TRANSPORT of CBR TRAFFIC on IP NETWORKS Tom Carr, Bruce Roe Wave7 Optics

TRANSPORT OF CBR TRAFFIC ON IP NETWORKS Tom Carr, Bruce Roe Wave7 Optics

Abstract

The success of future distribution While we can envision our final networks will depend on their ability to destination, the question remains, how support legacy services including do we get there? Clearly the distribution committed bit rate traffic. Most of this network cannot, and will not, be traffic is transported by the PSTNs over replaced en masse with a new IP-based T1 facilities. architecture. Pockets of new high-speed IP networks will be deployed and, over a This paper describes a technology, very long period of time, finally consign Time Division Multiplexing over IP, the old copper telecommunications plant which is capable of providing T1 circuit to the pages of history. This slow emulation over IP networks. evolution means that some legacy services must be supported on the new INTRODUCTION architectures.

Industry visionaries foresee Next One likely architecture for future Generation Networks that offer hundreds distribution networks is that shown in of megabits of bandwidth to the Figure 1. An Ethernet network of one or consumer, extensively or exclusively more gigabits extends over fiber from a using IP as the network transport Head-End to an active bandwidth protocol. This is an easy vision to management element. From there, the believe in, and one that we are Ethernet is extended over multiple fibers convinced will evolve into reality. to serve a pocket of customers. By using

Home Residential Gateway u Video Settop

Computer(s) Telephone(s)

u Residential SOHO Active Core | Gateway To/From Passive Optical Fax Video Settop Headend 12 Computer(s) Telephone(s) Business Gateway Passive Optical 96 Business Router

Video Settops

Computers PBX Telephones

Figure 1: Next Generation IP-based Distribution Network

WDM, RF digital and analog video can billing; and still more sophisticated be distributed over the same fibers. characteristics, such as caller Plain old telephone service (POTS) identification, call forwarding, and would be supported by using voice over conference calls. There are literally IP (VoIP) for transport. In this manner thousands of such telephony features, voice, video and very high-speed data with dozens of national and local could be offered over a single efficient variations. Available VoIP integrated network. Unfortunately, this alone will circuits can handle some, but not all, of not provide for a significant and the PBX signaling in use in the U.S.. lucrative portion of traditional legacy Converting PBX voice circuits to VoIP telecommunication services, committed could require the end-user to give up bit rate services. some useful or much needed features.

Obituaries have been written for The one common element of committed bit rate traffic such as private committed bit rate traffic carried by line service, international frame relay PSTNs is that they are primarily networks, and every other non-IP transported via T1s. Having the ability protocol, but the double-digit to transport T1s over IP, regardless of compounded growth rate for these the data or signaling protocol, would be services, particularly in the international an ideal solution for supporting legacy markets, continues. Some forecasts1 of services in a new IP environment. Such international frame relay, for example, a technology does exist and is called predict a compounded annual growth TDMoIP, Time Division Multiplexing rate of 14-16% at least through 2004. over IP. TDMoIP is a technology that Even X.25 networks still exist and combines features from Time Division continue to grow. The inertia of Multiplexing and IP to deliver migrating these networks to IP will be synchronous T1 circuits transparently fueled by sluggish economies, and the over IP networks. An individual channel falling prices of both T1 service and old within the T1 stream is not changed in technology equipment. Any new any way, nor is there any signaling distribution network, particularly those conversion. This technology would be limited in geographical scope, must used point-to-point, from the customer’s either accommodate these legacy premises to the Head-End. services or exclude large, profitable markets. It is simply not economical for TDMoIP OVERVIEW end-users to convert national or international non-IP networks to IP in a piecemeal fashion. A T1 frame is composed of 24, single byte time slots plus a single PBXs present another problem: synchronization bit, for a total of 193 signaling. Signaling consists of basic bits. Frames are transmitted at a rate of features such as recognizing that the 8000 per second, resulting in a data phone is off-hook, or needs to ring; the stream of 1.54 megabits per second. In more advanced properties required for principle, the simplest implementation of reaching the proper destination and TDMoIP simply encapsulates a number of T1 frames in an IP packet by tacking on the appropriate IP header. At the possible that new standards may be destination, the stream is then recreated formulated for this specific purpose. by stripping away the headers and reassembling the segments. It is Packetization important to note that TDMoIP transports the T1 circuit without any Primary issues to resolve include attempt at interpreting the data. This which IP protocol to use and how many process is oblivious to signaling, time T1 frames should be placed in each IP slots, or whether voice or data are being packet. Since there is no standard, any transmitted. This also implies that a data IP protocol could be used. Some, bit-stream using the entire 193 bit frame however, would clearly be inappropriate. can be supported. The end-to-end reliability offered by TCP, for example, is not useful for voice Standards packets, since re-transmitted voice packets will reach the receiving side out TDMoIP is essentially the IP of order, only to be dropped anyway due counterpart of the same service in ATM to delay constraints. A good choice of referred to as “Circuit Emulation protocol could be RTP and the Service,” (CES). While there are, as yet, associated RTCP, which in certain no standards2 for TDMoIP, such networks would offer better clocking standards do exist for ATM-CES. functions. Ultimately, for Ethernet Furthermore, since the performance networks, only UDP is fundamentally requirements for TDM are independent needed. of the method of transport, it is clear that the performance requirements for ATM- There are tradeoffs to be considered CES should be adhered to as closely as with selecting the number of T1 frames possible in TDMoIP. Nonetheless, how per IP packet. The fewer the frames, the to achieve that has not been greater the IP overhead, which will standardized, and will likely vary increase the amount of bandwidth between TDMoIP platform vendors. needed per T1. The greater the number Thus, the interoperability of equipment of frames, the greater the end-to-end from different vendors should not be delay, packet loss becomes more expected. onerous, and larger buffers are required. Larger number of frames per packet The TDM performance guidelines to could also exacerbate adaptive clock follow primarily relate to clocking. The wander. QoS demands that the number clock rate of the TDM stream should be of frames per packet be kept small stable to within +/- 32 ppm3 and wander despite the overhead penalty. At four should not exceed 80µsec per day4. frames per packet, this penalty is about Performance standards directly related to 50%, meaning a 1.54 Mb/s T1 would IP networks, such as the maximum require more than 2.3 Mb/s bandwidth in allowable packet loss, do not yet exist. IP. Even with this overhead, a 1 Gb These standards are far stricter than what Ethernet network is capable of is required when terminating TDM on supporting several hundred T1s. end-user equipment and it is entirely Signaling asynchronous network with no distributed clock. There are three primary types of signaling: in-band signaling, channel Timing is not provided in IP associated signaling, and common networks, thus synchronization must be channel signaling. None of these are achieved from an external source. This impacted by TDMoIP. In -band, as the can be accomplished by: a Stratum 1 name suggests, is signaling in the audio external master clock at each end of the band of speech. The ubiquitous ‘touch TDMoIP circuit; clocking from an tone’, or Dual Tone, Multiple Frequency external clocking distribution network; (DTMF) is an example of in-band or in-band clock recovery and signaling. Since these tones are encoded regeneration, i.e., adaptive clocking. in the T1 frame time slots, they are Stratum 1 clocks are so precise that T1 automatically carried over TDMoIP. streams timed by separate Stratum 1s will be synchronized. This is a relatively Channel associated signaling is also costly solution, although the prices, carried within the T1 frame time slots. particularly GPS-based Stratum 1s, have Specific voice bits are ‘robbed’ and the been declining recently. signaling bits are substituted. TDMoIP does not distinguish between bits used An external distribution network for for voice and data bits, thus this clocking is also an expensive solution, signaling is carried transparently. and severely compromises the entire concept of having a single network to Primary Rate ISDN signaling, PRI, is maintain. In this scenario, a separate a popular type of common channel network would be maintained just to signaling. The twenty-fourth time slot of send clocking signals to every end-point. the T1 frame is used to carry the signaling data for the other twenty-three In-band clock recovery and time slots. Again, since TDMoIP does regeneration, or adaptive clocking, is the not distinguish between voice and data, most cost effective and will meet the the signaling is carried transparently. requirements of customer premise equipment such as PBXs. In adaptive Clocking clocking, the source TDMoIP unit, which is clocked to a Primary Reference Clocking is the most difficult Source, simply sends the data to the problem to solve in deploying TDMoIP. customer TDMoIP unit. The customer There are several methods of clocking in unit writes data to the segmentation and any type network. These are: re-assembly (SAR) buffer and reads it independent Stratum 1 clocks at each with the local clock. The level of the end-point; a synchronous network in SAR buffer controls the output which the primary reference source frequency of the local clock by (PRS) clock is distributed throughout; an continuously measuring the fill level asynchronous network in which a around the median position and feeding network clock is distributed through out; this measurement to drive a Phase Lock and adaptive clocking in an Loop (PLL), which in turn drives the local clock. Thus, the local clock time for packetization and recreation of frequency is modified to keep the re- the T1 is less than five milliseconds. assembly buffer depth constant. When Delay is not an issue in TDMoIP unless the TDMoIP unit senses that its SAR each TDMoIP packet contains a great buffer is filling up, it increases the clock many T1 frames. rate. When the unit senses that the SAR buffer is emptying, it decreases the clock End-to-end round-trip network delays rate. Since the packet arrival rate is greater than 30 milliseconds could directly dependent upon the packet necessitate the need for echo transmission rate established by the PRS cancellation on voice circuits. Round- at the head-end, synchronization of the trip delays of more that 300 milliseconds TDM stream is maintained. will result in unacceptable QoS for conversational speech. This is the same The proper choice of buffer size can as for VoIP service. In both TDMoIP prevent buffer overflow and underflow, and VoIP, quality can only be provided and at the same time, control delay and assured on tightly managed (greater buffer sized implies greater networks with well-executed delay). The buffer size is proportional to prioritization procedures. the maximum packet delay variation. This variation should be determined by Packet Loss summing the delay variation of each network device in the circuit path. The Sequence bits are used to determine a sum of the measured delay variations lost packet condition. In the event of that each piece of equipment introduces lost packets, timing is maintained must be smaller than the maximum through the insertion of dummy packets packet delay variation configured on the carrying appropriate framing bits. It is TDMoIP unit. If not, underflows and possible to mitigate voice quality overflows will occur. This buffering impairments by repeating the frames that will also remove any jitter encountered preceded the lost packet. However, the by packets arriving at slightly different total loss of several contiguous T1 time intervals. frames would not significantly degrade a voice circuit. In the event a TDMoIP packet is lost during transport, a dummy packet is VoIP packet loss could be used as a transmitted by the customer TDMoIP guideline. Unfortunately, various studies unit in order to maintain clocking in the show that while some deployments can T1 output stream. Since the packet will sustain a 5% packet loss before realizing contain no customer data, it must still be a significant degradation of QoS, other considered a frame-slip, however, timing deployments can suffer less than .2% problems will be minimized. packet loss. Obviously packet loss must be minimized. As with network delay, TDMoIP and Network Delay this implies that well conceived prioritization methods must be used A T1 frame represents .125 within the IP network, with TDMoIP milliseconds of real time. Processing given the highest possible priority.

Prioritization Deployment

Properly prioritizing the TDMoIP A single, small end-user integrated packets will minimize network delay and access device can be configured to house packet loss. This is critical for a video port, Ethernet port, and multiple maintaining satisfactory QoS. By POTS and TDMoIP ports. The TDMoIP marking TDMoIP packets they may be packets would be routed through the easily identified and prioritized. This is network to a Head-End TDMoIP unit. A done through proper marking of the T1 circuit, identical to the end-user’s Type of Service (ToS) bits, and VLAN original T1, would be generated by the tagging and priority labeling according Head-End unit. to IEEE 802.1 p&q. Additionally, there is an assigned, IANA-registered UDP TDMoIP may be deployed in socket number for TDMoIP. These overbuilds by network service providers features simplify flow classification or in a greenfield environment. There through switches and routers. In a are three primary methods of tightly managed network, the QoS of provisioning the T1 service. The first is TDMoIP should be equal to that of a to simply port the T1 directly to a TDM traditional T1 circuit. service provider, essentially just leasing T1 service from another carrier and Supported Features reselling the service. If the Ethernet network provider has a Class 5 circuit TDMoIP is capable of supporting switch, then the T1s would be unframed T1, Super Framing (SF), terminated on that switch. Extended Super Framing (ESF), as well as Channel Associated Signaling (CAS) In a greenfield environment in which and Common Channel Signaling (CCS), a softswitch and PSTN media gateway including Primary Rate ISDN (PRI). are deployed for VoIP service, the T1s may be terminated on the PSTN There are two solutions for media gateway. supporting fractional T1s. The first is to put a large multiple of the individual The Figure 2 illustrates how a Head- time-slots in the same TDMoIP packet. End configuration of such a network. This would reduce the overhead penalty. However, as discussed earlier, this SUMMARY would also increase delay as well as create major QoS problems in the event of a lost packet. The second method is Future distribution networks must to transport the fractional T1 as though it support legacy services such as T1 to be were a full T1, filling the unused time- viable in the marketplace. If these slots with idle code. This would require networks are packet based, as is using the same amount of network expected, technology must be deployed resources for a fractional T1 as a full T1. that will emulate a TDM T1 circuit. We have presented a straightforward method of providing such a service through the use of Time Division Multiplexing over through the use of existing standards can IP (TDMoIP). Since the T1 stream is minimize delay and packet loss. The carried transparently in TDMoIP, any resulting T1 circuit emulation service PBX signaling protocol or data format can equal that of existing TDM could be accommodated. technology, and easily meet the requirements of end-users. Careful control of prioritization

ISP PSTN

MPEG-4 Aggregate Encoder Switch

IP TDMoIP Services Unit Soft- Voice Video Switch Gateway Distribution HEAD-END

IP Network Potential Signaling Links

Figure 2: Greenfield Deployment Head-End Architecture

1 2000 Perspective on the Telecom Marketplace, Provisioning of Private Line and Frame Relay Services: A Global Perspective 1999-2004; Insight Corp. 2 TDM over IP, Internet Draft August, 2001 3 American National Standards Institute (ANSI) T1.403.1995 4 ITU (CCITT) G.823 and G.824