • ATSC A/53 [12], the ATSC Digital Television Standard, defining audio, IMPLEMENTATION ISSUES video, and transport aspects for sources originating from digital terrestrial We conclude with a couple of implemen- broadcast; tation issues for consideration by system de- signers. • SCTE 54 [13], defining transport- related characteristics for digital cable; Isochronous Channel Bit-Rates needed to deliver them does not need to be any higher than the total data rate of the pack- Certain devices, such as for example disc ets actually present. For example, a partial TS or digital tape recorders, may not be able to might include one standard-definition A/V handle the data rates as high as those deliv- programming service. While the full TS carry- ered by a 256-QAM modulated carrier on ca- ing that service might arrive in a 38.8 Mbps ble. In 256-QAM mode, a given 6-MHz cable Transport Stream, the partial TS carrying the channel can deliver data at a rate exceeding 38 single program could be sent across a 1394 Mbps. A recorder capable of handling a isochronous channel allocated with a much maximum bit-rate of 20 Mbps, for example, lower bandwidth (perhaps 6 or 8 Mbps). would be able to record any single HD au- dio/video program, but would not be able to Analog/Digital Source Selection handle the rate of a full Transport Stream de- rived from a 256-QAM carrier on cable. For the foreseeable future, HD digital ca- ble set-top boxes will have analog outputs in For this reason, and for the fact that from addition to their digital ones. While it is pos- the user’s perspective the desire is to record sible (and becoming more practical every day) one program (not an arbitrary group of con- for the DCSB to digitally compress and en- currently broadcast programs), the source de- code services received via analog transmission vice is expected to create a Single Program channels, cost considerations may preclude Transport Stream (SPTS). The process of cre- this in the near-term. That means that when an ating a “partial” TS is straightforward, and is analog channel is tuned, the digital video out- described in IEC 61883-4 [10]. put from the set-top may cease. How is this situation handled? Partial Transport Streams are structured like regular MPEG-2 Transport Streams ex- ANSI/SCTE 26 requires the set-top to cept that not every 188-byte transport present signal to the device connected to its digital in the original TS is present in the “partial” output on the 1394 bus whether to take its TS—only those corresponding to PID values digital or its analog output. This is described of interest are included. For example, a partial in Sec. 4.11 of [5], and it involves processing TS may consist of TS packets carrying the an AV/C CONNECT command identifying Program Association Table (PID value 0), the which output plug (analog or digital) should Program Map Table section (PID value as be taken as the current output from the box. identified in the PAT), and one audio and one Compliance with the MPEG-2 profile of video track (PID values identified in the PMT EIA/CEA-849-A [4] (and hence DTV Link) section). PIDs associated with Program and also requires support for this analog/digital System Information Protocol (ATSC A/65) source selection mechanism. data may also be included. A DTV display typically has several ana- The Common Isochronous Packet in [9] log A/V inputs, perhaps labeled Video-1, describes a time-stamp mechanism to enable Video-2, etc. In order for the AV/C the packet timing of the original partial Trans- CONNECT command to have the desired ef- port Stream to be accurately reconstructed at fect, the user must configure the DTV in a the receiving end of the isochronous channel. setup menu to identify which set of A/V in- The important thing to note about partial puts should be associated with a given set-top Transport Streams is that the bus bandwidth box. CONCLUSION Appendix B of Further Notice of Pro- posed Rulemaking FCC 03-3, January 10, This paper has explored the specifics of 2003. the historic December 2002 NCTA/CEA [4] EIA/CEA-849-A, Application Profiles Agreement as it relates to the IEEE 1394 in- for EIA-775-A Compliant DTVs, Elec- terface on the digital cable set-top box, and tronics Industries Association and Con- the ramifications of the provisions therein. sumer Electronics Association, 2001. The features and benefits of the CEA-931-A [5] EIA-775-A, DTV 1394 Interface Specifi- protocol published this year by the Consumer cation, Electronics Industries Asso- Electronics Association were outlined. ciation, 2000. [6] AV/C Panel Subunit Specification 1.21, One hopes that the NCTA/CEA Agree- TA Document 2002009, 1394 Trade As- ment, along with the standardized protocols sociation, January 2003 they reference, will spark a frenzy of creativity [7] AV/C Command Set General Speci- among manufacturers to bring the conven- fication, 1394 Trade Association. ience and exceptional video quality of digital [8] EIA-799, On Screen Display Specifica- technology to a new family of home network- tion, Consumer Electronics Association, based Consumer Electronics products. 1999. [9] IEC 61883-1, Consumer audio/video Acknowledgements equipment—Digital Interface—Part 1: General, International Electrotechnical The author wishes to thank the members Commission, 1998. of the CEA R-7.5 Home Networking commit- [10] IEC 61883-4, Consumer audio/video tee, chaired by James Williamson of Sony equipment—Digital Interface—Part 4: Electronics, for the development of the CEA- MPEG-2 TS Data Transmission, Inter- 931-A protocol. The necessary extensions to national Electrotechnical Commission, the Panel Subunit specification were com- 1998. pleted in the A/V Working Group of the 1394 [11] ANSI/SCTE 43 2001, Digital Video Trade Association, chaired by Scott Smyers. Systems Characteristics Standard for Cable Television, Society of Cable Tele- REFERENCES communications Engineers. [1] ANSI/SCTE 26 2001, Home Digital [12] ATSC A/53B, Digital Television Network Interface Specification with Standard, Revision B, With Amendment 1, Copy Protection, Society of Cable Tele- Advanced Television Systems Commit- communications Engineers. tee, 2001. [2] CEA-931-A, Remote Control Command [13] SCTE 54 2003, Digital Video Service Pass-through Standard for Home Net- Multiplex and Transport System Standard working, Consumer Electronics Associa- for Cable Television, Society of Cable tion, 2003. Telecommunications Engineers. [3] Consensus Cable MSO-Consumer Elec- [14] IEEE 1394-1995, Standard for a tronics Industry Agreement on “Plug & High Performance Serial Bus, Institute of Play” Compatibility and Related Issues, Electrical and Electronics Engineers.
P2P, THE GORILLA IN THE CABLE
Alexandre Gerber, Joseph Houle, Han Nguyen, Matthew Roughan, Subhabrata Sen AT&T Labs - Research
Abstract INTRODUCTION
There is considerable interest in peer-to- P2P (peer-to-peer) file sharing applications peer (P2P) traffic because of its remarkable have grown dramatically over the past few increase over the last few years. By analyzing years and contribute a significant share of the flow measurements at the border routers of a total traffic in many networks. In this paper, Tier-1 ISP backbone that carry broadband we analyze flow-based measurements of traffic, we are able to study its properties. P2P broadband traffic spanning several months, has become a large part of broadband traffic gathered in the backbone of a large ISP and its characteristics are different from older network. We first develop an understanding of applications, such as the Web. It is a stable P2P traffic behavior from the viewpoint of balanced traffic: the peak to valley ratio broadband provider networks (earlier studies during a day is around two and the IN/OUT were based on a Tier-1 ISP backbone traffic balance is close to one. Although P2P viewpoint [1] and on a University edge- protocols are based on a distributed network viewpoint [2]). The study then architecture, they don’t show strong signs of describes some key issues and challenges in geographical locality. A broadband subscriber handling/controlling this traffic, and presents a is not much more likely to download a file potential solution approach. We begin with a from a close region than from a far region. description of these P2P systems
It is clear that most of the traffic is File Sharing Applications generated by heavy hitters who “abuse” P2P (and other) applications, whereas most of the Many popular P2P applications such as subscribers only use their broadband KaZaA and Gnutella are organized as connections to browse the web, exchange application-level overlay systems in which large emails or chat. However it is not easy to numbers of computers (called peers) across the directly block or limit P2P traffic, because Internet link together in a decentralized manner these applications adapt themselves to their via application-level connections. The environment: the users develop ways of predominant use of these systems is for sharing eluding the traffic blocks. The traffic that large data files (particularly music and video) could historically be identified with five port among the connected users. The data files and numbers is now spread over thousands of TCP associated metadata information (useful for ports, pushing port based identification to its searching content) are distributed across the limits. More complex methods to identify P2P different peers. A key difference with traffic are not a long-term solution, the cable traditional client-server systems is that each industry should opt for a “pay for what you host in a P2P system acts as both a client and a use” model like the other utilities. server of content. In contrast to the stable
configurations of traditional distributed
systems, the individual peers can frequently join and leave the P2P system.
The process of obtaining a file can be motivated by the desire to circumvent firewall broadly divided into two phases – query search restrictions as well as rate–limiting actions by followed by object retrieval. First, a user ISPs targeted at such applications - we shall specifies a query (e.g., a combination of name, discuss this more later in the paper. genre, artist name etc.), and the P2P protocol searches for the existence of file(s) that match Another recent occurrence has been the the query. The requesting peer receives one or development of tools that allow an end-user to more responses, and if the search is successful, explicitly select the SuperNode it connects to identifies one or more target peers from which [3]. This appears to be an attempt to improve to download each file. The search queries as the quality of the best-effort search process in well as the responses are transmitted via the the P2P system, for files that are not widely overlay connections. The details of how the distributed, but are geographically localized. search is propagated through the overlay is For instance, connecting to a SuperNode in protocol-dependent. In earlier P2P protocols Brazil may increase the chances of locating exemplified by Gnutella version 4.0, a peer Samba-related content. initiates a query by flooding it to all its neighbors in the overlay. The neighboring Data Collection peers in turn, flood to their neighbors, using a scoping mechanism to control the flood. In We have access to “flow-level” data about contrast, for newer protocols like KaZaA, as broadband traffic at the border routers of a well as for newer versions of Gnutella, queries large ISP. Flow-level data is considerably are forwarded to and handled by only a subset more detailed than data sets such as SNMP, of special peers (called SuperNodes in KaZaA, and at least this level of detail is needed to and UltraPeers in Gnutella). A peer transmits perform application classification. When an index of its content to the ``special peer'' to looking at these flows we can make a very which it is connected. The special peer then educated guess about whether the flow is associated with a Broadband consumer and uses the corresponding P2P protocol to from which region it originates. A region forward the query to other such peers in the typically ranges from an extended system. metropolitan area to a state. For the remainder
of this paper we focus on traffic that appears to Once the search results are in, the be associated with broadband subscribers. requesting peer directly contacts the target peer, typically using some variant of HTTP (the By flow, we mean a sequence of packets target peer has a HTTP server listening by exchanged by two applications. More precisely default on a known protocol-specific port), to we define a flow to be a series of uni- get the requested resource. Some new systems directional packets with the same IP protocol, use swarming download-- a file is downloaded source and destination address, and source and in chunks from multiple peers. destination ports (in the case of TCP and UDP traffic). The flow measurements used here are Although the earlier P2P systems mostly called Cisco Netflow [4]; they are used default network ports for communication, implemented in many of Cisco’s routers. The there is strong evidence to suggest that data collected about a flow (apart from the substantial P2P traffic nowadays is information above) are the duration, the transmitted over a large number of non- number of packets, and bytes transmitted, and standard ports. This seems to be primarily which header flags (SYN, ACK, …) were used in the flow. Measured flows are also Ports (1024-49,151), and the Dynamic and/or constrained in time (Cisco Netflow collection Private ports (49,152-65,535). sends flows from the router at 30 minute intervals), so there is a need to reconstruct the A typical TCP connection starts with a actual traffic from a single “connection”. After SYN/ACK handshake from a client to a server. reconstruction there will be one flow per The client addresses its initial SYN packet to connection – a potentially enormous volume of the server port for a particular application, and information. uses a dynamic port as the source port for the SYN. The server listens on its port for In order to minimize any performance connection. UDP uses ports similarly though impact on the routers collecting the flow without connections. All future packets in the measurements the measurements are based on TCP/UDP flow use the same pair of ports at sampled packets collected on the routers, the client and server ends. Therefore, in which then export the flows to aggregators. To principle the server port number can be used to reduce the huge data volume the aggregator identify the higher layer application using TCP further samples the flows using the smart or UDP, by simply identifying which port is the sampling algorithm [5] that is better suited for server port (the one from the well-known, or heavy tailed distribution, such as typically registered port range) and mapping this to an found in Internet flows. In addition, there is application using the IANA list of registered also an uncontrolled sampling due to port [7]. measurement packet losses. These three types of sampling can be estimated and corrected There are many barriers to determining and don’t affect our results that are based on applications from port numbers. For instance, the weekly or monthly average traffic well know and registered ports are not defined generated by hundreds of thousands of for all applications and this is typical of P2P broadband subscribers between May 2002 and applications. Further more, in some cases February 2003. server ports are dynamically allocated as
needed (for instance, one might have a control Identifying Applications connection on which a data port is negotiated).
Finally, the use of firewalls to block There are a number of ways one could go unauthorized and unknown applications from about identifying individual applications using a network has spawned work arounds within IP traffic. However, as noted, Netflow that have made the mapping from port number only keeps data on some aspects of flows. The to application ambiguous. most useful of these for application breakdowns are the source and destination port numbers, and the IP protocol number. The Despite this, a great deal can be said about protocol numbers used are well documented the mapping of port to application, though [6], with TCP being protocol 6, and UDP being obviously there will still be some ambiguity, 17. TCP, and UDP traffic also define (16 bit) and chance for errors. Note that both ports source and destination port numbers intended must be considered as possible candidates for (in part) for use by different applications. The the server port, unless other data is available to port numbers are divided into three ranges: the rule out one port. Well Known Ports (0-1023), the Registered The algorithm that we have adopted here chooses the server port by (1) looking for a well known port, (2) a registered port, or (3) We should note that we are not collecting an unregistered port which is known (from any information about URL’s, or individual reverse engineering of protocols) to be used by subscribers usage: IP addresses measured are a particular (unregistered) application. If both not related to individual subscribers, and we source and destination port could be the server, only view the bulk properties of the traffic, then we choose the most likely one through such as its distributions. ranking applications by how prevalent they are in detailed (packet level) traffic studies – for APPLICATION COMPOSITION instance, WWW is considered a high ranking application, as are email, and P2P applications. Overview
The result is a mapping from flows to Table 1 shows the application traffic applications, that while not perfect, has been composition for 2 broadband regions in May shown to be reasonably effective. The biggest 2002 and January 2003. For each of these problem is that there are still a substantial regions, we examine both the traffic coming number of flows which cannot be mapped to an from outside the region to some IP address application. We further classify these unknown within the region (referred to as IN) and the flows by the size of the flows: the category of traffic sourced within the region and destined most interest here is “TCP-big”, which consists for outside the region (OUT). For each time of unknown flows that transmit more than period and region, we display the per- 100kB in less than 30 minutes. application traffic volume in each direction as a percentage of the total traffic in that We shall argue in this paper that the TCP- direction. For a given application we also big traffic is primarily P2P traffic that is using show the traffic normalized by dividing by its unregistered ports unknown to us. P2P IN traffic volume for May 2002, in order to applications already use unregistered ports, and show the In/Out ratio, and the growth between the two periods. the structure of P2P protocols (with separate control and data traffic) allows data traffic to We note that in either direction the P2P be assigned to arbitrary ports. In the past the traffic forms a much smaller percentage of the major applications have typically used default overall traffic in January 2003 than in May ports (for instance 1214 for KaZaA) but in the 2002. TCP-big registered dramatic increases in recent past many efforts have been made to traffic contribution in both directions (10.5 constrain P2P traffic through rate limiting times for Outgoing and 6 times for Incoming) single ports or by blocking some ports at over the same period. The normalized figures firewalls, with the result that P2P users show that the P2P incoming and outgoing commonly use work-arounds. Where-ever we traffic are very similar for either of the 2 refer to P2P traffic we are using the traffic on months considered. Note also that the TCP-big the ports known to be directly associated with traffic in the 2 directions becomes much more P2P applications: we shall keep this separate balanced recently than earlier. For example for from TCP-big except where explicitly noted. broadband region X, the ratio between Also note that some P2P traffic may be incoming and outgoing TCP-big traffic misclassified into other application classes and volumes changes from 1.94:1 in May 2002 to so our estimates of the total volumes of P2P 1.12:1 in January 2003. traffic are conservative.
Broadband Region X Broadband Region Y Applicationx Mix (percentage) Normalized Consumption Applicationx Mix (percentage) Normalized Consumption May 2002 January 2003 May 2002 January 2003 May 2002 January 2003 May 2002 January 2003 OUT IN OUT IN OUT IN OUT IN OUT IN OUT IN OUT IN OUT IN All 100.0% 100.0% 100.0% 100.0% 1 1.65 1.97 3.2 100.0% 100.0% 100.0% 100.0% 1 2.19 1.83 4.08 ESP/GRE 0.4% 0.5% 0.6% 0.5% 1 1.98 3.12 4.3 0.4% 0.5% 0.3% 0.4% 1 2.71 1.7 4.67 OTHER 4.4% 3.7% 5.7% 4.5% 1 1.37 2.54 3.23 4.6% 3.2% 5.4% 3.4% 1 1.53 2.16 2.97 TCP-BIG 8.9% 10.5% 47.5% 32.5% 1 1.94 10.5 11.68 9.5% 11.8% 45.3% 32.1% 1 2.71 8.71 13.72 AUDIO/VIDEO 0.2% 1.6% 0.2% 1.6% 1 16.61 2.77 32.64 0.1% 1.5% 0.2% 1.5% 1 23.71 3.1 44.29 CHAT 0.7% 1.3% 1.0% 1.7% 1 3.08 2.93 7.93 0.7% 1.2% 0.7% 1.4% 1 3.81 2.02 8.67 FTP 1.0% 1.3% 1.0% 0.7% 1 2.22 1.91 2.4 1.4% 1.4% 0.4% 0.9% 1 2.24 0.56 2.64 GAMES 1.6% 1.2% 3.6% 2.5% 1 1.29 4.54 5.15 1.3% 1.2% 3.4% 2.4% 1 1.92 4.73 7.43 MAIL 1.7% 0.6% 1.1% 0.7% 1 0.6 1.26 1.28 1.0% 0.5% 0.9% 0.5% 1 1.13 1.71 1.88 NEWS 0.3% 7.3% 0.2% 5.3% 1 38.52 1.51 54.55 0.7% 17.5% 0.7% 14.6% 1 54.99 1.76 85.33 P2P 75.2% 45.6% 32.9% 20.6% 1 1 0.86 0.87 75.1% 38.5% 36.7% 19.5% 1 1.12 0.9 1.06 WEB 5.6% 26.4% 6.2% 29.4% 1 7.8 2.2 16.88 5.2% 22.8% 5.9% 23.5% 1 9.53 2.06 18.27 Table 1: Application Composition of two broadband regions in May 2002 and January 2003.
Time of Day Pattern EST, 7.00 PM PST). The P2P traffic exhibits less variability across a day than Web traffic. We next examine the diurnal behavior of The peak load is about 2 times the minimum P2P traffic. Fig. 1 plots the time series of the as opposed to 5 times for Web traffic. The incoming and outgoing traffic volumes (P2P, smaller variance in P2P traffic across a day web and TCP-big) for a given broadband may be a function of the programmed region across a week in February 2003. For download feature in P2P applications that each application, all the data values are allow users to specify multiple files in normalized by the mean per-hour incoming advance, that can be downloaded data volume for that application, averaged asynchronously by the P2P application. across that week. For Web, the outgoing traffic is
1.8 significantly smaller than (at most 20% of) the 1.6 incoming traffic, suggesting that the 1.4 broadband subscribers are mostly consumers 1.2 P2P OUT P2P IN of web data. In contrast, for P2P, the traffic in 1 Web OUT 0.8 Web IN the 2 directions track each other much more TCP-big OUT 0.6 TCP-big IN closely, across a day and across the week. 0.4 Another notable here is that the TCP-big 0.2 traffic distribution across time is very similar 0 to the P2P traffic. Also, just like P2P, the TCP-big traffic in the 2 directions are similar. 2/3/2003 0:00 2/4/2003 6:00 2/5/2003 2:00 2/6/2003 8:00 2/7/2003 4:00 2/8/2003 0:00 2/9/2003 6:00 2/3/2003 10:00 2/3/2003 20:00 2/4/2003 16:00 2/5/2003 12:00 2/5/2003 22:00 2/6/2003 18:00 2/7/2003 14:00 2/8/2003 10:00 2/8/2003 20:00 2/9/2003 16:00 These behavioral similarities are another indicator that the TCP-big traffic includes Fig. 1: Time of day pattern of P2P and Web some P2P applications. Finally for all 3 traffic. applications, we do not see significant All three applications exhibit similar variations across days and between weekdays diurnal behaviors with peak loads (in either and weekends. direction) around 2.00 AM GMT (10.00 PM
P2P LOCALITY
One of the potential advantages of P2P applications is that by distributing content, they provide the ability to download this content from locations closer to a user. It is therefore interesting to consider whether this really happens, and moreover to consider the question of locality in P2P traffic in general.
We approach this question by considering the simplest possible counter examples to localized traffic: the simple gravity model [8]. In this model, a packet entering the Fig. 2: Comparison of the real matrix elements to the estimated traffic matrix elements for a broadband ISP. The circles represent purely P2P network at S, makes its decision about its traffic and pluses represents the total traffic. The blue solid diagonal destination D independent of the arrival line shows equality and the green dashed lines show ± 20%. point. That is, the packet is drawn (as if by gravity) to destinations in proportion to the What does that tell us? Well the main volume of traffic departing at those locations. point is that the gravity model above explicitly excludes any notion of geographic, or The gravity model can be used to make topological distance. Therefore, as the predictions of the traffic volumes between measured traffic fits this model to some two regions based purely on the volumes extent, we may believe that neither P2P traffic entering and exiting at those two regions, by nor the traffic overall exhibit strong locality at the formula the regional level. A further, somewhat STT D T ,DS = in out subjective conclusion one might drawn from T the graph is that P2P traffic actually seems to where T is the total volume of traffic across fit the gravity model slightly worse, and so we S the network, Tin is the traffic entering the may hypothesize that P2P traffic shows more network at region S, and T D is the traffic locality than other traffic sources. out exiting the network at region D. Fig. 2 below To examine these hypothesis in more shows a comparison of the gravity model details we present Table 2, which shows the predictions for inter-regional traffic of a normalized traffic volumes between regions broadband ISP. The plot is based on Netflow for the P2P traffic. The table shows the traffic collected during one week in normalized probability that traffic originating September 2002; it shows traffic traversing from a particular region in one broadband the backbone between regions. The figure network, will depart from each region in the shows a scatter plot of the real inter-regional same broadband ISP (given it stays on the traffic versus the gravity model prediction, for same broadband network). Table 2 can be both P2P traffic, and the total traffic to the seen to have a number of almost identical broadband regions. One can see that in both rows (for instance the group of regions R1, cases the gravity model predicts the true R2, and R5 are very similar, as is the group traffic within about ±20%. R6, R7 and R8) indicating a complete lack of locality of traffic with reference to these regions. Other regions (specifically R3 and prevent us from seeing the intra-regional R4) are not dramatically far away, but rather traffic on a single broadband ISP, we can fall somewhere in between the other two gain a good view of this data by considering groups. the traffic between broadband ISPs. If locality were being exploited in P2P However the table also shows some applications, then one would expect traffic disparity between the groups of rows. This from ISP Y, region R to prefer going to ISP disparity is at its height when comparing the X, region R, rather than the alternative regions in the Eastern Standard Timezone regions. (EST), with those in the Pacific Timezone (PST). This is an indication of some degree Table 3 shows an example, giving the of weak locality in P2P traffic, at the “super- normalized probabilities that traffic from ISP regional” level. Y to X will go from regions M to R. Although the regions for the two broadband From/To R1 (PST) R2 (PST) R3 (MST) R4 (MST) R5 (CST) R6 (CST) R7 (EST) R8 (EST) R1 (PST) - 0.18 0.14 0.126 0.174 0.128 0.124 0.127 ISPs are slightly different, regions M3 and R2 (PST) 0.172 - 0.141 0.126 0.19 0.132 0.118 0.12 R3 (MST) 0.132 0.12 - 0.189 0.135 0.145 0.139 0.14 R7 are very closely matched as are M4 and R4 (MST) 0.107 0.111 0.182 - 0.124 0.163 0.155 0.158 R5 (CST) 0.161 0.18 0.136 0.132 - 0.135 0.127 0.129 R8. However, we see only very minor bias R6 (CST) 0.107 0.108 0.145 0.155 0.125 - 0.187 0.173 R7 (EST) 0.107 0.106 0.137 0.157 0.127 0.182 - 0.184 R8 (EST) 0.109 0.111 0.127 0.161 0.128 0.178 0.185 - towards traffic from M3 to R7 (compared to Table 2: Normalized inter-regional traffic matrix of broadband ISP X other EST regions), and similarly from M4 to weighted by P2P+TCP-big traffic (Longitude defined by the R8. Timezone). From / To R1 (PST) R2 (PST) R3 (MST) R4 (MST) R5 (CST) R6 (CST) R7 (EST) R8 (EST) M1 (MST) 0.133 0.121 0.157 0.125 0.118 0.111 0.089 0.146 This super-regional locality could arise M2 (CST) 0.121 0.095 0.114 0.158 0.117 0.145 0.094 0.156 M3 (EST) 0.12 0.114 0.12 0.138 0.119 0.128 0.14 0.122 for a couple of reasons (other than P2P M4 (EST) 0.11 0.115 0.109 0.137 0.135 0.119 0.133 0.142 M5 (EST) 0.117 0.115 0.133 0.135 0.129 0.12 0.121 0.129 applications explicitly taking advantage of Table 3: Normalized traffic matrix from broadband ISP Y to content locality to improve performance). broadband ISP X weighted by P2P+TCP-big traffic. Firstly, because of usage patterns Our conclusion is that, although there is (specifically the times at which a user is some evidence for weak locality at a large connected to the P2P network), there is a spatial scale, P2P applications do not yet slight increase in the likelihood that a search exploit such information on a large scale, and will find content in a local time zone. consequently, P2P traffic does not show Secondly, there may be a group of people strong signs of geographic locality. Recent within a super-region with content that is developments such as the KazuperNode tool slightly more relevant to the local super- [3]) provide methods for selecting the super- region. However, the data so far suggests that node to which one connects. On the one hand both of these effects are not dominant, and this could potentially increase locality if certainly there is no strong locality influence users tend to connect to nearby supernodes. such as might be seen if the main P2P On the other hand, there could be less applications exploited locality information. locality if users connect to supernodes in
different locations in their attempts to locate In both of the above examples the content. monitoring location limits our data to seeing only inter-regional traffic. Thus, one might argue, we are missing the key component in any study of traffic locality: the intra- regional traffic. While the data limitations
HEAVY HITTERS AND P2P during the week ending February 9th 2003. By subscriber, we mean an active IP address. It is well known in the broadband industry Even though the IP address is not statically that some heavy hitters consume most of the assigned (the user obtains an IP automatically bandwidth. We shall divide subscribers into via DHCP), in the networks we examined it classes by their total usage, and analyze their is “sticky”. That is, over a week a subscriber consumption characteristics such as the maintains the same IP address in practice, application composition and the traffic because the DHCP lease expires only after 4 balance per class. We define three groups of days and it is reassigned to him if it is still users: the heavy users who consume more available. However, the IP address than 1 Gbytes/day in average over a week, distribution doesn’t reflect exactly the the medium users who consume between 50 subscriber distribution since it misses the Mbytes/Day and 1 Gbytes/Day and the light inactive subscribers and the subscribers with users who consume less than 50 Mbytes/Day. a very low usage that may not be sampled.
User Distribution The six distributions in Figure 3 and 4 are quite consistent. In each case, the top 1% of We first compare the distribution of the IP addresses account for 18.6 — 24.4% traffic per subscriber. In order to see if there of the total traffic and the top 20% of the are consistent patterns we compare three active IP addresses account for slightly more regions, all at two different points in time: than 80% of the traffic. during the week ending June 26th 2002 and
Fig. 3: Consumption per percentile of IP addresses of three regions Fig. 4: Cumulative Consumption of three broadband regions during a during a week in June 2002 and a week in February 2003. The mean week in June 2002 and a week in February 2003. consumptions are around 140 Mbytes/Day/IP and the medians are roughly 30 Mbytes/Day/IP. composition of each group of users, as defined Consumption Characteristics earlier, with the average application composition that was studied earlier in this Since the median consumption is 4 to 5 paper. Indeed, in a close look at one of these times smaller than the average consumption, it regions Table 4 shows that the light user group is clear that the average consumption doesn’t (67% of the IP addresses) is still mainly reflect the behavior of most of the subscribers. browsing the web, exchanging email and This still holds if we compare the application chatting online. Its traffic balance – the IN/OUT ratio – is 4.8, which is far from the % of that group is lightly using one of these traffic balance of the heavy and medium user applications and it generates 1.1 % of the groups at 1.4-1.7 and 1.8, respectively. Table 5 outgoing News traffic and 1.8 % of the makes it clear that this class of subscriber is outgoing P2P traffic. not familiar with P2P or News since only 12.6 Week ending June 26th 2002 Week ending February 9th 2003 User Type Heavy Medium LightHeavy Medium Light Heavy Medium Light Heavy Medium Light Direction OUT IN OUT IN OUT IN IN/OUT IN/OUT IN/OUT OUT IN OUT IN OUT IN IN/OUT IN/OUT IN/OUT Normalized Traffic per Sub 266.8 445.5 27.0 48.9 1.0 4.8 1.7 1.8 4.8 288.3 415.1 26.1 47.8 1.1 5.2 1.4 1.8 4.8 AUDIO/VIDEO 0.1% 0.3% 0.1%1.9%0.4%2.7% 3.2 26.4 29.8 0.1% 0.5% 0.2% 2.2% 0.4% 2.6% 4.9 17.3 28.4 CHAT 0.2% 0.4% 0.6%0.8%2.9%2.0% 3.2 2.4 3.4 0.3% 0.6% 0.7% 1.2% 2.6% 2.3% 3.0 3.0 4.1 NEWS 1.1% 34.9% 0.5% 13.5% 0.2% 2.1% 53.6 54.1 55.1 1.0% 32.8% 0.4% 10.5% 0.1% 1.4% 49.6 46.6 46.2 MAIL 0.4% 0.1% 1.5%0.4%8.3%2.3% 0.5 0.5 1.4 0.1% 0.3% 1.3% 0.7% 8.1% 2.7% 2.7 0.9 1.6 FTP 0.7% 0.9% 0.6%1.1%0.8%0.3% 2.2 3.5 1.7 0.8% 0.7% 0.5% 0.8% 0.6% 0.2% 1.4 2.8 1.9 GAMES 0.4% 0.5% 1.5%1.5%2.8%1.0% 2.0 1.7 1.7 3.3% 1.9% 4.1% 2.7% 2.9% 1.0% 0.8 1.2 1.7 ESP/GRE 0.0% 0.2% 0.7%1.1%5.3%2.8% 6.9 3.0 2.6 0.1% 0.3% 1.0% 1.4% 6.0% 3.1% 5.6 2.5 2.5 P2P 87.4% 44.0% 82.3% 43.2% 18.5% 6.8% 0.8 1.0 1.8 37.7% 22.9% 29.5% 14.0% 7.0% 2.3% 0.9 0.9 1.6 TCP-BIG 6.9% 8.4% 3.3%6.3%2.4%2.5% 2.0 3.4 5.1 51.2% 30.5% 47.6% 29.3% 13.1% 6.8% 0.9 1.1 2.5 WEB 0.9% 5.3% 5.1% 26.6% 46.2% 71.6% 10.1 9.5 7.5 1.6% 6.5% 6.4% 31.5% 46.7% 72.3% 5.7 9.0 7.5 OTHER 2.0% 5.1% 4.0% 3.7% 12.2% 5.7% 4.3 1.7 2.3 3.9% 3.1% 8.2% 5.8% 12.5% 5.3% 1.1 1.3 2.1 Table 4: Comparison of the application composition of the heavy, medium and light user groups of a typical region.
On the other hand the heavy user group is Looking at the evolution of the traffic balance of Web traffic of the heavy users also mainly generating file sharing traffic. That leads to the conclusion that a more complex group is actually providing content to the rest phenomenon is happening. Indeed in June of the P2P community since its P2P traffic 2002, the web traffic balance of the heavy balance is below 1. Even though that subscriber users – 10.1 - was clearly higher than the web group accounts for only 2.9% of the subscriber traffic balance of the light users whereas, in population, it generates almost half of the P2P February 2003, that heavy hitter web traffic traffic (table 5). What is more surprising is that balance went down to 5.7, i.e. even lower than these P2P applications are not the only way for the one of the light users. This suggests that the heavy hitter class to download files. Only web traffic starts to be contaminated by a more 83.6 % of that group of users installed one of balanced traffic, namely P2P applications. these major P2P applications. This percentage Furthermore, the traffic balance per goes up to 96.7% if we take also Netnews into application is another evidence that most of the account. Finally the remaining 3.3 % chose traffic classified as TCP-big this year was other solutions that include FTP and actually what was classified as P2P last year. downloads from the Web. It is interesting to While the TCP-big traffic of the heavy hitters notice that Netnews and the Web are only increased enormously, its traffic balance means to download content but not to share it shifted from 2.0 to 0.9 and is now equal to the and so the traffic balance for these applications traffic balance of the P2P traffic that is still is very large: up to 50 bytes received for one classified as P2P. It is now high time to byte sent. understand why we are reaching the limits of Week ending June 26th 2002 port based identification of P2P traffic. Direction OUT IN User Class Heavy Medium Light Heavy Medium Light IP address Percentage 2.9% 30.1% 67.0% 2.9% 30.1% 67.0% Traffic Percentage 46.6% 49.4% 4.1% 41.6% 47.9% 10.5% LIMITING P2P TRAFFIC NEWS 68.6% 30.4% 1.0% 68.4% 30.5% 1.1% P2P 49.6% 49.5% 0.9% 46.2% 52.1% 1.8% TCP-BIG 64.9% 33.1% 2.0% 51.5% 44.5% 4.0% WEB 8.5% 52.2% 39.3% 9.8% 56.6% 33.6% The ability to accurately identify P2P P2P Users in that Class 83.6% 63.4% 10.1% 83.6% 63.4% 10.1% News Users in that Class 25.8% 12.4% 2.6% 25.8% 12.4% 2.6% traffic is a crucial requirement for News or P2P Users 96.7% 71.6% 12.6% 96.7% 71.6% 12.6% appropriately handling this traffic in the Table 5: P2P and News Users in a region having more than 100 000 network - through either traffic engineering, subscribers. provisioning, rate-limiting or pricing. However, P2P applications have evolved rapidly in a direction which makes accurate This conclusion is supported by the accounting of the traffic more difficult. In previous findings of this paper, but we shall particular, previously the applications used investigate in even more detail. Fig. 6 plots the default TCP ports, and it was possible to per-port traffic distribution for June 2002 and account for the bulk of the P2P traffic by February 2003, for the P2P or TCP-big ports monitoring a relatively small number of ports. for the 2 time periods. Note that in 2002, 60 % However, the current widespread use port- of that P2P and TCP-big traffic was hopping makes such mapping exceedingly contributed by only three ports. However, in impractical. We next present specific evidence February 2003, the traffic was much more of this trend and then discuss the implications uniformly distributed among a larger number of for managing this traffic. ports – the top 3 ports now account for only 20 % of the traffic. To get 60 % of the traffic we KaZaA Rate Limiting Experiment would need to monitor a larger number (1000) of ports. We first show an interesting case study which graphically illustrates how difficult it can Much more difficult is the task of mapping be to limit P2P traffic. In Fall 2002, a the traffic on these heavy-hitter ports to particular broadband region began rate limiting specific applications. Given the use of port- traffic on port 1214 (the default port for hopping by bandwidth-intensive applications KaZaA). Fig. 5 shows the IN traffic for web, like P2P, an important unanswered question is p2p and TCP-big for that region before and how much of the traffic on these ports can be after the rate limiting was initiated. Note that attributed to the IANA-registered applications, the P2P traffic decreases significantly after the and how much is P2P. rate-limitation was initiated. However, the TCP-big starts increasing and in 2 months has tripled compared to its value just before rate– limiting began. The web traffic (port 80, 8000, 8080) also increases over the same period. A reasonable explanation for the jump in the TCP-big traffic coincident with the rate limiting action on the KaZaA port is that the traffic spurt was caused by KaZaA traffic migrating to other ports that were mapped to TCP-big.
Normalized Traffic to Broadband Region X
5
4 Week Ending 07/28, Before
3 Week Ending 08/18, One Week Later Fig. 6: Distribution of traffic by TCP port numbers classified as P2P or Week Ending 09/08, One TCP-big 2 Month Later Week Ending 10/25, Two Months Later Given the limitation of port-based 1 accounting, one might try to develop 0 alternative techniques to accurately identify p2p web
TCP-big P2P applications. For example, additional Application information such as packet-level data,
Figure 5: Mutation of P2P traffic into TCP-big traffic. identification of SuperNodes etc. could help in developing signatures of P2P traffic. However, identify themselves as such and pay a price for P2P applications have exhibited remarkable the enhanced service they are receiving. ability to rapidly evolve to evade detection and Caps have been good to the industry and control. For example, many P2P applications take us part of the way to where we want to now encrypt their communications, making it go. However, P2P traffic is a relatively more difficult to reverse-engineer and/or “passive” phenomena. The requester can monitor such systems at the application-level. queue-up a set of requests for files then walk The above trends have important away. The file provider does not even need to implications for port-based traffic control of be at the serving PC. In this situation rate P2P applications. If the rate control is targeted capping will make the requests take longer, but to a few well-known P2P ports, a significant will likely not change the behavior of the P2P fraction of the P2P traffic will evade the limit participants. Fig. 1 enforces this point with the by hopping to other ports. The alternative is to lower correlation between P2P traffic with the track a larger number of ports that contribute times users tend to be at their PCs. significant traffic volumes and that are suspected to carry P2P traffic. The problem Attempts to manage P2P traffic explicitly with this approach is that (i) it may not be have met with little success. As illustrated in feasible to track such a large and potentially Figure 5, attempts to block standard ports of dynamic set of ports, and (ii) such a one P2P application only cause the user widespread rate control may adversely affect population to shift their behavior so that the the performance of many non-P2P users traffic reappears on other ports. Devices inside running valid applications on these other ports the network to block or significantly throttle – this would be undesirable for the broadband specific port numbers have questionable providers. economic return given the “slipperiness” of ports that P2P applications use and the risk that SERVICE EVOLUTION TO TAME THE valid applications also are using those ports. P2P GUERILLA Not that we should treat High Speed Data There are an assortment of approaches to Services as a classic utility, but let’s look at address the “problem” of P2P traffic. Let’s how other “utilities” handle the problem of review a few that may be applicable to the consumption hogs. Water, power, landline cable industry. phone utilities all have a “pay for what you use” model. There is no attempt in these Over the past few years many Multiple industries to limit the usage besides the System Operators (MSOs) have incorporated economic consequence of paying for what is “caps” into their service definition. These used. Cell phone providers put an additional service caps tend to be implemented by twist on this model and provide usage bands. controlling the rate at which data can flow into These bands allows a subscriber to sign-up for or out-of the network. The effect of these caps a usage band that best represents their need, is to limit the instantaneous peaks of on- but then gets charges for usage beyond what is demand transactions. This has started us included. With these revenue models down the path of keeping bandwidth hogs in consumption hogs are not “bad”, they are just check. Some MSOs are now adding “tiered big consumers. caps”. This allows the bandwidth hogs to User response to these revenue models may not be as bad as we may fear. Users will be concerned that this will raise their rates. more obvious controls, such as rate limiting Surveys suggest that many users, on the traffic on particular ports are shown to be average, feel they themselves are heavy hitters. ineffective, because they simply push the But Figure 4 suggests only 5% of the users are traffic onto alternate ports. A more practical creating 50% of the traffic. With strategic approach is to adopt a usage-based pricing selection of banding, the users will be approach, where the customers are billed for pleasantly surprised to find that they can buy the resources they use. one of the lower bands. There will be a small percentage of users (maybe the 1% that is REFERENCES causing the 20% of the traffic) that will not be happy with their new rates and will balk to [1] "Analyzing Peer-to-Peer Traffic Across other broadband services, but those are the Large Networks", S. Sen and J. Wang, ones that the cable industry can afford to lose. Proceedings of ACM SIGCOMM Internet Measurement Workshop, Marseilles, France, CONCLUSION November 2002. [2] "An Analysis of Internet Content Delivery In this paper, we examined a large set of Systems", S. Saroiu, K. P. Gummadi, R. J. flow-based measurements of network traffic Dunn, S. D. Gribble, and H. M. Levy, associated with broadband consumers, Proceedings of the 5th Symposium on spanning several months. Our analysis reveals Operating Systems Design and Implementation several interesting features. Firstly, they (OSDI), Boston, MA, December 2002. illustrate that broadband consumer traffic is [3] KaZaA supernodes: dominated by P2P applications. We further http://www.fasttrackhelp.com/kazupernodes/en look into the properties of the various [4] Cisco Corp. NetFlow Services and application classes, in particular the traffic Applications: patterns, and IN/OUT ratios, noting that P2P http://www.cisco.com/warp/public/cc/pd/iosw/i traffic has a much more balanced traffic oft/neflct/tech/napps_wp.htm pattern and IN/OUT ratio than applications [5] “Charging from sampled network usage”, such as the web. In addition we show that N.G. Duffield, C. Lund, M. Thorup. ACM geographic locality is not yet a dominant SIGCOMM Internet Measurement Workshop feature of P2P traffic. 2001, San Francisco, CA, November 1-2, 2001. The paper then considers the traffic patterns [6] Internet Assigned Numbers of user groups, showing that the well known Authority,protocol-numbers: 80-20 rule (80% of the traffic is generated by http://www.iana.org/assignments/protocol- 20% of the users) applies here, but moreover numbers that the group of heavy users actually tend to [7] Internet Assigned Numbers Authority, port- use different applications : they tend to numbers: http://www.iana.org/ generate more P2P and Netnews traffic, while assignments/port-numbers the group of light users tend to use more web, [8] “Fast Accurate Computation of Large- email and chat applications. Scale IP Traffic Matrices from Link Loads'', Yin Zhang, Matthew Roughan, Nick Duffield Finally the paper considers how one might and Albert Greenberg, to appear in ACM control the large volumes of P2P traffic that SIGMETRICS 2003. currently flood the broadband networks. The
PACKET NETWORK TOPOLOGIES FOR NEXT GENERATION VIDEO ON DEMAND AND SWITCHED BROADCAST SERVICE DELIVERY John Amaral and Paul Pilotte Artel Video Systems
Abstract Ring 100k - 120k Master This paper focuses on packet network Homes Passed Headend 400k - 500k architectures that are optimized for the Homes Passed 500 delivery of next generation Video-on- Homes Node Passed Demand and Switched Broadcast. The paper HFC Network explores the behavior of switched video Hub delivery networks that satisfy the growing 20k - 25k user demands for unique orthogonal Homes Passed sessions. A detailed analysis of video delivery infrastructure composition is undertaken. The paper discusses packet switching systems, optical transport, Layer 2 forwarding, QAM modulation and storage. A Figure 1 - Typical Cable Network hypothetical 300,000-subscriber VoD network is employed as the basis for THE MOVE TO VOD describing network behavior under several scenarios. The analysis culminates in a cost- The deployment of next generation video effective, extremely high capacity network on demand, EoD, and diverse content that dramatically increases bandwidth offerings by the MSO's in one form or resource utilization and provides dynamic another is regarded as a foregone conclusion. and agile program delivery. The disclosed MSO’s must deploy VoD services to topologies possess effective redundancy and counteract the competitive threats of Digital resiliency. Several practical examples are Broadcast Satellite (DBS). Most regard the considered with regard to the disclosed near-term rollout of these services critical to topologies; the examples include both reducing digital churn and to increasing “Everything on Demand” (EoD) and subscriber revenues. To date nearly all of the Switched Broadcast Services. The analysis is North American cable operators have predicated by the feasibility and practicality deployed video on demand. The fantastic of the described topologies. Considerations success of trials and early deployments has such as interoperability, cost, ability to motivated cable operators to accelerate deploy, and ease of use is taken into account rollouts. as important factors when describing the topologies. These service types rely on the ability of
the cable networks to deliver unique,
orthogonal video streams established by
dynamic user control. This facility behaves in
much the same way as the public telephony
network; a user connects to the network (lifts
the receiver), signals a unique switched communication path (dials the number) and of sufficient scale and density necessary to conducts a unique session (talks to the support thousands of interactive and unique desired party). One-to-one connectivity is content sessions, a volume of sessions necessary in telephony, and in data networks, requiring hundreds of Gigabits per second of because nearly every transaction context is bandwidth. unique; i.e. different content, at a different time, for a different reason. Based on The embodiment of such architecture must conservative estimates, 10 million unique and yield a system that is subjectively easy to use. simultaneous "one-to-one" streams will be It should also be scaleable in both size and deployed by North American MSO's in the capability. An optimal solution must have next five years. low capital and operational expense, high asset utilization, and manageable complexity. The move to session based video delivery will require a cost effective, high capacity EVOLUTION OF VOD NETWORKS switched packet network infrastructure, two way digital HFC plant and sophisticated Early Video on demand networks content processing, storage and management deployed video servers at the edge of the facilities. Fortunately, the 50 billion dollars network in a distributed model. The network spent for HFC plant upgrades and edge is the location in the hub where the bidirectional digital television capability QAMs interface to the HFC plant. At the underpin this endeavor. MSO's are left to time, low utilization, scarce content and focus additional spending to enable the head- limited investment in equipment made it end to hub network infrastructure for VoD. acceptable if not preferable to utilize resources in this fashion. In fact, some Today, the cost of deploying end-to-end MSO's still operate large distributed VoD networks capable of delivering such a large networks. Early deployments helped prove number of independent streams continues to the business cases for VoD. In addition, be prohibitive. The cost to provision a single important lessons about market behavior and stream (server, switch, transport, QAM, RF) consumer preference were learned. is around five hundred dollars (assuming a Experience gained during deployment of Gigabit Ethernet based delivery early-distributed VoD systems drove infrastructure). High equipment demand, and technical initiatives to further optimize VoD cost pressure from MSO's coupled with network design. technical advances and a growing competitive landscape is rapidly decreasing the per Distributed VoD systems are technically provisioned stream price. simple in that the delivery network is inherently localized. VoD servers sit in hubs, Accelerated deployment of advanced connected to QAM modulators and feed services relies on further advancement and channel groups dedicated to serve content. cost reduction of network components. This Generally, servers deployed for this purpose paper assumes equipment will inevitably reach had ASI output connections; in some cases direct QAM outputs. Asset distribution to acceptable price points. However, low servers varied from an intensely manual component costs are not sufficient to enable process, a technician driving from hub to hub large scale and effective VoD networks. A with a TK-50 tape in his hand, to more comprehensive architecture must be adopted automated systems having servers connected to effectively provision and manage networks by an out of band channel (ATM or IP network). This channel allowed content to be streams per ASI. When deploying small- inserted at a central location and copied to scale VoD networks with low peak load, the remainder of the VoD servers via FTP. provisioning services at this granularity is acceptable. As peak load demand increases, MSO's rapidly realized the shortcomings it is necessary to provision unacceptably of distributed VoD systems as they began to large numbers of ASI links. The low ASI deploy en masse. In a distributed transport capacity undersubscribes server architecture there is invariably a mismatch resources and physical transport assets between the playout capacity of the VoD (fibers, lasers etc). Replacing ASI transport server and the number of provisioned with Gigabit Ethernet transport can increase streams. This mismatch is due to the lack of the payload capacity by 500% to 1000% for granularity inherent in most VoD servers. the same cost. The second disadvantage is The unused playout capacity drove the cost that the payload containers carried by ASI per stream unacceptably high. Storage are MPEG-2 transport streams. MPEG-2 utilization was also poor because each server transport streams are not intended, nor in the network needed to be loaded with capable, of forming the basis for a switched exactly the same content. In addition network interconnect. The virtues of a packet operational expenses were unacceptably high switched interconnect for VoD will be because service personnel needed to travel to discussed in the next section. The effect of each hub to maintain and upgrade VoD these two properties is unacceptable cost per servers. An apparent solution was to bit transported. In order to reach the cost centralize video server assets in a common points necessary to proliferate VoD and head end and transport the streamed sessions move toward EOD, most believe that a to the hubs. The change to this "centralized" switched packet interconnect based on approach better matched server playout Gigabit Ethernet is necessary. capacity to stream demand, thereby reducing the amount of unused server capacity. Load GIGABIT ETHERNET AS A BASIS FOR sharing allowed storage to be arranged such VOD NETWORKS that the amount allocated to a title is proportional to the number of simultaneous The benefits of Gigabit Ethernet sessions demanded of that title. This resulted technology in transport systems for current in improved cost per stream because fewer and next generation VoD networks begin servers were needed to service the same with its ubiquity. Quite simply, lots and lots number provisioned streams. of Ethernet equipment is bought and sold each year. This assures the contributing ASI optical transport is routinely electronic components will remain deployed for VoD and SVoD by MSO's commodity. Ethernet’s plug-and-play today and has been an effective choice for interoperability makes networks deployed building low to medium scale VoD systems. with Ethernet easy and cost-effective to Rings and point-to-point ASI over DWDM install, manage and use. Gigabit rate optical networks connect VoD servers to QAM and electrical interfaces prevalent in the data modulators in remote hubs. communications world provide a high-speed, robust interconnect between servers, edge There are two inherent disadvantages of devices and processing elements connecting ASI transport systems. The first disadvantage to the transport infrastructure. Contemporary is that payload capacity of ASI is only about servers and storage systems are optimized to 216 MbpS. That comprises about 40 VoD operate in increments of 1 GbpS and make excellent use of off the shelf network maximum capacity. System cost is reduced by adapters. A 1GbpS Ethernet link can virtue of needing fewer lasers. This is transport up to 240 VoD streams, a suitable important because laser cost is the single increment of streams to deploy in large scale largest contributing factor to overall VoD VoD networks. Another advantage of transport costs. Layer 2 capability enables Ethernet is its media access control (MAC) aggregated links to be demultiplexed and layer. The Ethernet MAC layer is a data link delivered to the correct destination. Packet protocol with sufficient properties to form a switching provides full mesh connection sophisticated, and extremely scaleable packet between content sources and HFC switched network. Switching allows effective destinations allowing servers to load balance. bandwidth management thereby reducing per stream costs. Figure 2 illustrates the benefits of Layer 2 aggregation over Layer 1 transport. In the INTRODUCTION TO NEXT Layer 1 example, content from each VoD GENERATION VOD TOPOLOGIES servers can only be directed to the corresponding downstream QAM. What does the ideal VoD network look
λ1 16xQAM like and why? VoD networks provide service 600Mbps λ2 600Mbps λ3 VoD 600Mbps L1 L1 600Mbps 16xQAM Server Xport Xport to a geographical region approximately the 600Mbps 600Mbps size of a large city. They are high in 16xQAM transmission capacity, moderate in complexity 24xQAM and provide robust, reliable interconnect. 600Mbps λ1 900 Mbps VoD 600Mbps L2 L2 900Mbps 24xQAM Server Xport Xport They are typically deployed as overlay 600Mbps networks, and are not intended to replace existing broadcast infrastructure or data/voice 16xQAM networks. The motivation is to develop low 900Mbps λ1 600 Mbps VoD 900Mbps L2 L2 600Mbps 16xQAM Server Xport Xport cost per stream networks tailored for the 600Mbps 16xQAM unique properties of VoD traffic. The systems envisioned have basis in Ethernet as the data link protocol, no different from a Figure 2 – Layer1 vs. Layer2 Aggregation LAN. The departure between a standard Ethernet LAN and VoD network is the In the Layer 2 examples, content from all allowed distance between end terminals, three VoD server ports can be aggregated to intermediate path capacity, and optical route form a fully utilized wavelength, thus usage (DWDM). In order to develop a reducing fiber and laser costs. In addition system optimized for VoD delivery the video streams can be directed to any of the following capabilities must be employed. QAM devices from any server port on a packet-by-packet basis. Layer 2 Aggregation for Bandwidth Recovery Layer 2 Forwarding and Shared Wavelength Layer 2 (Ethernet MAC layer) switching Topologies can be used to combine multiple partially filled Ethernet links to build fully utilized Layer 2 forwarding allows the contruction optical transmission paths between headend of shared wavelength topologies. Video assets and hubs. Fully utilized optical links streams entering multiple inputs to a head ensure each transport laser is operated at end transport device can be aggregated and tagged with information which allows them
λ1 24xQAM to be discerned by hub end devices 900Mbps λ2 900Mbps λ3 VoD 900Mbps L1 λ4 L1 connected a shared optical path. Each device Server Xport Xport 900Mbps on the shared path can selectively receive 16xQAM 600Mbps any of the video streams available on that L1 Xport
path and forward them to the QAM. This 24xQAM 800Mbps optimizes bandwidth utilization by allowing L1 streams to be delivered to a number of Xport 16xQAM QAMs using only the bandwidth they 400Mbps L1 instantaneously require. This results in the Xport lowest cost per video stream.
Figure 4 - Layer 1 Point-to-Point Topology λ1 24xQAM 900Mbps 900Mbps Figure 4 shows a logically equivalent VoD 900Mbps L2 L2 Server Xport Xport topology using Layer 1. Note that the Layer 900Mbps 16xQAM 600Mbps 1 topology is point-to-point and requires four L2 underutilized lasers while the Layer 2 Xport 24xQAM topology with the shared ring uses only one 800Mbps L2 fully utilized laser. This results in a much Xport
16xQAM higher cost-per-stream compared to the Layer 400Mbps L2 2 solution. Xport Asymmetric Reverse Path Figure 3 - Layer 2 Shared Ring Topology Up to now the topology diagrams have An example of these benefits is evident in shown the forward video path only. There Figure 3. In this case Layer 2 switching may be a need for a reverse path to carry aggregates three server outputs and places control and management traffic from the hubs the video streams on a single wavelength of to the head end. This reverse path typically the shared ring. Multiple hub end devices are requires an order of magnitude less connected to this single wavelength. The hub bandwidth than the forward path. In a Layer end devices are configured to receive only the 1 system, bidirectional links must be used to video streams that are destined to their support a reverse path. In this case the associated QAMs. Ethernet allows this to be reverse path would have the same cost as the done automatically with no user intervention. forward path even though it does not carry Bandwidth to the QAM devices is allocated in video content and is underutilized. arbitrary proportions as required. This is Figure 5 shows an example of asymmetric extremely useful in multicast environments reverse path for hub to head-end where single copies of a video stream are interconnect. Since VoD traffic is presented to all hub devices, but selectively predominantly from head-end to hub-end with switched to QAMs based on user demand. comparatively low-bandwidth control traffic required for the reverse direction, Figure 5 shows a topology where a single wavelength transports the reverse path management information for all of the downstream QAMs to the head end over a single wavelength. This preserves all but one wavelength for deployment of diverse implementations forward path traffic. within a single large network. Furthermore, deployments may begin with a few Gigabit 24xQAM VoD L2 L2 Ethernet links and grow to many more as the 24xQAM Server Xport Xport 24xQAM number of provisioned subscribers increases.
24xQAM Equipment used for the early deployments L2 L2 VoD 24xQAM Server Xport Xport 24xQAM must be cost-effective for the initial small number of links yet scale and remain cost 24xQAM VoD L2 L2 24xQAM Server Xport Xport effective in the growing network. 24xQAM 24xQAM L2 VoD L2 24xQAM Switched broadcast Server Xport Xport 24xQAM Switched broadcast is an application Figure 5 - Asymmetric Reverse Path intended to solve the problem of delivering limitless content choices within the finite Layer 2 aggregation concentrates the limitations of the provisioned infrastructure reverse path control traffic from the hub onto by minimizing bandwidth utilization in the a single gateway element, which then places transport and HFC networks. By this means this reverse path traffic onto an available only program content being watched is optical wavelength in the ring. All other delivered over the network. Furthermore, wavelengths in the hub are receive-only and content watched by multiple users is do not require transmitter optics. Likewise, transported as a single copy over the optical at the head-end the wavelengths not used for network. Distributed Layer 2 switching reverse traffic are transmit only and do not performed by transport devices in the head- need receivers. This tailoring of transport end and hubs effectively utilizes bandwidth by optics to better match the forward and directing content needed by that hub over a reverse path bandwidths of VoD traffic can shared optical ring. Without a switched provide significant cost-per stream savings. broadcast solution, video content demand will eventually exceed the available bandwidth of The optimal VoD network makes the cable infrastructure. Deploying systems extremely efficient use of high capacity that can support switched broadcast services DWDM optical components and in particular which future-proof the VoD network for this optimizes the use of cost-effective solutions eventuality. available today. Efficiency is gained through fully utilizing each optical wavelength by a TRANSPORT OPTICS combination of link overhead minimization and Layer 2 traffic aggregation. Essentially The optical components and topologies of each wavelength is operated near theoretical the VoD transport network are designed to maximum capacity. provide the lowest cost while fulfilling the requirements for bandwidth, reach/distance, Scalability resiliency, and ease-of-use. Figure 7 shows a typical transport network for hypothetical 5- MSO network planners require hub model. networking equipment to be cost effective and scaleable for both large and small market Video streams from the VoD servers are regions. They also require that the passed through an Layer 2 Ethernet switch, equipment can satisfactorily support the video gateway elements which perform Layer
2 aggregation of video streams onto a SVOD 100 100 1000 5000 NPVR DWDM optical network, optical Titles Titles Titles Titles + multiplexors; single-mode optical fiber to SVOD Peak Load% of HP 0.5% 1% 1.5% 2% 4% 10% connect to an adjacent hub, optical splitters, Streams 100 200 300 400 800 2000 optical protection switches, optical #GigE’s@16QAM 0.6 1.3 1.9 2.5 5 12.5 #GigE’s@16QAM 0.4 0.8 1.2 1.6 3.3 8.3 demultiplexors, video gateway elements Table 1 - Optical Technology Comparison for performing Layer 2 deaggregation, and “5-Hub” Model Assuming 10% Subscription Gigabit Ethernet QAM devices. Rate
Bandwidth Reach / Distance
The VoD transport network is designed to The fiber distances between the head-end carry VoD streams of personalized video node and hub nodes in the VoD transport content for each viewer from VoD servers to network introduce optical loss and dispersion destination QAM devices. Each VoD stream which need to be accounted for to maintain requires 3.7Mbps of payload bandwidth error-free transmission. The optical losses quantized as 188-byte MPEG-2 video are introduced at connector boundaries, packets. Digital packet switching using splices, mux/demux elements, and Ethernet has become the defacto standard predominantly in the fiber itself. Network transport mechanism for VoD because of its engineers compute an optical link budget by ubiquity and low cost. The typical packet subtracting the optical receiver sensitivity and encapsulation scheme adds 3.4% overhead optical losses from the optical transmitter and uses 7 MPEG-2 packets (1316 bytes) minimum output power. A link budget with encapsulated over Ethernet (18 Bytes) over adequate margin ensures adequate optical IP (20 bytes) over UDP (8 bytes). Table 1 power at the receiver; while a link budget shows the number of VoD streams, which can with no or negative margin indicates the need be carried over a single fiber using various for an optical amplification device (an EDFA) transport mechanisms. in the optical path. EDFAs (Erbium Doped Fiber Amplifiers) use active Erbium-doped Table 1 illustrates the number of Gigabit fiber and a laser pump source to boost the Ethernet ports provisioned based on service optical signal in a fiber. EDFAs become offering and provisioned peak load. Note essential elements in a VoD transport that for all services above 100 Titles + SVOD network with medium to long spans between and a peak load greater than 1.5%, Gigabit hubs. Ethernet links serve as ideal containers. Dispersion in optical fiber will eventually Furthermore, 3G optics are sufficient for up result in unacceptably low bit error rates. to 5000 Titles while 10G optics remain highly Because dispersion cannot be compensated underutilized and therefore result in higher by EDFAs, the optical distance (also called cost per stream. reach) limit is determined by the dispersion characteristics of the optical transmitter. When the VoD distance requirements exceed the laser optical dispersion limit, network architects must either add O/E/O regenerators or add another head-end, both costly alternatives.
Resiliency An effective strategy for optical resiliency is to design protection for those elements Resiliency is the networks’ ability to whose failure would affect the greatest recover and sustain traffic in the presence of number of VoD users. Optical protection fiber cuts or equipment failures. VoD optical costs for VoD can be lowered by moving the transport networks should be designed with protection for fiber cuts from the Layer 1 resiliency in mind to eliminate single points of electrical layer (as is done with SONET) to failure and provide for redundancy of critical the optical layer. This eliminates the added components and a means for automatic cost of 1:1 or 1:N electrical interface detection and failover. The asymmetric protection. Ideally, individual optical nature of VoD and cost per stream pressures transmitter and receiver redundancy favor optical protection architectures which protection could be optionally provided, allow differing levels of resiliency which can allowing network designers to make the be chosen to optimally balance the cost per cost/resiliency tradeoff per wavelength while video stream against optical protection still maintaining 100% protection for fiber coverage and switchover times. cuts.
The telecom industry has provided very Ease of Use mature network topologies (SONET/SDH and 2-fiber and 4-fiber BLSR) and elements The operational expense of the VoD for achieving full redundancy of optical transport network is minimized and system equipment and fiber and very fast protection uptime maximized by providing an detection and switching times (less than 50 integrated network which is easy to install, milliseconds) that exceed the times needed for replicate, and maintain. The interoperability VoD. While these can be used to provide and ubiquity of Gigabit Ethernet has made effective resiliency for VoD transport they do the interface to the VoD server, video so at a high cost per video stream, add gateway elements, and QAM devices additional management complexity in inexpensive and easy to maintain. The optical mux/demux, splitters, and ADPs are managing an additional transport layer, and passive optics requiring no on-line dictate a costly symmetrical ring network management and have very high reliability. topology that does not match the inherent The EDFAs and optical switches can either asymmetric nature of VoD transport traffic. be deployed as unmanaged devices which Unlike telecom networks, cable networks do power-up in a default state or can be on-line not have strictly defined redundancy managed; for example via SNMP or a requirements and the ability to choose and Network Management System. optimize redundancy cost/performance is desirable. VoD transport networks that are Optical Elements designed for the same telecom resiliency goals (for example hybrid networks designed The optical elements used in the transport to carry both voice and VoD traffic), these network are well understood and have been topologies can be very effective. However, used in existing cable networks for years. while the burgeoning VoD network The laser transmitter performs the electrical infrastructure demands favor the lowest cost to optical conversion of data streams. To approach, other resiliency schemes become achieve the bandwidth required for VoD, more favorable. several optical channels are used per fiber
using Dense Wavelength Division Multiplexing (DWDM). The key parameters PACKET FORWARDING AND for laser choice are cost, optical reach, and SWITCHING SCHEMES overall bandwidth. 1G lasers achieve the lowest cost per laser, but the limited Previous sections discussed and bandwidth results in a relatively high cost per highlighted the subjective usefulness of Layer stream. 10G lasers have high bandwidth, but 1 and Layer 2 constructs and their relative are costly and have distance limitations due to benefits in developing a suitable VoD optical dispersion. 10G optics also have infrastructure. The following sections provide coarse cost granularity, requiring high a detailed description of how traditional incremental cost as additional bandwidth is packet based constructs are used to fulfill the added to the network. An optimal key capabilities described earlier. compromise of cost per stream and optical reach can be achieved using 3G lasers with Layer 1 Transport extended optical reach. For VoD network equipment not capable of Layer 2 and higher packet switching, Layer 1 provides a “direct wire” interconnect
$4 between the head-end and hub elements. 1 G Soln 3 G Soln Layer 1 transport provides physical path $3 10 G Soln connectivity between two or more endpoints on an optical link. Link information entering $2 the input ports of a Gigabit Ethernet Layer 1 transport device are interleaved and encoded $1 Relative Cost Per Stream Cost Per Relative on an optical carrier. Optical wavelengths may accommodate one or more Layer 1 $- 0 102030 signals. Time Division Multiplexing generally # of Gigabit Ethernet Ports accomplishes this. Layer 1 switching, commonly referred to as cross bar switching, Figure 6 - Optical Laser Cost Comparison offers some path flexibility by which source Optical receivers are used for optical to and destination Layer 1 ports can be cross electrical conversion of data streams. connected arbitrarily. Layer 1 devices, EDFAs (Erbium Doped Fiber Amplifiers) are however simple to provision, do not support used to provide amplification of all fractional link aggregation and therefore wavelengths on an optical fiber to offset fiber typically underutilize optical transport and connector losses and ensure that capacity. For example, a Layer 1 device sufficient optical power is available for connected to a 600Mbps VoD server will downstream receivers. Low cost passive have 60% optics and fiber utilization, while a optical mux/demux and splitter elements are Layer 2 device can achieve near 100% used in the optical fiber distribution plant utilization by aggregating multiple fractional from head end to hub. Optical switches are links. In addition, Layer 1 function does not used as protection devices to automatically provide shared wavelength-forwarding switch to alternate fibers in a ring topology. capabilities necessary for such schemes as switched broadcast. Layer 1 devices can only provide full duplex or simplex links. Operational complexity grows unreasonably with scale because transport assets cannot be forwarding to interfaces based on destination dynamically reallocated or shared. Also, information resident in the Ethernet packets, because there is no fractional rate support, switching is accomplished. Switching asymmetric traffic patterns cannot be utilized forwarding and classification are used in the for reverse trunking. presented topologies to develop several classes of flows. The flow types are the While Layer 1 interconnect simplifies the following. network topology for small-scale VoD networks, it fails to provide the benefits of Path routes bandwidth recovery, shared rings, and switched broadcast. Flows are grouped based on port affiliation and act like a virtual wire. Packets entering a Layer 2 Capabilities physical port on a Gigabit Ethernet transport device are grouped together and transferred The Layer 2 behavior of the proposed high to a corresponding destination port over the capacity transport network differs slightly fiber optic plant. from a conventional Ethernet LAN but is implemented in such a way that equipment Layer 2 groups costs are low and Ethernet interoperability is preserved. The following sections are a Layer 2 groups are formed by the technical tutorial on the salient Ethernet aggregation and dissemination of traffic to functions used to achieve the desired VoD and from multiple sources. Filtering is network features such as switched broadcast performed and decisions are made about and shared rings. which packets go where on a packet-by-
packet basis. The use of multicast addressing Layer 2 classification and forwarding provides capability to forward the single packet to one or more destinations. Layer 2 classification and forwarding are the principal operations by which packets Tunneling entering the transport domain are identified, organized and delivered to one or more Tunneling is a mechanism used to trunk destinations. The destination address equivalent flows to a common end contained in Ethernet packets are identified destination. Tunneling allows Layer 2 devices by the classification process. The result of this to forward aggregated flows on a shared classification process is a list of forwarding virtual medium. Packets are grouped destinations. The Layer 2 learning process together and transported through the network ascertains the forwarding destinations. The as "equivalent" flows. The terminal transport Ethernet frame under consideration is node disseminates the flows through encapsulated on the optical link and identified classification and delivers each packet to the as a frame addressed to the downstream described destination. device listening on the selected optical wavelength. The act of identifying Ethernet frames based on their destination address, aggregating them with equivalent flows, and presenting them to downstream devices is called forwarding. By virtue of selective Layer 2 filtering transport of more granular traffic flows such as reverse path traffic. Asymmetric reverse Layer 2 filtering works in conjunction with path forwarding is an example of Layer 2 the classification process and provides a flow aggregation that exploits the traffic mechanism to scope and restrict traffic flows. patterns in VoD transport networks. Filtering can be configured to drop frames based on destination address or matched Layer 2 asset provisioning and load balancing filtering criteria. A common use is to manage unknowns in the network. Unknowns are A significant advantage gained through packets for which the destination is not Layer 2 switching is the ability to ideally present, or is unreachable in the network. match content delivery assets to transport Unknown filtering is useful in defeating infrastructure capability. Switching allows broadcast storms and forwarding loops that equipment providing session fulfillment, such can result in service disruption due to as video servers, to reach any endpoint in the excessive bandwidth consumption. network. This allows servers to share the workload in delivering VoD sessions. Layer 2 add, drop, pass Layer 2 address learning and aging Layer 2 add, drop, pass capability allows the formation of packet rings and provides a This function utilizes Ethernet source basis for multicast and broadcast over a single address awareness and classification to wavelength. Layer 2 flows are injected and determine destinations for packets impinging terminated by members of the optical ring. the network (i.e. you don't have to know Multiple members of the ring can receive a which downstream port to plug the QAM singular flow. Shared optical wavelength modulator into learning will find which port paths, used in conjunction with "star over its plugged into). Learning also provides ring" paths, provide a way to organize information to forwarding/filtering functions transport bandwidth based on steady state to automatically determine which end points and transient load. Star topologies deliver the are reachable by the network. basis bandwidth which is the steady state load. Ring topologies provide a mechanism SUMMARY for bandwidth leveling and a shared medium for delivery of content assets, and switched The VoD architectures presented in this broadcast services. This hybrid approach paper can be summarized by considering the offers the ability to manage and mitigate high capacity 5-hub 300,000 homes passed transient bandwidth demand with little or no VoD network model presented earlier. The over provisioning. Both star and ring key architectural and cost advantages of this connections may exist within the same fiber model are high VoD stream capacity per or within tunnels on the same wavelength fiber, long optical reach, automatic optical protection switching, and a low-cost Layer 2 Path and Flow Aggregation asymmetric reverse path.
Layer 2 Path and Flow Aggregation are The optical transport system is used to concatenate traffic and fully utilize constructed as a hybrid ring/star architecture. optical paths. They also provide the ability to A northbound ring and a redundant utilize aggregation to virtual domains for southbound ring connect the head-end and five hubs. Each ring has 100GHz spaced DWDM 3Gbps wavelengths, which in demultiplexes the VoD streams from the aggregate carry, over 33,000 VoD streams. optical wavelength and switches individual Transmitter/receiver optical link budgets VoD streams to the appropriate QAM device. exceeding 32dB are achievable with low-cost This allows any VoD stream from any VoD laser transmitters and six spans each server to be directed to any head-end QAM exceeding 25km without the need for output. regeneration and using two EDFAs per ring. A lower cost per stream point could be Gateway elements vary in complexity from achieved with shorter reach optics, but would simple Layer 1 devices to fully featured Layer add more network design complexity and 1-4 devices capable of advanced features such limit the flexibility of deploying the same as switched broadcast. A Layer 1 device architecture in nearly all VoD deployment functions as a “wire” connecting entire areas. Gigabit Ethernet ports (with all VoD streams remaining intact) from the head-end Gigabit The duplicate rings provide 1:1 protection Ethernet switch to a hub-end QAM. Adding for fiber cuts anywhere along the transport Layer 2 capabilities into the gateway elements ring as well as failure of any of the optical allows individual Ethernet frames (and the splitters and EDFAs used on the ring. By VoD streams they are carrying) to be monitoring the recovered optical power at the individually switched and aggregated which destination, the optical protection switches gives much finer granularity in provisioning can failover without any user intervention. traffic from the VoD servers to the hub-end The topology can be constructed with an QAMs. This allows advanced features such asymmetric reverse path where the head-end as traffic aggregation and load balancing uses mostly transmitter-only optics and the which can save VoD costs by reducing the hub-end uses mostly receiver-only optics. required number of Gigabit Ethernet switch This reduces the optical transceiver and QAM ports. components (which can be 20% to 40% of the overall transport network cost) nearly in In conclusion for a VoD transport half. network to be cost effective and scaleable, it must be more than just a large pipe. Optimal In the hybrid ring/star topology, the use of fiber bandwidth and reducing the cost Ethernet switch at the head-end forms a star of transport optics are essential to reducing Ethernet connection between the VoD per stream costs. Layer 2 VoD stream servers and the QAMs. The gateway aggregation reclaims fallow bandwidth. elements then aggregate the VoD streams Asymmetry in the reverse optical path onto an optical wavelength, each of which is reclaims fallow fiber bandwidth and cuts the connected via the optical transport ring. A cost of DWDM transmitters and receivers corresponding QAM at a hub-end nearly in half.
Hub#1 L1/L2 Split QAM De Switch Head-End Switch Split Mux L1/L2 Switch QAM Ethernet L1/L2 Switch Switch Hub#2 L1/L2 L1/L2L1/L2 Split QAM SwitchSwitch De Switch Switch VoD Split Mux L1/L2 Server L1/L2 QAM Switch Switch
L1/L2 Switch Hub#3 L1/L2 Split QAM L1/L2 De Switch VoD Switch Mux Splitter Switch Server Mux L1/L2 Split QAM L1/L2 Switch Switch
L1/L2 Hub#4 Switch L1/L2 Split QAM De Switch VoD L1/L2 Switch Server Switch Split Mux L1/L2 Switch QAM L1/L2 Switch Hub#5 L1/L2 L1/L2 Switch Split QAM De Switch Switch Split Mux L1/L2 Switch QAM
Figure 7 - 5-Hub VoD Network Model
CONTACT INFORMATION:
John Amaral Artel Video Systems 237 Cedar Hill Street Marlborough, MA 01752 [email protected] 508-303-8200 x285
Paul Pilotte Artel Video Systems 237 Cedar Hill Street Marlborough, MA 01752 [email protected] 508-303-8200 x304 QUALITY OF SERVICE OVER HOME NETWORK USING CABLEHOMETM 1.1
Amol Bhagwat Cable Television Laboratories, Inc
value-added broadband services to their Abstract subscribers over any available home- CableHome™ 1.1, the latest version of the networking technology in a seamless and CableHome specification by CableLabs®, has convenient manner. The CableHome 1.0 defined a Quality of Service (QoS) solution Specifications were released in April 2002 for home networks. The key challenges in and certification testing of the products designing the CableHome 1.1 QoS system began in October 2002. CableHome 1.0 were: varying degrees of QoS support from specifications also gained international home-networking technologies, support for acceptance via International legacy home LAN devices, and backward Telecommunications Union (ITU) in 2002 compatibility with CableHome 1.0. The when ITU document J.191, that adopted CableHome team specified a priorities-based CaleHome 1.0 specifications almost entirely, QoS system in the CableHome 1.1 was consented as a fully approved ITU specification that addresses these key recommendation. challenges. The main functionalities of this The CableHome 1.0 specification QoS system are prioritized queuing, standardizes functionality for a residential prioritized media access, and provisioning of gateway device that simplifies manageability application specific priorities. The first of subscriber’s home network [1]. The functionality resides only in a residential following are some of the key features gateway whereas the later two are part of offered by CableHome 1.0: both a residential gateway as well as home 1. Remote configuration and LAN devices. Provisioning of application management of residential gateway in specific priorities is a very simple process. a secure manner. Hence the CableHome 1.1 QoS solution is 2. Hands-off authentication and easy to deploy and implement by cable provisioning of residential gateway. operators. In addition, since this design 3. Application and cable friendly allows legacy home LAN devices to co-exist standardized NAT/NAPT with complaint QoS-enabled devices it is a 4. Secure download of software images convenient solution for consumers too. 5. Firewall management and rule set download. 6. Remote home LAN devices visibility INTRODUCTION & BACKGROUND and connectivity tests 7. Local Name Service CableLabs, a research and standards 8. Protection of cable network from development consortium for the cable home network traffic industry, has initiated the CableHome project at the direction of its member cable television A follow on version of the specification companies. The project is aimed at is CableHome 1.1. In addition to enhancing developing a managed infrastructure that capabilities of the residential gateway, enables cable operators to offer high-quality, CableHome 1.1 extends its reach beyond the residential gateway to devices in the home residential gateway as well as in home LAN. CableHome 1.1 specifies a new set of LAN devices to be able to manage and functionalities for home LAN devices that obey these characteristics. enable several new key features including: 1. 2. Packet Forwarding and Queuing: This Quality of Service (QoS) over home is a functionality of a residential gateway networks, and 2. Device and services or a bridge in which packets arriving at discovery. In general new capabilities that multiple interfaces are to be retransmitted CableHome 1.1 enables are as follows: through another outgoing interface. This functionality needs to be enhanced so that 1. Standardized firewall configuration packet forwarding is performed to meet 2. Configuration file authentication the necessary QoS requirements. 3. Simple Parental Control 3. Management of QoS Characteristics: 4. Static Port Mapping This functionality deals with assignment 5. VPN Support of QoS characteristics to various devices 6. QoS over the home network and applications in the home and remote 7. LAN Management Messaging manageability of these characteristics. 8. Device and Services Discovery This functionality is a part of both This paper focuses on the home network residential gateway and home LAN QoS functionality designed in CableHome devices. 1.1. The paper first discusses key challenges There are two main QoS paradigms that involved in designing a generic QoS system can be utilized to provide the aforementioned that could be overlaid on any OSI layer-2 functionalities: parameterized (planned, home-networking technology in the second guaranteed) QoS and prioritized section. The third section describes in detail (differentiated) QoS. the CableHome 1.1 QoS solution and how cable operators can implement it. The Prioritized QoS: The prioritized QoS implications of this QoS solution are paradigm entails providing differentiated discussed in the fourth section and the last shared media access to the traffic based section presents the conclusions. on priorities and prioritized queuing and forwarding in a residential gateway and CHALLENGES IN DESIGNING A in a bridge. This mechanism does not GENERIC QoS SYSTEM OVER HOME provide performance guarantees for QoS LAN parameters such as bandwidth, jitter and A quality of service system over home delay. networks can be provided via three main Parameterized QoS: In this paradigm, functionalities: performance guarantees for QoS 1. Management of shared media access: parameters can be provided to the traffic When multiple devices are sharing the over the network. This is a planned same transmission media some approach for allocating resources on a mechanism is required to manage the network. Such planning is done based on access to this media. This involves the prior knowledge of resource manageability of various traffic QoS requirements of various devices and characteristics such as traffic priorities, applications in the network. bandwidth, jitter, and latency. In order to There are pros and cons for each of these implement such management a certain set paradigms and it was necessary that the of functionality needs to reside in a methodology chosen for CableHome 1.1 QoS Shared vs. Point-to-point Media solution satisfy the requirements set forth by cable operators for CableHome 1.1. The key With respect to QoS considerations cable operator requirements for CableHome different OSI layer-2 home-networking 1.1 QoS solution were: technologies can be categorized into two main categories: point-to-point technologies • It should be able to support legacy home and shared media technologies. For a point- LAN devices and best effort traffic such to-point technology there is a direct that they can coexist with new QoS- connection between two devices that are enabled devices. communicating with each other, e.g. • It should be OSI layer-2 home- Switched Ethernet. However, in case of networking technology independent shared-media technologies all of the devices • It should be software upgradeable from share the same media for all of their CableHome 1.0 communications. Most of the home- There were several challenges in networking technologies such as 802.11 fulfilling these requirements. The rest of this a/b/g, HomePNA, HomePlug, are shared- section is dedicated to discuss these media technologies. For such shared media challenges. technologies some mechanism is required to control how and when devices transmit data on the media. This can be achieved by Varying Degrees of Qos Support From employing either parameterized or prioritized Different Standards Based OSI Llayer-2 QoS paradigm. Home-Networking Technologies
The requirements for the CableHome 1.1 QoS solution mandated that cable operators Support for Prioritized QoS should be able to overlay the QoS system on Most of the standards based shared media any standards based OSI layer-2 technology. technologies- 802.11 a/b, HomePNA and This requires that the QoS system should be Powerline (HomePlug) have support for designed strictly at OSI layer-3 and above. priorities based QoS scheme. 802.11 a/b and Due to this fact such a system is dependent HomePNA supports 802.1p/q [2] priorities on the underlying home-networking while HomePlug has native priorities technology for its QoS support at the MAC support. In general, for these technologies, layer. However, the support for QoS in prioritized media access is accomplished by different standards based home-networking providing preferential media access for technologies varies from technology to higher priority traffic. The highest priority technology. It is essential to assess this traffic gets first opportunity to transmit its support in order to design a QoS system that data on the shared media and then, depending could be OSI layer-2 home-networking upon the bandwidth availability, lower technology independent and is still realistic. priority traffic gets subsequent opportunities (See Appendix 1 for information on QoS to send their data. support in leading standards based home- networking technologies.) Support for Parameterized QoS The amount of bandwidth consumed by higher priority traffic cannot be controlled by using a prioritized scheme. A parameterized scheme is necessary for such a control.
Parameterized QoS requires that the should not incur substantial inconvenience to underlying PHY/MAC technology be able to either customers or to cable operators when it deliver constant bandwidth and jitter. It is has to co-exist with legacy devices. very difficult to achieve this for home Additional cost for hardware or software networking technologies based on wireless, upgrades of legacy home LAN devices so phoneline, and powerline as underlying that they can coexist with QoS-capable throughput and jitter can be strongly CableHome complaint devices was influenced by rapidly changing interference. considered highly undesirable, e.g. requiring Perhaps due to these reasons, at the time a “QoS adapter” for best effort devices adds when CableHome 1.1 QoS system was being additional cost and is inconvenient for the designed, none of the standards based home- customer. Thus the challenge was to devise a networking technologies supported a truly QoS solution that will not require an upgrade parameterized QoS scheme. to the legacy devices and will make sure that best effort traffic from these legacy devices will not interfere with the traffic from QoS- Special Case of Ethernet: enabled devices in the home. Most existing Ethernet hubs in home LANs today do not support either a 802.1p/q Prioritized Approach priority scheme or a parameterized scheme and it is more than likely that Ethernet will A prioritized media access system can be not support these capabilities in future. overlaid on existing shared-media home However, when CableHome 1.0 is deployed networks. Even though most of the current in a consumer’s home, existing hubs are standards based home-networking likely to be replaced with switches that are technologies support prioritization of traffic, integrated in the CableHome 1.0 residential in general, these prioritization schemes are gateway devices. For switched Ethernet, not consistent and there is no central entity differentiated media access is not of much managing the priorities in the home. A value, in many cases; since traffic is residential gateway in the home LAN can essentially point-to-point and it is likely that perform the function of priority assignment, such a link is less contentious. Finally, on behalf of a customer, for various 100Mbps bandwidth seems to be sufficient to applications and devices, at the direction of address most of the needs of home cable operators. Thus, if priorities based QoS networking applications, especially when it is functionality is added to a residential for each point-to-point link. Hence QoS gateway and to new compliant home LAN functionality adds very little value in the case devices, then traffic originating from the of CableHome residential gateways that have these devices can utilize priorities to take Switched Ethernet interfaces. Thus while advantage of the prioritized media access designing CableHome 1.1 QoS solution capabilities of underlying OSI layer-2 home- Ethernet was considered as an outlier among networking technology. Traffic from legacy other available shared media home- non-compliant home LAN devices will networking technologies. continue to use best effort priority and therefore typically will not interfere with the
media access opportunities for prioritized Supporting Legacy Home LAN Devices and traffic from compliant devices. Thus with a Best Effort Traffic prioritized QoS system compliant A key cable operator concern was that CableHome devices as well as legacy non- newly designed CableHome 1.1QoS system compliant devices can co-exist in the home network without compromising the integrity controller (implemented in the residential of QoS for the applications that are taking gateway) and to follow its decisions before advantage of the QoS system. sending traffic over the network. However, Legacy (non-compliant) devices do not using available standards based OSI layer-2 home-networking technologies residential have a means of requesting and using media gateway cannot have any control on legacy access priorities for the packets. Thus these home LAN devices as to when they should devices cannot perform prioritized media send the traffic and how much; unless the access while transmitting their data. legacy devices are upgraded with hardware However, with manual (operator or or software addition that instructs them to consumer) set-up of priorities for legacy obey the admission controller in the network. devices in a residential gateway, it can be Thus, without this hardware or software instructed to perform prioritized media adapter, legacy home LAN devices will access for traffic that is destined to legacy interfere with the planned transmitting device. Thus prioritized QoS can be provided intervals of complaint devices and as a result for a sink-only legacy home LAN device, will compromise parameterized QoS system sinking traffic from a residential gateway. over the home network. Without appropriate Also with such manual settings a residential support for legacy devices from underlying gateway can perform prioritized queuing for OSI layer-2 home-networking technologies traffic to and from a legacy device going through the residential gateway. this limitation could not be overcome while implementing parameterized scheme. This particular limitation of parameterized QoS Parameterized Approach paradigm is very undesirable from cable operator and consumer convenience point of The parameterized QoS paradigm entails view. planned opportunities for media access and queuing. This requires that all of the devices and applications on the home network Software Upgradeability of Existing convey their requirements for various Cablehome 1.0 Devices parameters such as bandwidth, jitter, and One of the key overriding requirements delay to a centralized network controller such for CableHome 1.1 was that it be software as residential gateway. When a device or an upgradeable from CableHome 1.0. This application needs to transmit data over the would enable cable operators to upgrade network it sends a request to this centralized CableHome 1.0 residential gateway devices controller (termed as Admission Controller) in the field to CableHome 1.1 via remote [3]. Based upon the set policies and available download of a new software image, thus network resources, an admission controller avoiding the need for a truck roll. This gives either accepts or rejects the request in such a substantial cost advantages to cable manner that guaranteed QoS could be operators, allows new complaint maintained over the network. However, such functionality to evelove, and enables them to QoS guarantees can be provided only if all offer new products and services with the devices in the network obey decisions of the admission controller. CableHome 1.1. Thus a CableHome 1.1 QoS system should be such that any newly Through specifications, new complaint specified residential gateway features could home LAN devices can be instructed follow be added to the CableHome 1.0 device using the process of sending their QoS parameter just software implementation. requirements to the centralized admission
Upgradeability of Prioritized Approach admission decisions. Additional MIBs would If a prioritized QoS paradigm were to be also need to be implemented to store and manage the QoS parameters required for employed for CableHome 1.1 additional applications on the home network. The features that need to be added in the admission controller feature could be CableHome 1.0 residential gateway would implemented using Subnet Bandwidth be: prioritized media access, prioritized Manager (SBM) [3] and QoS parameter queuing and management of QoS priorities communication and reservations could be over the home LAN. To perform prioritized performed using the RSVP [4] network media access the residential gateway could protocol stack. Accurate estimates of the be software upgraded to set priorities for software footprint of these additional packets transmitted on the shared interfaces. protocol stacks for residential gateway Similarly, prioritized packet queuing and weren’t available. However, it was clear that forwarding could be accomplished through a comparatively it is heavier than the software upgrade. For queuing and functionality required for prioritized QoS forwarding functionality, additional network system. Thus it was uncertain if protocol stacks are not required but parameterized QoS system could be additional processing steps would be implemented on existing CableHome 1.0 required for existing CableHome 1.0 packet residential gateways by just upgrading forwarding process. Management of QoS software. priorities would require additional MIBs in the residential gateway to store priorities and additional software to manage and THE CABLEHOME 1.1 QoS SOLUTION communicate those priorities to various complaint home LAN devices connected to After analyzing different challenges in the residential gateway. The software providing in-home QoS as well as key footprint of this functionality can be requirements for CableHome 1.1 QoS minimized by using the same communication system, prioritized QoS paradigm was protocols as those used for the in-home chosen as the most appropriate solution. Due device management and discovery features of to the lack of adequate support for CableHome 1.1.Thus, the various required parameterized QoS from underlying OSI features of a prioritized QoS system can be layer-2 home-networking technologies implemented applying software upgrade to a parameterized QoS based solution for CableHome 1.0 residential gateway. CableHome 1.1 seemed unrealistic. In Preliminary estimates indicate that the addition, a QoS system based on a footprint of such implantation doesn’t seem parameterized scheme would potentially to be substantial enough to warrant a require additional hardware or software hardware upgrade. upgrade to legacy best effort devices to maintain the integrity of in-home QoS.
Taking into consideration these facts a Upgradeability of Parameterized Approach priorities-based QoS solution was specified If a parameterized QoS system were to be in the CableHome 1.1 specification. The later specified for a CableHome 1.1 residential part of this section describes in detail various gateway, implementation of an admission elements of the CableHome 1.1 QoS controller feature would be required as well solution. as a communication protocol to communicate parameter requirements and network
CableHome 1.1 QoS Architecture 1. Portal Services Element (PS): This is a logical element in the CableHome The CableHome 1.1 QoS architecture architecture [1] that represents consists of various logical elements and sub- CableHome specified functionality elements as shown in Figure 1: within a residential gateway device.
Apps Apps P PQCC BQCC BP B BP Q CH Q CH
Apps
Network ISCRG S Segment 2 Cable C O Head End D PS Q-Domain P Q QCS C QFM Network Segment 1 Apps PQCC BP B Q CH
Cable Home Network Network
FIGURE 1: CableHome 1.1 QoS Architecture
2. Boundary Point Element (BP): This is a the traffic originating from the PS as well logical element in the CableHome as for the traffic transiting through the PS. architecture that represents CableHome It is also responsible for communication specified functionality within a Home of QoS characteristics to various devices LAN device. within the home (described later in this section). 3. CableHome QoS Portal Sub Element (CQP): CQP is a sub element of the PS 4. QoS Boundary Point Sub Element (QBP): logical element. The CQP acts as a QBP is a sub element of the BP logical CableHome QoS portal for CableHome element. It performs priorities based compliant applications. Its primary media access for the traffic originating function is to enable priorities based QoS from the BP. It is also responsible for the for the devices within the home network. reception of QoS characteristics from the It performs priorities based PS. queuing/forwarding and media access for
In addition, these logical elements CableHome Queuing Priorities: described above contain QoS related functionalities (QFM, QCS, and QCS) that are Packets can be transmitted from multiple described later in this section. incoming interfaces to single outgoing interface in the residential gateway. Hence
CableHome Priorities each interface implements a queuing function. CableHome 1.1 defines the following In order to provide prioritized QoS for in- three different types of CableHome QoS home traffic passing through the PS, priorities: CableHome specifies prioritized queuing functionality per physical interface in the PS. 1. CableHome Generic Priorities A physical interface will have one or more 2. CableHome Queuing Priorities queues associated with it and each individual 3. CableHome Media Access Priorities. queue is designated with a certain queuing priority. This is defined as the CableHome CableHome Generic Priorities: Queuing Priority. The CableHome Queuing Priority needs to be identified for each packet CableHome 1.1 introduces the concept of to be transmitted on each PS interface so that Generic Priorities. This is primarily due to the the packet can be placed in an appropriate fact that OSI layer 2 priority approaches are queue. This CableHome Queuing priority is not consistent as the number of priority levels derived from the CableHome Generic Priority supported varies from technology to using the number of queues supported per technology. A generic priorities scheme gives interface on the PS. Implementation of cable operators a consistent approach, which number of queues per interface is vendor is abstracted from the particular OSI layer 2 specific. home-networking technology. In addition, this single generic priority can serve to indicate CableHome Media Access Priorities: both media access priorities, as well as queuing priorities (described below). This is the media access priority of a packet and is derived from its CableHome CableHome 1.1 defines eight CableHome Generic Priority based on the number of Generic Priority levels, 0 through 7, 7 being media access priorities supported by the highest and 0 being the lowest. Cable interface’s layer-2 shared media technology. operators assign one of these eight priorities Since the number of priorities supported by to an application. Application is identified different OSI layer-2 home-networking using an application ID, which could be an technologies varies, such mapping is IANA assigned port number for the necessary. CableHome Media Access Priority application [5]. Of the three types of values are logical levels that represent a level priorities defined by CableHome, a cable of preference that a packet should receive for operator sets only the CableHome Generic media access. Priority value for an application based on its CQP QoS Functionality: ID. The other two priorities - CableHome Queuing Priorities and CableHome Media The CableHome QoS Portal (CQP), which Access Priorities - are derived from this resides in the PS element, consists of two CableHome Generic Priority depending on the main functionalities as shown in figure 1: QoS capabilities of the hardware and software in Forwarding and Media Access (QFM) and the device. QoS Characteristics Server (QCS).
QoS Forwarding and Media Access (QFM): Access Priority. The packet is then transmitted on the shared media with the The QFM element provides the PS with a appropriate level of preference as mechanism to order and transmit packets out indicated by the CableHome Media of a PS interface to a LAN host according to Access priority. assigned priorities. The PS exercises QFM functionality on any packet that is transmitted QoS Characteristics Server (QCS): out of the PS on any LAN interface. The QFM performs following three actions on the packet once it is received in the PS: The QCS element provides a mechanism for the cable head-end to communicate 1. Packet Classification: The PS examines desired QoS Characteristics (for particular the destination IP address and destination applications) to the PS and then further to BPs port number of the packet. Using these in the home. In CableHome 1.1 QoS values the PS looks up a corresponding characteristics refer to priority information for CableHome Generic Priority for the different applications over the home network. packet from the classifier table stored in The overall functioning of the QCS is the PS database. If no matches are found explained below. for that destination IP and port, then the PS assigns priority 0 to the packet. 1. Application Priority information to the PS: The cable head-end provides mapping of application IDs to CableHome Generic 2. Prioritized Queuing: The PS then maps Priorities to the PS either using a PS CableHome Generic Priority of the packet configuration file or via SNMP MIB to CableHome Queuing Priority based on interface. This mapping in the PS serves the number of queues implemented for the as a master table in determining priorities interface on which the packet is to be for various applications or services on the transmitted. Multiple queues implemented home LAN. for the interface are designated with
different CableHome Queuing Priorities. 2. BP Application Information to the PS: The PS puts the packet in an appropriate The QCS receives information about the queue based on its queuing priority. The applications associated with a BP in the QFM polls all of the queues on each form of an XML message, called the interface according to their priorities to BP_Init message, which is sent using extract packets. The packets are extracted SOAP over HTTP [6]. This message from the queuing system by employing a contains the list of application IDs that a methodology of First in First Out with BP supports. It may also contain a list of Priorities, Highest Priority Queue First. destination IP address and port number pairs for which a particular application on 3. Prioritized Media Access: After the the BP likes to request destination specific packet is extracted from the set of queues priority. Such a request for destination IP associated with an interface, the packet and port specific priority is sent by a BP to needs to be transmitted on the shared LAN the PS after an application session has media with the appropriate media access been established. priority. The QFM performs the mapping of the CableHome Generic Priority value 3. Application Priority Information to the of the packet to the CableHome Media BP: Upon receipt of the application
information from the BP, the QCS Also if an application needs a specific consults the priority master table provided destination IP address and/or port specific by the cable operators and determines priority, then the QCC sends a request for appropriate priorities for different such destination IP and port priority in the applications on the BP. If there is no entry BP_Init Message, after the application on for a particular application in the priority a BP establishes a connection with another master table, then the QCS assigns a application. In addition, the QCC is default priority of 0 (best effort) for that responsible for communicating to the PS application. QCS also determines the any updates (addition/deletion) to the destination specific priority information as application information in the BP. After requested. Both these priorities are the PS sends updated application priority determined using applications IDs. The information to the BP, the QCC makes QCS sends this updated priority sure that priority information for information to the BP in the XML format applications on the BP gets updated using BP_Init_Response message (SOAP appropriately. over HTTP). The QCS also stores all of this updated priority information in the PS 2. Prioritized Media Access: Once the database (which is accessible to cable application on the BP starts operators via the MIB interface). communicating, the QCC uses the priority assigned to it for prioritized media access. Thus through these three main processes If a destination IP and port specific the QCS manages and communicates priority priority is requested then QCC uses information to various applications on the destination specific priority otherwise it home network uses the default priority assigned to the application. The QCC maps the QBP QoS Functionality: CableHome Generic Priority to CableHome Media Access priority based The QBP is a logical sub element of a BP on the number of media access priorities that resides in a CableHome compliant home supported by underlying layer-2 home– LAN device, termed as CableHome Host. The networking technology and then delivers QBP consists of only one QoS functionality: the packet on the shared media. QoS Characteristics Client (QCC). IMPLICATIONS OF THE CABLEHOME QoS Characteristics Client (QCC): 1.1 QoS SOLUTION The QCC has two main responsibilities: obtaining application priority information As explained in the earlier sections, the from the PS and using this priority CableHome 1.1 QoS system is a simple and information for prioritized media access. elegant solution that cable operators can These two functions of the QCC are explained provide on CableHome devices easily. The in detail in the subsequent paragraphs. only additional provisioning step that cable operators need to perform is provisioning of 1. Requesting priority information to the the application priorities master table in the PS: As explained earlier in the QCS residential gateway. Applications on section, the BP sends its application compliant CableHome devices will receive information to the PS in the BP_Init appropriate priority information that they can Message. The QCC entity in the BP is use for subsequent management of prioritized responsible for that message exchange. traffic flow. With CableHome 1.1 QoS, legacy home LAN devices can co-exist with take advantage of this solution in a convenient complaint QoS-enabled devices without and seamless manner, as it does not require collapsing QoS of the entire network. Thus it any additional hardware or software upgrade is a convenient solution for both consumers for legacy devices in order to maintain QoS and cable operators. Minimal additional over the home network. Additional functionality is required to implement functionality required to implement this CableHome 1.1 QoS on residential gateway solution is minimal. Hence it is attractive and and home LAN devices. cost effective for vendors to build products for The benefits of the CableHome 1.1 QoS this specification. Cable operators can solution are significant and compelling, provision QoS for their applications in the however there are a number of implications home network by a simple configuration of associated with the chosen approach. Since application specific priorities in the this QoS solution is based on a prioritized CableHome 1.1 based residential gateway. paradigm it does not provide absolute Thus, CableHome 1.1 QoS is a simple, cost- guarantees for QoS parameters such as effective and easy-to-use solution that enables bandwidth, jitter, and delay. It provides cable operators and consumers to take preferential media access to certain traffic, advantage of QoS over the home networks. classified as higher priority. Thus two traffic streams that have the same priority would AUTHOR CONTACT INFORMATION contend with each other for media access and thus they might get best effort treatment Amol Bhagwat between themselves. Also, if a top priority OSS Engineer, Broadband Access application consumes the entire bandwidth of Cable Television Laboratories, Inc the home network then it is possible that 400 Centennial Parkway access to the shared media could be denied for Louisville, CO 80027 all applications. Considering the bandwidth P: 303-661-3333 provided by typical home-networking F: 303-661-9199 technologies today and services that E: [email protected] CableHome 1.1 plans on enabling it seems that this scenario is unlikely. However, in ACKNOWLEDGEMENTS future if cable operators decide to offer bandwidth intensive services such as video The author wishes to acknowledge the distribution over the home network, this following individuals for their valuable scenario may occur and in that case the contribution, inputs and assistance. applicability of a prioritized QoS scheme may Ralph Brown, Steve Saunders, Kevin Luehrs, need to be revisited. [See Appendix 2 for Stuart Hoggan, Jennifer Doran, Liz Weeks typically expected QoS requirements for various applications and services]
CONCLUSIONS
CableLabs has defined priorities based QoS solution for its home networking project- CableHome 1.1. The CableHome 1.1 QoS solution can be deployed on any layer-2 home-networking technology that supports priorities. Cable operators and consumers can
APPENDICES
Appendix 1: Features summary of various leading standards based OSI layer-2 home-networking technologies
Home Specifications PHY Layer Data Rates QoS Networking and Standards Modulation Capabilities Technology Group Ethernet IEEE 802.3 Baseband 10Mbps,100Mbps, None & 1Gbps Wireless LANs IEEE 802.11a OFDM 54Mbps 802.11 e IEEE 802.11b DSSS 11Mbps Working group IEEE 802.11g OFDM 54Mbps (Prioritized & Parameterized Proposals) Powerline HomePlug 1.0, OFDM 10Mbps Prioritized Home Plug Powerline Alliance Phoneline HomePNA 1.0 & FDQMA 1Mbps & 20Mbps Prioritized HomePNA 2.0, respectively Home PhoneLine Networking Alliance
Appendix 2: Expected QoS requirements for various applications and services
Output Parameters of Input Parameters of Performance Testing Performance Testing Service Number Payload Packet Max Max Header of Rate (per Size Max PER Latency Jitter Type Streams stream) (bytes) (ms) (ms) HQ Voice 2 per call 64 kb/s IP/UDP/RTP 120 1.5*10 -3 10 +/-5 Calls MQ Voice 2 per call 8 kb/s IP/UDP/RTP 80 1.5*10 -3 30 +/-20 Calls HQ Video Conference 2 per call 1.5Mb/s IP/UDP/RTP 228 3.6*10-5 10 +/-5 Call 19.68 HDTV 1 IP/UDP/RTP 228 3.6*10-5 90 +/-10 Mb/s SDTV 1 3 Mb/s IP/UDP/RTP 228 3.6*10-5 90 +/-10 CD Quality 1 256 kb/s IP/UDP/RTP 360 5.8*10 -5 100 +/-10 Audio High Speed 1 10 Mb/s TCP/IP 1540 0 >100 >100 Data Med. 1 2 Mb/s TCP/IP 1540 0 >100 >100 Speed Data Low Speed 1 500 kb/s TCP/IP 1540 0 >100 >100 Data NOTES: 1. Voice Packet = (IP/UDP/RTP Header) + (voice payload) IP/UDP/RTP Header: 40 bytes (20 bytes IP Header + 8 bytes UDP Header + 12 bytes RTP Header) without RTP header compression. If RTP header compression is applied header reduces to 2-4 bytes. In this table we assumed no RTP header compression. Voice payload: variable size depending on codec, considering the end-to-end latency budget, typically 10-40 ms voice samples can be used. Given the Max Latency/Max Jitter in HN portion and to keep packet overhead to its minimum, we assume 10 ms voice samples for HQ voice and 40 ms voice samples for MQ voice. Video Packet = (IP/UDP/RTP Header) + (video payload) IP/UDP/RTP Header = 40 bytes (20 bytes IP Header + 8 bytes UDP Header + 12 bytes RTP Header). Video Payload size = MPEG packet size which is 188 bytes. Data Packet = (IP/TCP Header) + (Ethernet payload) Packet Error Rate (PER): is measured at MAC-SAP for packets delivered from MAC layer to higher layer. For data, since packets in error should be discarded and only error free packets are passed to the MAC-SAP, then for data PER=0.
REFERENCES [3] SBM, A protocol for RSVP-based Admission control over IEEE 802-style networks, IETF RFC [1] CableHome 1.0 Specifications, Cable Television 2814, May 2000, http://www.ietf.org/rfc/rfc2814.txt Laboratories, April 2003, [4] Resource reSerVation Protocol, IETF RFC 2205, http://www.cablelabs.com/projects/cablehome/specific September 1997, http://www.ietf.org/rfc/rfc2205.txt ations/ [5] IANA Port Numbers, [2] IEEE Std-802.1Q, IEEE LAN MAN Standards http://www.iana.org/assignments/port-numbers Committee, December 1998 [6] SOAP Version 1.2, W3C Working Draft, World Wide Web Consortium (W3C), June 2002, http://www.w3.org/2000/xp/Group/#drafts.
ROUTERLESS AGGREGATION: CONVERGING DATA AND TDM NETWORKS Benoit Legault Dir. Technology Advisory Board ADC
Abstract products and services (i.e. video), and expanding the product portfolio. MSOs interested in remaining Business services, historically an ILEC competitive and deploying new services monopoly, are one possible area of are faced with two network architecture growth or portfolio expansion, which options. Traditional routing-intensive have not been aggressively exploited by metro networks offer clear benefits for MSOs. data services, but fall short on the ability to converge data and TDM networks. Reducing CapEx and Opex Innovative MSOs interested in offering voice and other delay-sensitive services An MSO’s network represents the should deploy a network architecture largest portion of its investment. It is with fewer router: “routerless composed of three different segments: aggregation”. the access network, which starts at the hub or headend and terminates in the INTRODUCTION subscriber’s home, the metro network, which interconnects the hubs of a The current economy dictates that metropolitan or regional area, and the MSOs increase earnings and free cash backbone network which interconnects flow. This can be done by increasing the metro networks. revenue through the introduction of new services and increasing the level of Today, MSOs own and operate three profitability of current services through different metro networks: video, data reducing CapEx and OpEx. and TDM. Therefore, the opportunity exists for operators to significantly reduce their network CapEx and OpEx through converging the data and TDM networks.
The Requirements
Fig 1 - MSO Opportunity Current metro data network architectures can handle the Increasing revenues requirements of Internet Access, both residential and commercial. To allow MSOs can improve revenues through convergence in the future, metro increasing the market share of products networks must be able to support the and services with a potential for growth requirements necessary to deliver the (i.e. high-speed Internet access), following services: protecting market share of mature • Residential voice (individual POTS lines) • Business private networking • Business voice (T1s and T3s for PBX applications)
Voice Services
Voice services, both for residential and commercial customers, require high- availability networks with low latency, delay and jitter. The main attributes of high-availability networks are: Fig 2 - Current Metro Network Architecture • No single point of failure, both in the signaling path and in the call Voice services, whether individual path POTS lines or T1s and T3s for PBX • Fast network convergence to applications, require high-availability avoid dropping calls upon networks that guarantee service failures availability and sub 50ms automatic • Transparent software upgrades to switchover. Since the enterprise-class avoid downtime associated with routers currently used in traditional network maintenance metro network architectures do not offer carrier class availability, the classical Commercial private networking method of increasing the availability of services require that the MSO be capable these networks consists of installing of delivering Layer 2 “pipes” across the redundant edge routers at every hop. metro network. This requirement is This architecture has two major driven by two factors. First, many drawbacks when offering telephony business customers still carry legacy services: protocols, such as IPX, SNA, LAT, DECnet, Appletalk and others, on their • Since high-availability is networks. In addition, businesses do not provided at the IP layer expect or desire that their address plan through routing protocols, be impacted by the carrier’s network. TDM services can only be offered through circuit Current Metro Network Architectures emulation over IP. • Since OSPF’s convergence There are two main problems with time is far above the the traditional routed metro network traditional 50ms recovery architecture. First, it is composed of time of voice networks, enterprise-class elements, which do not circuit emulation requires support high-availability. Second, it is that operators implement exclusively composed of routers at every MPLS-TE for its fast hop, which complicates the offering of recovery features. private networking services. In addition, the traditional metro to the more complex MPLS switching. network routed architecture lacks native In a typical regional network, this support for business private networking architecture is composed of Layer 3 services, which can only be supported CMTSs at the subscriber edge (located through the introduction of new in the hub), and edge routers at the networking protocols. The only solution provider or MSO backbone edge is to emulate Layer 2 over Layer 3, (located in the regional head-end or which requires the configuration and regional data center). All other elements management of a number of protocols in between are Layer 2 switches. (L2-MPLS/VPNs (Martini), MPLS-TE, OSPF-TE extensions, CR-LDP, etc.) This architecture provides a number across the metro. The result is an of benefits and, as we will discuss, a extremely complex and costly network number of shortcomings that must be to own and operate. addressed. Lets start with the benefits.
Routerless Aggregation The Benefits
As stated previously, routed metro Faster – Cheaper networks pose two main problems: the lack of native support for TDM services Routerless aggregation is far more because of the absence of sub-IP layer cost effective than enterprise-class Layer path protection and the complexities of 3 aggregation for mainly two reasons. offering private networking services. First, carrier-class switches only require half the number of elements and interfaces, compared to the traditional network architecture, by eliminating the need to duplicate the elements. A number of manufacturers now offer Layer 2 switches with built-in redundancy at all levels, and support for transparent software upgrades. Note that, when combined, these features allow a single Layer 2 switch to provide equal or better overall system availability than a pair of enterprise-class routers. Second, Fig 3 - Routerless Aggregation Architecture on a side-by-side comparison, any particular router port is generally more The routerless aggregation metro expensive than its equivalent on a Layer network architecture solves both these 2 device. problems. This architecture uses carrier- class Layer 2 switches as the aggregation Simple support for private networks element in each hub and pushes the routing function to the edges of the VLANs have been used for years to metro network. It essentially applies the support private networking services. networking principle of routing at the Most PTTs and ILECs around the world edge and switching at the core through have been using this technology for well the use of Layer 2 switching as opposed over a decade, and continue to do so networks as viable network with much success. Through routerless architectures. Most of the reasons why aggregation, MSOs operate metro spanning tree was considered inadequate networks that can support Layer 2 point- still exist, and therefore, the author to-point or multipoint-to-multipoint shares the view that if spanning-tree “pipes” from anywhere to anywhere cannot be avoided in a routerless within the region. The private networks aggregation architecture, the architecture are simple to configure and manage, and should be considered incomplete and allow the MSO to support any Layer 3 problematic. protocol without getting involved with the subscribers’ Layer 3 address plan or On the other hand RPR, which is a even being aware of the transported new Layer 2 protocol that creates fault Layer 3 protocols. tolerant rings as an overlay of point-to- point GigE or Sonet links, allows the use PHY-layer protection of Layer 2 devices without resorting to spanning tree to manage redundant The carrier-class nature of these paths. Elements on an RPR ring are switches, when combined with interfaces provided with a single Layer 2 path to all that provide native support for TDM and other elements on the ring, such that packet-based services, allow the spanning-tree is never required to routerless aggregation architecture to manage the ring’s redundant paths. The provide native TDM services from RPR MAC layer handles interface and anywhere to anywhere in the metro. link failures transparently, such that Sonet and ATM are good examples of changes to the network’s links’ status are such interfaces. This architecture truly never apparent to any element’s Layer 2 allows the convergence of the data and (or Layer 3, for that matter) forwarding TDM networks in the metro, further table. reducing CapEx and OpEx. VLAN scalability limits The “Gotchas” And The “Fix-Its” The maximum number of supported Avoiding spanning tree VLANs on any given interface, per the standard Layer 2 header, is 4096. In Layer 2 networks usually rely on some cases, this limit poses a scalability spanning tree to manage redundant paths problem for MSOs, especially in in the network. Spanning tree provides medium to large size regions. slower convergence and is far less intelligent than routing to control and The routerless aggregation network manage redundant paths in a network. architecture proposes to solve this Spanning tree is known to cause outages problem by creating multiple Layer 2 through broadcast storms, constant RPR aggregation rings in the metro and flapping to administrative mode, and to joint these rings through the use of other problems that derive from its basic edge routers implementing Layer 2 operation. It is, in most cases, the main MPLS VPNs (Martini). This approach reason why many network architects addresses the scalability issues of have previously dismissed Layer 2 VLANs without introducing the complexities associated with logical overlay uses VLANs to create implementing MPLS throughout the two separate MAC domains over a metro network. The result is a very single physical network through the use scalable Layer 2 VPN solution that is of an edge router that can be attached manageable and has a level of anywhere in the ring. complexity that grows with the services’ level of success. RPR-Enabled Sonet
Impact of Layer 2 aggregation on OSPF This final section describes how routerless aggregation is a network The routerless aggregation network architecture that can truly enable the architecture essentially flattens the metro convergence of TDM and data networks network from a routing perspective. in the metro. As described earlier, RPR Flattening the metro has impacts on plays an important role in routerless OSPF, or any other routing protocol. The aggregation metro network architectures. most significant impact is that it Also, Sonet and ATM are essential to the increases the number of OSPF native coexistence of TDM and packet- adjacencies maintained by each router in based services over a single the network. If not factored into the infrastructure. Given the fact that RPR design, an oversized growth in OSPF can either use GigE or Sonet as its adjacencies will cause problems in the underlying physical network layer, RPR- operation of the network. Routers will enabled Sonet networks are, if suffer from performance problems, implemented properly, the true enabler convergence will be slow, and network of converged networks. stability will be negatively affected. Note that the maximum number of A proper implementation of RPR- adjacencies supported by any given enabled Sonet allows the MSO to carve router is vendor-specific. STS-1s out of the Sonet bandwidth allocated to the RPR interface to natively The scalability solution for VLANs support T1 and T3 services across the also solves the OSPF scalability issues metro. Products that support this feature associated with routerless aggregation. can provide pure circuit switched T1 and In large metro networks, where T3 interfaces for telephony applications, routerless aggregation would cause too along with GigE and 10/100BT many OSPF adjacencies if a single Layer interfaces for data applications on the 2 network was created, the network subscriber side, and use a single uplink should be partitioned into multiple sub- on the network side to carry all services networks through the use of edge across the metro. routers. This goal can be reached either by a physical implementation, or by CONCLUSION creating the partitions through a logical overlay. An example of a physical The routerless aggregation metro implementation is to create two rings in network architecture provides a number the metro and place an edge router at of key benefits over current metro their intersection point, physically networks architectures, which allow separating the two MAC domains. A MSOs to a la fois increase revenues and reduce CapEx and OpEx.
The architecture benefits MSOs by enabling:
• Data and TDM network convergence, with native support for TDM services (as opposed to circuit emulation) • Private networking services in a simpler and just as scalable manner as MPLS
The architecture avoids the age-old issues associated with link redundancy in Layer 2 networks through a new protocol: RPR. SEAMLESS, SCALABLE HDTV ROLL-OUTS OVER TODAY’S HEADENDS
Sylvain Riviere BigBand Networks, Inc.
Abstract MAINTAINING CABLE PLANT EFFICIENCY Offering HDTV programming is a WHILE ROLLING OUT HDTV necessity for cable operators competing with satellite and terrestrial broadcast This section justifies bit rate television alternatives. However, digital adaptation techniques as a solution to cable faces several challenges in overcome bandwidth challenge while broadly replicating the HDTV services scaling up HDTV roll-outs. It suggests currently available via alternatives: channel line-up scenarios and implementations that maintain video • Maintenance of equal video quality over cable. quality, satisfactory to broad- casters, content creators and Existing bit rate constraints prevent subscribers, while maintaining scaling up of HDTV service cable plant bandwidth efficiency The Advanced Television Systems • Compliance with carriage of data Committee (ATSC) gave birth to HDTV associated with programming and 8-VSB (8-level vestigial sideband) according to the PSIP standard modulations standards. At that time and openly accessible by consumer ATSC adopted the VSB transmission electronics devices because of its “large” bandwidth, which is needed to transmit HDTV • Responsiveness to programming programming off air. In December 1996, changes by on-air broadcasters the FCC approved those standards to whose content is being mapped replace the analog standards of the onto cable plant, such as when NTSC. The 8-VSB mode supports up to sudden shifts are made in use of 19.4 Mbps of content and drives the bandwidth from an HDTV feed to maximum bit rate allowed for HDTV several SDTV feeds streams.
This paper describes a MPEG-2 technologies have brought comprehensive way to roll out full encoders a long way during the last HDTV experiences in existing headends. decade to offer similar video quality at It suggests channel line-up scenarios lower bit rates benefiting from statistical that achieve bandwidth efficiency with multiplexing techniques. Such high quality content through advanced techniques take advantage of the video bit rate adaptation techniques, inherently variable bit rate of video effective multiplexing of PSIP data, and feeds, such that when multiple feeds are real-time intelligent responsiveness to combined, it is highly unlikely that most broadcaster changes. or all will experience intensive action simultaneously, and in fact bandwidth peaks for some will usually correspond The following table compares the with troughs for others. Unfortunately, maximum number of HDTV versus such techniques are predominantly SDTV feeds carried today using various employed for SDTV and little is modulation approaches. It also shows expected from statistical multiplexing of how broadcasters efficiently carry one HDTV feeds because only one or two HDTV feed versus other multiplexing feeds are carried per multiplex. alternatives.
HDTV SDTV Carried Carried with Theoretical at Near Statistical Modulation Rate Constant Multiplexing
Broadcaster 8-VSB 19.4 1
Cable 64 QAM 27.0 1 7-8
256 QAM 38.8 2 10-12
Satellite QPSK 27 1 8-12
8-QPSK 40 2 12-20
Cable operators face a severe HDTV rate shaping optimizes statistical challenge in scaling up HDTV while multiplexing techniques – it’s not all accommodating for over 19Mbps per about crushing bits! HDTV stream, as defined per ATSC standards. This is on top of the Rate shaping describes bit rate challenges of bringing together content adaptation techniques applied to MPEG- sourced from broadcast and satellite 2 encoded streams, to further enhance feeds, with their distinctive formats, onto bandwidth efficiency. This technique a single plant. can substitute for decoding-encoding operations that are expensive, space Unless massive efforts such as cable consuming and ultimately harmful to plant upgrade or aggressive analog content quality. reclaim are undertaken, the HDTV constant bit rate approach won’t scale up Accommodating various transport in a world of fast growing double-digit alternatives, rate shaping also adjusts the available HDTV programs. MPEG-2 bit necessary bit rate to “bridge” HDTV bit rate adaptation techniques, also called rate from satellite and off air delivery rate shaping, can address those onto cable plant. By doing so, HD rate problems. shaping removes fixed bandwidth allocation constraints imposed by ATSC standards. The technique considers and accommodates cable plant transmission capabilities for greater bandwidth efficiency, without the harmful effects of stay competitive with alternate sources. decoding and re-encoding. HD rate As mentioned earlier, content intensity shaping does not blindly steal bits to drives bit rates in statistical re- squeeze more into a channel, so video multiplexing operation. Because cable quality does not suffer for bandwidth operators have the freedom to choose efficiency. specific channel line-ups, they can optimize for certain scenarios that will In the case of three HDTV into one maintain video quality compared to 256QAM channel, the rate shaping alternatives. operation does not reduce all streams 33% to squeeze one more program. As an example, sports, movie and Instead bit rates are dynamically driven news content introduce different by incoming content complexity. complexities that drive the bit rate Depending on content, economic bit rate reduction allowances differently in reduction per program can be as low as HDTV. The higher the complexity is, the 10% and up to and beyond 50%, while less likely is bit rate reduction to occur, maintaining identical perceived video and vice versa. As a general rule, quality to original sources. avoiding excessive content within a re- multiplexing pool will allow optimum By taking multiple outputs of this video quality and greater bandwidth process from multiple sources and efficiency. packing them together, varying bit rate results in cumulative bit rate efficiencies Certain statistical re-multiplexers at desired video quality. This process, have the ability to combine both HDTV also called statistical re-multiplexing, and SDTV feeds. Benefits include outlines how cable operators can offering granularity at which the accommodate their own bandwidth bandwidth efficiency is reached efficiency by moving away from the independently from content complexity. near constant HDTV bit rate Combined with priority mechanisms, expectations traditionally imposed on there are cases where the HDTV bit rate them. can remain untouched while bandwidth efficiency is gained by removing stuffed Statistical re-multiplexing is utilized packet (nulls) for SDTV channels. This by cable operators for SDTV feeds technique is called pass through mode as already and can be applied to HDTV it exactly replicates the on screen HD feeds similarly. However because of the content and quality level as at the peak nature of available HDTV content and bit rate, while finding unused bandwidth its video quality emphasis, caution is in stuffed packets that can be allocated required when implementing HD rate to other traffic. shaping in the headend. Fig. 1 below outlines HD and/or SD Applying content intelligence in rate channel line-ups and their respective shaping. bandwidth efficiencies in the example of 256QAM cable plant. The efficiency When deploying rate shaping gain is the amount of content carried on technology in cable headends to gain a QAM channel versus a constant bit bandwidth efficiency, it is important to rate alternative of two 19.4 Mbps HDTV feeds. For purposes of this analysis, the The figure also suggests deployable general guideline applied is that six scenarios under the condition justified SDTV feeds conventionally consume the by identical subjective video quality same bandwidth as one HDTV feed. comparison to alternatives sources available in the field.
Channel line-up Subjective Video quality comparison 4HD Consistent Impact
3 HD+ 4SD Frequent Impact
No Impact 3HD (If line-up & content driven) 2HD + 4SD No Impact
1HD+ 8SD No Impact
20 % 30 % 50% 70 % 100 %
% of initial 256 QAM content carriage gained
Fig. 1: Bandwidth and quality impact of rate shaping and statistical multiplexing alternatives.
By appropriately implementing rate including compliance with industry shaping with proper channel line-up, standard agreements. cable operators can achieve up to 50 % bandwidth efficiency gains, while What is PSIP and why is it needed? preserving similar video quality to alternatives. Beyond 50%, video quality PSIP (Program and System can be impacted, and it is recommended Information Protocol) has been that operators make considered decisions standardized by the ATSC to allow about deployment. tables of information to be transmitted along with the associated video PSIP SUPPORT programming. PSIP data are originated ON THE CABLE PLANT by broadcasters and are required by some ATSC receivers to tune to the This section addresses cable-ready correct digital channel. television support through the implementation of the PSIP format, Set-top boxes used by cable operators analog or digital. In this way the analog do not require PSIP to tune to a channel, branding is preserved in the digital but the growing availability of retail world. market cable-ready digital televisions requires PSIP presence on the cable The electronic program guide (Fig. 2) plant to guarantee proper tuning. Besides is the navigation interface provided to tuning, PSIP tables contain other the user to tune to a channel, specifying important information about the timing and event description. PSIP programming, such as branding and carries tables that can be used by cable- electronic program guide data. ready televisions to build the electronic program guide. This function is often Consider a local station that offers seen as a benefit of converting to digital three different shows during the day. by the user, and will be provided Digital televisions allow viewing the automatically by cable-ready televisions analog service as channel 65, for as long as PSIP tables are available. The example, and the three others as ATSC PSIP standard requires that a channels 65-1, 65-2 and 65-3, even minimum of the next 12 hours of tough the digital broadcast is technically program information be available in on channel 66. The main number advance, although PSIP can offer up to indicates that those channels belong to 16 days of programming. the same broadcaster whether they are
Fig. 2: Electronic program guide representation.
Complying with NCTA-CEA agreement. electronic products with cable-ready televisions (like the one shown in Fig. The emergence of cable-ready 3), but the cable and consumer televisions, with their PSIP support, call electronics industries have taken to question digital cable interoperability important steps towards achieving with ATSC PSIP standards. There is a interoperability by establishing the large number of standards, agreements, NCTA/CEA technical and PSIP specifications and FCC rules that relate agreement. to the interoperability of consumer NCTA and CEA negotiations resulted from the content provider, and that it in an agreement that provides a would be prepared to support carriage of consistent set of standards that enable PSIP information when made available cable-ready televisions to be connected from the content provider in accordance directly to a cable plant without the need with the agreement. CEA agreed with for a set-top box. The agreement section the document. The agreement also relating to PSIP addresses television specified mandatory/optional PSIP receivers that do not have a security tables to carry on the cable plant while module (Type 1 Television). recommending standards to overcome implementation issues. For more During the negotiation, the cable information on NCTA\CEA agreement industry made clear that carriage specific to PSIP, refer to ATSC A/65 assumes the availability of PSIP data and SCTE DVS-097 standards.
Fig. 3: Mitsubishi cable ready TV – WS55909.
PSIP agreement implementation in the ACCOMMODATING headend. BROADCASTER MULTICASTING HABITS Cable operators will need to obtain necessary hardware and software to This section addresses how off-air implement the NCTA/CEA agreements. multicasting can impact cable service, Whether two broadcast signals or more and proposes an alternative approach to are combined in a single transport stream multicasting through automatic response. for delivery using 256QAM or 64QAM, the re-multiplexing operation (with or What is multicasting? without rate shaping) will require rebuilding the PSIP information. Multicasting occurs when broadcasters suddenly shift use of Sent in-band with the video, PSIP bandwidth between various table implementation requires a platform combinations of HDTV feeds to SDTV that can combine both video and data. In feeds. The broadcaster, in an effort to some cases, rate shaping can be required accommodate formats and leverage to make room for data when bandwidth content availability, often disrupts the of incoming video consumes what’s multiplex several time a day switching available. from one channel line-up to another. Stream characteristics also change on the fly depending on the operation done at broadcaster sites. While those sudden changes are transparent to an ATSC although it had content earlier, receiver, cable architecture does not increasing call center activity accommodate multicasting transparently, dramatically. Even worse can be the possibly impacting service. subscriber who becomes frustrated by the service to the point of cancellation, Possible impact on services with without ever placing a call. Generally multicasting. the cable operator is blamed for service loss although it is a consequence of Depending on broadcaster and site, broadcaster multicasting. impacts on cable services and their consequences varies, and can include the Channel line-up confusion. following: The switch between SDTV and Loss of service even though content is HDTV format streams brings confusion present. as far as channel line-up structure and how HDTV streams could be grouped Stream characteristic changes disrupt together in logical channel line-up overall equipment performance, starting numbering. with the decoder that may need to be retuned to the channel to update stream SDTV format and HDTV format with characteristics. This can be the case upconverted SD content (black bar). when SDTV feeds replace HDTV feeds. In some cases headend equipment may A particular manifestation of channel lose the incoming identification line-up confusion is when the user sees information (same program number but standard definition 4:3 ratio on a 16:9 different packet ID), preventing ratio television. In one case the user can automatic restoration even though the stretch the image of its television to content is restored. avoid the black bar caused by SDTV content ratio (see Fig. 4), in the other Customer satisfaction issues. case the television will not allow it because the feed is already in HDTV If a channel disappears, its session format although SDTV content was already mapped on the cable plant is still upconverted. Too much time with black maintained whether the program is there bars on screen also risks image burn-in or not. The user tunes to a black screen on the television. Fig. 4: Example of standard definition stream stretched by television to 16:9 screen.
Satellite and broadcaster alternatives channel number in a line-up whether the provide a clear indication of what type format is HD or SD, indicating whether of feed is broadcasted to avoid this or not television stretching function is confusion. An alternative for cable is to enabled. organize channel line-up per stream format to avoid the confusion, SUMMARY independently from the content when multicasting involves both types of Because of the realization of FCC streams on same channels. timetables for terrestrial digital broadcasting launch and satellite Solution to accommodate multicasting in progress and aggressive marketing, cable existing cable architecture. is particularly challenged to compete effectively in providing HDTV services. An intelligent re-multiplexing Elements such as constant bit rate platform capable of real time encoding, PSIP data tables and real time responsiveness to broadcasters’ sudden broadcaster switching between formats changes is needed prior to the usual are inherently unfriendly to the cable cable session management systems. plant. Strategic positioning of this re- multiplexing can preserve static channel This paper has shown that an line-ups independently from intelligent multiplexing platform can multicasting events and act as a shield overcome obstacles through HDTV rate during eventual stream characteristic shaping; capabilities to combine HDTV change. and SDTV within the same channels; regenerating PSIP tables for cable in In response to program loss, a graphic compliance with the NCTA/CEA message can be substituted for the agreements; and recognizing and program disappearing to offer an accommodating broadcaster format indication to the user that a channel will shifts. These techniques enable the cable return or is not available, and suggest industry to provide subscribers with a alternative programming locations. Also, complete and fulfilling HDTV smart session management can redirect experience. The cable industry can then streams dynamically to the proper leverage its own inherent advantages, such as its balanced access to both local in HDTV and industry competitiveness and national feeds. Cable operators can can be enhanced. be liberated from defensive positioning
Sylvain Riviere Director of Product Marketing BigBand Networks, Inc. 475 Broadway Redwood City, CA 94063 650 995 5024 [email protected] http://www.bigbandnet.com SECOND GENERATION POINT OF DEPLOYMENT (POD) INTERFACE FOR MULTI-TUNER CABLE RECEIVING DEVICES
Joseph W. Weber David Broberg CableLabs
Abstract The operation of the POD module is shown in Figure 1. Digital content is Under the current agreement between CE received via a QAM tuner and sent to the companies and cable operators, uni- POD module. Premium content that the directional “digital cable ready” televisions customer is entitled to view is decrypted using may soon be offered to consumers. Cable the network’s conditional access system. A customers could be able to receive premium dedicated out-of-band communication channel digital content without the need for a set-top (either one-way from the cable network to the box through the use of a conditional access device, or two-way) is required in order for point-of-deployment (POD) device. the POD to connect with the conditional access system. As these single-program POD devices begin to be deployed, CableLabs along with Premium content is then re-encrypted its members and the vendor community is within the POD module with the open developing a next-generation POD capable of standard POD copy protection method defined providing multiple streams of premium digital is SCTE 41 [2]. Authenticated devices are content. This will enable new devices and able to decrypt the POD copy protected expanded services such as picture-in-picture, content for display or recording via a 1394 watch-and-record PVR, and home networking interface with 5C (DTCP) copy protection. of multiple displays within the home. CableLabs has defined a number of This article looks at the features of the different consumer premise devices that second generation POD and some of the interface with POD modules [5]. These technical details of how it will operate. include one-way digital televisions and sophisticated two-way set-tops with the OpenCable middleware (OCAP). Appendix A THE POINT-OF-DEPLOYMENT (POD) is a list of the currently defined OpenCable MODULE defined devices. In the future corresponding versions compatible with a multistream POD The Point-of-Deployment (POD) module, will be defined as well. as currently defined by SCTE 28 [1], SCTE 41 [2], and OpenCable™ specifications [3,4], Single Streams provides a common format for decrypting premium MPEG-2 content delivered via a The current POD specifications were built cable network. As a result of the use of open upon the National Renewable Security standards the consumer premises equipment Standard (NRSS-B) [6]. As such they were can be independent of the conditional access designed to work on a single multi-program system used on that particular cable plant. transport stream received via a single QAM tuner. While it might be possible to multiplex make it difficult to share a common out-of- multiple transport streams into a single band transmitter. Therefore it is necessary to transport stream, it would require define a new POD device and interface sophisticated hardware on the part of the capable of supporting multiple transport receiving device. In addition, the current out- streams from multiple QAM tuners. of-band signaling methods used on the POD
OpenCable Host Device RF Cable QAM MPEG Graphics Tuner Decoder Audio/ Out-of- Video Band RX/ Outputs TX
CA POD Decrypt Copy Protect
POD Module
Figure 1 - Current POD and Host Operation Diagram MULTIPLE STREAMS Even more sophisticated host devices might serve multiple displays within the home Expanded Features via a home network. A central home server device may have multiple tuners (at least one A POD capable of supporting multiple per display device) as well as a HDD for transport streams would enable a number of storing content for later viewing. Content is new digital cable devices. A multi-stream then delivered via the home network to the POD would be capable of decrypting multiple various display devices. premium programs located on different multiplexes received by multiple QAM Requirements tuners. Perhaps the simplest feature enabled is picture-in-picture (PIP) where one program is In collaboration with the cable operators, displayed in a window overlaid on another CableLabs has defined a tentative list of program. Multiple tuners and a POD capable requirements for the second generation POD of decrypting multiple programs are required device that supports multiple transport for a device offering this feature. streams. These requirements are subject to change as we develop the technical If the host device contains a hard disk specifications. drive (HDD) for temporary storage of MPEG streams, then multiple tuners and a multi- 1) The Multi-Stream POD specification stream POD enables the ability to watch-and- should encourage the development of record, or to record two shows retail digital cable set-top and simultaneously. terminal host devices through the use of OpenCable and other publicly available specifications. Just as the current open standards for the POD 6) The Multi-Stream POD interface shall module are enabling multiple provide a discovery mechanism for a companies to offer digital-cable ready Multi-Stream capable host to discover devices, the multi-stream POD should how many simultaneous Transport enable even more innovative products Streams and PID decrypts the POD utilizing multiple tuners. scan handle. This allows the Host to manage POD resources and prioritize 2) The interface shall provide sufficient which programs to send to the POD. bandwidth for a maximum data rate which supports the payload from up 7) The multistream POD specification to six simultaneous 64 QAM transport shall allow for the use of multiple streams, or up to five 256 QAM PODs in a single device. This would transport streams, or any combination allow for any future products that that is below the maximum data rate. might need more stream support than a single multistream POD can handle. 3) The Multi-Stream POD shall be able This extensibility ensures any future to decrypt multiple Programs from a devices that contain even more tuners single Transport Stream as well as than anticipated today could be multiple Programs from multiple supported. Transport Streams, up to the resource limitations of the POD (see 8) All two-way multistream Hosts shall Requirement number 6 below as contain a cable modem that supports well). the DSG specification [9]. All multistream PODs shall support DSG 4) The Multi-Stream POD shall be out-of-band. backward compatible with the Single- Stream POD. It shall appear as a 9) Every transport packet that enters the Single-Stream POD in a Single- POD from the Host will be returned Stream host device. This will enable to the Host in the same order, and the second generation PODs to be with a fixed delay. used in current single-stream POD devices. Cable operators can Given these requirements, a POD-Host transition to the new PODs and still interface specification is being developed that support the deployed single-stream meets the requirements while also creating a devices. viable commercial product. CableLabs is working with cable operators and consumer 5) Multi-stream PODs shall support both equipment manufacturers to create these traditional QPSK out-of-band specifications. methods and out-of-band data delivered via cable modem. The POD Second Generation POD Operation will indicate to the Host which OOB The multi-stream POD builds on the method to use depending on what the current POD by providing a command cable plant supports. As a result, interface, out-of-band communication second generation PODs will be able interface, and transport stream ports. The to be used on any digital system in command interface will use the same layering North America now and in the future. of objects as the current POD. Like the current POD, the multistream The multi-stream POD will use the same POD will provide a single transport stream HOST-POD communication method as the input port and a single output port. As a result present POD. The Host and POD the transport streams from multiple tuners will communicate via a series of Application be multiplex into a single stream before Protocol Data Units (APDUs). Most of the entering the POD. The POD will decrypt any APDUs for the multi-stream POD will be selected and authorized premium content identical to the ones for the current POD as programs. Transport packets are returned to defined in [1,3]. This will allow developers to the Host device in the same order they are re-use most of the software from single stream received. The Host then de-multiplexes the PODs and devices. Some APDUs will be various streams and provides them to the modified to include identifiers to indicate to various services requesting them. A block which transport stream they apply. For diagram is shown in Figure 2. In the example, the CA_PMT() APDU is used to multiplexer section, the sync byte within the indicate which encrypted programs are to be MPEG transport packet header will be de-crypted by the POD. In the multi-stream modified by the Host in order to uniquely POD the CA_PMT() APDU will be modified identify the different transport streams. The to indicate which program as well as which host de-multiplexer would then use the sync transport stream contains that program. byte to identify packets returning from the multi-stream POD.
OpenCable Host Device QAM MPEG Tuner 1 Decoder RF PIP Cable QAM MPEG Circuitry Multiplexer De- Audio/ Tuner 2 Decoder multiplexer Video QAM Outputs Tuner 3 HDD Out-of- Band RX/ TX POD CA Copy Decrypt Protect
Multi-stream POD Module
Figure 2 - Sample Multistream POD and Host Diagram
Further Study Appendix A – OpenCable Host Devices CableLabs will continue to work with its CableLabs defines the following member and interested vendor companies to OpenCable host devices to operate with the complete and publish the multi-stream POD current POD: specification. Interested companies can contact CableLabs for information on how to OCAP: OpenCable Application Platform become part of the drafting team for the [7,8]. A common middleware specification multi-stream POD specification. based on Java for interactive television applications. References OpenCable Set-top Box: Cable box with 1. ANSI/SCTE 28 2001, “Host POD ability to decrypt digital tiers. Includes one- Interface Standard (formerly DVS way and two-way capable boxes, requires 295)” POD for decryption of cable provider 2. ANSI/SCTE 41 2001, “POD Copy services. Outputs include RF, DVI, Protection System” Component outputs and 1394. 3. “OpenCable Host-POD Interface Specification”, OC-SP-HOSTPOD-IF- OpenCable TV: Cable ready TV with ability I12-030210 to decrypt digital tiers. Includes one-way and 4. “OpenCable Point-of-Deployment two-way capable TVs, requires POD for (POD) Copy Protection System”, OC- decryption of cable provider services. SP-PODCP-IF-I09-030210 5. “OpenCable Host Device Core Advanced OpenCable Set-top Box: Functional Requirements”, OC-SP- OpenCable Set-top Box with a cable modem HOST-CFR-I12-030210 for two-way services. 6. EIA-679-B, “National Renewable Security Standard” Advanced OpenCable TV: OpenCable TV 7. “OpenCable Application Platform with a cable modem for two-way services. Specification (OCAP) 1.0”, OC-SP- OCAP1.0-I05-030210 OpenCable HD Set-top Box: OpenCable Set- 8. “OpenCable Application Platform top Box that supports decoding of High Specification (OCAP) 2.0”, OC-SP- Definition TV. Can be either one-way or two- OCAP2.0-I01-020419 way, includes new outputs such as DVI and 9. “DOCSIS Set-top Gateway (DSG) HDMI. Interface Specification”, SP-DSG-I01- 020228 OCAP 1.0 Set-top Box: Supports all OCAP compliant applications, and is two-way. Reference Acquisition OCAP 1.0 TV: Supports all OCAP compliant SCTE: Society of Cable Telecommunications applications, and is two-way. Engineers, 140 Philips Road, Exton, PA 19341 OCAP 2.0 Set-top Box: Supports all OCAP Telephone 800-542-5040; compliant applications, two-way, with a cable E-mail: [email protected] modem (Advanced OpenCable Set-top Box). URL: http://www.scte.org
OCAP 2.0 TV: Supports all OCAP compliant CableLabs: Cable Television Laboratories, applications, two-way, with a cable modem Inc., 400 Centennial Parkway, Louisville, CO (Advanced OpenCable TV). 80027 Telephone: 303-661-9100 URL: http://www.cablelabs.com SWITCHED BROADCAST CABLE ARCHITECTURE USING SWITCHED NARROWCAST NETWORK TO CARRY BROADCAST SERVICES
Gil Katz Harmonic Inc.
Abstract A switched broadcast architecture allows operators to offer a virtually unlimited number Bandwidth is a precious resource in any of broadcast programs while freeing costly cable network. Today, Cable MSOs broadcast bandwidth for revenue-generating services. hundreds of digital channels over HFC The switched broadcast model dynamically networks to all cable subscribers. These switches a broadcast channel “on”, via the channels occupy a sizable part of the plant RF narrowcast channel to which the subscriber is spectrum, yet at any given moment, most connected, when a subscriber attempts to tune channels remain unviewed. Significant RF in the channel. Operators can also add an spectrum can be reclaimed by switching “less array of specialty or targeted channels without popular” broadcast channels according to increasing their systems’ broadcast channel user demand. spectrum or capacity.
A narrowcast switched video network for The theory behind switched broadcast VOD services is already in place in many relies on typical Pay-TV viewing behaviors. large cable systems today. This switched Within a particular service segment or node, network provides digital video content to only a handful of channels are being accessed subscribers on demand, occupying bandwidth at any given time. In other words, numerous only when a title is requested and sent over channels are not being watched and there is a the HFC. lot of bandwidth that could be saved or used for other services. This paper will discuss the existing switched narrowcast network architecture as Many cable systems already have a a scalable, cost-effective, flexible and narrowcast video network for VOD services “switched broadcast ready" network. in place. The most common network architecture for VOD uses standards-based commodity GigaBit Ethernet (GbE) switches to build a cost-effective switched network. INTRODUCTION Edge QAM devices with standard IP interfaces serve as gateways between the Bandwidth is a valuable resource for cable standard Ethernet/IP network and the HFC, operators. They are constantly trying to and enable QAM sharing for multiple leverage their existing infrastructure to expand services. Switched broadcast is just another their broadcast channel line-up and offer as type of service that can use that narrowcast many revenue-generating services as possible. video network infrastructure and share the IP Possible new services include video-on- network and QAM resources. demand (VOD), high-speed data or voice- over-IP (VoIP).
The diagram below describes a typical VOD system running in parallel to a broadcast network.
This paper presents a solution using an standard IP solution, simply needs to join a existing VOD infrastructure to enable a multicast group in order to get the broadcast switched broadcast services overlay at service. The IP router will send the service to nominal cost to the operator. The discussion the client once it is part of the group. will outline any required additions to an existing broadcast infrastructure in order to Using the same approach, the IP edge support a switched broadcast application. QAM device will join the multicast group following a subscriber’s request to receive a THE SOLUTION switched broadcast service, and will forward the service to the QAM feeding the set-top
The solution presented here has been box’s Service Group. adapted from a standard Multicast IP solution. For several years, Telco operators have been A single edge QAM device can serve providing video services over their standard IP multiple service groups. Its MPEG-2 networks. The video content may be provided multiplexing core enables service on demand and transmitted to a specific multicasting at the edge of the network by Unicast address, or broadcast in Multicast IP duplicating the content and streaming it groups into the IP network. Each client, in the simultaneously to multiple destinations or QAMs. THE SWITCHED BROADCAST SERVICES they are currently, and some will be sent to the switched narrowcast network. Content intended for switched broadcast The same devices that were used before for can be any standard definition and/or high- rate shaping will now be used for VBR to definition video programming, or data content CBR conversion. A new device should be such as games or electronic program guides. added, converting the switched broadcast The programs may be locally encoded or streams into SPTS (Single Program Transport received off the air. The channels should be Stream) over Ethernet/IP/UDP frames. The carefully selected as “less-viewed” programs IP/UDP frames are identical to those being relative to the entire broadcast domain in generated by the Video Server. A single 1-RU order to guarantee efficient use of network converter box costing less than $10,000 resources. For example, these channels could supports hundreds of SPTSs! be niche programming, ethnic programming or local interest channels. Most of the modifications needed to enable ENABLING SWITCHED BROADCAST switched broadcast services happen in software, such as the Switched Broadcast The concept proposed in this paper application on the STB and the SRM (Session suggests the use of existing components Resource Manager). The SRM is responsible owned by the MSO to keep the capex for sharing the network and QAM resources investment nominal. Looking at the diagram between the VOD services, the switched below, which represents the modified broadcast services and other future services. architecture, it is clear that not much has been The diagram below shows the suggested added. Using the same broadcast feeds, some architecture. will be forwarded to broadcast channels as SO, HOW DOES IT WORK? encoded content, generated in different regions, to all systems within a cable network. Each channel selected for switched Doing so would require no extra bandwidth. broadcast will be transmitted into the Instead, only standard IP connectivity between streaming IP network as an SPTS over a the systems is required. dedicated IP/multicast group (multicast address). The QAM edge device emulates the Virtual VOD (V-VOD) RF portion of the network to an IP network and treats the switched broadcast channels as Virtual VOD can be seen as an improved standard IP/multicast services throughout the version of NVOD services using the network - from the video source to the narrowcast network. It still enables a level of subscriber’s set-top box (STB). VCR functionality, while dramatically reducing the bandwidth consumed by the Once a “switched broadcast” program is NVOD service over the broadcast network. selected from the program guide, the STB Another significant benefit is that V-VOD forwards the request to the SRM. The SRM requires less streaming capacity than regular identifies the service group (SG) where the VOD services. request originated, and checks the bandwidth availability of the QAMs feeding this SG (or For example, ten two-hour movies that zone). start every five minutes (or twelve times each hour) require 10 x 24 = 240 streams. At 3.75 The SRM provisions the appropriate edge Mpbs per stream, 240 streams require a total device for the relevant multicast address, and of 900 Mbps bandwidth. A single standard sends information regarding which QAM VOD server will suffice. The VOD server will should receive the stream. Upon provisioning, stream all 240 streams regardless of actual the edge device will “join” a multicast session user demand. However, as in the case of and re-multiplex the stream into the switched broadcast programs, the streams will appropriate MPEG transport stream/QAM. be dropped at the IP switch connected to the The SRM sends an acknowledgement to the VOD server and will not consume HFC STB, providing the QAM channel and the resources unless requested by a subscriber. program ID. In this architecture, switched V-VOD In general, the process is nearly identical to leverages the narrowcast QAM access to carry the way a subscriber selects a VOD service. NVOD streams based on demand. While the The primary difference between a switched- concept does not require any in-band channels broadcast stream and a VOD stream is that for NVOD, it does enable V-VOD so that each streaming does not originate directly from a subscriber will not have to wait more than five server or other storage device. Rather, the minutes for a movie to start. content is simply streamed off the broadcast services. nPVR APPROACH
LEVERAGING THE BROADCST Network PVR (nPVR) is already providing SERVICES broadcast channel programming per user demand. In the nPVR model broadcast Other applications can be integrated into a channels are being recorded on Video Servers, switched broadcast infrastructure. An operator the same Video Servers providing VOD could easily provide all available locally- services. The content is being provided to the lower prices on Video Servers as well as the user using the same narrowcast network narrowcast network, enabling niche infrastructure used for VOD services. The programming over nPVR infrastructure. advantages of the nPVR solution are clear: it enables VCR control as well as very targeted Again, The argument here is about minimal Ad insertion. So, why wouldn’t we use the investment for the niche programming, and nPVR model, which provides prime broadcast reuse of existing infrastructure. We are content on demand, for the more niche proposing a migration path from broadcast to programming? The answer relates to storage switched broadcast (provided as another cost and capacity, and to the narrowcast source into the switched narrowcast network) network cost. Nevertheless, the migration to to nPVR. Everything on Demand (EOD) will drive
SUMMARY devices such as switches and IP edge QAM devices. Existing VOD systems based on A switched broadcast architecture enables these devices are scalable, cost-effective, an unlimited expansion of the broadcast flexible and “switched broadcast ready.” channel line-up while freeing up precious bandwidth for other revenue-generating Minimal capital investment is required to services. New services could include Virtual enable switched broadcast services on a VOD and local content distribution. switched narrowcast network built for VOD services. The same QAM and IP network The solution introduced in this paper resources can be shared between the relies on standard “off the shelf” IP/GbE different services. TAMING THE PEER TO PEER MONSTER USING SERVICE CONTROL
Michael Ben-Nun P-Cube Inc.
Abstract (e.g., Napster). Completely decentralized This document explains the increasing P2P has no central server (e.g., Gnutella) to bandwidth and network capacity planning provide search capabilities due to the fact challenges peer-to-peer file exchange that the clients search amongst themselves. applications cause Internet Service Other variations of P2P provide application Providers. It discusses how Service Control specific networks (e.g., KazaA) and some – the concept of statefully tracking network utilize an open standard (e.g., Gnutella and usage and enforcing advanced subscriber, OpenNAP) to allow clients share all sorts of application and destination differentiated content. All of these applications allow policies – is key to resolving the peer-to- individual users (conveniently shielded by peer traffic issues within existing network the anonymity of the network) to share files infrastructure. over the Internet. These files often contain copyrighted materials (e.g., songs, movies, software, etc.) that no commercial content provider could legally afford to publish.
PEER-TO-PEER AFFECT ON NETWORK CONGESTION Due to this simple file sharing method, Napster, which is considered to be the first The Evolution of Peer-to-Peer (P2P) P2P application with mainstream appeal, was an immediate success among Internet Understanding the relatively short history users, especially those with high-speed of P2P applications and its underlying Internet connections. A court ordered technologies is critical to the comprehension shutdown of the Napster service did little to as the impact it has on broadband IP decrease the amount of P2P file swapping networks. Internet based P2P is a relatively activities, rather it can be argued that the new technology, which allows for the added publicity probably achieved the creation of decentralized, dynamic, and opposite effect and the popularity of P2P anonymous logical networks for information applications has increased ever since. With exchange using the public Internet. In new P2P clients and applications released to “traditional” client/server model a well- provide more functionality and ease of use, known source provides content and P2P traffic comprises a large part of Internet information to requesting clients, whereas in bandwidth usage. The popularity and use of P2P, applications utilize various techniques different P2P clients is varied and can be to allow users to search and share content determined by a variety of factors. Some between themselves. There are several clients are more popular in certain different P2P technologies and architectures geographies (such as Winny which has wide that evolved from the most basic type – one spread acceptance in Japan), while others that has a central “coordinating” server have a strong following among the utilized for content searches between clients “distributors” of specific types of material.
Peer-to-Peer Incurred Congestion and the influence P2P technologies have on these parameters. P2P applications are P2P clients, due to their numbers and increasing in popularity and constitute a intensive need for network bandwidth are growing percentage of network traffic. causing significant network congestion. These applications are so popular that a new With less bandwidth left for other network term has been coined to describe the more traffic, this results in a reduction of the avid users of these technologies. Often overall broadband experience for other referred to as “bandwidth hogs” or “abusive subscribers on the network, and raises subscribers,” these users are using their network capacity, planning, and broadband network connections to generate management issues. Every IP network is a disproportional amount of network traffic built with assumptions about usage, which and significantly contributing to network in turn is used to analyze and compute the congestion. necessary amount of network capacity and resources needed to support a given subscriber base. P2P applications are The following charts, produced from different from traditional client/server analyzing the usage of a particular network, applications in the way that users run them serving HSD cable subscribers uncovers the and how the applications use the network. alarming truth: Approximately 70% of The table below provides a glimpse of some network bandwidth is being used by P2P of the parameters used by service providers, applications. their importance for planning the network,
CONTROLLING PEER-TO-PEER (a) Identify, account and report on P2P TRAFFIC: TECHINCAL usage. REQUIREMENTS (b) Control the bandwidth these applications consume. With the growing amount of P2P traffic, there is a clear need to address the link congestion and bandwidth issues The following section provides detailed it creates. To solve the problem, service- technical requirements that a solution must providers must use a solution that is able provide. to: Technical Requirements new ones emerge, the underlying protocols used to carry the peer-to-peer traffic change When attempting to identify and frequently. The solution must be quick to control P2P traffic, it is important to adapt to new protocols, and provide new remember the underlying technical identification mechanisms. requirements from a proposed solution. Once the requirements are fully understood they could be used to Note that the importance of the above- evaluate possible solutions. The unique mentioned identification requirements technical requirements that need to be increase in complexity and number with the addressed are: growing speed of the development of new peer-to-peer applications/protocols. Even today, P2P applications use well-known IDENTIFY: ports, assigned to other network uses (such Ability to classify traffic based on as port-80 for web-browsing), and they are layer3-7 parameters: Peer-to-peer constantly migrating to these port numbers applications do not utilize well-known in an attempt to masquerade as ‘traditional’ port numbers, and thus cannot be network activities and thereby avoid classified by simply looking at IP packet detection. Hence, simple analysis based on headers (IP addresses, TCP port- port-numbers leaves most of the P2P traffic numbers, etc.). Rather, deep inspection unaccounted for, and will not truly address of packets, including the identification of the problem. layer-7 patterns and sequences must be supported. CONTROL: Ability to maintain bi-directional Ability to control bandwidth at various flow state: In order to identify a isolation levels & granularities: To control the particular flow of packets as peer-to- bandwidth impact of P2P applications it is peer, carriers cannot inspect each packet necessary to provide a network control within that flow to make the mechanism for different levels of isolation and identification. The solution that control. The solution must provide the means performs proper identification of P2P to control bandwidth at “subscriber traffic must ensure that once a particular granularity”, whereby it limits the total flow (e.g. a TCP connection between amount of bandwidth each subscriber can two hosts) is identified as P2P, all consume. It must be able to control the packets on that flow are tracked, and bandwidth of particular flows, so as only the treated as such. Of critical importance is P2P identified traffic of a particular subscriber the ability to tie between both directions is limited, while the rest of that subscriber’s (i.e. upstream & downstream) of a flow, traffic is left unaffected. since in many cases the initial identifying pattern resides in a packet sent from one host, yet the majority of Ability to enforce time, destination and traffic can flow in the other direction. subscriber differentiated policies: To control the bandwidth congestion cause by P2P, and enforce various control policies, Ability to provide quick turn-around while maintaining the necessary flexibility for new P2P applications: As peer-to- to actually implement these on real-life peer applications constantly change, and subscribers, the solution must provide the means to create differentiated capacity to support the total number of enforcement schemes (or policies) based subscribers served by the network links, for on time of day, destination and both existing subscriber numbers today, and subscriber. Specifically, the ability to for forecasted growth. create different enforcement packages for different subscribers must be supported. APPROACHES TO CONTROLLING PEER-TO-PEER TRAFFIC
Ability to maintain subscriber level With the technical requirements in mind, quotas: In order to control P2P traffic in the following section explores possible a persistent manner for each subscriber, solutions to identifying and controlling peer- the solution must provide the to-peer traffic. infrastructure to maintain a usage state for subscribers, and account for the total amount of P2P traffic over time. As an Using Router/Switch QoS Mechanisms example, the ability to maintain the total Existing routers, switches or similar amount of P2P traffic each subscriber network devices contain various types of has consumed on a traffic classification and QoS mechanism, daily/weekly/monthly basis, and apply which could potentially be used to control different bandwidth quota based P2P bandwidth. consumption restrictions based is key to moderating the use of the network. Note that while the issue of However, as these devices were not controlling and enforcing P2P bandwidth designed to address these issues, they do not consumption is crucial for maintaining a provide the following capabilities: congestion-free and predictable They do not provide Layer 3-7 traffic broadband network, it can cause classification. Nor do they maintain state customer expectation issues, as the across packets flows. current subscriber-base is unaccustomed to imposed limitations on its high-speed data access. Therefore the above They are not “subscriber-aware” and flexibility is mandatory as service- cannot provide subscriber differentiated providers create the policies best suited enforcement for their subscriber-base.
As a result, switches and routers do not Support high-speed network rates, provide the means by which the peer-to-peer and subscriber-capacities: As today’s traffic can be identified, and network usage broadband networks are built to sustain policies be applied to it. Additionally, as the significant traffic loads, the solution QoS mechanisms in switches and routers must support today’s network interfaces attempt to deal with link congestion and and traffic rates. Typical broadband bandwidth distribution, they do not provide networks use Gigabit Ethernet and OC the necessary subscriber-differentiated interfaces with high throughput. In policies, required to control the peer-to-peer addition, the solution must have the traffic once identified. Using DOCSIS 1.1 classification of the traffic and the subscriber it is mapped to. The DOCSIS 1.1 specifications, contains many features and capabilities The following diagram depicts the to control bandwidth utilization, and internal operations of a service control offer differentiated services to platform. subscribers. However, by itself the DOCSIS 1.1 specifications cannot fully On step (1), the platform classifies each address the issue of controlling P2P packet received into a stateful, bi-directional applications. This is due to the fact that flow. DOCSIS 1.1 does not: On step (2), the platform performs dynamic stateful reconstruction of the Provide the mechanisms to classify application (layer-7) message exchange in traffic based on layer-7 capabilities, or the flow, and identifies the application used maintain state for bi-directional network by each (peer-to-peer, web, mail, etc.) flows. Provide the required bandwidth On step (3), the platform maps each such control isolation and granularities. flow into a particular subscriber. Typically DOCSIS 1.1 provides the means to there is a many-to-many relationship, in control traffic at a defined flow which many application-flows are mapped specification (typically a combination of to many subscribers. layer3-4 parameters). However, as mentioned above, to fully control P2P On step (4), once the traffic has been bandwidth consumption, there is a need classified, identified and mapped, it is to implement various layer of bandwidth accounted for on a subscriber basis. control, which the DOCSIS 1.1 Subscribers’ state is updated according to specifications does not attempt to the traffic they transmit or receive, and this address. impacts (along with the their assigned policies) the final bandwidth enforcement policy (5) applied. As a result, while DOCSIS 1.1 is a potential key component in service On step (6), the selected policy is differentiated high speed data networks, translated into packet level decisions, it does not provide the mechanisms to indicating how the actual implementation of control the peer-to-peer abuse problem. the bandwidth restriction is performed.
Ultimately the total bandwidth consumed Using Service Control Platforms is reduced through control implemented in the service control platform and the overall A Service Control Platform is defined network congestion is reduced to a level as a platform that is able maintain state acceptable to the network provider. for each network flow, classify it according to layer3-7 parameters, and implement various bandwidth shaping and control rules, based on the Recieved Packets john123 f1: peer-to-peer f2: web-browsing ... mary612 fN: peer-to-peer (1) Classify (2) Classify flow (3) Map Packet to to application by application- flow layer3-7 patterns flow to and messages subscirber
john123: john123: Transmitted Packets Total Daily "unlimited P2P: 44MB p2p" WEB: 12MB mary612: mary612: Total Daily "limit p2p to P2P: 14MB 64kbps" WEB: 33MB
(6) Perform (5) Select ( 4) Maintain Bandwidth Enforcement Subscriber Shaping & Policy State & Quota Control
CONCLUSION Users utilizing the network for P2P traffic are typically residential The combination of Peer-to-Peer subscribers, unaccustomed to enforced applications’ aggressive use of network bandwidth restrictions. As such the resources and the growing popularity of control mechanisms required to contain P2P is straining broadband networks, the affects of P2P, while avoiding and causing congestion, operational subscriber alienation due to rigid costs, and user satisfaction issues. Using policies, are not provided by standard Service Control, P2P traffic can be QoS mechanisms. This means that precisely identified and controlled, so as commonly deployed switched and to contain its affects on the network routers cannot act in the same capacity without influencing other applications as Service Control Platforms for the and network users. Furthermore, the purpose of P2P monitoring and control. P2P consumption has a different A complete and effective solution network behavior and usage pattern than requires the combination of P2P typical “common” network applications identification flexibility, traffic control, that require differentiated bandwidth quotas, and subscriber awareness – the (such as enterprise based SLA and QoS). main building-blocks of a Service Control Platform.
Visit P-Cube on the web at: www.p-cube.com. 388 Oakmead Parkway, Sunnayvale CA 94085. •Tel: 408-720-7770 •Fax: 408-720-7772
Parameter Importance for network Influence of Change caused by planning traditional P2P applications applications
Networks are A typical residential P2P applications asymmetrical in nature: user uses the network encourage users to the amount of traffic that for downstream share files, and a a network can sustain applications. These typical peer serves applications (e-mail, upstream (i.e., from the gigabytes of files. web browsing, etc.) subscribers to the generate a larger This causes a drastic network), is different amount of change in the from the amount it can downstream traffic upstream/ sustain in the opposite for each downstream ratio, direction. The ratio corresponding and as a result Upstream / Downstream Traffic Ratio required between these upstream request, congestion on the two directions is in direct and service providers upstream link (due to correlation to the have come to rely on individual users’ requirements of the this ratio to model increased uploading applications using the network capacity of files).
network. Networks are built with a specific ratio which, if incorrect, may cause high rates of congestion and unutilized capacity. Parameter Importance for network Influence of Change caused by planning traditional P2P applications applications
Service providers The time of day As P2P applications typically assume an and percentage of are usually used to average duration of activity expected upload or download network use per for residential large, multi- subscriber per day, and broadband megabyte files, they (based on subscriber subscribers is are typically left profiling) peak use rooted in the unattended for days periods. A service premise that a at a time while the provider would typically typical residential application constantly be able to predict and customer uses the attempts to download account for network network only when a list of files. At the “rush hours” and “lulls” the subscriber is same time it can periods of network use. physically present serve as a search This subscriber profiling and actively using node for the P2P is based on assumptions the connection. network and serve that residential home Such is the case multiple file requests Time of Day and users primarily use the when web of other peers. This Percentage of Activity network during weekends browsing, reading creates a never and evenings, and that e-mails, etc. ending, high volume telecommuters and small stream of network offices use it primarily activity throughout during business hours. the day. Sudden or sporadic changes in these patterns For example, a may cause congestion student’s computer during certain hours that with a broadband were not evident before. connection can compete with telecommuters for vital network resources during business hours while the student is at school.
Parameter Importance for network Influence of Change caused by planning traditional P2P applications applications
The costs associated with Traditional uses of P2P traffic has serving each network the data network increased the amount packet and connection are mainly OnNET of traffic between can depend on the (email, nntp, web- users in a significant location of the peer of the proxies), with a way. When two or subscriber. Carefully small percentage more P2P clients start crafted peering being OffNET. using the network agreements with other This small they form a direct network providers can percentage of connection to help reduce the amount of traffic is for exchange the file. Traffic Destination and traffic, and hence the cost content that is Whether the clients Peering points of expensive transit located at sites use the same or connections. Furthermore external to the different providers is local traffic (often network providers not a determining referred to as OnNET) domain. factor in how the P2P that does not leave the connections are service provider’s own made. P2P file backbone network, is exchange has significantly lower in cost significantly than traffic that does increased the (OffNET). potential for OffNET traffic.
No matter the topology Traditional P2P applications are and architecture of the applications have a mainly used to share network, there is a finite large “time-to- large binary files that amount of bandwidth consume” factor: A have a much lower available for all its users, small web-page “attention-per-byte” and certain over- can take several ratio. A three-minute subscription assumptions minutes to read, a song is usually 3-5 Estimated Traffic Volume are used when planning single e-mail megabytes. A 10- the capacity of the message might take minute movie can be network a number of hours hundreds of to process. This megabytes long. determines how the Each piece of content traffic volume for that is served is each type of traffic/bandwidth content served. intensive.
THE THREE DIMENSIONS OF HOME NETWORKING
Doug Jones YAS Broadband Ventures
Abstract
Home networking is a lot more than network in their homes to serve their needs. “wired vs. wireless.” Home networking also This network will connect many and various includes both the command/control devices for services such as entertainment, language to search for and play content communication, energy management and within the home and defining how home control. applications are run and managed. Keeping an eye on all three dimensions is needed to To accomplish all this, the home network promote plug and play interoperability of will need to be easy to install, easy to home equipment. connect devices to, easy to upgrade, adaptable, low cost, and secure. Quite a wish list indeed, but we are just at the beginning INTRODUCTION of home networking.
Business Case This paper discusses in technical terms It started with Personal Video Recorders specific functions needed on the home (PVRs). Now those PVRs can both be network in order to have different devices networked within the home, and over a high- and services work together. All of the topics speed connection to sources outside of the discussed in this paper are available today home. The home is becoming a source of from more than one supplier, though not stored content and services. always in an interoperable fashion. In order to have widely available plug and play With content in the home, users will need interoperability, the industry will have to not only better and faster connectivity within converge around a select few of the the home, but also the means to search the initiatives currently available. home for content and play it back to the audio/video device of their choosing over a Three Dimensions reliable home network. This paper discusses three distinct dimensions of home networking, all of Home networks will interconnect both which are critical to understand from both a entertainment devices and general technical and a business perspective. computing devices. This will allow services Understanding what happens in these three like a home calendar and shopping list to be dimensions is important to understanding integrated onto the same platform as how home networking really works. entertainment services. The three dimensions are: Just like cable is providing many services Datalink Technology, e.g., 10/100Base-T, over one network (video, voice, data), it IEEE 802.11a/b/g, HomePNA, IEEE 1394, makes sense that users will want a single HomePlug, etc. These are all technologies that move electronic bits. Some are wired Overall System View and some are wireless; some are The discussion begins with a simple synchronous and some are asynchronous; network consisting of two devices, one that some work over existing wires and some has content (PVR) and one that wishes to require new wires. Each has inherent search through and play a piece of content advantages and disadvantages that will be (Set Top Box [STB]). These devices are discussed. shown in Figure 1.
Interoperability Software to do functions such as device discovery and content search and playback. All of these functions are PVR STB sometimes known as “plug and play.” This Television very important dimension defines how the home devices learn about each other without Figure 1 – Simple System for Home prior provisioning and how they work Networking together to offer useful services. In order to use these “plug and play” protocols, the Note that a television is also shown in devices are connected using some kind of Figure 1. The STB could be embedded in datalink (wired or wireless) to carry the bits the television but the two are shown as that make up the message used by the separate devices here because STB/TVs are interoperability software. Suites of protocols not generally available (yet). As well, the such as UPnP™ (Microsoft), Rendezvous PVR could have been embedded in the STB (Apple), JXTA™ (SUN), DENi, etc., solve (or visa versa). Note also that the STB is not these issues in their own ways. Other necessarily connected to a HFC Cable companies such as Ucentric, Motorola and network, rather, the STB is just connected to Scientific-Atlanta are creating solutions in the PVR and converts content to a form that this area. While many claim to be can be viewed on the television. “standards” or standards-based, the reality is there is a very fragmented marketplace right This example alone is useful as new now with several solutions vying for the top paradigms for devices on the home network position. This is the key area to ensure will arise as multiple suppliers enter the service interoperability among devices on market and consumer choice begins to the home network. dictate what devices do. Traditional suppliers and traditional devices will have to Applications Frameworks to allow the adapt. development and execution of machine- portable applications within the home. For the time being, assume the PVR and the STB are connected via some kind of A well-known framework is datalink. The datalink technology is PersonalJAVA™ although there are many important from the standpoint that maybe others. These environments allow an the STB can only accept analog video, or application to be run on any of several perhaps it can only accept digital video. The computing platforms, for instance a Personal datalink technology has to support carrying Digital Assistant from one supplier and a data in the format that is needed to be useful. multimedia personal computer from a If digital video is used, the system has to different supplier. know if the datalink supports a 3.5 Mbps Standard Definition (SD) bit rate or a 20 PVR STB Mbps High Definition (HD) bit rate. Television If the datalink is a dedicated connection Hello ! (I'm a STB) between the PVR and the STB (that is, not Hello ! (I'm a PVR with content) shared with any other device) then maybe it Tell me all your content does not need to support Quality of Service Here is my content (Metadata) (QoS). If the datalink is shared by other Play this content devices, perhaps QoS is needed to ensure the streaming content bit stream sent to the STB arrives there Pause/FF/RW/Stop uninterrupted and without loss. Raw bit rate, support for QoS, and the ability to be shared Figure 2 are some of the characteristics to consider when studying datalinks. The first two lines in Figure 2 (the “Hello’s”) are what allow devices from When the PVR and the STB are first different suppliers to recognize each other networked together, they will use some on a home network. While these two lines method to discover each other. That is, are highly simplified, they show that when devices on the home network should be able devices are connected to a datalink they to learn of each other without the need to be advertise what kind of device they are and explicitly programmed by the user. In this what kind of services they offer. This way devices from various suppliers can be information is generally broadcast around placed on the same network and learn about the home network to allow all the other each other without user intervention. This devices on that network to learn about each complex task is handled by interoperability other without user intervention. software running on all the devices on the network. The latter exchanges shown in Figure 2 look somewhat similar to a Video on Once discovered, the STB has the Demand (VOD) session; searching content capability to search the content on the PVR and controlling playback. The home will and play some of it back, including pause, have its’ own sources of content and devices fast-forward, and rewind capabilities. on the home network need the capabilities to Highlights of this process are shown in search for content and play it back. An issue Figure 2. to be solved includes allowing a user to
To complete the exchange shown in seamlessly search for content on both in- Figure 2, the PVR and STB have to talk the home devices as well as devices in the same interoperability protocols. Regardless service providers network. of supplier, if the devices on the network use the same interoperability software then they In order to introduce the third dimension can coexist on the same home network and of home networking, applications, a new function in a plug and play manner. piece of equipment is added to the system diagram as shown in Figure 3. DATALINK TECHNOLOGIES Broadband Network Home Network Introduction STB Application Television server Datalinks are technologies that move bits
PVR on a wire, or in the case of wireless, across the air. The bits could represent MPEG frames, Ethernet frames, IP packets, ATM Figure 3 cells, or some other type of protocol data unit. The datalink does not care what it This new piece of equipment is labeled carry’s, rather it just makes sure the bits get “Application Server” for lack of a better from one end of the connection to the other. term. It could be a STB, or a home media There are dozens of datalink technologies center, but it has the characteristic that available; every day examples include wired applications from a service provider can be 10Base-T and wireless IEEE 802.11b. placed there and function as expected Additional examples are given in later regardless of what the physical device is. sections. The subscriber can use these applications to add value to their home networking Home networks can have more than one experience. Also out of convenience, the type of datalink, for instance using a wired high-speed connection to the home is shown connection to get to various rooms within a connected to the application server. home but then using wireless within the room. If one wanted too, the Application Server, STB, and PVR could all be collapsed into The datalink has to carry two basic types one device. The cable industry has an of information, interoperability software example of such a device and it is called the messages and content. The datalink does not OCAP hardware platform. This is a care about this type of information, it is just monolithic, fully integrated, higher cost moving bits around and the bits can be just solution. What if the hard disk in the PVR about anything. But there are some datalinks needs to be changed out? What if more that will do a better job than others. For memory is needed to run applications? What instance, if the datalink can only support 1 if the user wants to add HDTV outputs on Mbps, it will not be able to carry digital the STB? In the monolithic case, the entire video at a rate of 3.5 Mbps. device has to be switched out. With a modular home networking approach, the When all said and done, the two key items user can mix and match equipment from of concern with a datalink are having several suppliers and upgrade the equipment enough throughput (bits per second) to carry as driven by either personal choices or the services and easily being able to make changes in available technology. The three connections in the rooms where subscribers dimensions of home networking provide the want them. necessary connectivity tools and Two of the most debated aspects of environments to allow devices from multiple datalinks are “wired vs. wireless” and bit suppliers to interoperate on a home network. rate. These will be discussed in the
following sections.
Wired Datalinks need to give one service better treatment One of the key questions is, which wire? than another when there is congestion on the Datalinks exist for home power wiring datalink. With enough raw throughput (HomePlug™), for home telephone wiring available, an arguably simpler datalink can (HomePNA), for home data wiring (Cat-5 be offered, one that does not need QoS. On 10/100Base-T, USB, IEEE 1394) and for the other hand, there are different types of optical fiber (SPDIF). There are even QoS too, specifically prioritized versus several companies working on turning the parameterized. Parameterized QoS is more in-home coax cable into a datalink. complex, having many parameters to guarantee exactly the QoS needed for that The pro’s of wired datalinks include particular service. Prioritized QoS is higher speeds, more consistent speeds, and relatively simpler, giving certain services the knowledge that there is a physical higher priority than other services on the connection. Con’s include the possibility of datalink. Prioritized QoS does not give having to install new wires. guaranteed bandwidth, but sometimes a simple higher priority is sufficient to support Wireless Datalinks the needed service. Of course if QoS is These technologies include Bluetooth™, needed, that means some services will get HomeRF, the IEEE 802.11 series, Magis better treatment than others, and the ones Networks, etc. These datalinks have the that get the “less better” treatment may not benefit of “no new wires,” however, they be happy. may not have the bit rate needed to support several streams of HDTV along with high- INTEROPERABILITY SOFTWARE speed Internet applications like peer-to-peer and gaming. Introduction Interoperability software enables plug and The pro’s of wireless datalinks include no play architectures where devices and new wires and relatively higher speed services can be introduced into a network throughputs that are coming to market e.g., without configuration hassles. In addition, IEEE 802.11g. Con’s include speeds that interoperability software is an important step can fluctuate based on distance and concerns toward eliminating manually installed over security. drivers, relying instead on standard interfaces to put devices in touch with other Datalink Bit Rate devices and the services they offer to the There is a debate of how much bandwidth network. is needed on the home network to support services that consumers want. A key driver Interoperability software is a very to get at the answer will depend on the important step, one that will free the compression used for video services. consumer from having to understand the Entertainment quality video takes a fair technical details of each piece of equipment amount of bandwidth, in the megabits per and manually provision them to interoperate. second range, that must be delivered The protocols and procedures included with consistently and reliably. the interoperability software allow devices on a home network to learn each other’s If there is not enough raw throughput at capabilities and to interact in a way to the datalink, QoS technologies may be provide services. needed. QoS is needed when there is the Interoperability software can work to contact each other and request content completely in the background; the user does and services. not have to configure the equipment to make it work. These protocols are device-to- Service Discovery device, and can occur autonomously once a Service discovery provides a means for device is connected to the home network. devices to advertise the services they can Other devices on the home network that offer to other devices on the home network. understand and talk the same For instance, a device could advertise that it interoperability software will respond stores content, or that it is a printer, or that it autonomously as well. is an audio and/or video playback device.
There are two main architectures for Consider the case where a user wants to interoperability. One is peer-to-peer, where playback audio to a set of speakers in the all devices communicate with all other kitchen. Service discovery information is devices on the network to learn about used to create a list of playback devices that services. Individual devices create their own is categorized by audio speakers. The user internal database of the devices and services would choose the speakers in the kitchen. In available on the network. A second the case of looking for a networked printer, architecture is centralized, where a device is the user would call up a list of all printers. “elected” to be the centralized repository of From there, the user could further information. This centralized device then differentiate the printers based on service periodically broadcasts its presence for new characteristics such as black and white devices coming on the network and all the printers versus color printers. while aggregates information about all other devices and services on the home network. Playback Control Other devices then query the centralized Once all the devices are connected, and device when they need specific information all are discovered (including their services), about other devices and services on the the fun can begin for the consumer. network. Interoperability software also includes the mechanisms to search and play content. Home networking interoperability Searching can be implemented seamlessly software provides several key functions, across devices. When a user requests a list including device discovery, service of all the digital photographs available, they discovery, and playback control. These are presented a list of all the digital photos issues will be described in the following on the network, regardless of the specific sections. device on which they are stored. This is the scenario that is closest to the VOD service Device Discovery of today, except that the service is entirely in Device discovery is the process by which the home. With content on various storage a device learns about other devices on the devices in the home, other devices can home network. At the device level, the search that content and play it back using information exchange is on the order of normal controls such as fast-forward and Ethernet Media Access Control (MAC) rewind. To continue the VOD analogy, its addresses and Internet Protocol (IP) like having movies cached both locally and addresses. By learning the addresses of each centralized. When the user searches the device on the network, the devices are able VOD titles, they are not aware of where a particular title is stored, just that it is The application framework consists of the available for viewing. Application Environment, the APIs, and the individual applications. This software is Industry Initiatives portable to different types of devices As stated earlier, various industry running different operating systems. initiatives, such as UPnP and Rendezvous, are promoting their solutions for Individual applications are written to run interoperability software. There are no less within a specific application framework. The than eight initiatives out there and several framework itself is ported to various suppliers are developing additional operating systems. In this way, the solutions. application, which the service provider cares about, works on various supplier boxes that The key about all these initiatives is that have implemented the application they are not interoperable. While they framework. For instance, a particular generally all use the same underlying application framework (including all the standard protocols and procedures (e.g., applications) should be able to run on a DHCP, SLP, SOAP, XML, etc.), they do so device with an Intel processor running a in ways that are not interoperable. As Microsoft operating system or a device with explained a bit later in this paper, a SUN processor running the Solaris CableHome is positioned to take advantage operating system. It is this device portability of the innovations in this space. that makes an application framework so powerful. The operator no longer has to APPLICATION FRAMEWORK worry about the underlying hardware device, rather, the operator just manages the Description applications that run on that device. The application framework is the third dimension of home networking and allows Application Environment applications to be device independent. That The application environment is a place is, an application can run on any appropriate where software applications are allowed to device regardless of supplier, operating run. The environment supports both one or system, underlying hardware, etc. The more programming languages that relationship of the Application Framework developers use to write applications and to the rest of the device is shown in various tool to manage those applications. A Figure 4. widely recognized application environment is PersonalJAVA™, which is an Individual Applications environment for applications written in the
Program Order a JAVA™ programming language. VOD etc. Guide pizza ...
Application Programming Interfaces (APIs) Application The application environment includes Framework
Application Environment tools that are specific to the programming language. This support can include complex technical tasks such as transaction Operating System management, state management, resource pooling and security checks for all the Hardware (HDD, memory, Input/Output, etc.) applications running within the environment. This is very technical stuff the Figure 4 consumer should never know about, but it ensures the integrity of the applications there are many groups working on a running within the environment, especially consensus solution that hopefully can be since several applications can be running presented in the not too far distant future. simultaneously. CableHome™ Positioning Application Programming Interfaces (APIs) The CableHome initiative, which seeks to APIs can be thought of as standard extend cable-delivered broadband services subroutine calls that developers use to create throughout a customer’s home, issued key applications. These calls include everything specifications for the 1.0 version in April from mundane tasks such as writing to 2002. Also in 2002, CableHome achieved memory to more exotic tasks such as international standardization at the drawing a graphics overlay on a display International Telecommunications Union device. The APIs are the “Rosetta Stone” of (ITU). portable applications. The underlying operating system and hardware speak a low- Whereas most Interoperability Software level machine programming assembly industry initiatives are concentrating language. The applications speak a high- specifically on connectivity within the level programming language. The APIs home, CableHome started with a focus of translate between these two languages first connecting the home to a high-speed allowing humans to write in a high-level data connection over cable using a programming language and allowing the residential gateway. This is illustrated in machine to operate on low-level assembly Figure 5. language. Home Network Speakers High-speed access Security and Rights Management connection CableHome MP-3 player Residential Personal No discussion of home networking would Gateway Computer be complete without including security and CableHome 1.0 STB rights management. focus Television
Security is fairly straightforward, the data Figure 5 carried on the network should be private such that it cannot be snooped, especially in It is expected that the other initiatives will the case of wireless datalinks. eventually focus on the high-speed connection as well, and CableHome in turn Rights management is an important topic is beginning to focus on solutions for device that is currently being debated not only and service discovery and the accompanying within the cable industry, but also by other security issues of rights management and industries as well including broadcasters, content protection. consumer electronics, content producers and satellite. There are several examples or Among the residential gateway features rights management technologies available, provided by CableHome 1.0 are mechanisms including DFAST (Dynamic Feedback for secure remote management and Arrangement Scrambling Technique) and configuration of residential gateway HDCP (High Definition Copy Protection). capabilities that include DHCP, DNS, NAT/NAPT, LAN test tools and a firewall. For the purpose of this paper, suffice it to say that rights management is needed and CableHome key features: CableHome is a tool to extend cable- delivered services to the home. - CableHome helps ensure customer privacy because it does not provide cable Summary operators with the ability to probe or Home networking is a complex topic. configure consumer devices, such as There are three clear and distinct dimensions PCs, within the home. that comprise home networking, and each dimension provides a new set of - CableHome also helps protect customer technologies and a new set of suppliers. privacy by providing cable operators with the tools that assist in mitigating unauthorized snooping of Wi-Fi based Author Contact home networks. Doug Jones YAS Broadband Ventures - Consumers only use this additional 300 Brickstone Square CableHome functionality (e.g., remote Andover, MA 01810 diagnostics) if they have voluntarily voice: +1 978.749.9999 x226 chosen to subscribe to a cable operator’s [email protected] home networking service. CableHome equipment will be deployed in those households where consumers have elected to pay for this additional functionality because they see value in having the cable operator manage the technical complexity associated with the deployment and operations of a residential gateway.
THEFT OF SERVICE IN HIGH SPEED DATA SERVICES: A WAY TO DEAL WITH THIS DIFFICULT PROBLEM
Jonathan Schmidt PerfTech Bulletin Services
Abstract BACKGROUND
The residential HSD/Internet services A lot of effort has been devoted to the environment is experiencing rapid control of HSD network components to technology changes. The providers’ eliminate service theft from modem cloning management tools have not kept pace. That and spoofing. That has left one remaining has exacerbated the vulnerability and extent service leak — the connection behind the of service theft. Consequently, the tools cable modem. must change. Some have, as in the newer set-top boxes; DOCSIS 1.1 CMTS/modems Stopping Modem Spoofing and Cloning use strong certificate-based authentication to prevent service theft through modem Control of the subscriber modem has cloning and spoofing. At the same time, been achieved through certificate-based home networking equipment has become authentication in both configuration and data very inexpensive, is common even in MSO transfer in DOCSIS 1.1. offerings, and has filled computer store shelves. With this equipment, the home network devices become anonymous and can exist in large numbers behind a single modem. Similar to the downstream portion of the analog video cable plant, high quality Internet access can be replicated in a tree- and-branch architecture behind that modem and serve many users. WiFi products enable this architecture to be implemented with invisible, wireless links that are also open to opportunistic taps. These links are Figure 1: Certificate-based security in not susceptible to on-site inspection to check modem data transfer and BIN file for redistribution. authentication
This paper presents an in-band Stopping Address Theft and Disruptive CPE communication channel that can target all Address Configurations workstations on suspected accounts with unblockable screen alerts, which will utilize Control of the subscriber CPE network these invisible links to become an essential address configuration has also been tool in combating such theft. implemented. With Cable Source-Verify, the DOCSIS MAC domain is protected against rogue CPE configurations for unassigned IP addresses or duplicate IP/MAC addresses. In the first years of HSD Internet provisioning, some attempts were made to detect the number of computers behind a gateway. An example can be found in the paper "A Technique for Counting NATted Hosts," by Steven M. Bellovin of AT&T Labs Research. The paper concludes that the technique is imprecise in the single case, and may only be of use in estimating workstation populations behind home gateways.1 Indeed, the mechanism doesn’t Figure 2: Cable Source-Verify control of work at all with some gateways, such as Address assignment on DOCSIS network Nortel’s Instant Internet. It is also easily domain confused by internal LAN activity such as Windows intra-network activity. The Theft of Service Behind the Modem difficulty in definition and detection of the NAT gateways caused very casual, if any, A single home networking NAT enforcement of anti-gateway provisions. gateway can be inserted between a network of PCs and the HSD modem and can appear Today, for example, Time Warner Cable to the network management as a properly offers sales, support, and installation of configured, single user. However, it can multiple user home gateways and home serve high quality Internet access networking equipment. Home networks are anonymously to many users both inside and a source of new subscribers and of outside the account residence through a tree additional product revenue. of wired and wireless links. Those users who are outside the primary residence are In Comcast's recently acquired AT&T engaging in theft-of-service according to the Broadband networks, they promote the sale Terms of Use clauses in most MSO service of multiple IP addresses (up to 4) for extra agreements. cost as a solution to the problem of using multiple computers on a single modem. The Terms of Use agreement doesn’t appear to THE RAPID GROWTH OF HOME forbid the use of NAT gateways. Computers NETWORKS BEHIND THE MODEM that aren’t “directly connected to the modem” are not supported. However, one The home gateways provide security, of their current FAQs provides a anonymity, and the support of multiple configuration for a LinkSys NAT gateway. workstations through many-to-one address translation, NAT. Gateways Pass the Knee of the Adoption Curve Early MSO Rejection Became Acceptance The rapid rise in the rate of installation Early MSO service agreements included of home gateways has followed the drop in wording that would preclude the use of prices ($500 to $50 in 5 years) and the multiple computers through a gateway. improved simplicity afforded by a synergistic and automatic configuration of
the gateway-HSD and the PC-gateway compound annual growth rate (CAGR) interfaces. to $10.3 billion in 2006.3
WiFi Revenue through 2006 Microsoft is Focusing on Home Networks (in billions)
Microsoft now adds protocols to their 12 operating systems and to their hardware 10 gateways to expand the services that can 8 pass through the address translations. 6
4 Microsoft intends to incorporate everything from the refrigerator to multi- 2 0 media components into the home network 2002 2003 2004 2005 2006 behind such gateways. This would indicate a continued, increasing population of home Figure 3: Estimated growth in revenue in networking installations. WiFi market through 2006
WiFi Increases Home Networking Use AN ENVIRONMENT INVITING Additional demand for gateways has REDISTRIBUTION been fueled by the similar reduction in price and greater simplicity of wireless LAN Redistribution is Simple to Wire up and with components, mostly the standard called WiFi, Easier than with Analog TV WiFi. Home gateways, WiFi components, Additional home computers that had hubs, switches, cables, and PCs are very been beyond easy reach of wired Ethernet simple to configure in a home network to are easily and inexpensively integrated with give all PCs access to the Internet. In fact, the home network with these wireless LAN for the most part, the components configure components. themselves. An infrastructure required for redistribution within a multi-tenant building, WiFi Is Encountering a Very High Growth for example, can be constructed entirely Rate with parts and technical assistance from a Radio Shack® store. • Shipments of WLAN products increased more than 100 percent in 2002, and will Redistribution Can Support Many Users continue growing as the stellar wireless market performer for the next several The high inherent bandwidth of the HSD years. Internet connection and the bursty data • Average equipment prices overall demands of the typical browsing or e-mail dropped by almost 28 percent, while user allow these environments to redistribute revenues increased by more than 50 high quality Internet access to dozens of percent in 2002 to $2.6 billion2 simultaneous users. • The WLAN equipment market will continue growing at a 43 percent
Wireless LANs Offer More Opportunity subjects the account holder to no additional billing if not caught. WiFi access points can be connected to home gateways entirely within the MSO are expected to offer tiered broadband service agreement. These access services based on consumption and that may points, when connected to a home gateway, reduce the illegal practice both because of can provide anonymous and uncontrolled the potential for extra charges to the tiered access to the account’s Internet connection services account holder as well as the over a wide area as large as hundreds of feet conversion of some participants to being around it. subscribers of the more attractive, less expensive, low consumption service. The gateway user’s assumed network anonymity combined with a widespread “the However, collecting additional revenues Internet is free” attitude adds encouragement from high-bandwidth users accustomed to to those who are inclined toward service “free” access could be problematic unless redistribution or an opportunistic wireless addressed before tiered services are tap. Users of WiFi have easy access to introduced. programs such as “NetStumbler” and “MiniStumbler” that provide the user with a graphic display of all the locally accessible, DETECTING REDISTRIBUTION unencrypted WiFi networks and the necessary “SSID” to sign on (SSID: Service Redistribution is, indeed, a problem Set Identifier, a 32-character unique since each instance represents one or more identifier attached to the header of packets potential subscriptions that are lost. sent over a WiFi network). NetStumbler is used with PCs and MiniStumbler is used It does not appear that there is a method with pocket computers. These programs are to attack the problem by prevention either in used by ordinary hobbyist computer users configuration or warnings in the terms of and are not restricted to hackers. Open WiFi service. Therefore detection and response to networks in populated areas don’t remain a detection represent the remaining option. secret for long. Detection with Network Equipment The Billing Method Doesn’t Discourage Redistribution Detection of likely incidents of service theft should be much easier than with analog The billing method for most HSD cable TV theft since the multiple users Internet access accounts is a fixed rate for a behind the gateway create a protocol stream fixed bandwidth high-speed connection. that has a variety of signatures that indicate There is no limit on consumption and, the number of individual users. therefore, no extra cost to the account holder for extra use. This method is unlike that for Although, as shown in the Bellovin water and electricity, for example, but is AT&T Labs paper, identification of similar to the downstream portion of an individual accounts with multiple users is analog TV cable plant and can be expected imprecise with standard network devices to similarly encourage theft. Redistribution that observe at L3, specialized devices could be used to identify such accounts when
inspecting other layers. For example, data • Blatant advertising of wireless “WiFi passing through a device similar to a hotspots” at the location of a residential firewall could identify accounts exhibiting broadband subscriber many simultaneous connections or the use of • Mass service cancellations at a multi- many versions of browsers, or the checking tenant location. of many different e-mail accounts. • Alert service representatives who spots Although it is possible within the character trends of returned modems in a of the protocol, equipment to perform such concentrated area. checks has not been brought into such • A serendipitous tipster service. REAL SERVICE THEFT, VAGUE Detection of likely redistribution INDICATIONS, INVISIBLE LINKS situations is not necessarily an indication of a definite redistribution event. It is only a The Vulnerable Revenue is Real violation if users outside the account residence are accessing the Internet through “Comcast has found that the Internet it. business has become even more profitable than providing basic cable service. The Drive-by Detection of Open WiFi Points capital costs are lower: a cable modem costs $50 compared with a television set-top box The default installation of most WiFi at $225. And churn - the rate at which components establishes the network without customers cancel their service - is far lower encryption. These networks can often be for Internet service than for video.”4 The detected with simple PC equipment and HSD service and subscriber base is a major freely available software such as asset. Vulnerabilities can chisel away at the NetStumbler. When detected in this way, revenue and do it well before the loss is the drive-by inspection PC can also detect if measurable with current tools. the network has an Internet connection through the HSD network of the MSO. The Service Theft Problem is Real
Interestingly, unsecured WiFi equipment The NCTA “2000 Report on Cable is not in violation of the Terms of Use Industry Lost Revenue” indicates theft of agreement in force with the subscriber and, video services is 10% throughout all service therefore, is not an indication of a definite levels from basic to premium.5 These event of redistribution. It is only a violation figures occur despite the fact that it is illegal if users access the Internet through the to possess cable descramblers in 32 states. network from points outside the account residence. HSD networks face corner-store availability of the complete set of Other External Indicators equipment, legal to sell and own, required to implement an extensive wired or wireless re- There are also other external flags to distribution system. indicate potential redistribution activity such as:
The Indications are Frustratingly Vague
The network indicators of service re- distribution, although potentially more varied than that for cable TV, can be difficult to extract and would often flag a benign situation. Figure 4: Sample bulletin
Ambiguous evidence for potential Immediate Messaging to Every PC User Is service redistribution requires a more Not Available with Traditional Means controlled response than the traditional service disconnection and personal Up to now, there has been no reliable confrontation. mechanism to reach anyone on a PC with or without a gateway. Most of a provider’s Invisible Links are Difficult to Spot subscribers don’t have a known e-mail address and of those who use the provider’s On-site inspection, the major tool of e-mail address, few of them check it cable TV service theft control, is clearly regularly. A recent Forrester report ineffective in the case of HSD propagation determined that over one-third of all e-mail through wireless links. users change their address over the course of a single year. The variety of incompatible instant messaging systems is not useful, and REAL TIME MESSAGING TO ALL not just because of the incompatibility. SCREENS BEHIND A TARGET MODEM Many providers are opposed to
maintaining a software component on This paper describes the utility of an in- subscriber PCs. Software components band communication channel that can target rapidly degenerate to nonfunctionality. all workstations on suspected accounts with And, a sizable portion of subscribers will unblockable screen alerts. It can utilize blame any malfunction of their computer on these invisible links to reach all hidden the provider-installed software. workstations. It is an effective new tool for this new problem. Other Communications Channels are
Unreliable and Untimely Unlicensed Users Will Know They Have
Been Discovered Telephone calls go unanswered, door
knocks are expensive and have many other Informational, personalized warnings are problems, and bill stuffers are mostly effective in dissuading service theft without thrown away, unread. the unpleasant, confrontational contact that
normally accompanies service theft Creating a Mechanism to Allow the MSO to situations. By using the providers’ own Use their Own Channel to Communicate channel, it becomes a way to communicate with Subscribers with both the unlicensed users as well as
inadvertent opportunists who grab service The normal Internet Protocol does not through the anonymous Ether of WiFi. provide for any communications to a
browsing subscriber except for the pages
that are requested from the destination site • Security controls to guarantee access or from pages, in turn, linked from the functions from only authorized destination site. personnel and locations • Delivery to individual targeted users or The constraints of the carrier-grade described groups with a characteristic requirement of the MSO limits the determined by a database or subnet opportunities for a solution and defines the • Proof-of-delivery receipt requirements: • Delivery of fully interactive screens in order to provide abuse warnings and, in • Absolutely no software or configuration the case of virus contamination alerts, a change required at the subscriber PC screen that also includes the remedy • Operational on any workstation, PC, Macintosh, Linux, or other device with a The facility to accomplish this type of browser even through wireless links communication that meets the above • Transparent to DOCSIS versions requirements was developed in 2002 and • Operational in the presence of any deployed testing has begun by Perftech gateway or firewall or pop-up blocking Bulletin Services and the MSO software WideOpenWest. • Installation in the provider network while the network is operating The installation consists of router- • Failsafe operation in that any failure in attached devices called Bulletin Directors the delivery system does not affect the that are distributed to aggregation routers normal operation of the network • Non-disruptive to the network operation and throughput
There will be A brief outage tonight at midnight NOC Backbone Hub/Home
Provider Bulletin Manager There will be a brief out- CMTS age tonight at midnight H F C
Subscriber Bulletin Director 100 Middleware Servers and Back Aggregation Office systems Router
Internet
Figure 5: Sample network configuration
and a management device that administers small when compared to the cost of on-site the policies for bulletin delivery. visits and personal confrontation with the account holder. Installation takes place while the network is up and running using a fail-safe A Broadly-Applicable Support Tool layer 4 redirection only of upstream port-80 traffic. None of the downstream traffic or Unlike cable TV or telephone service other upstream traffic passes through the where the subscriber's premises device is a director devices assuring both performance fixed, passive entity, HSD Internet and privacy. There is no measurable impact broadband access is a complex technical on network throughput. partnership between the provider’s network As expected in the initial deployment, and the subscriber's computer and installation was made into an operating configuration — all of which is required to network and the network performance was sustain Internet accessing activities. unaffected. Test bulletins are successfully Immediately available provider-subscriber delivered to the targeted accounts. communication is clearly a requirement for proper support of such a system. How It Will Be Used For those subscribers who do not know The communication can be used to warn that their personal WiFi network is being and stop likely abusers by dispelling the abused and hacked and, therefore, insecure, illusion that “they don’t know I’m here.” the revelation is welcome. Every unlicensed screen will display a selected alert whether inadvertently or Advance-warnings of system outages intentionally involved in the theft of service. would be more than welcome by users who Warnings can become sufficiently insistent have incorporated the Internet access into that surfing becomes difficult. critical work-at-home activities, stock trading, or auction participation. It would The result can be expected to turn many also be welcome at the provider’s support unlicensed users into paying subscribers. call center where the call floods that An analysis by Time Warner Cable as accompany such outages would be presented in "Cable And Broadband significantly diminished. Security Case Study: The Broadband & Internet Security Task Force"6 demonstrated Alerts should be issued about temporary that nearly 1/3 of all discovered unlicensed outages in subsystems such as e-mail cable TV connections were turned into servers. Informed users are not likely to be paying subscribers. Unauthorized re- participants in call floods and are likely to distributors can also be turned into be less agitated and more understandable commercial accounts when they need to when kept informed. continue the re-distribution. Instances of disruptive network operation An Economical Solution due to virus contamination in a subscriber’s PC can demonstrate the value of immediate, The cost of this solution is a small. interactive communication to handle this Amortized over several years, the cost per difficult support problem. In this case, the subscriber is in cents per year. That is very subscriber can be delivered an alert
immediately upon being seated at the PC REFERENCES and browsing. The alert would contain an explanation along with a button that would be a remedial link to decontamination 1 Steven M. Bellovin, “A Technique for facilities. Counting NATted Hosts,” from AT&T Labs CONCLUSION Research, Nov. 2002. 2 Forward Concepts, Feb. 2003 Theft of service is going to get worse. 3 InStat/MDR, Feb. 2003 Support problems are going to get more 4 Saul Hansell, “In Broadband, Comcast Lets severe and complex. The subscriber Users Find Their Own Flourishes, ” New population is going to keep growing at a York Times, March 17, 2003 very high level. The provider’s own 5 “2000 Report on Cable Industry Lost communications channel is the key tool in Revenue,” NCTA, March 20, 2003 the control of all these issues. 6 "Cable And Broadband Security Case Study: The Broadband & Internet Security Task Force, Tap Audit," Time Warner CONTACT INFORMATION Cable, Syracuse, NY Division
Jonathan Schmidt Vice-President Business Development 210-349-7152 [email protected]
TIERED DATA SERVICES TO ENHANCE CUSTOMER VALUE AND REVENUE: WHY ARE DOCSIS 1.1 AND ADVANCED QUEUING AND SCHEDULING REQUIRED?
Gerry White Senior Director of Advanced Technology Network Infrastructure Solutions Motorola Broadband Communications Sector
Abstract INTRODUCTION
High-speed data services are making Initial DOCSIS deployments have significant revenue contributions to generally followed a “one size fits all” model. Broadband Operators as service penetration To date the success of this approach in terms rates are now averaging over 10% and of service penetration and revenue generation exceeding 30% in some markets. Still, there has been considerable such that cable modems are large potential customer segments that have become the preferred means of Internet can be captured by offering data service tiers access for a significant section of the that provide a better match between population. customers needs and income levels and the In order for Multiple System Operators features of the high-speed data service (MSOs) to attract a wider customer base and required. For example, business customers reap additional rewards from their investments are willing to pay much higher rates for a in Hybrid Fiber Coax (HFC) and IP service with greater bandwidth and service infrastructure it will be necessary to go level guarantees. At the other end of the beyond this initial audience and provide more spectrum, customers who are paying $25-30 tailored services to specific target populations. per month for dial-up services would jump at Thus high-end service offerings can be created the chance to have an “always on” for business customers offering guaranteed connection at speeds that are double standard services at bandwidth (and price points) dial-up rates. comparable to T1 and business DSL offerings. This paper will discuss how measurable data At the other end of the spectrum there are service tiers can be delivered to customers many dial up customers for whom current through Quality of Service (QoS) control cable modem service is too expensive. enabled by the use of the DOCSIS 1.1 Providing a rate limited entry-level service for specifications and advanced queuing, these users can add to the customer base with scheduling and congestion control techniques. an immediate revenue impact and also create a pool of broadband addicts, which can be Additionally the paper will present a targeted for later upgrade to premium mechanism to support overbooking of network services. resources in a tiered environment, enaling cost effective deployment while ensuring that Ultimately all networks are a shared guaranteed service levels are met for each of resource and rely on statistical multiplexing the service tiers. between users to provide cost effective service. In the case of the HFC network the shared resource extends to the customer premise, while for a DSL network the customer link may be dedicated copper pair. have been the subject of research and In both cases the infrastructure network is standardization efforts resulting in the shared and statistical multiplexing is used to differentiated services (DiffServ) model for reduce costs. Thus to ensure fairness between aggregated QoS and the integrated services users (even for a best effort single service tier (IntServ) model for per flow QoS [References network) QoS must be provided to some 1 and 2]. degree, although it may be relatively At the edge of the network bandwidth is simplistic. When tiered service offerings are typically scarce and expensive to provide. enabled QoS becomes more complex as it Devices in this domain, such as a CMTS, deal must not only provide fairness between users with a bounded number of traffic flows (e.g. within a service tier but also differentiate to a maximum of 8000 flows in each direction between tiers. in DOCSIS). In this region QoS can be The MSO community is in the fortunate provided most effectively at a per flow level. position of having the tools to provide tiered For each flow a traffic specification and a services readily available. In fact in many flow specification are defined. The traffic cases they are already deployed, waiting to be specification defines a classifier to identify the enabled. packets that belong to a specific flow. DOCSIS 1.1 provides the mechanisms to Typically this is a masked set of fields based deliver tiered services over the HFC network on the content of the packet header such as IP and when this is combined with suitable addresses, port numbers, DSCP markers, etc.. backbone technologies the MSO can deliver The flow specification defines the QoS end to end service tiers which compete parameters to be applied to the flow effectively with any in the market. (bandwidth, latency…). This mechanism was defined by the IETF as the IntServ architecture and was adapted to provide the TRAFFIC FLOWS basis for DOCSIS 1.1 QoS [Reference 3]. At the customer level the focus of interest DOCIS 1.1 provides for a signaling is on individual applications, the user wants to mechanism to set up new flows, for admission make a phone call, connect to a corporate control functions and for isolation between network or simply surf the web. In general flows. The CMTS has the primary each application will involve multiple traffic responsibility to provide QoS in the DOCSIS flows. For example a VoIP client will realm by implementing admission control and exchange control packets with a call providing isolation between flows based on management system and voice traffic with the the upstream and downstream scheduling far end called system. These traffic flows may mechanisms. take different paths through the network and In the core of the network bandwidth is each will require QoS to be provided. The relatively abundant (and cheaper) but the concept of traffic flows and of providing QoS infrastructure is shared between tens of to each flow is central to both DOCSIS and thousands of clients. Thus core switches and backbone QoS mechanisms. routers must support hundreds of thousands of flows. If these devices were to operate at an individual flow level the amount of data to be AGGREGATE VS PER FLOW QoS maintained would be massive and systems Two distinct mechanisms are typically would simply not scale so that an aggregated used to provide QoS on an end-to-end basis mechanism is needed. To achieve the scaling over large IP networks. These mechanisms required packets entering the DiffServ domain are classified into one of a limited number of mapped to a service identifier (SID) and in the behavior aggregates (64 max.). DiffServ downstream direction to a service flow defines a field in the IP header of the packet identifier (SFID). The cable modem (CM) known as the DiffServ code point (DSCP). and CMTS cooperate to assign the required Systems at the edge of the DiffServ domain QoS to each flow and to ensure that it is met. mark packets with the code point desired The MSO metro networks that connect the before transmission into the domain. All CMTS systems are typically based on gigabit packets with the same DSCP to be transmitted Ethernet, Sonet or RPR physical on a given link are considered part of the infrastructure. Historically an IP routing behavior aggregate and are to be treated in the infrastructure ran on top of this possibly same manner. The DSCP defines the required providing QoS based on the DiffServ model. behavior for each packet so that no per flow Newer networks replace the pure IP routing state must be maintained. Routers within the model with one based on MPLS DiffServ domain use the DSCP to determine infrastructure. Both types of network provide the QoS desired by the packet and apply the the MSO with the capability to provide appropriate queuing and scheduling aggregated QoS based on traffic engineering algorithms to achieve the required QoS. and provisioning. In either case the Multi Protocol Label Switching (MPLS) CMTS/ER provides the transition point [Reference 3] has recently emerged as another between the QoS domains. The major issues, mechanism to provide aggregated QoS in which must be resolved at this demarcation metropolitan and core networks. It defines a point, will be considered later in this paper. mechanism to set up label switched paths (LSPs) between endpoints at the edges of the QoS MECHANISMS MPLS network. Packets entering the network are assigned to a particular path and a label The ability to deliver QoS involves four added to the packet identifying the LSP to be key functions: used. This label is used by the MPLS routers • Classification of packets to in the core of the MPLS network to forward determine which flow a packet is the packet to its destination endpoint. The part of and the appropriate service core network does not need to examine the level for each traffic flow packet headers beyond the label and can • Policing of traffic to prevent flows therefore focus on switching traffic to its from getting higher than agreed destination as quickly and efficiently as upon service levels possible. Each LSP can be associated with a defined forwarding equivalency class (FEC) • Buffering to ensure that queues are which defines specific QoS parameters so that created to contain packets during the MPLS network can provide a mesh of periods of congestion QoS enabled paths. • Scheduling to enforce packet In a typical MSO network environment the transmission in accordance with transition point between per flow and QoS policy. aggregated QoS domains occurs at the In order to provide QoS successfully the intersection of the HFC/DOCSIS and metro/IP CMTS/ER must provide this functionality for networks. The HFC access network is based both HFC and metropolitan network on the DOCSIS 1.1 protocol, which provides environments. QoS to applications on a per flow basis. In the upstream direction an application flow is QoS IN THE HFC NETWORK 2. The CMTS schedules upstream The DOCSIS 1.1 specifications provide transmission for the SID based on its QoS parameter set and the traffic history QoS for the cable access network. They define of the SID. This is the most complex enhancements to the Media Access Control and the critical step for providing (MAC) protocol of DOCSIS 1.0 to enable upstream QoS. more sophisticated access methods over HFC access networks by adding the following: 3. The CM transmits the packet to the CMTS • Packets are classified into service flows based on their content. Thus each 4. The CMTS receives the packet and application can be mapped to a unique reclassifies it based on CMTS service flow. configuration. In general the CMTS is • the first trusted device in the network Network access (upstream and and should not rely entirely on user or downstream) is scheduled per service CM packet classifications. flow using one of a number of defined scheduling mechanisms including 5. The CMTS maps the packet into the QoS constant bit rate, real-time polling, non scheme for the MAN (mark traffic for real-time polling and best effort. differentiated services forwarding, map traffic to MPLS LSP tunnels, map traffic • Service flows may be configured to physical interface) through management applications or created and deleted dynamically in 6. The CMTS queues the packet to the response to the starting and stopping of egress link based on the required QoS applications. 7. The CMTS implements its network • Fragmentation of large packets is interface scheduling algorithms and required to allow low latency services transmits the packet to the link when it to operate on lower-bandwidth reaches the head of queue. upstream channels. Packets received by the CMTS from the These features provide the basic tools for network are treated as follows and shown in QoS management. They allow applications to Figure 2 request QoS changes dynamically and allow The CMTS receives the packet and providers to isolate multiple data streams from classifies it into a downstream service flow each cable modem, set-top box or MTA. identified by a unique SFID. DOCSIS 1.1-based systems can therefore potentially deliver the ability to allow 1. The CMTS enforces policing on application-specific QoS treatment within the maximum rate if required. HFC access network for each traffic flow. 2. The CMTS queues the packet to the Packets transmitted into the network from egress link based on the required QoS a host system are treated as follows. The 3. The CMTS implements its downstream interaction between the CM and the CMTS scheduling algorithms and transmits the upstream scheduler is shown in Figure 1: packet to the link when it reaches the 1. The CM filters each packet and classifies head of queue. it into a service flow identified by a unique SID. 4. The CM forwards the packet to the host. QoS IN THE MAN/WAN mechanisms employed within the CMTS/ER There are two primary methods for must maintain the QoS during this transition. providing QoS control in the regional The problem is complex due to QoS network, packet based and connection-based. requirements constantly changing as service flows are created and deleted dynamically. In the packet based case, individual flows are policed and marked at the edge of the The metro to HFC boundary is also a network using a DiffServ DSCP marker in the capacity transition point. In the downstream IP header, so that aggregated flows are direction packets received from gigabit speed delivered to the network core with each flow optical links must be transmitted onto megabit tagged for the appropriate QoS treatment. The capacity DOCSIS networks. Thus the DiffServ standard defines the code points to CMTS/ER must implement congestion use and the per-hop forwarding behavior to be management based on the conformance of the applied to each marked packet. subscribers and applications to their Service Level Agreements (SLA’s). Connection-based QoS can be implemented using MPLS. In an MPLS The queuing and scheduling mechanisms network a number of paths are established used within the CMTS/ER to implement the between the end points of the network. Each transition between the per-flow HFC and path can be traffic engineered to provide a aggregated metro domains will determine how defined level of QoS. All packets on the path successfully QoS can be delivered. The key share the same forwarding equivalency class concepts required to provide this transition (FEC) consisting of the MPLS end point and successfully are: the QoS parameter set. As with the DiffServ • Queuing and Scheduling case the packets from the individual flows are • Congestion Control policed and marked at the edge of the network. In this case the marker is an MPLS label that is prepended to the packet and QUEUEING and SCHEDULING identifies the path through which the packet Three queuing and scheduling will traverse the MPLS network. Each path is mechanisms will be considered; FIFO, class referred to as a label switched path or LSP. A based and per flow. single LSP, with a defined FEC can support multiple flows and thus provide the necessary aggregation mechanism. FIFO Queuing First-In-First-Out (FIFO) queuing is both a queuing and a scheduling mechanism. In FIFO QoS AT THE BOUNDARY queuing, all packets are stored in a single In order for applications to see real queue and are transmitted in the order that benefits, QoS must be provided on an end-to- they are received. FIFO queuing is easy to end basis. Thus the QoS-enabled traffic flows implement and it requires little configuration. from the HFC access network must be Unfortunately it does not provide any support mapped to the QoS mechanism(s) used in the for the differing QoS levels required by regional or backbone networks. The diverse applications. In a FIFO scheme CMTS/ER at the boundary must be able to packets for a low-latency service can be perform this mapping and implement the QoS queued behind those from high bandwidth mechanisms for the two domains. To deliver services and must wait for these to be QoS at a per service flow level this must take transmitted. place for multiple flows at wire speed. The Class-Based Queuing flow is served at the rate defined in the service Class-based queuing (CBQ) attempts to level agreement. avoid this problem by sorting the traffic into To assign flows without reservations to a different classes by examining the packet and queue a method known as stochastic queuing trying to determine the type of traffic to which is used. In stochastic queuing the parts of the it belongs. Once the packet has been packet header that are the same for all packets classified, it is placed in a FIFO queue that of a flow, such as the source and destination contains only other packets of the same type. IP addresses and source and destination port Each per class FIFO queue can be serviced numbers are fed to a hash function that is used according to configured policy to provide the to map the packet to a queue. This ensures behavior required. Thus a FIFO that is that all packets for the same flow are mapped serviced frequently so that it is usually empty to the same queue (to avoid miss ordering). It of packets could provide a low latency, low also eliminates the need to configure loss service such as VoIP. Similarly a FIFO bandwidth shares per-class, and consequently that is kept full can provide a service such as avoids the miss-allocation caused by varying bulk file transfer, which requires high usage patterns. If the system can support bandwidth but can tolerate moderate packet more queues than there are flows then most loss. flows either have their own queue or share it In theory, this allows the FIFO of each with a small number of other flows. To class to provide the desired type of service, support stochastic per-flow queuing the but in practice there are a number of problems CMTS/ER must support thousands of flows with this approach. It requires a constant, (with DOCSIS 1.1 a CMTS can support 8000 heavy configuration burden because the flows in each direction per DOCSIS domain). operator has to configure the allocation of service to the different classes e.g. 1/10 for e- CONGESTION CONTROL mail, 1/10 for voice, 1/3 for Web traffic, etc. Given that the data rate in the MAN will In a dynamic network environment, the class- be significantly greater than the capacity of an based queuing method of allocating service is HFC link it is also important to consider impractical because the allocation is congestion control mechanisms, especially as independent of the number of users of a given applied to the downstream traffic. Three class. CBQ does not provide application mechanisms to handle congestion will be isolation, as although each queue contains discussed; tail drop, Random Early Detection traffic from a similar application type (RED) and Longest Queue Pushout (LQP). (e.g.VoIP), flows from multiple users are mixed within the queue. Tail Drop and RED Per-Flow Queuing (Refer to Figure 3) Per-Flow Queuing (PFQ) solves the With all packets sharing the single queue problem of providing isolation between for simple FIFO or all packets of the same application flows by assigning each packet type sharing the same queue for CBQ there stream to its own queue. Those queues for are limited options for congestion control. flows with QoS reservations are served at The simplest scheme is simply to drop packets their guaranteed rate while flows without from the tail of the queue when no buffers are reservations are served in a round robin or available for them. This takes no account of fair-share manner. Thus the queue for each the content of the packet and hence it’s potential value or of whether the flow to by definition the longest queue is that which is which it belongs is in compliance with its exceeding its allocation by the greatest service level agreement. Mechanisms such as amount so that traffic is automatically RED attempt to solve this problem by discarded from those applications which are randomly dropping some of the arriving non-compliant with their SLAs. This occurs packets when the queue starts to fill. The without the need to configure congestion intent is that the end systems using a control. windowed flow control, such as TCP will Further details on PFQ performance can notice the packet loss and slow down. Thus it be found in [Reference 5]. is dependent on the end systems of the flow to slow down transmissions to solve the congestion problem. While this might work OVERBOOKING TIERED SERVICES in a well-controlled environment such as an In order to create a commercially viable enterprise network it is unlikely that the end network the operator must rely on statistical systems will be as cooperative in a public multiplexing. Not all users are active at the network. FIFO queuing does not provide a same time so that the network can be over mechanism to ensure isolation for well- subscribed to reduce the cost per user. In a behaved flows from miss-behaved single tier best effort network overbooking is applications generating heavy loads. relatively simplistic. Subscribers are added to As the number of flows sharing the FIFO the network until a local heuristic, such as the queue increases this problem becomes worse. number of cable modems per upstream or a The limitations of class-based queuing and defined traffic load, is reached. In a tiered RED often result in the inappropriate network overbooking becomes a more discarding of packets. Since traffic flows are complex process and should be applied at lined up in shared queues, it is impossible to each service tier as well as on each network isolate and discard those flows that are interface. The amount of over booking is exceeding service level agreement (SLA) dependent on the types of services to be guarantees before discarding traffic flows that offered (e.g. guaranteed vs. best effort) and on are staying within their SLAs. Packets are the traffic patterns of the users within each therefore discarded randomly from shared service tier. queues as they become full. Service Classes Longest Queue Pushout Responsibility for providing QoS to a (Refer to Figure 4) service flow resides with the CMTS, which must be configured with the parameters In a per-flow queuing environment longest necessary to control this operation. In order to queue push out (LQP) is the mechanism that simplify this configuration and help best meets the congestion control operations staff retain their sanity the concept requirements. LQP allocates buffers to the of service classes has been introduced. individual flows as required until 100% of the DOCSIS 1.1 has defined a set of QoS buffer pool is used. When no buffers remain parameters, including maximum sustained and and a new packet is received, LQP discards minimum reserved traffic rates, and a way for traffic from the flows, which are the longest associating specific QoS parameter values to queues. The scheduling system is transmitting service flows. It has further incorporated the from these per-flow queues at a rate that concept of a service class name so that service matches the QoS assigned for each flow. Thus flows, when being created, may be assigned their QoS parameters by referencing a service their guaranteed minimum reserved rates are class name. in excess of the configured MAB for the Thus an operator may define a number of service class. To control the amount of service classes. Individual service flows will overbooking, a configurable overbooking be assigned to a service class and all flows factor the Configured Active Percent (CAP) is belonging to that class provided with a provided. The CAP is an estimate of how defined quality of service. Service classes can many service flows, expressed as a be supported for both downstream and percentage, are likely to be active upstream directions. simultaneously. For example, if the CAP for a service class is set to 20 percent then it is In order to facilitate overbooking the estimated that only 20 percent of the service concept of the service class can be extended flows belonging to that class will be active by introducing two additional service class simultaneously. Therefore, 5x (1 / 0.2) parameters, Maximum Assigned Bandwidth overbooking would be allowed. A CAP of 100 (MAB) and Configured Active Percent percent means that no overbooking will be (CAP). With these additional parameters the allowed. A CAP of zero percent means that operator can explicitly control overbooking unlimited overbooking is allowed. for each service tier and for the network interface in total. With this scheme all service flows must be assigned to a service class. If a ADMISSION CONTROL service class name is not present in a Registration Request message generated by a Admission control is a process wherein CM, then the CM service flows are assigned the bandwidth requirements of a service flow to a default service class. are checked to verify that admission of the service flow to a service class does not exceed
the class’ MAB after accounting for the Maximum Assigned Bandwidth allowed level of overbooking. Service flows Maximum Assigned Bandwidth (MAB) are created during modem registration or specifies the amount of bandwidth a service through dynamic service messaging. A CM class is permitted to consume on an interface. registering with primary service flows should It is expressed as a percent of the total be permitted to register regardless of whether interface bandwidth capacity. The MAB of a the admission of its service flows would service class is applied during admission exceed its service class’ MAB. In this case control to determine whether to admit a new however the service flow would be admitted service flow and again by the scheduling in a ‘Restricted’ state meaning that the service algorithms to provide a class-based weighting flow will not be provided any guaranteed to the scheduler. Any unused portion of a minimum reserved rate. Service flows created class’ bandwidth may be used ‘on demand’ by via dynamic service messaging will be other classes which have a traffic load in rejected if admission of the service flow excess of their own MAB. would cause its service class to exceed its MAB. Configured Active Percent Examples of MAB and CAP operation can be seen in Figure 5 and Figure 6. In both Since not all service flows are active examples the bandwidth on the interface is simultaneously the service level classes shared between two service tiers a best effort feature permits customers to overbook service class and an enhanced service class such as a classes. Overbooking means admitting service flows to a service class such that the sum of business service tier. In the first example SUMMARY three flows are active in the enhanced class Providing tiered service offerings has the with no flows active in the best effort class. potential to extend the target audience and revenue stream for high-speed data services. Initially the total bandwidth is below the Extending existing best effort DOCSIS to MAB for the class so that all flows receive support service tiers with QoS guarantees their guaranteed bandwidth. At this point the requires MAB is reached so that requests for additional flows to be set up would be rejected by • The use of the DOCSIS 1.1 protocol admission control (or could receive best effort extensions bandwidth as defined by operator policy). As • Sophisticated scheduling and flow 3 bursts to a higher data rate, above the congestion control mechanisms in the guarantee, it will share available bandwidth CMTS. from the best effort service class if this is • A means of overbooking, which is available. service tier aware. The second example shows the same 3 All of these features are available in CM and flows active with all MAB consumed so that CMTS systems, including much of the no further flows could be admitted. Flow 3 currently installed equipment base. then terminates at which time bandwidth consumption falls below the MAB for the class and additional flows could be added.
Classifier
Active SIDs
Classifier
(Inactive flows) Active SIDs
... SID1 SID2 SID3
Max rate token bucket policing Class Table
Default Max rate Traffic rate token buckets High Priority Max rate Resvd rate DOCSIS upstream scheduler Schedule upstream transmission opportunities for each SID in DOCSIS MAP messages to CMs
Packets forwarded to network interface scheduler for transmission
Figure 1 Interaction between CM and CMTS upstream scheduler
Classifier
(Inactive flows) Active flows
... Flow1 Flow2 Flow3
Max rate token bucket policing Class Table
Default
max. rate Traffic rate token high priority buckets max. rate resvd rate Down stream scheduler
Packets re-ordered for transmission
Figure 2 Downstream Packet Scheduling
Output Forwarder/ Queues Classifier Queues Scheduler
Incoming Delivering Packets Packets
Entire Class is penalized not just Allocated buffer space per class Misbehaving (dark) Flow
Figure 3 Class Based Queuing with RED
Queues Forwarder / Scheduler Classifier
Incoming Delivering Packets Packets Misbehaving Flow can be Selectively Penalized Allocated buffer space per flow
Figure 4 Per Flow Queuing and Scheduling with Longest Queue Push out
Interface
Best effort class
MAB for enhanced class Flow 3
Enhanced class Flow 2
Flow 1
Contends as best effort
Figure 5 MAB Operation -flow 3 exceeds reserved rate
Interface
MaxAssigned Flow 3
Flow 2
Flow 1
Flow 3 ends
Figure 6 MAB Operation -reserved rate flow 3 ends REFERENCES About the Author 1. S. Blake, D. Black, M. Carlson, E. Gerry White is Senior Director of Davies, Z. Wang, W. Weiss. An Advanced Technology within the Network Architecture for Differentiated Infrastructure Solutions Group of the Services (RFC 2475). Motorola Broadband Communications Sector, and he was previously CTO of 2. R. Braden, D. Clark, S. Shenker RiverDelta Networks, Arris Interactive, and Integrated Services in the Internet LanCity. He has presented papers and Architecture: An Overview spoken at dozens of cable industry events (RFC1633). worldwide and has published many trade 3. DOCSIS 1.1 Radio Frequency press articles on cable industry issues. White Interface Specification. is the co-author of several patents and 4. E. Rosen, A. Viswanathan, R. Callon. articles on data communications technology, Multiprotocol Label Switching and he holds a BSc. (Honors) from Architecture (RFC 3031) University College in London. 5. J.C.R. Bennett and H. Zhang Hierarchical Packet Fair Queuing Author Contact Information Algorithms Gerry White Senior Director of Advanced Technology Network Infrastructure Solutions Group Motorola Broadband Communications Sector ### 3 Highwood Drive Tewksbury, MA 01876 (978) 858-2300 [email protected] ISBN 0‐940272‐01‐6; 0‐940272‐08‐3; 0‐940272‐10‐5; 0‐940272‐11‐3; 0‐940272‐12‐1; 0‐940272‐14‐8; 0‐ 940272‐15‐6; 0‐940272‐16‐4; 0‐940272‐18‐0; 0‐940272‐19‐9; 0‐940272‐20‐2; 0‐940272‐21‐0; 0‐940272‐ 22‐9; 0‐940272‐23‐7; 0‐940272‐24‐5; 0‐940272‐25‐3; 0‐940272‐26‐1; 0‐940272‐27‐X; 0‐940272‐28‐8; 0‐ 940272‐29‐6; 0‐940272‐32‐6; 0‐940272‐33‐4; 0‐940272‐34‐2; 0‐940272‐35‐0; 0‐940272‐36‐9; 0‐940272‐ 37‐7; 0‐940272‐38‐5; 0‐940272‐39‐3; 0‐940272‐40‐7; 0‐940272‐41‐5; 0‐940272‐42‐3; 0‐940272‐43‐1; 0‐ 940272‐44‐X; 0‐940272‐45‐8; 0‐940272‐46‐6; 0‐940272‐47‐4; 0‐940272‐48‐2; 0‐940272‐49‐0; 0‐940272‐ 50‐4; 0‐940272‐51‐2; 0‐940272‐52‐0; 0‐940272‐53‐9; 0‐940272‐54‐7
© 2015 National Cable and Telecommunications Association. All Rights Reserved.