THE COMPLETE TECHNICAL PAPER PROCEEDINGS FROM:

“BUT THE DEMO LOOKED GREAT” A TECHNICAL PRIMER ON SETTOP SOFTWARE DEVELOPEMENT

Stephen Johnson Coach Media

Abstract this knowledge, cable operators and other gatekeepers can ask the right questions and Creating effectively designed software— thereby create—or procure—better for any platform—is a complicated and time- applications to meet their ever-expanding consuming business. Creating software for needs. digital settops has encountered challenges unique to the cable industry: scarce MIDDLEWARE: FLEXIBILITY VS. applicable design precedence, a thankfully SPEED temporary mania of inflated expectations for “interactive television,” and a paucity of Middleware refers to a software basic technical knowledge about how application that runs programs—but also runs applications are created and deployed. While on top of an existing operating system (OS). the first two issues have largely faded, the Think of a web browser, e.g., Internet last continues to hamper effective application Explorer™ running “on top of” Windows™ deployment. Without a general and you get the idea. A middleware understanding of the issues involved in settop application uses code—or even just files of software design, cable operators will text or formatted data—written according to inevitably experience frustration and disap its own languages and syntax rather than that pointment with their deployments. of the OS. The diagram below illustrates This paper addresses five topics in how these various applications stack up: software design applicable to digital settops, SETTOP SOFTWARE STACK providing a brief technical background of DATA & GRAPHICS each issue followed by their individual effects on software performance—and ultimately MIDDLEWARE APPLICATION(S) viewer comprehension and acceptance. DATA & GRAPHICS While illustrated by real-world examples, MIDDLEWARE NATIVE this discussion does not focus on a particular APPLICATION(S) settop box manufacturer or set of features; OPERATING SYSTEM rather, the principles discussed here apply to all platforms. The topics cover the following: Schematic diagram of typical digital settop box • Middleware Usage software stack

• Memory Allocation Current Examples • Settop Performance

• Viewer Interface Details How and where is middleware used? The • TV Display Technology most prominent use today is probably the OpenCable™ Application Platform (OCAP), While not comprehensive, this list which uses middleware called the Java nevertheless covers a range of technical Virtual Machine™ (JVM). Java™ was issues collectively having a large impact on originally developed by Sun Microsystems as successful software deployment. Armed with an Internet-friendly programming language that could execute code using a JVM on any faster. Sometimes much faster. The reasons operating system. OCAP applications, for this are simple: fewer instructions to therefore, are written in Java and could translate and direct, customized code. Think potentially run on any settop operating of the stack diagram above; when a system; the JVM would handle all the middleware program runs its code it has to be underlying operating system routines. continually translated for the operating system. The difference is as dramatic as While OCAP remains the most prominent speaking English to an English speaker middleware example, the settop box versus speaking Italian to the same speaker development world has also seen the and attempting to translate it in real-time. introduction of middleware applications from While the translation will generally work, no companies like OpenTV, Liberate, and one can reasonably argue it will be as fast or Microsoft. With so many companies offering as clear. middleware solutions to settop development, why consider any competing strategy? While it’s certainly the most important Indeed, in addition to the promise of OCAP reason to avoid middleware development, middleware offers some very tempting lack of speed alone unfortunately doesn’t advantages. exhaust the disadvantages. Middleware is also very large. Since it translates The Best of All Worlds? instructions for many different operating systems—and needs to be resident in the Start with labor. Given the latent settop’s memory to execute commands at popularity of Internet-based programming a anything approaching reasonable speed—by relatively large pool of trained candidates has necessity it often requires generous amounts become available to code middleware of precious memory. Many middleware applications—at least of the browser-based applications simply cannot run on older or JVM variety. Since middleware digital boxes due to memory restrictions. programmers don’t write code directly to the operating system, developers are additionally In addition to its other difficulties, free to “script” quick applications and test middleware platforms also aren’t—alas—as them without enduring the rigor of compiling flexible as advertised. Even Sun’s touted and linking code. Think of an HTML Java, once marketed as “write once, run scripter doing a quick web page layout anywhere” has occasionally been derided as compared to a C++ programmer writing “write once, debug everywhere.” Even proprietary code and you can appreciate this discounting the jokes, Java developers often difference. find they have to adapt code for JVMs to their particular use. With deployment flexibility and an available labor pool on its side, what So use of middleware—a key decision disadvantages could there be to middleware point in settop software development—really development? Why suffer through hiring comes down to flexibility versus speed, with programmers to work in a very specific and a sizeable caveat on the former and little unforgiving development environment—on argument over the latter. code that can’t be used elsewhere? In a word: speed. Programs written without middleware overhead simply run MEMORY: BE LEAN, MEAN, AND PLAY Developers recognize these issues and WELL WITH OTHERS devise appropriate strategies to save memory space, but the resulting tradeoffs are far from Whether an application uses a middleware painless. Memory-related challenges platform or not (see previous section), it still confronting developers—along with loads into the settop’s memory when accompanying strategies and tradeoffs—are launched—not unlike how a program loads noted in the table below: on your PC. Compared to an average PC, however, settop memory is severely limited. Challenge Strategy Tradeoff(s) Large application • Store unused • Interactive The first issue confronting application size (including components performance memory usage, therefore, is simple: how supporting data) on server when much room do you need? Many developers vs. limited • Distinguish uploading memory between new data equivocate on this issue—and not without launch-ready good reason. The raw code size specified for and full an executable application may be misleading application for at least three (3) reasons: Large application • Store unused • Interactive graphics vs. components performance • The additional data the application limited memory on server when requires (e.g., program guide data) might • Compress uploading graphics new graphics be even larger than the code that • Use settop- • Graphic manipulates it; based degradation • Graphics and other large non-code graphics • Customized graphic components might not be included (for programmin good reason) in an application’s memory g footprint; and Multiple • Limit • Application • An application may be sharing memory applications number of Swapping space with several other applications— residing in simultaneous • Extensive about which it has no knowledge—and memory running Quality applications Assurance might be unstable at any size. • Certify (QA) deployed Why do developers need to address these applications issues? Program guide data, for example, can be enormous. One day of TV listing data To save memory many developers keep a uses up to 250 kilobytes of memory. Doing subset of application code, data, and/or the math it’s not hard to see why loading two graphics stored on a remote server and only weeks of guide data (14 x 250K, or 3.5 load it when required (e.g., on viewer megabytes) into settops with free memory request) to do so. While this strategy saves sizes of a few megabytes presents some space, the process of swapping in code and difficulties. Graphics are arguable worse: data might create some awkward one high-resolution (uncompressed) full performance delays when contacting a screen background requires almost a remote server. Delays become acutely painful megabyte of data. And lack of vigilance when loading graphics as viewers wait for an about several different applications interface to “arrive” and a screen is not yet (regardless of size) residing in memory has visually “complete.” very high costs: settop crashes (and reboots) correlate very well with this situation. A related memory-saving strategy a settop’s memory creates challenges of its effectively divides an application into parts, own. Furthermore, these challenges directly allowing a smaller version to launch (e.g., affect the cable operator since individual initially display on-screen) while the full developers may have no prior knowledge of version is stored remotely (e.g., on a server, other applications with which their programs as described above) and downloaded later. need to co-exist. While requiring some programming subtlety, this strategy mitigates performance delays As noted above, having a large number of by: 1) launching application faster; and 2) applications in memory often creates settop allowing other applications parts to be loaded crashes for a variety of technical reasons into memory while the initial launched beyond the scope of this paper. With this application runs. unpleasant fact in mind, cable operators need to address two challenges in this area: 1) Compressing graphics also saves space, keeping applications from running but tradeoffs beyond the obvious—potential simultaneously as much as possible; and/or visual degradation—should be considered. 2) testing multiple application combinations Depending on the compression schemes (e.g., before deployment. The first challenge places MPEG, JPEG, BMP, IMG) supported by the restrictions on application features and user settop operating system or middleware, interfaces while the second potentially graphics may require decompression time to imposes testing costs on the operator. be properly displayed—adding to performance delays. If compressed graphics Settop boxes will probably never have can be shown as-is the inherent low- enough memory for every deployed resolution NTSC display standard for application—or combination of applications. television often hides ugly artifacts—to a Even with the current tradeoffs operators point. The rapid deployment of High have deployed dozens of stable and robust Definition (HD) capable settops—and their interactive products. Intimate knowledge of demand for high-resolution imagery—will these issues often makes the difference likely eliminate this advantage very soon. between deployable and “demo” applications. The process of creating graphics directly from code available in the settop’s operating SETTOP PERFORMANCE: FASTER IS system (if available) offers the promise of GOOD, ROOMY IS BETTER avoiding speed and compression issues altogether. Already available from the OS, Many standards are available to gauge settop-generated graphics display very settop hardware performance. The previous quickly and require no compression. Creating section discussed memory management; this these graphics, however, requires hyper- section covers two other issues: processor specific programming skills and generous speed and permanent, or non-volatile, development time; working at the level of memory. individual screen pixels is not uncommon. Compared to similar hardware in PCs, Memory management issues unfortunately settop microprocessor speeds are aren’t limited to those relating to application frighteningly slow. Even accounting for size; keeping several applications— settop box economics, the numbers are regardless of their size(s)—simultaneously in startling: for example, the MicroSparc processor in a baseline Scientific Atlanta Settop hardware performance therefore 3000-series settop clocks in at 166 boils down to both suggestive and definitive Megahertz. In non-geekspeak that’s about numbers. Raw processor speed can be four to six times slower than an Intel or AMD overrated—but all the memory in the world microprocessor in a cheap PC. probably isn’t enough . Fortunately, raw processor speed pales in VIEWER INTERFACE: THE DEVIL IN importance to how well a settop is optimized THE DETAILS to use it—and how well developers make use of it. Current settops are generally optimized Even when a settop application runs to play digital video; relatively speaking, the quickly, uses memory efficiently, doesn’t tax processing power devoted to running the network or local storage, it still may not interactive applications doesn’t need to be look very good. Obviously this is often due nearly as fast as that for a PC. Settop to aesthetics, but this paper leaves that applications display graphics and data to hornet’s nest to the legion of TV graphic viewers and respond to interactive commands designers. Rather, the primary focus here is but have little need for the number crunching on the technical details of the viewer power PCs require for their word processing interface display. While this subject is nearly and spreadsheet tasks. inexhaustible, a survey of three (3) general areas—graphics, video, and on-screen text— The amount of available memory—both suffices as a primer. volatile (lost without power) and non-volatile (or NVRAM, maintained without power) in a Getting Graphic settop actually has a stronger impact on application speed and stability than its While graphics may be compressed in microprocessor clock speed. NVRAM several creative ways, their resolution is contains data that a developer doesn’t want ultimately dictated by the settop’s operating his application to lose if the settop loses system. Resolution is actually made up of power. Pay per View purchase (secured) data two separate parameters: size and color is an obvious example; settops have been depth. The supported sizes and color depths storing this data since the days of analog of some typical settop boxes are shown boxes. Applications also use NVRAM for below: parameters like viewer preferences (e.g., favorite channels) and code-critical data Settop Class/ Graphic Color Depth (e.g., patches and updates). Manufacturer Resolution Explorer 2000/ 320 x 240* at 72 16 bits Scientific Atlanta dpi NVRAM is also expensive and therefore Explorer 8000/ 640 x 480** 16 bits scarce. The aforementioned Scientific Scientific Atlanta DCT-2000/ 352 x 240 Up to 8 bits Atlanta box contains all of 2000 bytes of it. Motorola Developers who run out of room are forced DCT-5100/ 704 x 480 Up to 24 bits to store this important data on a server, via Motorola *640 x 480 supported, but not typically used due to memory sending it at regular intervals and correlating constraints** Typical use; up to 720 x 480 supported the data with individual settops. (Motorola) settop boxes presents a developer Video: Scale with Caution with some intriguing differences. Notice from the table above that resolutions and Settop applications have little control over color depth vary on both systems on almost a full-screen video playback resolution; the box-by-box basis. Developing for lower-end MPEG video compression standard (used in settops has attendant difficulties: low all digital settops) uses its own display resolution graphics must be recreated for the parameters. However, when video plays back same application on a higher-end box. at less-than-full-screen, e.g., in a quarter screen “window”, the settop application Graphic resolution is often dictated by the chooses the “scaled” video window size. largest image size accommodated, then scaling back if necessary. For example, the Since settops scale video by simply settops listed above use so-called “full” and removing pixels, video playback in windows “half” resolutions, meaning the fullest at non-fractional sizes can introduce visual resolution uses twice the horizontal and anomalies: “pixelated” objects and jagged vertical pixel count as the lowest. By some diagonal edges. In this case the relatively simple visualization one can see these labels low resolution of NTSC television doesn’t are somewhat misleading: a full resolution help and often hurts: degraded video images image (at the same color depth) actually still move at 30 frames/second and are more requires four times the memory capacity as a likely than graphics to be noticed by viewers. half resolution image (see below). What’s Your Type?

"Half" Resolution Explorer / 320 x 240 DCT / 352x 240 "Full" Resolution Like graphics and video, on-screen type Explorer / 640 x 480 must abide by resolutions largely set by the DCT / 704 x 480 settop operating system. Unlike these other visual criteria, however, type resolution is

determined by how it is rendered to the Comparison of “Half” and “Full” Graphic Resolutions screen rather than the number of pixels or color depth of its characters. Use of low resolution graphics is often an attractive tradeoff when memory space is a Most operating systems support both consideration. In this case graphics are vector-based and anti-aliased fonts; the expanded (“stretched”) at twice their difference can be seen in the images below: horizontal and vertical resolution to fill the screen. Fortunately the NTSC television display standard—which relies on flickering interlaced lines rather than pixels—often compensates for the low resolutions artifacts (or “stretch marks”). f

Vector-based Anti-aliased Bitmapped

Vector-based fonts rely on mathematical consequences—but the resulting applications formulas to print characters with smooth can look pretty ugly. Interfaces that flout edges, not unlike a standard laser printer. technical constraints exhibit overly bright Although these formulas take up very little colors, distorted images, and illegible or off- space (approximately 32 kilobytes per screen text and graphics. typeface; bold and italic versions not Two basic technical constraints on included) many settops exclude them because displays come straight from the arcane world developers don’t tend to use the same of television post-production: title safety and typeface—and including more than a few color limits. creates storage difficulties. Better Safe Than (Really) Sorry Anti-aliased fonts typically take up less ii space than vector-based fonts—and Commercial televisions do not show an surprisingly look better at some resolutions— entire video signal; rather, the TV monitor but have two non-trivial drawbacks: 1) they itself visually cuts off the outside edges. must be adjusted—sometimes pixel-by- Images and colors at the edge of the display pixel—at low resolutions to read properly; appear to “bleed off” the border of a TV. and 2) the “fuzzy” edges of their characters The reasons for this are complex, but this must be legible against every background on type of display has some advantages, e.g., which they’re rendered. allowing certain unsightly video artifacts like the vertical blanking interval, or VBI, (where If an application cannot use either vector- closed-captioning information is transmitted) based or anti-aliased fonts, settops typically to be hidden off-screen. provide several bitmapped fonts as a fallback. Characters from this font type have To complicate matters, TVs differ— very rough edges but some typefaces sometimes drastically—in the actual amount rendered in this fashion can be reasonably of signal they cut off; that is, two TVs may legible and acceptable for limited application display different amounts of the same signal. use. Televisions uses two cut-off display

conventions: title safe, meaning the area TV DISPLAY TECHNOLOGY: where no TV will ever cut off any portion of “ART IS NOTHING an image and action safe, meaning the area WITHOUT LIMITS”i where on-screen action (e.g., something

moving) must be contained. Action safe is Since settop applications focus primarily slightly larger than title safe since viewers on displaying images to viewers (and are not especially bothered by moving occasionally receiving feedback) developers objects occasionally disappearing at the need a thorough understanding of television edges of the screen; their eyes can visual technology. Ignorance of these compensate for the constraints doesn’t necessitate catastrophic movement. The image below shows how the Settop developers wrestling with color safety areas work on a standard 4:3 television should be aware that televisions—unlike display. standard computer monitors—are not calibrated to saturation values. A PC display typically uses colors in mixed values of red, Tonight green, and blue (or R/G/B); these are color saturation values, describing a linear scale of how much of each color is included to create Clear 55 a final value. (This technology largely mimics the process used for printing, which uses a mixture of four colors: Cyan, Magenta, Yellow, and Black, or CMYK.)

News Television, however, create colors based John Smith Weather 15 15 on chroma, luminance and other values that are not direct mixes of saturated colors. This TITLE SAFE process makes television sets very sensitive ACTION SAFE to sharp changes in brightness and contrast, especially with highly saturated colors, e.g., Comparison of Full-Screen Signal, Action Safety, and Title Safety rich red, pure black and white. Offending this sensitivity leads to displays that bleed or So how much should images be reduced “buzz” at the edges of colored areas of text. to “fit” into these safety areas? Many rules- And strong contrasting colors aligned of-thumb exist, but a good conservative vertically create a bowing effect, e.g., the standard is 20% less than full-screen for title vertical separated image appears to curve safety and 15% less for action safety. inward or outward, depending on its on- screen placement. NTSC: Never Twice (the) Same Color Armed with technical knowledge about As for color limits, televisions display a television display constraints, developers rather narrow range. Historical precedent is should also consider conforming to less- at work here: color usage was nearly an formal design limitations: afterthought when televisions were initially • Conform to standard television graphic developed. When black-and-white television conventions. These include how screen was the standard, video was recorded at a objects move or animate, how images are speed of 29.97 frames-per-second; the rendered against other images, and how remaining 0.03 cycles (to fill-out the 30 screens of information or video transition frames/second standard) were reserved for from one to the next (e.g., via visual the (future) color spectrum. This range— effects like “wipes” and “dissolves”). 0.03 Hz—is not large, and favors blue and • Avoid inactivity. While an interface that green hues at the expense of other colors. must be addressed by a viewer may be “held” for a short time, never forget television is a visual, moving medium. • Make text BIG and BOLD—and ensure messages stand out from the background on which they’re rendered. Asking the right questions before This paper merely scratches the surface of development begins keeps application the rich subject of TV display technology; requirements and expectations in perspective. the principles delineated here only aim to While demonstrations often provide a nice starting a design on a firm technological preview of application features, knowing the foundation. technical tradeoffs in deployable settop software is crucial now and will soon be CONCLUSION: EMBRACE YOUR indispensable. The number of settop LIMITS—YOU’LL ALWAYS HAVE technologies—and software to support THEM them—shows no signs of abating.

We can all dream of settops with i unlimited memory and blisteringly fast A paraphrased quote, ascribed variously to Beethoven, Picasso, and Goethe, among many others. processors supporting perfectly flexible middleware, full-resolution graphics and ii Note this discussion presumes a standard, interlace- pixel-perfect video. But until then we have to scan, analog display television. Televisions make choices. The limitations involved in displaying a true digital or high definition signal— settop software development won’t go away whether progressive- or interlace-scan—are outside the scope of this discussion. tomorrow. Even if they did—mirroring the evolution of the PC development world— Stephen Johnson is technology consultant new limitations would surely replace them. specializing in television interface design. He can be reached at [email protected]. “NOW A WORD FROM OUR SPONSOR” TV RATINGS MEASUREMENT IN A DIGITAL WORLD

Robert A Luff Chief Technology Officer Nielsen Media Research, Inc

Abstract measurement technologies requiring totally new TV measurement techniques to be Last year approximately $45 Billion developed and rapidly deployed. dollars exchanged hands between programmers/ networks operators and the This paper will provide a brief historical advertisers. The value of a particular 30 TV measurement technology overview; detail second commercial is directly related to the the current measurement technology and total number and demographics of the issues caused by various digital over-the-air, viewers that were tuned to and likely saw the cable, and satellite distribution techniques as programming segment that contained the ad. well as the plethora of new digital consumer And in the US, one company, Nielsen Media electronic devices. The paper will also Research, has had the primary task of describe the new Universal Metering independently measuring and reporting TV Initiative (UMI) technology that is being viewing statistics continuously since deployed by Nielsen to measure all analog or television broadcasting began more than 50 digital over-the-air, cable-TV, and satellite years ago. The Nielsen ratings has become viewing, including capability for all forms of the currency of the TV industry. time-shifted viewing.

The core technology used to measure TV viewing has been based on determining what IMPORTANCE OF TV RATINGS frequency/channel the home TVs, cable MEASUREMENT TECHNOLOGY settop or satellite receivers are tuned to using direct RF frequency measurement of The extraordinary success of television, the tuner’s local oscillator, or indirect including the proliferation of broadcast and channel segregates such as, reading LED or cable networks, TV stations, program on-screen channel indicators, or a number of suppliers, and the tens of billions of dollars in other channel determination based annual advertising expenditures and cable technologies. However, the adoption of programming fees necessary to fuel this digital transmission technologies plus new growth has had an invaluable tool-- an digital based consumer electronics independent measurement system which equipment in the home are creating new directly samples the viewing audience and challenges to frequency/ channel based TV reports the information back to the measurement technology. Multiplexed marketplace on a daily basis. This program distribution, time-shifted viewing, information—the Nielsen ratings—is the xVOD, IPTV, and home wireless video currency of television. This year, more than networks are blurring the traditional concept $45 billion exchanged hands in the TV of “channel” and are causing a Global industry based on this currency. obsoleteness of all channel based TV While nearly everyone has heard of The TV technical community will want to Nielsen ratings, few recognize the enormity become knowledgeable about Nielsen’s new of the task or the complexity of the overall Universal Metering Initiative A/P system-- measurement technology system. But now But first, some TV measurement basics to that the US and the rest of the World are better understand TV measurement shifting to a digital based TV system a totally requirements and why a new system is new “digital friendly” Universal Metering necessary in the digital TV World. Initiative (UMI) using active and passive audio (A/P) technology and new Nielsen TV Ratings 101 Audio and Video Encoders (NAVE) must be rapidly deployed. In its simplest form, TV ratings require detailed measurement data of two equally The new A/P system will work equally important broad areas-- what is being viewed well on analog or digital channels, including and the demographic detail of who is in the HDTV and time-shifted viewing. The new audience. While a full treatise on TV NAVE real-time encoder encode both the measurement definitions and methodically audio and the VBI just before over-the-air issues is beyond the scope of this paper a few transmission or network distribution with an basics are necessary in order to understand assigned constant source ID code and a the current analog based system and why a time/date stamp about every two seconds in completely new digital friendly system must the active program audio using a patented be rapidly deployed. Four sections are technique that makes the codes non-auditable presented as background information: to the ear but easily detected by in-home Definitions; Sampling; Privacy, and; The monitor equipment and Monitor Plus sites. Nielsen TV Diary (The Diary section comes The NAVE encoder also encodes the later in the paper.) standard AMOL codes that are necessary with the older measurement technology Definitions during the transition period and with the new system until a new all digital metadata Universal Estimate (U.E.) Total persons or standard is adopted and used by the industry. homes in a given population, e.g., TV households in the US. The new audio encoding technology is very robust and will withstand almost any Rating Percentage (Average Audience) The normal “bad practice” that would still percentage of the Universal Estimate produce a reasonable picture and audio to the home. (U.E.) viewing a TV program during the average minute: The new A/P system has for the first time Audience in a TV measurement redundant Rating % = ______measurement technologies in case of failure Universal Estimate or operator error of the primary active audio encoding system. The new system will automatically default to the redundant passive signature mode.

Share Percentage of TV sets (or persons Hawthorne Effect Even the tiniest of viewing) tuned to a program: inadvertent change to the overall measurement environment introduced by Rating either the measurement equipment itself or Share = ______just the impact of respondents knowing that

HUT they are being measured can cause bias or errors in the collected data. (For this reason, HUT Number of homes using TV using “doctored remotes” or exchanging the household TV with similar pre-metered TVs PUT Number of persons using TV is not a acceptable option.)

Total U.S. Rating (%) Average audience of Source Detection/ Reporting In today’s total U.S. households more competitive TV industry a program/ network can arrive and be viewed in the Reach Number of different homes/people home from several alternative sources: exposed at least once to a program or direct-off-air reception; cable-tv; satellite, or commercial. Also referred to as Cume. perhaps the internet or telco based delivery options in the future. In addition, in order to Average Frequency.Average number of accurately measure HUT nearly everything times a home or person is exposed to a connected to the TV that could generate program or a commercial. video must also be identified and reported, including video games, VCRs, PVRs, and Average Audience Projection/Impression DVD players. The audience expressed in numeric rather than percentage form: Automated Measurement of line-up (AMOL) A defacto Nielsen industry standard of Average Audience = Rating X Universal Estimate inserted codes in the VBI for the purpose of electronically identifying the source and real- Gross Average Audience (GAA Rating) time/date stamp of its transmission. Sum of the percent of households tuning (or

persons viewing) during the average minute Monitor Plus Sites (M+) Nielsen maintained of each telecast of the program, including monitoring sites in 354 of the largest cites repeat telecasts during the report interval. across the US and Canada. These sites

collect and store 24/7 all audio and one frame Gross Rating Points (GRPs) Sum of all per second of video from nearly all TV ratings for a program in a schedule. stations and cable/satellite networks. The

stored audio and video are used (and Cost per Thousand (CPM) The cost to augmented by stations line-ups) as the “as expose a commercial/ program to 1,000 transmitted” data base and used extensively people or homes: with both the current People Meter/ Mark-II Media Cost CMP = ______X 1,000 and new Universal (Digital) Metering Initiative A/P system. Impressions Designated Market Areas (DMAs) There “I don’t watch enough TV for Nielsen,” or “I are currently 210 individual “local” DMA only watch…” are two often heard reasons markets defined by Nielsen. TV viewing selected viewers feel they shouldn’t behavior can vary between local markets, participate. But inclusion of all randomly especially in early morning and late night selected households/viewers is essential to viewing patterns. have accurate projectable data.

Sampling An accurate, projectable TV The homes are first contacted by phone. measurement system depends on a random (Nielsen is exempted from the “Do Not Call” audience sample that is representative of restrictions) and follow-up with letters and viewers as a whole. It is important that the person-to-person contact as necessary, random group mirror the behavior and including using callers and field staff with characteristics of the overall population not special language/ dialect skills to properly only at the aggregate total level, but also communicate and put the person at ease. within each of the narrower defined This is especially true in the case of some demographic breakdowns important to minorities or non-citizens who might be programmers and advertisers. By using initially unfamiliar with Nielsen and the random probability samples, Nielsen can importance of this research to them and project the viewing in its samples to the others with similar background that they will entire population and sub-classifications of now represent in the sample. There is a very viewers. small monthly payment that is not tied to their amount of viewing. This process can be quite complex in terms of practicable implementation. It Privacy starts by using the US Census Bureau’s decennial (updated annually) census counts Privacy of viewing data is of utmost of all housing units in the nation. Using this concern to all involved, especially to the data Nielsen randomly selects more than sample households. Nielsen’s privacy record 6000 small geographic areas (blocks in urban and privacy controls are unparalleled and areas and their equivalent in rural areas) and surveys continue to show extremely high dispatches surveyors to each area to levels of public trust of Nielsen and its enumerate and list housing units. Housing processes. Beginning with the first contact, it units are then randomly selected within each is explained how the viewing data is sample area. This process ensures that each collected and that a particular household’s/ household in the population has a known person’s viewing data is combined with the chance to be selected. aggregate with no individual or household identity. All of Nielsen’s and industry TV Once a household is selected it is viewing, including demographic breakdowns important to the sample quality that they is derived from data processed from various agree to participate and stay as a sample aggregate buckets of data. The household household for the research period (one week members sign a release allowing Nielsen to for diaries; two years for People Meter and use their viewing data in this way. This Local People Meters). While the overall process adds to ensuring that all data process defines “alternates,” any deviation collected meets state and federal privacy laws from the true initially randomly selected and regulations. households detracts from the sample quality. Households are also requested to maintain alternative technologies are being developed a low profile that they are a Nielsen sample that would require less or no involvement by household while they are actively in the the household viewers to maintain accurate program. This is to avoid any unlikely audience data. A detailed discussion of these targeting of the household by media players technologies is beyond the scope of this of mailings or incentives that might influence paper but include: Infrared, RF tags, Voice the household’s viewing/ reporting data. Print/ Recognition, as well as others.

People Meter Currently there are People Meters installed in about 5000 randomly selected The heart of today’s National TV households across the US-- and the number measurement system is the People Meter, an of National Sample metered homes will electronic metering technology that measures begin to increase to more than 10,000 starting both programs viewed and logs who is in the this year. home viewing audience. The People Meter is actually a combination of a People Meter Sample households agree to be a Nielsen which logs who is viewing and a Mark-II sample household for a two year term and to companion device that measures which allow Nielsen engineers come to their home channel is being tuned. and install People Meters and its associated Mark-II measurement equipment on every The People Meter is a small relatively flat TV, and its associated peripheral devices in profile unit with a row of red and green LED the house, including even small portable indicator lights visible to the TV viewers rabbit ear TVs on the kitchen counter or in which is used to log in/out who is in the room the garage workshop. These installs can take watching TV. The People Meter and its up to two days and require that all Nielsen wireless remote has simple single buttons for measurement equipment and associated inputting which family members are in the wiring be completely hidden/ out-of-sight in room viewing. There are also provisions for order to minimize any long lasting logging in visitors and their demographic Hawthorne effect. No data collected from a data (age & M/F). The When the TV is newly installed Nielsen sample household is turned-on all LEDs begin flashing signaling used for the first thirty days to allow the the viewer(s) to log-in which family impact and newness of the installation and members are viewing. Any family member being a Nielsen sample household dissipate. can log in all others viewing. Periodically Research has shown that normal pre- and after channel changes the LEDs will installation viewing returns well within that begin flashing for an up-date on which period. family members are still viewing. MARK-II The People Meter is placed so the LEDs are visible while watching TV, generally The companion measurement device right on the TV or the cable settop, during installed in a National People Meter sample installation and is usually the only visible household is the Mark-II. The Mark-II was sign of Nielsen’s measurement equipment first engineered in 1978 and has remained the after installation. Because of the possible work-horse of measuring what Hawthorne effect resulting from the current program/channel is tuned for over 25 years! People Meter, various stealth people meter Indeed, it is hard the imagine a device performing such a complex task still The Nielsen TV Diary accomplishing its basic goal after so many years in such a dynamic industry. Its Mainly to address cost, diary strength is its basic simplicity, and the measurement is used to collect viewing forward thinking of its original designers. information from sample homes in every TV The Mark-II while having a power-supply, market in the US. Each year more than 2 UART, dial-up telephone modem, and million paper diaries from households across modest memory and processing capability- the country are processed for the sweeps has survived so well because its visionary ratings periods. The standard November, designers embraced a designed concept of February, May, and July sweep months larger supporting up to eight additional external markets include; October, January, and specialized data collecting probes. March, as well.) Additionally, two Mark-IIs can be daisy- chained together for up to sixteen probes, if In approximately 50 large local markets, necessary. More than 120 specialized Mark- not using the Local People Meter, Nielsen II probes have been designed over the years uses the Mark-II device to collect just set- to allow for measurement of virtually every tuning information on a daily basis. This TV and peripheral device such as, VCRs, daily household tuning data is cost DVDs, dual decks, video games, etc. effectively augmented at least four times a year with demographic viewing data which is Full installation of a People Meter/ Mark- collected from another sample of households II in a typical sample household can take a which maintain a paper-viewing diary for one full day and some homes with higher than week. Each household member in the sample average number of TV sets and/or peripheral is asked to write down what programs and equipment can take two full days. channels they watched over the course of that Obviously, a Nielsen service call must be one week. Then the diary and the Mark-II set scheduled anytime the household decides to tuning information from the separate samples move their equipment, add a new TV are merged to produce the standard “sweep” viewing location, or add any new component reports. to their viewing systems.. TECHNICALLY DIFFICUILT The Local People Meter (LPM) HOMES (TDs)

The Local People Meter is a new Nielsen Despite the enormous flexibility of the initiative to deploy essentially the same People Meter/ Mark-II and its various probes, National People Meter technology within sometimes a home is encountered that has a separate sample of homes starting in the new TV or peripheral device that does not yet largest of the Nielsen DMA local markets. It have technical measurement solution. If will use the People Meter to collect audience even only one device can not be properly viewing data and the Mark-II technology to measured, the household is declared measure TV-tuning data. “Technically Difficult” and no data from the household can be accepted.

TDs would generally occur when new to the But today both cable and satellite TV market devices with either very different industries are capable and have inadvertently circuitry or functionality are purchased and downloaded Global software up-dates to all brought home. Obviously, systematic settop boxes in their local system or network elimination “early adopter” households from that suddenly changed their channel line-ups the overall viewing database would possibly or various internal settop settings enough to introduce a bias in the data if not addressed. suddenly cause a market wide Accordingly, close ties are maintained with incompatibility with the current Mark-II program distributors, consumer electronics measurement technology. manufacturers, and regulators to gain early awareness and prototypes of new products As taxing as this trend has become, that may challenge existing measurement digital TV will over time completely break solutions. Over the years Nielsen has the current measurement channel-based maintained the percentage of Technology system. It’s not that plain digital TV breaks Difficult households to less than .5%. the measurement system, its what the new digital TV domain supports as standard Digital Challenges to the Current System operating practice at the distribution and consumer level that breaks the system over Even without the impact of digital TV, in time and forces a new system. The two recent years the friendly “cat & mouse” game primary causes are: of staying ahead of the challenges of measuring TV viewing and all the peripheral Multiplexed Program Transmission CE equipment (DVDs, Dual Deck devices) began to take on a significantly more The current measurement systems used complex, intrusive, and costly posture. The almost universally are based on a core trend of CE equipment becoming smaller, channel/ frequency determination multi-purposed, and feature rich as resulted measurement strategy. But in digital TV the in a general CE trend toward massive multi- FCC and economics encourages several purpose LSI silicon device-on-a-chip designs separate content streams (programs) be where access to critical TV measurement multiplexed and transmitted over one taps/ test points is not longer available. frequency/channel. When the current measurement system is attached to a digital Development of special Nielsen software, TV tuner tuned to a multiplexed channel it reverse engineering, or ASIC solution simply knows the channel, not which of the development is very expensive, especially for four or five multiplexed programs is actually the very small number of homes it would being viewed. Accordingly, channel based likely be used. Solutions for some new CE measurement systems are not longer equipment are approaching in excess of deterministic is a digital TV World where $500K each with no upper end in-site. multiplexed program distribution will be Further, while most new CE devices, commonplace. regardless of popularity takes years after introduction to reach a penetration level Time-Shifted Viewing sufficient to cause any impact with the Nielsen sample data thereby providing And the second of the “one-two” digital reasonable development time for a solution. TV punch is the meteoric raise time-shifting capabilities in the home either as a feature of a stand-alone TiVO type CE device, built-in decoding equipment in the home or Monitor to an up-graded cable or satellite box, or as a Plus sites. Since the encoded data is in the virtual time-shift capability supported by the active audio of the program it automatically cable headend server. Indeed, it is cable’s rides along wherever the program and its unique ability to significantly accelerate audio goes. It is very robust and can survive deployment of time-shift capability in certain almost any “bad practice” until the content markets that has resulted in a corresponding itself is nearly un-viewable. acceleration of Nielsen new digital friendly Universal Metering Initiative and especially The “back-up” redundant passive its ability to measure time-shifted viewing. signature system does not add anything to the program audio. It simply uses the program UNIVERSAL METERING INITIATIVE audio itself to generate a corresponding thin data stream signature that is unique to that More that seven years ago Nielsen program audio. At the Nielsen Monitor Plus accessed the direction of the TV industry and sites master signatures are generated as consumer electronics and determined that a needed using a special algorithm and stored completely new metering system would be 24/7 for any TV station or cable/ satellite necessary to address the capabilities and network, as needed. In the home, the A/P TV needs of a more digital empowered and measurement equipment is connected completive TV industry. A complete and internally to the TV audio line-level out and through review of all measurement active codes containing its source ID and technology options was considered. In time/date stamps about every two seconds addition to all the previously discussed during the viewing are collected from all concerns/ issues, Nielsen also was aware that viewed programming. If the viewed significant segments of the industry desire programming is not properly NAVE encoded Nielsen to develop the ability to measure for whatever reason, the A/P home “out-of-home” viewing. monitoring equipment automatically begins to generate and store its thin stream of Active/ Passive (A/P) passive signatures derived from the unidentified audio using the same algorithm The result was the development of a as used at the Monitor Plus sites for a match totally new concept in TV measurement that when all viewing and audience data is relies primarily on audio to determine forwarded to the central operations facility by programs viewed. However, because of the dial-up modem later at night. reliability needs, a patented active and passive audio system (A/P) that supports a Nielsen Audio and Video Encoder (NAVE) unique and first time redundancy in TV measurement was developed. The A/P The NAVE encoder is a critical part of the system relies on active encoding using the new TV measurement system. Its purpose is Nielsen patented psychoacoustic masking to automatically insert a source ID and technology that “actively” encodes a special time/date stamp undetectable to the human code into the program audio in such a way 3hearing into the program audio. It also that it conceals itself from normal hearing, inserts the current AMOL VBI format source yet easily detected by special monitoring/ and time/date stamp information to add in the SUMMARY transition period where both the current AMOL dependant and the new A/P system Digital TV is triggering a global shift to must be supported. The Nave encoder will new TV measurement technology solutions. on average insert a full psychoacoustic In the US, Nielsen Media Research, Inc. has audio masked source ID and time/date Initiative (UMI) in which its audio based stamp code about every two seconds. redundant Active/Passive (A/P) system is already being deployed. This system has the It is important that station and network capability of measuring both analog and engineering community understand the digital TV, including HDTV and time-shifted critical importance of maintaining the NAVE viewing. encoder in the active audio and video program stream. While the passive system The new NAVE audio encoding is very can act as a failsafe emergency back-up, robust and generally will survive all but the using the primary active codes for viewing worst “bad practices.” Both the current detection is more accurate and requires less AMOL used with today’s Mark-II metering processing/memory resources. technology and new NAVE audio coding will be required during a multi-year transition Cable channels where the original signal period. is either from off-air/ fiber delivered local TV stations or from national satellite feeds Since most of the programming carried by which will already be encoded will not need cable will already be encoded with the new a NAVE encoder. At some point when cable NAVE psychoacoustic encoding technology desires to measure xVOD it will need to by the original local TV station or TV, cable, either require that the programming be satellite programmer/ network-- a cable already NAVE encoded or locally encode the system will not need to encode much if any material itself before delivery to viewers. of its programming. If cable operators eventually desire xVOD viewership measurements either the program supplier or the cable operator will need to NAVE encoder the content and provide Nielsen with the content name details.

ADVANCED VIDEO CODING FURTHERS HD SERVICES

Carl Furgusson TANDBERG Television

Abstract have reduced the bit rate required to digitize an acceptable broadcast quality picture into the The use of Advanced Video Coding to home. Figure 1 shows this bit rate reduction enable bandwidth efficient HDTV Video on over time for MPEG-2. Demand and broadcast television is gaining in popularity as technological advancements in MPEG-2 compression continue to slow. While First Broadcast MPEG-2 has been the undisputed leader in 8 MPEG-2 Encoder video compression for the past decade, 7 Statistical algorithms such as Windows Media 9 Video B Multiplexer i 6 Enhancements and MPEG-4 Part 10 are emerging at a time t 5 Enhanced Motion Estimation when consumers are requiring more broadcast R 4 channels and VOD services in HD. a Noise t Reduction 3 e Advanced Pre- As the subscriber volume grows, using processing (M 2 MPEG-2 compression for HD will result in b p 1 bandwidth bottlenecks in the networks. To solve s) this problem, increasingly efficient compression 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 algorithms will be required to satisfy this Year consumer demand. Algorithms such as WM9V and MPEG-4 Part 10 have the potential to Figure 1 – MPEG-2 bit rate for standard outperform MPEG-2 by up to 50%, meaning definition broadcast quality picture over time. that a 50% data reduction to deliver the same quality picture is possible — taking full A high definition television (HDTV) video advantage of network bandwidth and increasing picture has 4-6 times more information to be revenue generation possibilities. This paper will processed than a standard definition picture discuss advanced video coding techniques and (horizontal resolution x vertical resolution x how they apply to HD to enable bandwidth frames per second), as the bit rate for a efficient deployment of HDTV for both standard definition picture reached the 3- broadcast television and VOD. 4Mbps (Megabits per second) range in 1998 then it became feasible to deploy HDTV EVOLUTION OF MPEG-2 BIT RATES digital video services within the bit rates available on terrestrial, cable and satellite The MPEG-2 standard [1] reached transmission platforms. commercial deployability in 1994 and has since been widely adopted for standard We can see from the curve of Figure 1 that definition television (SDTV) broadcast and on- each new technique and technology leap has demand services, becoming the defacto returned less of a bit rate efficiency professional broadcast digital video improvement as we approach the fundamental compression standard. Since 1994 advances in limits of performance of the MPEG-2 technology, systems design, implementation compression algorithm. With continuing efficiency and video pre-processing techniques development we are likely to only achieve a further 10% MPEG-2 efficiency improvement bit rate between a number of channels to at a commercially viable cost. This places a optimize picture quality verses bit rate. As the limit on the potential bit rate of MPEG-2 based bit rate for a “broadcast quality” video is such standard and high definition services in the a subjective matter a range of bit rates are future to achieve an acceptable broadcast indicated in figure 2. picture - in the order of ~1.75-2.0Mbps for SDTV and ~8-10Mbps for HDTV. It should also be noted that different performances can be achieved between real- Competition from satellite platforms, who time encoding products, required for live TV are launching 30-40 channel HDTV services, broadcasting, and products using non-real time in addition to increasing demands from the encoding processes, that can be employed for customer, for example HDTV Video-On- the creation of stored assets, such as VOD Demand (VOD) content, place continuing movies. pressures on cable broadcasters. The bandwidth available to the home is limited on MPEG- Advanced Advanced any system, including cable, and while “all 2, Video Video digital” system upgrades and reducing the today Coding Coding techniques, techniques, homes per node try to overcome the system st capacity limits they carry with them an 1 promise investment and infrastructure cost penalties. A deployments over time major improvement in coding efficiency would HDTV, 10-16 6-10 Mbps 4-6 Mbps enable many more services to be deployed real- Mbps across existing infrastructures, and this is the time (1) promise of advanced video coding techniques HDTV, 6-8 3-5 Mbps 2-4 Mbps such as MPEG-4 part 10 [2] and Windows non Mbps Media Player 9 Video [3]. real- time SDTV, 2-4 1-2 Mbps 0.5-1 Mbps BITRATES OF ADVANCED real- Mbps VIDEO CODING time

SDTV, 1-3 0.5-1.5 0.5-1 Mbps The nature of the video compression non Mbps(2) Mbps algorithms being used in any of these coding real- techniques being discussed means that the time perceived picture quality of a constant bit rate video service will vary depending on the (1) As evidenced by content samples on Microsoft website, complexity of the picture. A “talking heads” www.microsoft.com sequence, typical in news programming for (2) Note, 3.2Mbps transport (~2.8Mbps video) typically used for MPEG-2 VOD assets today example, is relatively simple to encode – there is little motion or detail changes between Figure 2 – Advanced Video Coding Techniques vs frames – whereas sport material involves both MPEG-2 bit rates, by service resolution. high motion and high detail changes between frames, testing the limits of the coding While there are differences in the algorithm implementation. The reverse of this techniques used in the MPEG-4 part 10 and is also true – a constant quality picture requires Windows Media Player 9 Video coding a variable bit rate. Other system techniques, algorithms, resulting in some differences in such as statistical multiplexing, try to make video coding artefacts, it has been shown advantage of this by flexibly sharing a pool of through subjective and objective testing [4] that both these techniques produce averagely • Additional block sizes for intra prediction equivalent quality over a range of bit rates and within I Picture. programming material types, therefore no • In-loop deblocking filter, provides no distinction between coding algorithms is made benefit at higher bit rates but has clear in figure 2. It should be noted that broadcast advantages at bit rates lower than video applications are most likely to use the traditional MPEG-2 operating ranges. “Main Profile” of MPEG-2 part 10 or the “Advanced Profile” of Windows Media Player • Improvements in motion estimation 9 Video. including Tree-structured motion segmentation with a selection of block sizes and 1/4-sample motion TECHNIQUES THAT ENABLE compensation. THE SAVINGS • Additional coding efficiencies from range The new advanced coding techniques of alternative integer transform block sizes, achieve their efficiency improvements over the definition of multiple variable-length MPEG-2 in a number of areas: coding (VLC) tables enabling Context- based Adaptive VLC-based entropy coding • More efficient syntax – saving of bits from (CAVLC), and for MPEG-4 part 10 the fewer and shorter headers, greater range of option for Context-based Adaptive Binary allowable picture sizes, and carriage Arithmetic Coding (CABAC). mechanisms within native IP and with

MPEG based transport streams. • Removal of more information from the CARRIAGE OF ADVANCED CODED picture – improvements in coding techniques VIDEO IN A CABLE SYSTEM and motion prediction tools, including the addition of a more flexible selection of block The advanced coding techniques have sizes. several options of carriage mechanism for transmission. • Hiding artefacts – as bit rates are pushed lower there is an increased likelihood of MPEG Transport Stream blocking artefacts on difficult scenes. Tools such as loop filtering hide the The packetization of advanced coded video blocking artefact edges making them less within the traditional MPEG transport stream perceptible to the viewer, therefore carriage mechanism enables Advanced Coded allowing the bit rate to be pushed lower. services to be broadcast on any existing network carrying MPEG-2 services today.

Using MPEG transport stream also provides a In comparison to MPEG-2 some of the number of other advantages: changes that yield the greatest savings are: • Timing Reference - MPEG transport • Multiple intra-prediction modes for timing references ensure the Field/Frame adaptive coding at picture synchronization of video, audio and time level, and Field/Frame adaptive coding at critical metadata services when the macroblock level (MBAFF) – provides for digitized services are presented to the more flexibility when a picture sequence customer to view. exhibits a mixture of progressive and interlaces characteristics within a • Ancillary data carriage - Other MPEG field/frame. transport related standards also provide a definition for the carriage of ancillary efficiencies over MPEG-2 combined with next services such as vertical blanking interval generation transmission networks will enable data, such as Closed Captioning and V- cable operators to further provision advanced chip information. HDTV services to their customers.

• Audio codec choices – there are multiple

audio codecs defined within various REFERENCES standards and specification which utilize

MPEG transport carriage today. Combined 1. ISO/IEC 13818 International Standard, with the commonality of the MPEG timing "Generic Coding of Moving Pictures reference this enables an operator to mix and Associated Audio Information" and match video and audio codecs to meet their needs. 2. JVT, N5555: Draft Text of Final Draft International Standard for Advanced Video Coding (ITU-T Rec. H.264 | IP Carriage ISO/IEC 14496-10 AVC), Pattaya, The packetization of Advanced Coded March 2003 video services within a native IP transport 3. Microsoft Digital Media website: removes unwanted overhead (header tax from www.microsoft.com/windows/window Advanced Coded Video packets over MPEG smedia/ transport encapsulated into IP frames) when 4. In-depth review of advanced coding Advanced Coded Video services are intended technologies for low bit rate broadcast to be transported over Internet Protocol based applications, J.Bennett & A.Bock networks. (TANDBERG Television), IBC 2003

conference proceedings. CONCLUSION

The major improvement in coding ACKNOWLEDGEMENTS efficiency that are being achieved with first The author would like to thank Ping Wu generation MPEG-4 part 10 and Windows and Alois Bock, TANDBERG Television, for Media Player 9 video codecs will enable many their input and assistance with this paper. more services to be deployed across existing Windows Media 9 Series is a trademark of infrastructures. The efficiency improvement Microsoft Corporation. will further improve as more research and development resources and time improve each CONTACT INFORMATION generation of product implementation. While the support for MPEG transport stream For further information regarding this paper carriage provides an answer for existing please contact: networks, the native IP carriage for next generation networks will become more TANDBERG Television Inc prevalent over time. The IP carriage 12633 Challenger Parkway, Suite 250 mechanism allows a migration to an IP based Orlando, FL, 32826 core network, or even a Switched Digital Tel: +1 407 380 7055 Video over QAM network to achieve Email: [email protected] infrastructure cost savings or provision a greater number of services over a bandwidth limited edge. The combination of compression ADVERTISING INSERTION AND VOD: A MATCH MADE IN HEAVEN

Jay B. Schiller, Senior Vice President, Market Development nCUBE Corporation

Abstract • Recapture digital viewers watching VOD and not watching broadcast cable Video-on-demand (VOD) is now a standard • Support geographically targeted and 'must-have' service for cable operators. It is household addressable ads highly strategic in reducing churn and growing • Provide 100% measurable advertising premium digital subscriptions. It also provides data a great financial return on invested capital through movies on demand and subscription Historically, cable operator’s and cable VOD (SVOD). Local ad sales, meanwhile, networks’ ad sales businesses have not continue to provide cable operators significant leveraged VOD although some have cash flow and are an excellent promotional experimented with it. This is primarily due to vehicle for cable operator’s various services multiple factors including: 1) most VOD including VOD. As viewers migrate to on- content is pay-per-view movies and premium demand, local and national ad insertion must cable programming, which is not normally ad- migrate as well in order to sustain and grow supported programming and does not have cable ad sales’ revenue. In addition, VOD and local breaks; 2) a lack of critical mass of digital cable platforms offer unique viewers; and 3) a lack of acceptable audience opportunities to target advertising to measurement data available (and the newness households, as well as interact with ads, both of the medium) to the advertising industry. of which have the potential to grow ad sales businesses for both operators and cable However, these factors are all changing. networks far beyond today's TV advertising Ad-supported cable networks such as ESPN are market share. becoming VOD programmers themselves. In addition, a number of new programmers are appearing only on VOD services, such as ADVERTISING INSERTION AND VOD MagRack, and may consider ad-supported models. The number of VOD households has VOD streams are unique, one-to-one TV rapidly increased as well, to 16.5 million of the sessions with each household. VOD ad 23 million digital homes (66.6 million basic insertion is local ad insertion but in VOD homes) now, and is growing to more than 38.2 content. VOD ad insertion appears to be million by the end of 2008.1 VOD usage and unique and advantageous for advertisers and ad VOD audiences are also growing weekly. For sales operations in that it has the potential to: instance, nCUBE has seen on average 10-15 VOD session starts per digital subscriber per • Use local breaks in VOD content month in the field. Lastly, ad agencies, media similar to traditional local breaks on buyers and advertisers are being exposed to the broadcast cable (called ad replacement) medium. However, they will need to value • Create new local breaks before and after new audience measurement data and will need VOD content (called playlists) to be able to process it with their back office

1 Kagan VOD & iTV Investor, July 23, 2003 systems, which they rely on to plan effective ad • Detecting local breaks campaigns and to value advertising. Nielsen • Inserting ads in VOD content Media Research, the leading audience • Targeting ads by geography measurement service, has published a VOD • Household addressability in the future system interface specification, which is being implemented into VOD vendors’ systems and An overview of today’s ad insertion may address advertisers’ needs. technology and today’s VOD technology follows. These two sections are helpful for This paper presents both technical those unfamiliar with the technical aspects of requirements and an implementation for each and also set the context for extending ad advertising insertion around VOD content (i.e. insertion into VOD. playlists) and in local breaks in VOD content (i.e. ad replacement). VOD content includes Today’s Ad Insertion movies, SVOD, free VOD and network PVR. The solution leverages existing ad sales Local cable ad sales is a mature and growing organizations, research, and traffic and billing business in the U.S. contributing significantly systems as a starting point. However, the to cable operators’ revenue and net income. Ad solution scales into the future with household sales organizations sell TV commercial time level addressable advertising using newer designated for local insertion on cable networks traffic and billing, and research systems. There to ad agencies and advertisers. Ads are sold for are several other advanced advertising particular networks, day parts, programs, or applications being implemented, tested and groups of networks. Each cable network speculated by nCUBE and others which are not provides 2 to 3 minutes per hour for local discussed in this paper including: insertion, or enough to run 96 ads each on average per day. Typically, 15 to 25% are used • Advertiser-sponsored free VOD using for cable operator marketing purposes. There long form ads and programming on the are more than 80 ad supported networks with VOD service roughly 40 carried on analog and 40 carried on • Interactive advertising for direct digital cable. Most ad sales revenue occurs on response, surveys, coupons, the top 20 networks carried in the analog personalized messages and more domain. However, operators are rapidly • Linking from traditional ads to long moving toward 40 channels of ad insertion and form programming on the VOD service are also beginning to migrate away from analog • Advertising on digital video recorders cable and toward all digital cable.

These applications may be the topics of Cable operators use traffic and billing subsequent papers. This paper will talk (T&B) systems to enter orders, schedule spots specifically about extending today’s ad and bill advertisers. T&B systems track the insertion business and technical operations into number of ads that are sold and thirty second VOD. Of particular relevance in combing ad ads available to be sold (called “avails”) for insertion operations with VOD systems are: each network, and produce schedules for the ad insertion systems so it knows which ads to • Organizational responsibilities, play. The schedules list ads to play during each architecture and workflow local break on each network. Based on • Traffic and billing system configuration schedules, the ad insertion system determines and back office integration which ads need to be encoded, distributed, and removed, as well as whether any are missing. deployments, with both standard definition Once played, the ad insertion systems log each television (SDTV) and high definition ad played and returns the logs to the T&B television (HDTV) naturally supported for ads system so that advertisers may be billed. and programming.

Nearly 100% of all ad insertion systems in Analog and digital switching each have use today use digital ad content based on unique requirements. Analog switches must be MPEG 2 content locally encoded at 4 to 6 accurate and fast time-wise such that the switch Mbps constant bit rate (CBR). Digital ads, into the local break is not late so that the mostly thirty seconds long, are distributed as beginning of the network’s underlying ad is not data files and are stored on video servers’ hard shown or heard. And they must not switch out disk drive arrays located in headends. During of the local break too late so that the network’s local breaks ads stream from the video servers programming following the local break is not and are switched in place of ads in the cable stepped over, although cues may also be sent to network feed. Local breaks in ad-supported signal the end of the break which may prevent cable programming (e.g. CNN) are identified this; in which case the local ad would be cut by the presence of cue signals detected by the short. They must also not switch too slowly ad insertion system. Historically, cue signals such that black video results. Lastly, analog consisted of DTMF tones carried in a second switches typically contain audio automatic gain audio channel in each ad-supported cable control and level matching circuitry to ensure network receive at each headend. Cable that audio levels of the networks and ads networks are migrating to newer MPEG cue match. packets specified by SCTE in the SCTE 35 (formerly SCTE DVS 253) standard. The Digital switching is done by ad splicers SCTE 35 cue packets multiplexed into the which are statistical bitrate re-multiplexers MPEG 2 program stream along with other typically located in headends for efficiently elements such as video, audio and data. The creating digital multiplexes with digital cable cue packets signal local break positions and programming. Ad splicers have the same time insertion splice in and out points. accuracy requirements as the analog switches. Audio level matching is a more complicated Once the cue is detected, ads are streamed issue however for DPI than for analog and from video servers at the appropriate time. cooperation among cable networks, splicer Today, they are typically decoded to analog manufacturers and cable operators are required and switched into analog broadcasts prior to to ensure consistent audio levels before and distribution into the cable plant and distributed after switching. to viewers’ homes. With digital cable growing in popularity, and with cable operators Local cable is unique in that it can insert migrating to all digital cable, ads may also be advertisers’ commercials across a metropolitan streamed from video servers and switched into area (i.e. “market”), and can also “target” ads to digital broadcasts where decoding to analog geographic regions within a market (i.e. occurs in digital set-top boxes in viewers’ “zones”). This flexibility offers advertisers homes. When digital ads are inserted into tremendous efficiency in reaching relevant digital cable programming, as opposed to viewers, while maximizing ad sales revenue for analog programming, it is called Digital operators. With digital set-tops, local cable ad Program Insertion (DPI). DPI represents the sales operators are experimenting with next significant wave of ad insertion “addressable advertising” in which different ads are played to different households (and workflow and monitoring functions required to individual set-top boxes) based on each manage multiple ad insertion systems from a household’s lifestyle segment. Addressable single location. The key functions are: advertising offers tremendous efficiency since ads are directed to the most relevant households • Import traffic schedules from T&B (e.g. has children) versus targeted advertising • Encode and distribute new ads and in which ads are played which are relevant to schedules to headend insertion systems household characteristics of a geographic • Alert of missing ads and system failures region (e.g. incomes >$100K) or just the • Remove expired ads geography itself (e.g. near Bob’s Ford) • Monitor breaks and insertions • Retrieve verification logs and export to An example of today’s ad insertion system T&B is depicted in Figure 1, below. In Figure 1, the SkyVision HQ manages the day-to-day

Satellite Analog Broadcast w/ Broadcast DTMF Cue Tones

Traffic &Billing Head End System SkyVision Reporting MDS Traffic & HE Video (SCSI) SplicerSplicer Billing Schedule Schedule SkyVision Head End GUI File Distribution SkyVision Kit SCTE/30 (TCP/IP) HQ Video (DVB-ASI) SkyVision SplicerSplicer Encoder Splicer Headquarters

Digital Broadcast w/ SCTE 35 Cue Packets

Satellite Broadcast

Figure 1. A schematic diagram of nCUBE’s SkyVision digital ad insertion system deployed for both analog and digital cable. Analog MDS switch decoders and digital DPI splicers shown in a headend with nCUBE’s SkyVision HE ad server. Ad insertion servers and switches in multiple headends are managed centrally with the SkyVision HQ and MPEG encoder.

Today’s Video on Demand workflow is not too dissimilar from ad insertion and both workflows can be automated or With VOD, subscribers select programming performed manually, as the situation demands. from menus or their programming guides and An example of today’s VOD system is depicted request and control video streams. VOD in Figure 2, below. In Figure 2, the nABLE HQ manages the day-to-day workflow and • Alert of missing content and asset monitoring functions required to manage metadata errors, and system failures multiple VOD systems from a single location. • Remove expired content The key functions are: • Publish asset metadata to VOD menus • Authorize subscribers and post charges • Import asset metadata and content files to subscriber management systems from asset distribution systems (billing) (catchers mitts) • Monitor buys, sessions and bandwidth, • Distribute new content to headend VOD and report data systems

Catcher's Mitt

Asset Metadata & ContentFiles Head End

nABLE Reporting & nABLE GUI Authorization GUI nABLE RT nABLE Session Set-up HQ SMS / Video SMS / Billing Billing / ContentFile Billing n4x Server STB Authorization Distribution SplicerSplicer Stream Headquarters Control VOD System

Asset Metadata

Figure 2. A schematic diagram of a typical deployed nCUBE nABLE VOD system. The n4x video server and nABLE RT session and resource manager are shown in a headend. (On demand application server and client that provide the subscriber user interface are not explicitly shown but the functionality is implied in HQ, RT and STB since actual configurations vary depending on the deployment.) VOD systems in multiple headends are managed centrally with nABLE HQ.

Before getting into the specifics of VOD ad headends prior to each title’s availability start insertion it’s important to understand the date, which are contractually set by the content process of requesting a title and setting up a provider and included in the asset metadata. VOD session since this is when ads are selected The metadata is validated and published the on- and inserted in and around VOD content, in this demand application’s VOD menus and/or implementation discussed below. Prior to digital cable programming guides where selecting a title, the asset metadata and content subscribers can navigate and make selections. files are imported from the catcher’s mitts. The The metadata includes content provider, start content files are typically encoded at MPEG2 date, end date, title, category, description, 3.75 Mbps (CBR SPTS) as specified in rating, actors, price and more. Other files such CableLabs VOD content specification. They as DVD cover graphics may accompany a title are distributed to the VOD servers in the and are also retrieved from catcher mitts, and published as part of the menu information seen has a capacity of 240 VOD streams. If each by subscribers along with the metadata. VOD stream can serve 10 digital subscribers, Content files are removed once their end date is then a 24 MHz service group can serve at least past. 2,400 VOD enabled digital homes. In a market with 240,000 digital homes, there would be 100 Once a title is selected, the on-demand service groups. The homes sharing a service application in the set-top box (STB) initiates a group are typically clustered in the same session set-up request to the VOD system in the geographic region making service groups headend. The commonly used protocol choices relevant to geographically targeted advertising, implemented for session set-up (including as discussed later. A service group with 2,400 teardown of a session) today are RTSP and digital subs represents about 7,000 cable homes DSM-CC. In the session set-up request at 34% digital penetration and about 10,000 message, the STB communicates the requested homes passed. Assuming that there are 500 title’s asset ID, and the STB’s MAC address homes passed for each optical node in an HFC and its service group. The MAC address is plant, a service group consists of twenty optical unique to each STB. The service group nodes all sharing the same VOD video RF identifies the population of STBs, which share spectrum. As digital penetration grows and as the same group of 6 MHz VOD channels VOD usage increases, service groups will (QAM modulators and RF upconverters) each either get larger in MHz or smaller in optical with a capacity of 10 VOD streams, or 10 nodes or both. simultaneous one-to-one TV sessions. Service groups are essential in economically setting up Using the STB’s MAC address the VOD VOD sessions and streaming. They are also system checks the household’s authorization relevant to VOD ad insertion as explained later. for the requested content and posts any charges. Based on the service group passed to the VOD In a cable plant there is a fixed amount of system from the STB, the VOD system’s RF spectrum for video equal to a fixed number session and resource manager determines the of 6 MHz video channels. Analog cable TV best way to route the video stream and allocates requires a 6 MHz video channel for each cable the required bandwidth (e.g. 3.75 Mbps) from network broadcast on analog. With digital the video server and through the digital cable cable TV, ten or more cable network broadcasts plant. Then, using the asset ID, it instructs the share a single 6 MHz video channel. Digital video server to stream the requested content. and analog cable networks (e.g. CNN) are Using the session set-up protocol (RTSP or broadcast through the entire cable plant and are DSM-CC), it provides the STB with both the available, with varying service tiers, to frequency of the 6 MHz channel (RF channel subscribers simultaneously. For VOD, of the QAM modulator) and the MPEG2 however, broadcasting all simultaneous program number carrying the requested video. viewers’ VOD sessions through the entire cable The STB tunes to the frequency and program plant requires too much capacity and is not number and the viewer sees the stream. The economically viable. Therefore, each 6 MHz viewer can then pause, fast forward, rewind, channel assigned to VOD is assigned to a stop and resume the stream. When doing so the service group. Each service group serves a on-demand application in the STB particular subset of the VOD capable homes in communicates a command to the VOD system any given deployment. Service groups can be using either RTSP or DSM-CC protocols. any size but 24 MHz or four 6 MHz channels each are common. A 24 MHz service group VOD Ad Insertion video recording (nPVR) succeeds with ad- supported programming. VOD ad insertion There are two primary applications of VOD creates a viable business model for content ad insertion applications: 1) playlists, and 2) ad providers’ ad sales operations and may be part replacement. With playlists, ads are inserted of the business model that encourages cable around VOD content such as movies, TV networks to contribute content. programming and other on-demand content. When ordering a title, such as a movie, the Technically, both playlists and ad viewer sees ads prior to the movie itself. Ads replacement are nearly identical. One main may also be present at the end of the title and difference is that playlists can be applied to even after resuming from pause. Playlists are content that may or may not have pre-existing important for advertisers as well as for local breaks and embedded ads in those local promotional purposes for cable networks and breaks when it is loaded onto the VOD server. cable operators wishing to promote other Ad replacement, on the other hand, specifically products. Similarly, thirty-second ads can be applies to those local breaks and replacing sold and inserted ahead of and after their VOD those embedded ads. programming. In order to insert ads in and around VOD Today, VOD programmers and cable content, a number of operational and technical operators edit promotional ads and paid ads details must be implemented: into the beginning and end of their VOD content making it part of the same content file. • Organizational responsibilities, If they choose to change the ads for a title, the architecture and workflow entire program must be re-edited and • Traffic & billing system configuration redistributed. Playlists, however, allow the ads and back office integration to change dynamically on the VOD server and • Detecting local breaks therefore do not require editing in advance or • Inserting ads in VOD content redistributing the entire program. Only the new • Targeting ads by geography ads need to be distributed. • Household addressability in the future

Ad replacement allows ads to be inserted in ORGANIZATIONAL RESPONSIBILITIES, local breaks in VOD programming while it is ARCHITECTURE AND WORKFLOW streaming to a household. Ad replacement is similar to playlists in the way video segments Organizationally, different people are can be sequenced and it leverages much of the responsible for ad sales and cable system same technology. operations. With VOD ad insertion, where the VOD system is also the insertion system, each For playlists and ad replacement to work groups’ roles and responsibilities may be there must be local breaks in content and there unclear. The best way to manage this is for ad must be a way for local breaks to be detected. sales personnel to manage the workflow for Very little VOD content today has local schedules, ads and logs just as they do today commercial breaks or any breaks at all. Many but treating the VOD system as the ad insertion cable operators believe that that will change system along with their traditional ad insertion over the next few years as the number of cable systems. However, cable system personnel programmers supplying VOD content grows. should continue to own and manage the VOD This is especially true if network personal system and the VOD programming. This is very similar to how DPI ad splicers are dealt With this organizational and architectural with today. The splicers insert the digital ads arrangement, ad sales operations remain into digital programming but have another relatively unchanged. Using their existing ad function with respect to the cable system. management application in their HQ, they still: Simply put, ad sales personnel should continue to manage ads and cable system personnel • Import traffic schedules from T&B should continue to manage the VOD system • Encode and distribute new ads and itself, including storage and stream schedules to headend insertion systems requirements. • Alert missing ads and system failures • Remove expired ads Figure 3, below, shows an architecture for • Monitor breaks and insertions VOD ad insertion in which both the ad • Retrieve verification logs and export to management and VOD management systems T&B and headquarter facilities remain unchanged from today’s ad sales and VOD.

Traffic & Billing Head End System

Reporting SkyVision Traffic & SplicerMDS Billing HE Splicer SkyVision Schedule Schedule

GUI File Distribution Head End SkyVision Kit HQ SkyVision SplicerSplicer Encoder Splicer Headquarters

File Distribution Ad Sales Operations

Cable System Operations Schedules Reporting Head End

nABLE Reporting & nABLE GUI Authorization GUI nABLE RT nABLE Session Set-up HQ SMS / Video SMS / Billing Billing / Content File Billing n4x Server STB Authorization Distribution SplicerSplicer Stream Headquarters Control VOD Asset Metadata & System Catcher'sMitt Content Files

Asset Metadata

Figure 3. nCUBE’s VOD ad insertion architecture and management responsibilities with nCUBE’s SkyVision ad insertion system and nABLE VOD system and n4x video server. Ad sales operators view VOD as another inserted and manage ads, schedules and logs similar to traditional insertion systems. Traffic & Billing System Configuration and • VOD regions (e.g. East and West) Back Office Integration • VOD categories (e.g. movies, SVOD) • VOD playlist and VOD ad replacement Today’s T&B systems and the existing • VOD titles (e.g. Larry King Live) schedule format could be used for VOD ad • VOD content provider (e.g. ESPN, insertion initially. However, new formats and Warner Home Video) interfaces will be needed eventually. The schedule interface is a flat ASCII file today. Each virtual network can be scheduled The T&B system creates a daily schedule for independently of the others and do not share each cable network in which ads are being inventory. With virtual networks, schedules inserted. The inserter uses that day’s schedule can be created which vary by region, category, for encoding ads, distributing ads and playing title, content provider and more. ads in the correct program or day part. The schedule includes break windows and ad names Since all VOD virtual networks are part of (a.k.a. spot IDs) and ad positions in each break. the same physical VOD systems, the total number of impressions is the sum of the From a T&B perspective, VOD does not impressions of each of VOD virtual networks have a fixed number of breaks in a fixed configured. An impression report from the number of cable networks. With traditional ad VOD system for each virtual network assists in insertion, the number of breaks and their break predicting inventory for each. Day parts could lengths are configured into the T&B system, in be used in T&B and for impression reports, but advance, for each program and day part of each initially a 24-hour day part probably makes cable network. That way the T&B systems and sense. In the T&B schedule, all ads would the ad sales organization can track the number occur in the same 24-hour break window each of avails sold and the inventory available. With day. VOD, there are not a fixed set of avails to configure in advance; nor are there specific day Once virtual networks and day parts are parts. defined, and avail inventory estimates are entered, the ads can be priced and sold and The best method to create and track orders can be entered into the T&B system. inventory for VOD in T&B is to estimate the The current T&B schedule lists each ad to play number of sessions, and hence impressions, during each day for each network, or virtual expected in the future based on historic session network in this case. That could be quite a long reports from the VOD system. The estimate list if there are one or just a few virtual can be refined over time with regular reporting. networks and a lot of playlists and ad replacement opportunities. This is an area of Instead of cable networks, VOD virtual future optimization for T&B interfaces. Instead networks will need to be defined in the VOD of repeating ads, specifying the frequency of system. The T&B system will publish a daily occurrence for repeated ads would be more list for each VOD virtual network just as it does efficient and real time interfaces would be an for each cable network today. The simplest even better solution. case is to create a single virtual network (call it VOD1) for VOD. This may be sufficient in the Once the schedules are published, the VOD beginning. More sophisticated virtual networks ad insertion system plays and logs the ads for which could be implemented, and in any reporting and billing. With VOD today, each combination, include: session is logged in the VOD system’s database for later reporting. Advertising content on • Viewer set-top ID (MAC address) VOD is logged just the same. These logs may • Viewer account number be returned to the billing system and used for • Viewer service group billing and reporting to advertisers. The billing • Session date system can simply count the ads played against • Session start and end times the ads ordered and determine if the contract • Session stream activity and times (e.g. was met. If not, it can schedule ‘make goods’ pause, fast forward, resume, etc) in the future in which it schedules enough additional ads to fulfill the commitment (sold By combining data from the VOD system for each virtual network) in the future within with other quantitative and qualitative data, the flight of the contract. Alternatively, it’s advertisers and ad agencies can learn about conceivable that VOD ads can be sold on a per their viewers as a group, and cable operators impression basis in which each time the ad can use the data for billing purposes. Nielsen’s plays the advertiser can be charged. involvement in VOD should help make ad agencies and advertisers more comfortable with The biggest billing issue with VOD ad the medium and its value in the future. insertion is determining the value of the ads in the first place. VOD is a new medium for ad Cable operators could use the STB ID and sales managers, and advertisers and their account number to obtain viewers’ home agencies are used to TV measurement and addresses and their zip codes. Zip codes are traditional delivery methods. A number of more specific than service group and yields reports will be needed from the VOD system more accurate information about viewers when for use by cable ad sales operators to show their combined with qualitative research data. A customers what was delivered. Today’s T&B number of market research databases and systems and ad agency back office systems applications, such as Claritas, provide may not make much sense of VOD’s one-to- qualitative characteristics about geographic one impression data given that they were areas by zip code, which enhances the VOD designed for tradition TV advertising which is and Nielsen data. Further, numerous databases based on reaching a certain size and type of TV provide very specific information about audience and is very much based by Nielsen’s particular households which can be generalized measurement data. and made anonymous in such a way as to be permissible for use with advertisers when However, the data collected by VOD combined with the VOD data. systems shows very detailed, exact viewer information about every individual impression. Privacy law prohibits cable operators from And, when combined with other research and sharing data about specific viewers, without measurement data presents a detailed view of their permission, with third parties. However, the audience that can be inspected down to generalized qualitative and quantitative individual households. The data available from information about the characteristics of groups the VOD system for each ad in each VOD of viewers rolled up and made anonymous, can session includes: be shared with third parties. (Cable operators should check with their legal counsel regarding • Name of advertiser privacy law.) • Name of ad • Spot ID of ad content viewed Detecting Local Breaks For ad replacement, using SCTE 35 cue packets, the standard is already in place and Before local breaks can be detected, they starting to be used in traditional ad insertion. need to be there in the first place. Breaks may The cue packets signal breaks through MPEG be created differently for playlists and for ad packets multiplexed into the program stream. replacement. Playlist breaks may be defined by In VOD, cue packets would be present in VOD content providers or by cable operators or content loaded onto the server and used to by both. Ad replacement breaks, however, are identify local break positions and insertion always defined by the VOD content provider. splice points. The VOD system may use cue Both could use cue packets and metadata to packets in real-time for ad insertion or may use define break positions and lengths. It is likely them during content ingest and translate them that playlist breaks would use metadata where to a proprietary internal metadata format or the as ad replacement breaks would use cue standard future metadata format so that the packets. This is because it is much simpler to VOD server does not have to filter for cue use metadata when creating ad avails before packets in real time; it can use metadata just as and after content by either programmers or it does with playlists. cable operators. In addition, CableLabs is considering support for playlists in the next Inserting Ads in VOD Content release of the VOD metadata standard which today’s VOD content providers, cable operators In order to support playlists and ad and VOD vendors all support. Using metadata replacement, VOD systems need a way to and supporting playlists is likely to evolve first assemble multiple MPEG video files and need because of the industry momentum as well as to be able to switch in and out of MPEG video the fact that today’s content does not have local files seamlessly from viewers’ perspectives. breaks. Since ad replacement applies to nCUBE’s VOD video server leverages content, which already has local breaks in it, it nCUBE’s Logical Content technology to is more likely to have cue packets already in dynamically assemble video files for playlists the content when loaded onto the VOD server. and ad replacement in video streams in real- time. nCUBE’s Logical Content technology For playlists, the metadata needs to identify supports seamless streaming of multiple and be applied to: MPEG2 files playing back-to-back with zero black frames and with fast forward and rewind • Content provider, category, and title, across the boundaries. Logical Content also hierarchically (e.g. all new releases) allows MPEG2 video files to seamlessly switch • Break position (e.g. before and after) from in and out splice points switching from • Break length (e.g. 60 seconds) one MPEG2 video to another at any MPEG • Break owner (e.g. network, operator, group of pictures (GOP) boundary in real-time marketing, shared, etc.) while playing from the video server with zero black frames. The files are never compiled into The playlist metadata can be implemented a single static file but are assembled during by content providers in advance as part of what streaming for each viewer’s session. That way is pitched to the VOD system. The metadata an ad can be changed dynamically for can also be augmented locally by the cable targeting, addressability or new schedules. operator after it is pitched using the VOD asset Logical Content is illustrated in Figure 4, management system. below.

Content As Seen By Subscriber M N O A B C X Y Z G H

Old Ad

A B C D E G H M N X Y ESPN Sports Center VoIP Promo Kraft Ad

Figure 4. nCUBE’s Logical Content technology for dynamic VOD ad insertion with playlists and ad replacement applications.

To create playlists or perform ad 3. Choose the proper T&B schedule based replacement, the VOD system needs to perform on the virtual network (e.g. action the following steps during play out after category in East San Francisco). receiving the VOD session set-up request and requested asset ID (which is in the session set- 4. Select the next ad(s) on the T&B up message): schedules (or start with the first ad on the schedule) so that the ad lengths add 1. From the asset ID, determine break up to the break length. positions and break length specified in the asset’s metadata and synchronize 5. Play the ad(s) at the proper splice point the breaks with the stream position using Logical Content technology and (assuming any cue packets were return to programming after break translated into metadata when the title (unless the break occurs at the end). was loaded). 6. Log the ad verification and other 2. Determine the virtual network based on reporting data (e.g. advertising, ad the criteria discussed in the T&B name, time, etc.). section, above (e.g. category selected, region where STB is located, etc.). Lastly, it should be noted that VOD content is encoded at maximum of 3.75 Mbps with AC-3 audio as specified by CableLabs and that ad sales typically encodes at 4 to 6 Mbps and may or may not have AC-3 audio (DPI requires the VOD system in the session set-up request AC-3). Although VOD systems can message. The service groups represent accommodate multiple rates, ads at higher bit geographic regions consisting of about 2,400 rates would require more VOD bandwidth than digital subscribers based on the example in the operators have budgeted in their plant design. section discussing today’s VOD system, above. Using service groups, an ad zone can consist of Targeting Ads by Geography as few as 2,400 digital subs although in practice the zone may consist of multiple service groups Since a single VOD server may serve a large combined together. Once the service group is geographic region, it may make sense to select identified, the VOD system needs to associate different ads depending on the viewer’s it with a zone. The VOD system must location. It is very common for ad sales hierarchically list the zones associated with organizations to zone their operations today for each service group since smaller zones may be this purpose. There are two primary nested in larger zones. The VOD system then applications for targeting ads to different needs to choose the zone. This could be done geographic zones within a market: 1) some randomly or sequentially. advertisers, such as a dry cleaner, are only interested in marketing to a particular part of Targeting ads to geographic regions can be town and do not need to advertise to a wider accomplished with VOD ad insertion and geographic region, and 2) some advertisers, today’s T&B systems; however, the application such as an automobile manufacturer, want to pushes today’s T&B systems beyond their advertise to a wide geographic region but want intent and it is cumbersome. Newer T&B the ads themselves to be different in different systems, or enhancements to current systems, parts of town. This second approach is branded will surely be required to take full advantage of AdTag and AdCopy by Adlink in the Los VOD ad insertion and targeting. Angeles market, and others are deploying it now as well. Lastly, regions, or zones, will be Household Addressability in the Future defined hierarchically as well. Small zones can make up larger zones and the larger zones can Removing the geographic constraints of make up an entire market. targeted advertising and addressing ads to specific households dramatically changes the Targeting ads to zones can be dynamics of TV advertising. Household level accommodated in VOD ad insertion by addressability combined with the one-to-one determining the viewer’s location and matching TV sessions of VOD provides an ideal ad it with a schedule for that region. In addition to medium. Advertisers have the potential to the process outlined in the previous section deliver their message to households in which about inserting an ad, the VOD system needs to their product or service is most relevant (e.g. determine the viewer’s region prior to step 3 >$100K income, with kids). One house may and then use the regional information to select receive a diaper ad and their neighbor may the proper schedule for the proper virtual receive an ad for heart medicine. network. Keep in mind that virtual networks can be defined for regions as discussed in the Advertisers can also measure the delivery to T&B section of this document. each set-top. Household addressability with VOD ad insertion could move TV advertising One way to determine the region is to use towards the highly lucrative direct mail the VOD service group information passed to business in future. During a VOD session request, the VOD CONCLUSION system receives the STB’s unique MAC address, which can be used to select the ad that Local ad insertion is evolving to include best matches each household’s lifestyle. VOD. This requires support for VOD ad Existing T&B systems fall far short of insertion in both VOD systems and ad insertion supporting this application. Several companies systems. Local ad sales, cable marketers and have developed addressable advertising content providers can take advantage of VOD systems and databases for tracking household ad insertion and both playlists and ad lifestyle information as part of or as extensions replacement can be used to create new to T&B for household addressable advertising inventory. VOD streams are unique, one-to- with DPI system. Using these new addressable one TV sessions with each household giving it advertising systems, the VOD system can tremendous potential for targeting and request the ad to play for a particular STB in household addressability. real-time through a new software interface. The format for a new real-time T&B interface Operationally, VOD advertising should be is not yet standardized but is commonly an extension of today’s ad sales operations and understood as being needed for this and other has much of the same workflow, monitoring applications. and general ad sales responsibilities. Existing T&B systems can be used in the beginning but The addressable advertising systems can use will need to evolve in support of the new a variety of data sources to determine a medium and make the most use of one-to-one household lifestyle profile including: TV. Methods for detecting and describing breaks need to be agreed to and standardized so • Qualitative data for neighborhood that content may flow across different vendors’ characteristics systems and different cable operators. • TV viewing including VOD (quantitative data) VOD ad insertion appears to be unique and • Household credit card usage advantageous for advertisers and ad sales • Grocery store shopping behavior operations in that it has the potential to: • Use local breaks in VOD content Based on a variety of factors, theses systems similar to traditional local breaks on can determine a household profile and can broadcast cable simply tell the VOD system the ad to play. The • Create new local breaks before and after VOD ad insertion system still needs to know in VOD content advance what ad content is needed in order to • Recapture digital viewers watching distribute the ads to the VOD servers, monitor VOD and not watching broadcast cable ads and breaks and report back logs. However, • Support geographically targeted and today’s flat file daily ad schedule currently household addressable ads imported for T&B systems goes away. In • Provide 100% measurable advertising addition, verification logs would likely be data returned in real-time over a new interface as well. VOD advertising and other advanced advertising technologies and business are unique to cable and should be fully exploited. nCUBE is a leader advertising insertion, VOD and advanced advertising today. ALL-DIGITAL: WHETHER AND HOW TO GET THERE Nick Hamilton-Piercy Rogers Cable Inc. S.V. Vasudevan BigBand Networks, Inc.

Abstract THE CASE FOR ALL-DIGITAL

This year witnesses the first launches of There are three basic drivers most “all-digital networks” by major cable commonly cited by operators as their operators. These initiatives can free business motivations for eventually abundant bandwidth to support hundreds of replacing all current programming additional TV channels, faster Internet distributed in analog format with all-digital service and more rapid launches and scaling content. These drivers cumulatively promise of advanced video services such as video on- to save costs, increase per-subscriber demand and high definition. They also usher revenues and enhance cable’s in less expensive set-top boxes and other competitiveness. They are: subscriber devices, based only on digital tuning, and have promise for delivering 1) Bandwidth optimization: Reclaimed better video quality. analog spectrum will be used by bandwidth-hungry HDTV, VOD, and However, the path to the all-digital other advanced video and data services. network does not happen with a simple flip Analog broadcasting currently occupies of a switch. Implementation can be quite more than half of all cable spectrum, so disruptive to subscribers, as each one will this saving will significantly alleviate now need to have every one of their the capacity constraints impeding cable television sets and possibly VCRs connected operators’ abilities to scale up offerings to a digital set-top box (unless these devices across the full range of these services. are digital cable-ready). This scenario has the potential to become an operational 2) Set-top box cost reduction: Removing disaster, to say nothing of the imposition the analog tuner and conditional access many subscribers are likely to feel with from today’s hybrid analog/digital specific devices forced everywhere they boxes significantly reduces the cost and want to watch television. complexity of the devices. Expanding this concept to today’s dual-tuner Consideration of all-digital cable digital video recorders further expands certainly has merit, but methods and savings with the elimination of two alternatives should also be weighed. There analog tuners as well as the MPEG-2 are other emerging techniques for encoder chips required to digitally write conservation of bandwidth that can be the analog video to disk. employed before, in parallel with, or instead of going all-digital. And if all-digital is 3) Improved picture quality: Although the pursued, there are several options to level of compression utilized can make minimize the cost and disruption of the claim dubious, competitive service achieving it. providers like direct broadcast satellite (DBS) television have used “all-digital”

1 in marketing campaigns to claim DIGITAL SIMULCAST superior quality over cable signals. The quality achieved in digital programming Digital simulcast, simply put, is the is determined by the quality of the replication of the analog tier as digital encoder and how much bandwidth is programming. Most systems, especially allocated per stream, but converting to those upgraded to 750 MHz in the recent all-digital could provide sufficient years, broadcast at least 80 channels of capacity to in fact make an overall analog programming (basic and enhanced improvement in picture quality, basic tiers) to their customers. Using including expanded HDTV carriage. advanced RateShaping bit-rate adaptation Further, VSB-NTSC analog technology, these programs could be programming distributed by cable can encoded and statistically multiplexed into as have poor display quality on large- few as six 6MHz channels, or within 36 screen digital televisions, especially MHz of spectrum. Although having noticeable when tuning from digital upgraded to 750 MHz does not guarantee services to analog services, further that there is sufficient spectrum, especially reinforcing subscriber perception with the contention of emerging services problems versus DBS. such as VOD and HDTV, the amount of spectrum required can usually be attained These drivers are indeed persuasive. with some bandwidth engineering. Better picture quality would help cable retain and gain subscribers in the churn There are several perspectives regarding battle with DBS and other video service the optimal topology for digital simulcast alternatives. Likewise, more innovative programming distribution, from one of pure services available on a bandwidth- centralization on a national scale, to a very optimized plant keeps subscribers distributed methodology. In either case, the interested while driving incremental operator’s primary goal is to replicate the revenue gains such as on-demand program entire analog tier in digital form and utilize orders. The lower costs associated with a broadband transport networks such as more efficient plant, as well as with less Gigabit Ethernet to distribute this expensive subscriber devices, also of programming to the hub or zone level of the course contribute to an enhanced bottom network. This transport technology will line. require that an adaptive switch or router be used to translate some, if not all, of the But as stated above, with the benefits, all- digital programming from an ASI transport digital conversion brings challenges as well. protocol to a unicast IP protocol. In addition to massive churn of subscriber equipment, other current cable analog All digital feeds, including the insertable operations must be considered and primary network feeds, are delivered to the transitioned, such as the current multi-billion hubs/zones where they can be received by a dollar local advertising insertion practice of statistical multiplexer that supports Gigabit cable operators. There is increasing Ethernet input. All incumbent analog ad realization in the industry that all-digital insertion equipment would need to be conversion must be evolutionary more than replaced or upgraded to support digital revolutionary, and several alternatives can program insertion, and the multiplexer alleviate efforts and expenses. would splice the ads into the now-digital programs.

2 The multiplexer then duplicates all the programming streams, one will be subjected digital streams that will need to also be to RateShaping, placed into a multiplex and distributed as analog programming over the digitally delivered within QAM-modulated last mile to those customers who only spectrum; the other will be decoded to subscribe to the basic or enhanced basic tier, analog baseband, NTSC-modulated and or who have additional televisions with no combined onto the spectrum reserved for the digital set top box. For each split pair of analog tier.

Digital simulcasting allows all-digital conversion while preserving utility of analog customer premises devices by combining both analog and digital versions of basic and expanded basic programming, with transport and splicing performed digitally.

Once simulcasting is implemented, any that only subscribe to the basic or enhanced of the operator’s customers that are digital basic tier would continue to receive their subscribers (and therefore use a digital set- programming as they always have, but they top box or digital cable-ready television) can access the digital version of the same could now receive all of their programming programming with a very inexpensive set- digitally including the digital versions of top box that only includes a digital tuner. programming that is broadcasted in analog. The proposition of improved quality might They thus no longer require the expense of be sufficiently compelling for a portion of analog tuning in their set-top boxes and these customers to adopt the use of such a other television devices. Those customers set-top box. As more subscribers access the

3 digital versions of programming, legacy Local cable advertising could also be analog devices can be churned out of the more profitably operated by simulcasting, system to achieve an effective transition to regardless of whether all-digital cable is all-digital over time. eventually achieved. Cable operators currently derive multi-billion dollar annual Simulcasting is a viable and controlled revenues from local advertising, almost path towards the deployment of all-digital entirely on analog programming, but have cable networks, and also provides the cable plans for significant growth, as for one thing operator with ancillary benefits. In fact, their stake is a disproportionately small some of these benefits are already being share of all television advertising. Current seen as sufficient to motivate several cable and emerging digital splicing standards and operators to engage in simulcasting for its Gigabit Ethernet transport of content can own sake, whether or not an all-digital make the practice easier and more future is ultimately effected. economical, flexible and scalable. All insertable programming can be digital, Simulcasting enables lower cost digital including analog-sourced content that is video recorders (DVRs) by eliminating the encoded in the headend. Once in digital need for components that support analog form, programming can be centrally programs. With greater familiarity and groomed and multiplexed and distributed by easier availability through service providers, Ethernet transport to hub locations that more North American pay television correspond to the zones for insertion. subscribers are opting to own, and use, a Servers can be situated at those locations, DVR. 25% of all new DBS subscribers now and splice in the zone-specific advertising receive a DVR and this has been an area of on receipt of the cue tone in the transported competitive advantage. But the cable program stream. In simulcasting, at this industry has now introduced its DVR point an analog copy of the program, with offerings, which are being well received. In the inserted advertising, would be decoded large part due to the necessity of two analog and NTSC-modulated for plant tuners and two digital tuners to support transmission. There are alternative simultaneously watching and recording simulcasting and digital insertion models either programming tier, and an encoder to that can also be considered, and overall, the digitize and store analog content, the DVRs greater efficiencies of working with digital as provided by traditional cable set-top box content for transport, storage and splicing vendors are relatively more expensive than also indicates savings of millions of dollars those being used by the DBS providers. in a typical cable system. Some experts in the industry, though, estimate that cable operators could save as Simulcasting presents an overall much as $100 per set-top box if they, too, attractive means of gradually converting could provide a purely digital DVR. Using towards an all-digital cable network. It this potential saving, coupled with a set-top forestalls the disruption of a singular, box (sans DVR) that is also purely digital massive digital switchover in favor of more could provide another $50 in savings. This gradual means. Ancillary benefits in areas could save millions of dollars in a cable such as DVRs and local advertising can system that has hundreds of thousands of drive economic benefits that make homes passed. simulcasting worthwhile regardless of whether or when all-digital is eventually attained. Ironically, the goal of an all-digital

4 network is to liberate programming application that can be dynamically bandwidth consumed in its inefficient downloaded to any digital set-top box, the broadcast analog form, but the first step of network detects subscribers’ channel simulcasting begins the process by placing selections. Among programs designated to a an additional demand on cable spectrum. switched broadcast pool, only those The primary hurdle is to free sufficient requested from within a node or service spectrum for both digital and analog group are distributed within it. When a versions of basic and advanced basic subscriber switches to a locally unwatched programming, and several techniques can program, this is recognized, and a headend- accomplish this. based server dynamically allocates a frequency within the viewer’s area will now SWITCHED BROADCAST carry that program, and instructs an MPEG- aware switch and the set-top box of how to Like all-digital, switched broadcast transmit and receive this program. If a receives attention as a means of providing subscriber switches to a program already more programming and launching more being watched in the node or service group, services. Switched broadcast could be an the existing transmission is joined. Once all alternative to all-digital, or an aid towards subscribers in the area tune away from a achieving all-digital by freeing up the given program, that program’s session can spectrum required to implement analog- be dynamically discontinued and the digital simulcasting. Switched broadcast can bandwidth that was used now becomes even play a valuable role in the available for other switched broadcast infrastructure of an all-digital network once sessions. The mechanics of implementing it’s fully realized. switched broadcast are further explained in the paper Planning for and Managing the Switched broadcast combines broadcast Rollout of Switched Broadcast Services techniques of sourcing live programming presented by Time Warner Cable and along with on-demand techniques of BigBand Networks at the 2003 SCTE distribution. Through a thin client Conference on Emerging Technologies.

5

Switched broadcast is consistent with all-digital’s efficiency objectives, and can facilitate analog-digital simulcasting, by only carrying those digital programs being watched in an area (1 and 3 in the illustration above).

A major cable operator has completed a 50% of the number offered), then 94% of field trial of switched broadcast on a the time all requested programs would be contained pool of 10 specialized programs. provided, and over 99% of all demand In the analysis of channel requests during would have been met. These specific trial this trial, it was determined that the ten parameters are not a representative sample programs could have fit in the capacity for and the nature of statistics will certainly seven without suffering any blockage. If make the results even better with a larger they were fit in the capacity for six, then and more diverse set of programs than the 99% of the time all demand would be met 10 that are were used. Further detail on the whereas 1% of the time there would be one results of this trial can be found in the paper blocked program out of the six requested, Switched Broadcast: Statistics from the which means that more than 99.8% of all Field presented by BigBand Networks at the demand would be met. And if enough 2003 SCTE Cable-Tec Expo. capacity for 5 programs were provided (i.e.

6 The initial results of the limited trial were Switched broadcast video services encouraging enough to warrant larger scale can be used to offer many trials with more programming and larger more broadcast programs, using subscriber bases, which are being conducted less bandwidth, than traditional as this paper is being written. While final broadcast services. A typical 750 analysis is not complete, the statistical MHz cable plant could theoretically indicators are that among the few hundred offer over a thousand broadcast digital programs to subscribers, programs typically digitally broadcast, some compared to a few hundred programs subset of the most popular programs, using traditional broadcast. perhaps approximately 100, could be conventionally broadcast and the rest could The benefits of switched broadcast can be be most economically provided by including amplified by extending the practice of them in a switched broadcast pool. Based on dynamically allocating bandwidth and knowledge of subscriber viewing behavior, associated resources to all services according this programming can most likely be to subscriber use, not just the digital provided at a bandwidth capacity reduction broadcasting pool. Plant protocols such as on the order of 50% with virtually no 256QAM modulation apply across digital blocking of programs. Furthermore, with services, so the same modulator and channel switched broadcast, cable operators could that is primarily used for digital broadcast dramatically expand the selection of programs at a time when they are popular, programming they provide to include such as a weekend afternoon with many live worldwide international fare, increasingly sporting events, could be used primarily for niche programming topics, live events in VOD content when that is popular, as in the sports, performed entertainment, education, same day’s evening, when subscriber professional gatherings and other genres. preferences switch to saved entertainment These are similar benefits to those of all- content such as movies, and likewise for digital conversion. Switched broadcast could interpersonal communications (including be considered as an alternative or a video conferencing) or business data services complement to the all-digital network. when their traffic demands peak.

Both carrying programs only in digital While the long term benefits are enticing, form and switching them dynamically so in the near term, switched broadcast can that they only use bandwidth where and assist in facilitating digital simulcast when required provides dramatic bandwidth deployments. The basic and advanced basic efficiency gains. This derived capacity programs are most likely too popular to expansion future-proofs the network for the sensibly fit in the switched broadcast pool, migration of all programming to higher but freeing the six channels that may be quality levels including HDTV, and the required to digitally simulcast 80 programs, migration to all on-demand availability. could be achieved by taking approximately the bottom third of digital programs and Confirming the bandwidth gain promise switching them. Switched broadcast should of switched broadcast, the paper Modeling conservatively achieve a capacity gain of Switched Broadcast Video Services 50% when applied to such programming, so presented by Cable Television Laboratories switching 160 digital programs should at the 2004 Winter Conference states: provide the bandwidth required to simulcast the 80 programs already carried by analog.

7 NEXT-GENERATION DIGITAL 256QAM modulation. MPEG-2 now TECHNIQUES exceeds a decade of useful service and more efficient video encoding techniques are Other trends in the handling and emerging. H.264/MPEG-4 Part 10 promises processing of digital video content are also to enable the carriage of as many as five worthy of consideration for their potential HDTV programs per channel at constant bit for bandwidth conservation. These can be rates, and presumably several more with used to further enhance the efficiencies from conversion to variable bit rates and switched broadcast and all-digital, and aid in RateShaping practices. However, the implementation of digital simulcasting. RateShaping of MPEG-4 encoded programs is unlikely to achieve the extent of RateShaping techniques leverage bandwidth savings possible with MPEG-2 as statistical multiplexing benefits for variable MPEG-4’s encoding is already highly bit rate video programs by observing efficient. Moving to a new video encoding cumulative bandwidth behavior, and when format or transmission format requires channel capacity is strained, selectively replacing digital access equipment, with the throttling back allocation to selected corresponding capital expenses. IP encoding programs in order to maintain service while and transmission schemes could be maximizing quality. RateShaping algorithms simulcast with MPEG-2 (and for some have progressed to gain 50-100% capacity programs analog too if digital simulcast is gains with typical multiplexes of 15 SDTV being used). Switched broadcast programs or 3 HDTV programs per 6 MHz architectures could also be applied so that an channel, or higher, with imperceptible MPEG-4 stream of a program is only compromise in picture quality if care is broadcast in an area when an MPEG-4 taken in selecting appropriate program subscriber requests it, and likewise for other content type in the multiplexed services. formats. The bandwidth efficiency gains of Current developments, including tools for emerging encoding techniques do provide more intelligent assembling of channel savings in requirements of equipment such groups, should further boost these as VOD storage, DVR storage, and efficiencies, but at some point algorithmic transmission equipment, since more limits will be reached for RateShaping of programs per channel means more channels MPEG-2 video. An advantage of per headend device. There are also benefits RateShaping is that it is MPEG-2 compliant, in extension of broadcast content to more necessitating no changes to headend, plant subscriber devices, including PCs, and in or customer premises equipment, with abilities to leverage other IP practices such algorithmic advances generally achieved by as flexible encryption for open conditional software upgrades to existing grooming and access and XML tags to profile subscribers, multiplexing equipment. services and other entities. Further bandwidth efficiency gains beyond RateShaping may be achieved by utilizing different coding formats than today’s conventional MPEG-2 video over

8 CUSTOMER PREMISES DECODING

Eventually an evolution to an all-digital distribution, likely with advanced video encoding, is inevitable. In the nearer term simulcasting may ease the transition and avoid forcing an accelerated conversion of customer premises equipment. At some point, the forced change in the home environment must be confronted. Television receivers tend to have 15 or more years of usage before they are discarded and most Encoding technique advances promise cable operators are loath to retire a customer substantial near-term efficiency gains as provided STB until well past its depreciated all-digital complements or alternatives. life. Most customer-owned television receivers and VCRs require an analog Like MPEG compression, QAM signal. modulation is a practice that will continue to achieve bandwidth efficiencies in Provisioning a digital set-top box for succeeding generations. Use of 1024QAM every analog cable client is complicated and would improve channel capacity for digital expensive. From a cost standpoint, cable content by 25% over 256QAM. Even greater operators would prefer to avoid equipping bandwidth efficiencies would be achieved if every television in the household with the used in conjunction with advanced encoding full functionality of digital tuning, such as MPEG-4. Incompatibility with recording, picture-in-picture, interactivity current subscriber equipment would, like and other features unless the customer finds advanced encoding, be disruptive but a the added functionality sufficiently migration is doable using the switched compelling to produce incremental revenue. broadcast stream approach as previously Even the provision of relatively low-cost described or through simulcast of formats if devices that only perform digital tuning can spectrum is available during the transition. be prohibitive if several are required for each home. Subscribers can get antagonized Techniques such as MPEG-4 encoding as well. From their perspective, a change is and 1024QAM ultimately dovetail well with being forced on them, perhaps placement of other trends related to all-digital to achieve a the CPE being incompatible with built-in network of much higher overall capacity furniture, and without any perceived with richer functionality available to increase in the service value proposition – subscribers. But, like other means to convert no added services, just the same to that they to all-digital, these come at some cost and were already receiving in analog. Some may effort, and their merits must be weighed in argue that the digital transition introduces this light. reductions in the quality of service as measured by a metric such as channel- change latency, which is historically slower

9 for digital programming. Offering additional might provide enough benefit to offset at services, such as VOD, and demonstrating least some of this issue. improved picture quality of all channels

Use of a network interface unit to decode digital video preserves existing analog devices within customer premises while allowing for the functionality and bandwidth efficiency gains of all-digital broadcasting and services in the cable facilities and plant.

Alternatively, migration to an all-digital decode programs and distribute internally in plant with analog homes or at least some analog form. This requires the provisioning analog devices in homes might more of decoding equipment only once per home economically be achieved if homes have a rather than at every device. The NIU device single point of digital to analog decoding would need configuration for digital service and analog video distribution is effectively transparency and to pass the 2-way maintained just in the home environment. interactive signals for services such as VOD, Much like how circuit telephony has been high speed data and voice over IP. The provisioned on cable plants, a video network concept could be extended to a shared NIU interface unit (NIU) could be placed at a in a multi-dwelling unit, gated community point of ingress to the home in order to or neighborhood. Such consolidation

10 reduces equipment and maintenance costs. CONCLUSION The distribution could be agile across multiple physical methods including Conversion to all-digital enables wireless to further alleviate expenses. significant business and competitiveness gains for cable operators, but is on its Decoding digital video at or near surface a daunting proposition. Several customer premises enables the cable techniques that leverage characteristics of operator to use all-digital in order to gain digital content, described herein, can effective capacity at the most constrained alleviate the process. Analog-digital part of the network: the coaxial cable simulcasting, switched broadcast, distribution plant. Alleviating that implementation of technical digital video environment is the most important advances, and decoding content at customer accomplishment towards avoidance of the premises are all tools for gradual attainment most expensive technique of securing of all-digital without significant service capacity expansion, namely embarking on disruption or expense spikes. By further rebuilds and upgrades. Securing implementing several or all of these sufficient bandwidth is much more alternatives, operators can enact an eventual economical at the destination points of the migration towards truly all-digital cable plant, where decoding to analog can be services, from content origination to supported. This practice could deepen subscriber consumption, while attaining penetration of all-digital to the many of the goal’s promised benefits along neighborhood, curb, home or device while the way. optimizing economics of subscriber retrofitting and minimizing disruption and annoyance for customers. As more digital televisions migrate into these homes and analog units cease to operate from age, the role of the video NIU becomes redundant, and all-digital the whole way through the BigBand Networks, Inc. network to the client device emerges as an 475 Broadway eventuality. Redwood City, CA 94063 650 995 5000 http://www.bigbandnet.com

11 ANALYSIS AND PREDICTION OF SET-TOP-BOX RELIABILITY IN MULTI-APPLICATION ENVIRONMENTS USING ARTIFICIAL INTELLIGENCE TECHNIQUES

Louis P. Slothouber BIAP Systems, Inc.

Abstract the limited computational resources available. In such environments, applications that oper- We present an Artificial Intelligence based ate flawlessly by themselves may wreak method for improving the reliability of soft- havoc in a system where limited memory, ware applications, especially in digital cable CPU cycles or network bandwidth must be TV set-top-box and other embedded environ- shared amongst several applications. Such ments. Initially a small finite state model of software incompatibilities can occur inter- the software system and all relevant applica- mittently and be difficult to trace. Often the tions is constructed to define all user input actual source of a fault is elusive and may not events and application states of interest. A be obviously related to the point of failure. small set of expert system rules is then defined that analyzes state transitions in testing data. The results of such inter-application fail- When these rules are applied to actual testing ures can be disastrous in a highly distributed data a quantitative measure of suspicion is service-oriented industry, like the digital cable assigned to all event transitions in the origi- TV industry, where one software failure might nal finite state model. Analysis of this anno- be replicated throughout all devices in a sys- tated model can then uncover the source of tem, rendering the service (i.e., TV) inoper- otherwise intermittent inter-application fail- able for millions of customers. Despite the ures. very real possibility of adverse interactions between applications, it is often logistically impossible or prohibitively expensive to test INTRODUCTION inter-application interactions prior to installa- No amount of software testing can guaran- tion in an actual deployment environment. In tee the quality of an application outside of the the case of digital cable TV applications, environment in which it is tested [4]. How- software vendors rarely have access to each ever, it is often prohibitively expensive or dif- other’s products, nor can they easily afford ficult to exactly replicate the software and the cost of the hardware and software hardware environment during testing that will infrastructure required by another vendor’s be present once an application is deployed. product. Laboratory environments provided This is especially true in some embedded en- by a system operator do help this situation, vironments, like those found inside most digi- but individual systems are often quite unique tal cable TV set-top-boxes (STB). in terms of hardware and software versions, the mix of deployed applications, and other Software applications designed to operate subtle but important differences. in such constrained computational environ- ments must provide a highly reliable, quality, Inter-application testing in an environment interactive user-experience while simultane- of many distributed, embedded devices —like ously coexisting with other applications — that found in a digital cable TV system— in- usually from multiple vendors— and sharing volves a normal installation of the application on a live system or lab. The actual testing STEP 3 - Gather test data from a number process is rarely fully automated because —at of randomized testing trials. least in the TV environment— application results are often visual, and human interpreta- STEP 4 - Apply the rules from step 2 to the tion is required to detect defects. Further, the testing data from step 3 to annotate the model services provided by such applications (e.g., from step 1 with certainty factors. TV programming) often have relatively long durations, so testing cycles may be prolonged. STEP 5 - Analyze the annotated model Given these limitations and the large variety produced in step 4 to deduce likely sources of of possible adverse interactions between ap- failure and problem user event sequences. plications, no practical testing plan can be ex- pected to uncover all problems [4]. Each of these steps will be discussed in de- tail below. We present a different approach. Rather than attempting to design and execute large, STEP 1 - The Finite State Model comprehensive testing plans our method gath- The first necessary element in this process ers test data from a series of random user 1 is the definition of a simple finite state model events, and employs a combination of Artifi- that abstracts the relevant system and applica- cial Intelligence techniques to automatically tion states along with the user input events analyze the data and deduce the likely sources that cause transitions from one state to an- of faults and defects. other. Initially, this model may be quite sim- The method we present has been employed ple, with only a handful of states representing successfully to identify the sources of inter- the applications involved. Later, once the application failures and to predict system reli- method deduces specific states as the sources ability during a recent software trial with a of failure, these problem states may be ex- leading digital cable TV operator. The exam- panded into more complex sub-models to fur- ples presented below are abstracted from that ther refine the source of the failures. Table 1 real world case study. below depicts a finite state model that will be used later in Example 1.

METHOD In this table each row depicts a state with the state name in the first column followed by The method we present is designed to de- the state transitions for the five user key-press duce the sources of software failures from the events named at the head of the columns. sequences of user actions that are likely to Events that have no effect in the various ap- induce those failures. It is composed of five plications have no transitions listed in the cor- distinct steps, as follows: responding rows of the model. Pictorially, this same finite state model can be depicted by the STEP 1 - Define a finite state model that directed graph in Figure 1. The “Power Off” abstracts the relevant software system, appli- state is the start state of the model, but no cations, and user events. states are identified as final states. The only final state is the “REBOOT!” or failure state, STEP 2 - Define a set of IF-THEN rules with associated certainty factors that identify 1 sequences of events and states that tend to Also known as a Finite State Automaton (FSA), lead to failure. Finite State Machine (FSM), or Deterministic FSA. [2] which is accessible from all other states, and nite state transitions, will assign a numerical is therefore not shown. value to all transitions that estimates the like- lihood of that transition is plays a part in an Given a finite state model of a system, test inter-application failure. For example, using results for that system may be represented by the finite state model of Table 1, one sample a sequence of (state, event, state) triples that rule might be:

Model User Events IF (transition doesn’t appear in a sequence) States Pwr Key A Key B Key Exit Key Guide Key THEN likelihood is -0.8. Power Off Watch TV Watch TV Power Off App1 IPG App IPG App Power Off App2 App3 Watch TV Watch TV App1 Power Off Watch TV IPG App Notice that the state transition: App2 Power Off Watch TV IPG App App3 Power Off Watch TV IPG App REBOOT! Power Off (PowerOff, Pwr Key, Watch TV)

Table 1. Simple finite state model for Example 1 appears in the testing sequence above. Thus, define the transitions from state to state the sample rule above would not apply to that caused by a sequence of user input events. For transition for that sequence. However, the example: transition:

(Power Off, Pwr Key, Watch TV ) (App3, Exit Key, Watch TV) (Watch TV, A Key, App1 ) (App1, Guide Key, IPG App ) is not found in the test data sequence, and (IPG App, A Key, App2 ) therefore it is highly unlikely that the transi- (App2, Exit Key, REBOOT! ) tion is involved in the failure. This reasoning is quantified and represented by the likelihood This depicts a sample transition sequence that value of -0.8 in the sample rule. culminates in failure. All sequences are as- sumed to start in the initial state (e.g., “Power A collection of IF-THEN rules defines a Off”) and end in the final, failure state (e.g., simple form of Artificial Intelligence expert “REBOOT!”). system, a programming paradigm that is par- ticularly adept at encoding and applying im- STEP 2 - IF-THEN Rules & Certainty Factors precise expert knowledge about a very narrow topic [1,3]. In this case, the rules assess the Next, a set of simple IF-THEN rules is de- likelihood that the transitions found in testing fined that, when applied to a sequence of fi- sequences play a part in software failures. Most real world expert systems contain hun- Guide Power Key Key dreds or thousands of rules, and require spe- Power Watch Guide Off TV Exit (IPG) Key cialized software language support. However,

A the expert systems defined in this paper are Key Guide Exit A Exit Key quite simple, containing only a handful of Key Key Key B Key Guide simple rules, and are easily implemented by Key conventional programming languages. Exit Key App 2 App 1 App 3 Similarly, the “likelihood” numeric values in rules are actually certainty factors (CF), a

Guide mechanism for quantifying uncertainty Key [1,3,5]. Unlike probabilities, which are often Figure 1. Graph of finite state model of Example 1 difficult to apply to real world situations, cer- tainty factors are quite easy to implement. In almost any type of failure. Two of the most addition, certainty factors can also quantify a important are: lack of knowledge, which probability values cannot. Certainty factors are real numbers RULE 1: ranging between -1.0, meaning total disbelief IF (s1,e,s2) does not appear in a sequence in a conclusion, to +1.0, which denotes total THEN CF(s1,e,s2) = CF(s1,e,s2) ⊕ -0.9 belief. Any value in between denotes some measure of uncertainty. For example, a cer- RULE 2: tainty factor of +0.5 means the corresponding IF a test sequence enters a state si N times conclusion is probably true, while a certainty THEN ∀j | (sj,e,si) is in the sequence factor of –0.9 means it is almost certainly not CF(s1,e,s2) = CF (s1,e,s2) ⊕ (+0.2 / N) true. A certainty factor of 0.0 denotes no knowledge either way (i.e., unknown) and is Because we are only interested in transi- used to initialize all CFs. tion sequences that end in failure, and we as- sume that the number of inter-application

To combine two certainty factors CF1 and problems is small, it is logical to assume that CF2, that are derived for the same transition a transition that appears in the test data may by two different rules, we use the following be related to that failure. Unfortunately, this formula [1,5]: deduction doesn’t help much, because many transitions that are not related to the failure CF1 ⊕ CF2 = CF1 + CF2* (1 – CF2) if both are positive may also be in the test data sequences. How- = CF1 + CF2* (1 + CF2) if both negative ever, by reversing that same logic, a transition CF + CF = 1 2 otherwise. that does not appear in the test data is most 1− min(CF , CF ) 1 2 likely not related to the failure. This is codi- fied in RULE 1, where the “most likely not One important reason for choosing cer- related” translates into a CF of -0.9. tainty factors to represent potentially inconsis- tent test data is that certainty factors are both RULE 2 is based on the supposition that a commutative and asymptotic [1,3,5]. The for- state that appears many times in a test data mer means that it does not matter in what or- sequence that ends in failure has more oppor- der we combine certainty factors, the same tunities to cause problems. A small CF value result is obtained. This implies that it does not (i.e., +0.2) is thus distributed amongst the matter what in order our rules are applied, transitions in the test data that exit from the which greatly simplifies implementation. The suspect state. asymptotic property implies that as new evi- dence is found to support (or discredit) a con- Other rules are more effective for certain clusion, we increase (decrease) the CF value types of failures. Multiple, different sets of incrementally. Fore example, if two distinct rules may be applied to the same test data se- rules both generate a strong belief in a given quences to deduce different types of errors. transition (e.g., 0.7 and 0.8) we will tend to Consider RULE 3, which is effective when believe somewhat more strongly (e.g., one application is suspected of starving an- 0.7+0.8*[1–0.7] = 0.94). The asymptotic other of a computational resource (e.g., a property also keeps certainty factors nicely memory leak). between -1.0 and +1.0.

Some of the rules that have been most use- ful to date appear to be generally applicable to RULE 3: may also be converted to the necessary transi- IF transition t appears in the test data tion triples, as long as the test data still repre- THEN CF(t) = CF(t) ⊕ (+0.5 * (1 - P(t)) sents the current system and applications. It is understood that the initial model may be quite In this rule, P(t) denotes the probability esti- abstract and simplistic, and that many of the mate any random transition will be t. While a actual states and events exhibited by the sys- true probability value would be difficult to tem and applications are not represented by calculate, an acceptable estimate can be de- the model. But the test data must be modified rived easily as follows: to fit the model, removing extraneous transi- tions if necessary. Let t = (s1, e, s2). Let Tfsm be the number of non-empty transitions in the finite state Ideally, no strict testing plan will be used model, and Ts1 be the number of transitions to drive the sequence of user events that are that lead to state s1. Finally, let Ts2 be the input to the tested system. Rather, a random sequence of user input events is preferred. number of transitions that leave state s2. We then define P(t) as follows: There are several reasons for this counterintuitive preference. First, a non-

Ts1 random test plan embodies an implicit bias P(t) = toward one or more a priori results. While this Tfsm *Ts2 is a good thing if the bias is in the right Finally, consider RULE 4, which helps to de- direction, test results would be useless if the duce the state in which an inter-application bias is in the wrong direction; important state failure originates, even though the actual fail- transitions might never be seen. Second, ure may occur many transitions later. because the rules apply to a domain of uncertain data, and are statistical in nature, we RULE 4: suspect that many of the most effective rules IF transition t is one of the last N transitions work best with random test sequences. in the test sequence Finally, in some of the supplementary THEN CF(t) = CF(t) ⊕ +0.5 probability analyses performed on the test results, the mathematics require randomized This rule is predicated on the assumption that test sequences. the detrimental situation precipitated by the Implicitly, all test sequences will begin in first adversely interacting state is severe the start state. For all practical purposes only enough to produce a failure soon after. For test sequences that end in a failure state are of example, if one digital cable TV application interest. Because most failures are intermittent corrupts the video heap in a set-top-box, a in nature, little useful information about fail- crash will often occur the next time something ures can be deduced from a sequence that changes on the screen. does not fail.

STEP 3 - Gather Test Data STEP 4 - Apply Rules to Test Data

The next step is to gather test data in the This step is a straightforward application form of state-event-state transition triples as of the rules from step 2 to the testing se- defined in the finite state model. The actual quences gathered in step 3. All certainty fac- testing may be performed at any time. Old tors are initialized to 0.0 before applying any testing results generated for other purposes rules2. As rules are applied the certainty fac- (4) States that actually fail tend to have a tors associated with the various transitions of high entry CF and a low exit CF, be- the finite state model (e.g., Table 1) are modi- cause the likelihood of failure de- fied accordingly. Once all rule processing is creases after passing through the state. complete, two certainty factors should be (5) Most importantly, states that are computed for each state (e.g., s). The first likely to be an original source of fail- combines the certainty factors for all transi- ure, but seldom fail themselves, tend tions leaving state s for another different state. to have a much lower entry CF than Similarly, the second certainty factor is a exit CF, because the likelihood of combination of the certainty factors of all failure increases after passing transitions that are entering state s from other through the state. This type of result states. Ignore transitions from state s back to is often very hard to find using con- itself. ventional testing techniques. STEP 5 - Analyze the Results EXAMPLES This is a very interesting step, and at the time of this writing, we continue to find new The examples in this section present results and interesting results in the data produced by derived via the method presented above. The this process. However, several observations test data used has been generated by probabil- are generally applicable to all annotated finite istic simulation software to simplify the prob- state models: lem for purposes of example, while retaining (1) Positive CFs (e.g., ≥ +0.2) suggest the salient features of real world test data. that a transition is somehow associ- Example 1: Memory Leak ated with a failure. (2) Negative CFs (e.g., ≤ -0.2) suggest The test data generator was constructed to simu- that a transition is not associated with late the system defined by the finite state model a failure. of Table 1 and Figure 1. Five applications of varying characteristics, including two system (3) Many transitions with positive CFs applications and three third-party applications merely provide a path connecting the were simulated. Each application had different original source state of the failure to memory usage patterns and requirements. Each the state in which it fails. These states exit from state “App2” to another state generated do not appear to contribute to the a small simulated memory leak of random size. failure. Along such transition paths The simulation was sensitive to both memory the transition CFs tend to increase. exhaustion and fragmentation, and would enter However, the corresponding entry the “REBOOT!” state whenever insufficient and exit CFs of the states along this memory was available for an application to func- path tend to be nearly equal. tion.

Test data files containing 500, 1000, and

2 2000 legal transitions were generated by the If substantial prior evidence exists to implicate one or more transitions, the initial certainty factors simulation, each file containing a variable may be initialized accordingly. Regardless of the number of test sequences that end in failure evidence small initial certainty factors are recom- states. mended. RULES 1, 2, and 3 were applied to these tions, CF values are distributed amongst the test files to annotate the states and transitions transitions on the paths from “App2” to the of the finite state model. Optimum results failing states. Additional evidence pointing at were obtained by the files containing 1000 state “App2” as the source of the failure are transitions. Files with fewer transitions re- the entry and exit CF values for that state. No- sulted in less clear distinctions between high tice that the entry CF is significantly less than and low CFs. Files with more than 1000 tran- the exit CF. From this evidence we conclude sitions tended to wash out the CFs, so that all that the memory leak originates in “App2”. values were approximately +1.0 or -1.0, thus obliterating valuable information about rela- Additional simulations were generated that tive certainties. assigned the memory leak to random applica- tions to remove any experimental bias. The The resulting annotated finite state model results were similar, clearly pointing to the appears below in Table 2. Entry and exit CFs offending state in each case. for the states are shown in Table 3. Example 2: Adverse Interaction Pwr Exit Guide CF(s, e) Key A Key B Key Key Key Another test data generator was con- Power Off 0.000 N/A N/A N/A N/A structed to simulate the system defined by the Watch TV 0.000 0.766 N/A N/A 0.649 finite state model of Table 4, below. In this IPG App 0.000 0.734 0.175 -0.604 -0.932 example, we expand the finite state model App1 0.000 -0.998 -0.982 -0.899 -0.980 App2 0.000 N/A N/A 0.984 0.977 from Table 1 to add additional sub-states and App3 0.000 N/A N/A -0.934 -0.964 transitions within the previous “App1” state. REBOOT! 0.000 N/A N/A N/A N/A The simulation then generated random fail- Table 2. Transition CFs for Memory Leak ures with a 25% chance whenever state “App1.2” was entered sometime after exiting from state “App3”. In other words, the system CF(s) CF(s) CF(s) simulates an inter-application failure with the ENTRY EXIT source of the failure in “App3”, but the actual Power Off 0.000 0.000 Watch TV -0.999 0.917 failure occurring eventually in “App1.2”. IPG App -0.999 -0.878 App1 -0.998 -0.998 Test data files containing 1000 valid transi- App2 0.734 0.998 tions were generated by the simulation. Each App3 0.175 -0.996 file contained many actual test sequences end- REBOOT! 0.000 0.000 ing in a failure state. Table 3. State entry and exit CFs Model User Events States Pwr Key A Key B Key Exit Key Guide Key Notice that the high CFs associated with Power Off Watch TV transitions out of state “App2” (i.e., +0.98) Watch TV Power Off App1.1 IPG App IPG App Power Off App2 App3 Watch TV Watch TV indicate that this state is almost certainly re- App1.1 Power Off App1.2 App1.4 Watch TV IPG App lated to the failure. The “Watch TV” and App1.2 Power Off App1.3 App1.1 Watch TV IPG App App1.3 Power Off App1.4 App1.2 Watch TV IPG App “IPG App” states also have several substantial App1.4 Power Off App1.1 App1.3 Watch TV IPG App CFs associated with transitions. These transi- App2 Power Off Watch TV IPG App App3 Power Off Watch TV IPG App tions tend to be on the path from state “App2” REBOOT! Power Off

to the actual failure transition. Because fail- Table 4. Simple finite state model for Example 2 ures happen in a variety of states and transi- RULES 1, 2, and 4 as defined above were CONCLUSION applied to this test files to annotate the states and transitions of the finite state model. Digital cable TV systems, and other simi- larly large, distributed computing systems The resulting annotated finite state model ap- present unique difficulties for application pears below in Table 5. Entry and exit CFs for vendors and system operators. Time and re- the states are shown in Table 6. sources for inter-application testing is often severely limited, even though the hardware Pwr Exit Guide and software resource constraints within these CF(s, e) Key A Key B Key Key Key computing environments make them suscepti- Power Off 0.000 N/A N/A N/A N/A ble to inter-application interactions and fail- Watch TV 0.967 0.151 N/A N/A 0.998 ures [4]. We present a new method that ap- IPG App 0.000 -0.220 0.179 -0.220 0.999 plies two simple techniques from the field of App1.1 0.000 0.999 0.924 -0.166 0.040 App1.2 0.000 0.998 0.147 -0.083 -0.083 Artificial Intelligence to the problem. Rather App1.3 0.000 -0.089 0.724 -0.111 -0.543 than generating large complex test plans and App1.4 0.000 0.398 0.998 -0.468 -0.169 lengthy testing programs, the sometimes in- App2 0.000 N/A N/A -0.104 0.106 consistent results from a relatively small App3 0.000 N/A N/A 0.529 0.266 quantity of randomly generated tests produces REBOOT! 0.000 N/A N/A N/A N/A sufficient information for a small A.I. expert Table 5. Transition CFs for Memory Leak system to deduce various points of failure. In particular, we have demonstrated how the of- CF(s) CF(s) ten asymptomatic sources of inter-application CF(s) ENTRY EXIT failures can be deduced. Power Off 0.000 0.000 Watch TV 0.999 0.998 This method has been applied to a real IPG App 0.998 0.999 world case in the digital cable TV industry, App1.1 0.563 0.999 App1.2 0.999 0.998 and has successfully discovered a previously App1.3 0.998 0.254 unknown memory leak in another vendor’s App1.4 0.916 0.998 application, and also identified an operating App2 -0.220 0.001 system anomaly that can cause exhaustion of App3 0.179 0.654 video memory and a subsequent system crash. REBOOT! 0.000 0.000 Table 6. State entry and exit CFs CONTACT INFORMATION Again, notice the higher CF values in tran- sitions for state “App3” the disparity between Dr. Louis Slothouber entry and exit CFs for this state. This evi- Chief Scientist, dence again correctly suggests that state BIAP Systems, Inc., “App3” is the original source of the inter- www.biap.com application failure. [email protected] REFERENCES

1. Durkin, J., Expert Systems: Design and Development, Macmillan Publish- ing Company, New York, NY, 1994.

2. Hopcroft, J.E., and J.D. Ullman, In- troduction to Automata Theory, Lan- guages, and Computation, Addison- Wesley, 1979.

3. Rich, E., and K. Knight, Artificial In- telligence --2nd ed., pp. 231-239, McGraw-Hill, Inc., 1991.

4. Schulmeyer, G.G., and McManus, J.I., Handbook of Software Quality Assur- ance --2nd ed., Van Nostrand Reinhold Publishing, 1992.

5. Shortliffe, E.H., and B.G. Buchannan, A Model of Inexact Reasoning in Medicine, Mathematical Biosciences, vol. 23, pp. 351-379, 1975.

CAPACITY PLANNING FOR CABLE HIGH-SPEED DATA SERVICES✝ K. R. Krishnan, Martin Eiger, Arnie Neidhardt, and Tamra Carpenter Telcordia Technologies, New Jersey

Abstract cable operators need a new set of algorithms to determine the capacity The engineering of cable networks for IP- requirements for providing acceptable based voice and data services presents new levels of QoS for the services [1,2]. In this planning challenges to cable operators. paper, we describe capacity-estimation Unlike the broadcast video services for algorithms for IP services and the traffic which cable networks have traditionally models on which they are based. The been designed, the traffic of these new traffic models offer a mathematical services is directed to individual description of the traffic of the IP services, subscribers. We describe algorithms for and the capacity-estimation algorithms estimating capacity requirements to determine resource requirements, at support IP-based services at acceptable various points in the network, to support levels of QoS as well as the traffic models traffic loads at specified QoS levels. The on which they are based. The algorithms algorithms account for the efficiencies that account for the efficiencies realized from are realized from the statistical the statistical multiplexing of independent multiplexing of independent traffic streams traffic streams of different subscribers. We of different subscribers for each service provide examples of their use in (multiplexing the streams of heterogeneous investigating various "what-if" scenarios. services is fraught with problems, as By combining the capacity estimation pointed out later, and is not attempted). We algorithms with methods for deriving the show by means of examples the use of the parameters of the traffic models from algorithms for investigating various "what- network measurements, one could create a if" scenarios, including the trade-off monitoring and planning system for between QoS guarantees and network provisioning IP services on cable resource requirements, and the projection networks. of network capacity requirements for various scenarios of demand growth. INTRODUCTION

The services considered in this paper The engineering of cable networks for are Voice-over-IP and High-Speed Data. IP-based voice and data services, as We present mathematical models for the defined in CableLabs PacketCableTM and traffic streams of these two services and DOCSIS specifications, presents new determine the bandwidth requirements (in planning challenges to cable operators, the upstream and downstream directions) since the traffic of these new services is at various points in the network to meet directed to individual subscribers, unlike specified levels of QoS. The mathematical the broadcast video services in traditional models characterize the random cable networks. To be successful in offering these IP services to subscribers,

✝ Copyright 2004 Telcordia Technologies, Inc. All rights reserved.

fluctuations of traffic rates in typical sessions NETWORK ARCHITECTURE of each service, and enable us to determine the capacity requirements as a function of The architecture of a cable network is traffic loads and QoS specifications. This designed for the efficient distribution of capability lends itself for use in a “what-if” broadcast television services. Figure 1 tool to answer various questions, e.g., what shows the typical two-level hierarchical QoS levels that can be supported on the structure of cable networks. For broadcast existing network, where would capacity television, the signal feeds enter the augmentation be required, either for the network at the head-end, from which the current demand or forecast demand. traffic is transported over a high-speed backbone ring to various distribution hubs We consider the application of our (each of which is the site of one or more algorithms to two examples. The first Cable Modem Termination Systems example shows the trade-off between (CMTSs)). Each hub sends the traffic to capacity requirements and QoS constraints. each of its subtending fiber nodes, which In the second, we consider the evolution of then distribute the signals to individual demands over a 5-year horizon and show homes over coaxial distribution networks. the capacity savings achieved by the proposed algorithms, which account for multiplexing efficiency, in comparison Head-End Ring with linear extrapolations that fail to account for the multiplexing gain. Head-End

In principle, the parameters of the Hub mathematical models of traffic can be Hub estimated from traffic measurements that

are collected at a fine enough time- Hub Hub resolution. However, such detailed measurements may not always be practical or economical in all networks. If the model parameters can be estimated from the routine operational traffic measurements in a network, then the capacity-estimation FN1 FN2 FN3 algorithms could become part of an integrated monitoring and planning system. Such a system will enable a network operator to plan and install new capacity before existing capacity runs out. The integration of parameter-estimation methods with capacity-estimation Figure 1: Architecture of Hybrid Fiber Coax (HFC) Cable Network algorithms is a subject for future studies. The introduction of IP-based voice, a cable network to support IP-based data, and video services over this services requires that the traditional architecture raises new issues for method of designing for signal integrity consideration: within a certain bandwidth range must be combined with a method for ensuring that a) Since the traffic for these services is dedicated to individual subscribers, this the bandwidth at distribution hubs and on portion of the system’s requirements for the backbone ring is sufficient to handle bandwidth is determined by the number of traffic loads and deliver a given set of subscribers served and their traffic services at desired levels of performance. We now turn to the methods for estimating demands. these capacity requirements. b) The routing of downstream traffic to the right subscribers is more complicated CAPACITY DESIGN than in traditional broadcast television networks that offer tiered packages of Load Estimation video channels. c) The routing of upstream traffic from The load of a service at a CMTS at a subscribers on a shared medium is a new hub can be determined in two steps: problem that is absent in traditional In the first step, we estimate the average broadcast television networks (although number of simultaneously active users for some cable systems carry upstream traffic the service in the cluster of subscribers in support of pay-per-view video services). served by the fiber node or nodes associated with the CMTS. Typically, we IP-based services are accommodated consider the load during the network busy within the distribution network by hour for the service. Given the number of dedicating some bandwidth spectrum for homes served by the CMTS, the digital traffic. Typically, the upstream penetration of the various IP services for traffic occupies the spectrum from about 5 that cluster of homes, and the activity level MHz to 42 MHz [3-4], while one, or for a typical subscriber for each service, possibly two, 6 MHz channels at higher we can determine the average number of frequencies are set aside for downstream simultaneous users. digital traffic. At the distribution hub, analogue and digital traffic are combined In the second step, we combine this in the downstream direction and separated estimate of the number of simultaneous in the upstream direction. Within the traffic streams with information about the backbone network, analogue and digital shape or intrinsic characteristics of the traffic are carried on separate facilities. typical traffic stream generated by a user of that service, to determine the aggregated For the digital traffic of the IP-based traffic of the service passing through the services, each CMTS is a point of CMTS, in each direction, in the busy hour. aggregation for the traffic of all the Similarly, the aggregation of the traffic subscribers served by it, while the streams of all hubs gives us the aggregate backbone ring aggregates the traffic of all traffic of the service on the backbone ring. the CMTSs at the various hubs. Designing

Capacity Calculation Voice-over-IP

For each service, the bandwidth a) Traffic Model requirements at the hubs and on the ring are determined on the basis of the required For the traffic produced by a single levels of QoS and the aggregated traffic at Voice-over-IP call, we use an ON-OFF the corresponding points in the network, model for the traffic rate, with the rate thereby taking advantage of the efficiency alternating between a constant peak value of capacity utilization arising from the during ON-intervals (which correspond to statistical multiplexing (superposition) of talk spurts), and zero during OFF-intervals the individual traffic streams. The (which correspond to intervals of silence). algorithms for these calculations of bandwidth requirements are presented The peak rate resulting from the below for the two services considered in standard digitizing of voice-samples at the this paper: Voice-over-IP (VoIP) and Data. Nyquist rate is 64 kb/s, and is then subject

to modification by the coding rate, and by The traffic aggregation, however, is the overhead involved in forming IP considered only for traffic streams of the packets from segments of talk spurts. The same service and same levels of QoS, i.e., ON and OFF intervals are treated as we multiplex only within each class of random variables of exponential traffic, and not across traffic classes, distribution. Thus, the parameters because the multiplexing of streams of characterizing this model for Voice-over- different characteristics and QoS IP traffic are: requirements may offer no benefits, and, in fact, might require additional control τ = mean ofduration a talk spurt mechanisms to ensure that each class of on traffic receives its proper QoS. Therefore, τoff = mean ofduration a silent period we adopt the conservative rule of merely adding the separate bandwidth R = coding rate requirements of each class of traffic to h = oflength ofheader IPan packet arrive at the requirements of the combined c = gpacketizin" policy" traffic of heterogeneous streams. = ofduration talk - spurt segment

payload-packet IP as collected as IP payload-packet IP SERVICES P = ratepeak h We now present the traffic models, QoS R += c parameters, and capacity calculations for

Voice-over-IP and Data. The model and The first two parameters, τ and τ , QoS parameters for these two services, on off together with some representative default pertain to the characteristics of voice values, are summarized in Table 1. traffic, for which extensive experimental studies have produced typical default values that can be used in the absence of user input. The coding rate R and the coding policy c (which affect customer- perceived quality of voice-connections) are of simultaneous calls that must be parameters of the coder, and thus assumed supported at the CMTS is the smallest known. The header length h pertains to the integer N for which ),( ≤ bANB , IP protocol that is implemented. where ANB ),( is the Erlang-B blocking function [5]. This is the number of calls for b) QoS parameters which the VoIP traffic model above would

be used to determine the required For VoIP, the QoS parameters are packet- bandwidth at a CMTS, for specified QoS loss rate and delay-jitter. In addition, since constraints on packet loss, delay, and jitter. VoIP is a real-time connection-based service, subject to blocking if the network d) Bandwidth Calculations cannot meet the packet-level QoS constraints, the probability of such blocking As an example of bandwidth becomes an additional, connection-level QoS calculations using the VoIP model, we constraint. present below the formula [6] for the bandwidth L required for supporting N c) Load Calculations simultaneous calls at a loss rate of r , given

Consider the calculations for homes a buffer of size B .

(customers) subtending a given CMTS. We Define want to know the total offered load A in Pτ τ offon  1  Erlangs due to these customers in the busy φ = ln  ; hour. Let C be the number of customers, B +ττ offon )(  r  and let θ be the penetration factor for the τ on Voice-over-IP service. Then, Cθ is the f = +ττ number of subscribers to this service. If u offon is the utilization factor for a typical = Pfm subscriber, i.e., the fraction of time during the busy-hour that a typical subscriber Then, = LLMinL ],[ , where would spend on voice-calls if the 21 subscriber suffered no blocking, then the offered load per subscriber is u Erlangs.  ()2 +−+− 411 φφφ f  1 = NPL   Then, the total offered load  2φ  equals = θ uCA Erlangs, which can also   be viewed as the average number of simultaneous calls that would be in 2 NmL [ r −−+= π ] − mPNm )()2ln(ln2 progress during the busy-hour, if there were no blocking. Corresponding expressions can be derived for the bandwidth required to meet We assume that there is admission jitter constraints (often, the jitter constraint control for voice calls (i.e., a new call is treated as a bound on the maximum attempt would be blocked if sufficient delay, experienced when the buffer bandwidth cannot be provided to it), and allocated to the service is full). The that the probability of a call attempt being maximum of the bandwidths determined blocked should not exceed a specified by the loss and jitter constraints is then the valueb . To meet the blocking criterion for bandwidth required to meet all the QoS the offered load A , the minimum number constraints. Data c) Load Calculations a) Traffic Model The demand will be specified in terms of the requirements of the aggregate data In pioneering studies at Telcordia stream that has to be supported. Then, just Technologies, high-speed data traffic was as in the case of voice calls, we arrive at shown to be characterized by burstiness = θuCA as the average number of over many time scales, a phenomenon simultaneous data sessions in progress, known as "long-range dependence" [7-9]. where the penetration and utilization A fluid model known as Fractional factors now pertain to data service. If the Brownian Motion (FBM) [6] was shown to traffic of a single session is described by be capable of representing the aggregated the FBM process ,,( Ham ), the aggregate traffic of a large number of independent streams, such as those generated by users traffic of A simultaneous and independent downloading files from the World Wide sessions is given by ,,( HaAm ) . Web. The FBM model is a Gaussian model d) Bandwidth Calculations with stationary increments, and is specified by the parameters Ham ),,( , where As an example of bandwidth calculations with the FBM model, we m = mean traf fic arrival rate present below the formula that determines a = peakedness parameter the mean delay d when FBM traffic H = Hurst parameter, with 5.0 H ≤≤ 1 Ham ),,( is offered to a link of bandwidth L [6]: The peakedness parameter a describes 1 the variance of the fluctuations in traffic − K θ  1  rate at the time scale used in the model d 1 +Γ=  , description, and H characterizes the L  θ  persistence of correlation in traffic rates with time lag, i.e., is a measure of long- where range dependence, with H = 5.0 − mL )( 2H corresponding to short-range dependence θ ),1(2 KH =−= , and H > 5.0 corresponding to long-range − )1(2 − HHam 2)1(2 HH dependence. It can be shown that the and Γ z) theis( Gamma function multiplexing of n independent FBM sources, each described by ,,( Ham ) , gives We can invert the formula to determine rise to the FBM process ,,( Hanm ) . the bandwidth L required to achieve a given mean delay d by doing a binary b) QoS parameters search for L .

For data sessions, we assume that there is no admission control, and hence no EXAMPLES blocking constraint to be considered. The a) QoS and Capacity Requirements QoS parameters are, therefore, packet loss rate and mean delay. This example illustrates how the capacity-estimation algorithms can be used as a “what-if” tool to determine the effect of the QoS levels specified for the Voice- the bottleneck here) is 0.856 Mb/s. Thus, over-IP and Data services on the capacity the total bandwidth needed for the requirements. upstream channel is 2.256 Mb/s, which exceeds the available upstream channel Network bandwidth of 2.2 Mb/s. Thus the network cannot support the services at the We consider a symmetric network of 4 performance levels in QoS-1. hubs, each with one CMTS. At each CMTS, the upstream traffic has a channel We next consider the following set of of bandwidth 2.2 Mb/s, with a buffer of less stringent QoS parameters (QoS-2): 300 kbits, and the downstream traffic has a channel of bandwidth 27 Mb/s with a buffer of 3 Mbits. The bidirectional ring VoIP: Connection blocking = 1.0% has a bandwidth of 24 Gb/s, with a buffer Maximum delay = 10 msec of 3 Mbits. Bit-loss rate = 10% Load

Data: Average delay = 100 msec At each CMTS: Bit-loss rate = 5% VoIP = 30 erlangs

Upstream data rate = 0.5 Mb/s The bandwidth for VoIP is now 1.2 Downstream data rate = 0.5 Mb/s Mb/s, while that for the upstream data is The parameters for the traffic models for 0.777 Mb/s, for a total bandwidth VoIP and data are taken to be those given requirement on the upstream channel of in Table 1. 1.977 Mb/s, which is smaller than the given channel bandwidth of 2.2 Mb/s. QoS Thus, the existing network can support the two services at the performance levels in We first consider the following choice of QoS-2. QoS parameters (QoS-1):

VoIP: Connection blocking = 0.1% b) Multi-Year Capacity Planning Maximum delay = 10 msec This scenario deals with capacity Bit-loss rate = 5% planning over a 5-year horizon, under a given forecast of load evolution, and shows the benefit of taking account of the Data: Average delay = 50 msec statistical multiplexing that occurs in Bit-loss rate = 1% aggregating the traffic streams of different subscribers for the same service. The For the loads assumed, the bandwidth network is the same as in the previous required at each CMTS to support VoIP is example, and the forecast 5-year load 1.4 Mb/s, while the bandwidth required for evolution is given below in Table 2, along the upstream data (which turns out to be with the results of calculation for the total upstream bandwidth at each CMTS, using provide acceptable levels of QoS. There is the performance levels specified in QoS-1 a need for a new set of algorithms for above. planning and provisioning new IP services on cable networks. Once again, the bottleneck is the upstream channel bandwidth, which In this paper, we have described traffic remains adequate to support the loads for models and algorithms for estimating the chosen QoS parameters in Years 1-4, capacity requirements to support IP but becomes inadequate in Year 5, services at specified levels of QoS. We according to the results of the capacity- model the digital portion of the cable estimation algorithms appearing in Row 3 network in terms of resources and their of Table 2. capacities, along with mathematical models that capture the characteristics of Suppose, on the other hand, that one the traffic of the IP-based services that merely looked at the loads and bandwidth these resources must accommodate. The requirements in Year 1, and estimated the capacity-estimation algorithms determine bandwidth requirements in the future years resource requirements at various points in by linear extrapolation of the bandwidth the network to support traffic loads at requirements in Year 1. The results of such specified QoS levels. Where sufficient extrapolation (Row 5 of Table 2) would resources are present, they are partitioned lead one to the false conclusion that the among the different types of traffic. network runs out of capacity even for Locations with insufficient resources are Year 3. The comparison of bandwidth identified and the shortfall is determined. requirements determined by the capacity- The capacity-estimation algorithms can be estimation algorithms with those calculated used for investigating a wide variety of by linear extrapolation shows that the "what-if" scenarios, including the trade-off penalty in failing to exploit multiplexing between QoS guarantees and network gain increases with increasing load. So, resource requirements, as shown in the this scenario demonstrates the potential examples that we consider. benefit of the algorithms in deferred capital investments, by the explicit The capacity estimation algorithms accounting for multiplexing gain in account for the efficiencies that are calculating capacity requirements. realized from the statistical multiplexing of traffic streams of different subscribers for CONCLUSIONS AND FUTURE the same service. By considering a WORK scenario of multi-year evolution of subscriber demands, we demonstrate the The engineering of cable networks for capacity savings achieved by the IP-based voice and data services presents multiplexing efficiency built into our new planning challenges to cable algorithms, in comparison with the operators. The traffic of these new services approach of linear extrapolation of is directed to individual subscribers, unlike capacity requirement in the number of the broadcast video services for which subscribers, which could lead to gross cable networks have traditionally been overestimation of capacity requirements. designed. To be successful in offering More accurate estimates will lead to better these IP services to subscribers, the strategic decisions on when, where, and network must have sufficient capacity to how to offer new services. For mathematical models of traffic to REFERENCES be useful, one must be able to determine proper values for their parameters, to [1] P. Bates, T. Carpenter, Y. obtain a reasonable fit to the traffic being Chandramouli, J. DesMarais, M. Eiger, K. R. Krishnan, and A. L. Neidhardt, described. In principle, the model “Planning and Provisioning for Cable parameters can be estimated from traffic Internet Services”, National Fiber measurements collected at a fine enough Optic Engineers Conference, time-resolution. However, such detailed September 2002, Dallas, Texas. measurements may not always be practical or economical in all networks. If the model [2] J. T. Chapman, “Multimedia Traffic parameters can be estimated from the Engineering for HFC Networks,” routine operational traffic measurements in White Paper, CISCO Systems, 1999. a network, then one could derive the inputs [3] Cable Datacom News, Overview of to the capacity-estimation algorithms from Cable Modem Technology and traffic measurements and load projections, Services,http://www.cabledatacomne creating an integrated monitoring and ws.com/cmic/cmic1.html. planning system. Such a system will [4] W. Ciciora, J. Farmer, and D. Large, enable a network operator to plan and Modern Cable Television install new capacity before existing Technology, Morgan Kaufman capacity runs out. The integration of Publishers, Inc., San Francisco CA, parameter-estimation methods with 1999. capacity-estimation algorithms is a subject for future studies. [5] R. B. Cooper, Introduction to Queuing Theory, 2nd ed., New York: North It would also be desirable to expand the Holland, 1981. system to propose remediation when it [6] K. R. Krishnan, A. L. Neidhardt, and finds that demand exceeds network Y. Chandramouli, “Traffic Models for capacity. We envision optimization Cable Network Services”, Telcordia algorithms that will propose ways that Technologies, 2001. cable operators might add network [7] J. Beran, R. Sherman, M. Taqqu, and capacity at minimal cost, determining W. Willinger, “Long Range appropriate QoS parameters for viable Dependence in Variable Bit-Rate Service Level Agreements, and Video Traffic,” IEEE/ACM implementing controls to meet them. Transactions on Networking, 43, pp. 1566-1579, 1995. We also have to investigate whether other services such as Streaming Video, [8] M. Garrett and W. Willinger, “Analysis, Modeling and Generation of Video-on-Demand, and Video Games can Self-similar Video Traffic,” Proc. be represented in terms of existing traffic ACM SIGCOMM, London, September models or will require the construction of 1994, pp. 269-280. new models and corresponding capacity- estimation algorithms. [9] W. Willinger, M. Taqqu, W. Leland, and D. V. Wilson, “Self-similarity in High-speed Packet Traffic: Analysis and Modeling of Ethernet Traffic Measurements,” Statistical Science, 10, pp. 67-85, 1995. Table 1: Traffic Model and QoS Parameters (with reasonable default values)

Voice-over-IP Data • Jitter tolerance (0.01 sec) • Loss tolerance (1%) QoS • Loss tolerance (0.5%) • Average delay tolerance (0.05 sec) • Blocking tolerance (0.1%) • Talk spurt (1.004 sec) • Peakedness2 (61000 bits/sec2H–1) • Silence (1.587 sec) • Hurst parameter (0.85) • Packet header (416 bits) Traffic Shape1 • Analogue-to-digital coding rate (6.4 kb/s – 64 kb/s) • Voice frame size (0.010 – 0.030 sec) Intensity1 • Erlangs • Bits/second

Table 2: Five-Year Evolution of Loads and Capacity Requirements

Y1 Y2 Y3 Y4 Y5

VoIP Load (erlangs) 15 20 25 30 35

Upstream Data Rate (Mb/s) 0.25 0.30 0.35 0.40 0.45 Upstream Bandwidth Requirement 1.415 1.664 1.948 2.113 2.385 (Mb/s) (X) Linear extrapolation from Y1 1.415 1.821 2.228 2.635 3.041 (Mb/s) (X) (X) (X) Overestimate in extrapolation 0% 9.5% 14.3% 24.7% 27.5%

Entries marked with an “X” exceed the upstream channel bandwidth of 2.2 Mb/s

1 Traffic shape and intensity are given independently for upstream and downstream traffic. The parameters should assume the same values in the two directions for symmetric services, such as voice. 2 Peakedness is measured in bits/sec2H–1, where H is the value of the Hurst parameter. CONTENT SUPPLIER’S PERSPECTIVE: HOME NETWORKS AND RIGHTS MANAGEMENT

Robert M. Zitter, Craig D. Cuttner Home Box Office

Abstract subsequent distribution windows such as home video, pay-per-view, etc. – each Copyright Protection, Digital Rights window and its attendant revenue stream is Management, Home Networks and other critical to the total revenue stream that buzzwords are all swirling around in a makes theatrical movie production a viable frenzy that has seemingly pitted content ongoing business. providers against operators and all seem to make consumers appear to be felons. If video programs are available on the Internet at no charge, once the Internet is This paper defines the scope of the issue connected to a digital home network, and discusses the pro- and con-attributes of consumers will be less likely to pay for the a Home Network. services MVPDs offer.

Beginning in the first-steps of initial Distribution Security implementation of “simple” Copy Control Information (CCI) and how that may Over time, cable distribution technology evolve into sophisticated Digital Rights has evolved to thwart the capabilities of the Management (DRM) systems. “consuming public” to circumvent the collection of fees and protect revenues. BACKGROUND Non-standard channel (“mid-band” placement for pay channel security in the Content Value and Cable Value ‘70s) was succeeded by block converter devices and, later, by the cable-ready TV, The ‘sale’ of access to video content has which necessitated trapping. Higher value been a hallmark of the cable television premium TV led to channel scrambling and industry since its inception. Technology fixed (programmed) descrambling STB’s. used in the distribution, securing the Pay per view required addressability – and distribution and the content itself has the consumer’s ability to record high-value evolved over time as has the value of the content led to analog copy control (a/k/a content being distributed. Cable, and other Macrovision). forms of multichannel distribution (MVPDs), are actually the ‘tip of the Each of these steps was necessitated by iceberg’ related to a multi-faceted time- an evolutionary requirement to continue to value food-chain that impacts all aspects of collect revenue to keep growing the cable content. and content industries.

Time-value of content and distribution Likewise, the “robustness” of systems are not new concepts. Hard cover books was driven to sophistication requirements by become mass-market paperbacks and later the day’s environment. Security evolved motion pictures. Theatrical movies have from virtually-none (since early TVs did not generally tune RF non-standard channels), Consumers increasingly want all of their compromised by simple “orange wires” in home electronics products to integrate and set-top-boxes, to sync-suppression provide a ‘greater good’ by using common scrambling – and, in fact, it was not always control systems and making content clear that the consumer electronics products available in different rooms, on different were designed to respect the necessity of devices – including computers, portable cable security – early versions of digital- devices, and, perhaps, vehicles. chassis analog television receivers were able to defeat very sophisticated sync- This need for integration means a suppression scrambling. network – and the “Home Network” connectivity promise soon available to Although there were limited cases of consumers is both a potential revenue source “redistribution” of video programming (e.g. for enhanced services, provisioning, an apartment building with ‘channel-3 management, and maintenance – and a sharing’),. the physical audit capabilities of threat to the distribution-revenue just as has the cable operator, and simple physical occurred in decades past. limitations of single-premium-channel distribution, were adequate limits on the The stakes today, unfortunately, are spreading of content redistribution. much higher.

TODAY’S LANDSCAPE While a crude ‘channel-3 network’ might have provided a few like-minded apartment Building on the premise that the “on/off” dwellers a single-channel of HBO in the nature of conditional access (traps, 1970’s – one copy of a motion picture on the addressability, or other forms of access Internet has devastating impact on the denial) was appropriate to the consumer’s revenue stream before, during and after the then-current abilities to circumvent those cable window. The impact will be more systems. Further building on the ability to devastating now that the value of cable use simple analog Macrovision as an content has significantly increased – and appropriate means to prevent analog VCR more of that content is original-to- cable, as copies in the PPV window – where does that its first exhibition window. leave us today? First Steps in Security – Basic Conditional Digital cable-ready receivers will soon be Access able to be directly-attached to cable systems; digital VCR and DVD burning devices are The FCC and industry economics have growing in popularity, and the fastest- dictated the first steps to continue the growing segment of the (IT) computer evolution of cable security – the one-way industry, the ‘Media Center PC’ – promotes digital cable ready television is a reality. a video program guide, Digital Video Recording and DVD-burning. Using contemporary industry-standard digital encryption techniques, the Clearly, the next evolution in conditional CableCARD™ can take the best-of-breed access, content security – and potential new conditional access systems deployed today revenue streams are upon us. and integrate it into a consumer-purchased host device. As required by the Digital Millennium Copyright Act (DMCA), no used to ensure that the host device properly consumer-accessible ‘in-the-clear’ streams adheres to the PHILA principles. are available courtesy of DFAST encryption. 2. Compressed Firewire (1394) – the It is important to note that the current ‘5C’ (DTCP) copy protection regime is implication in the nature of consumer specifically tailored for high-speed behavior is that “in the clear” is a direct tie compressed applications. to “distributed on the Internet” without recourse. It is the need to encrypt content 3. Uncompressed DVI (also HDMI) – securely to both maintain its value and retain the ‘HDCP’ copy protection regime is legal protections under DMCA. specifically tailored for uncompressed digital signals such as might be transported CableCARD, however, is only the first- between a host device and display device. step evolution of an integrated TV / set top – and it alone does not provide functionality These copyright protection systems for the Home Network as consumers would protect, generally, a source-to-sink deploy it. relationship and convey only ‘basic’ copy control states. (Generally, signals in this In a consumer environment, more context, originate in a “source” device and connections than cable-to-host / display are are consumed in a “sink” device. A set top envisioned. Modular components (set top to is an example of a source, a TV display is a plasma display; recorder to display; media sink.) storage to display) begin to form the basis of a series of output to input links – all digital – The ability to separately enable and hence all needing some form of copyright disable items 2 and 3 (above), is referred to protection to prevent ‘unprotected copying as “selectable output control” which was a and redistribution.’ Each type of interface point of considerable discussion during the has a defined and approved copy protection Plug and Play negotiation. system and is a overall part of the ‘Plug-and Play Agreement’ tied to a set of principles Second Step – (Unprotected Interfaces) and rules that govern access to and use of cryptographic keys that control system Analog video, in many contemporary behavior. consumer devices today, is routinely converted from analog to digital in a very Second Step – (Protected Interfaces) high-quality manner. Once that conversion has occurred, all of the concerns about As secure digital interfaces require copyright protection, Internet redistribution, copyright protection regimes, there are etc. equally apply to the converted signal. systems appropriate to the major interfaces: This is the so-called “Analog Hole.”

1. The CableCARD itself – although NTSC composite analog video is not intended for consumer access, the pins “protected” only to the extent that of the card interface could be intercepted Macrovision may be applied as a part of the during their transport of high-value content ‘copy never’ state. CGMS-A may also be – thus, the DFAST encryption system is present to signal other copy control states, but may be removed by some consumer devices, thus it is not, by itself, sufficient for authoritative signaling. Further, the “low- Home Network Topology quality” nature of NTSC limits the quality of Secure copies that may be made from either the Video original analog or digital copies thereof. Home TV Display Network

Component analog video interfaces are

higher quality than NTSC, and can support DVR very-high quality signals up to high- “The Internet” definition. Considering the weaknesses of CGMS-A noted above, and the fact that only certain resolutions and formats of component analog support a vertical There are two other “competitors” that blanking interval for carriage of copyright currently are working to be the key control information, there are concerns providers of that home network: about very high-quality analog to digital conversions of component analog. When 1. Consumer Electronics (CE) component interfaces are “unprotected” manufacturers that look at device some content owners necessitate lowering connectivity as an extension of their the picture resolution to that of NTSC, so devices’ functionality – and hope to leverage that analog to digital conversion is less that product benefit into their offerings; effective – thus called “down-resolution” – which is also a point of discussion in the 2. Information Technology (IT) two-way negotiations. manufacturers, recently growing into the traditional CE space by selling Copy protection states are discussed entertainment hardware, have looked at more fully below. home networking as an extension of their data infrastructure – and the increased FUTURE NETWORK FEATURES functionality is also a point of desired benefit. Third Step – Protected Home Network 3. Of course, it is desired that the How do we evolve the current secure and MVPD be included in this list – as a very revenue-generating MVPD distribution key provider with decades of expertise in the infrastructure into a similarly secure (often transport of high-value content – and a key called “trusted”) part of the content participant in the revenue chain that benefits distribution infrastructure? Considering that from the secure transport of content. it is a forgone conclusion that consumer home networks will exist, this is an Currently, neither of these two non-cable important task at hand – as will be discussed potential home network providers are for the remainder of this paper. involved in the revenue stream nor the protection of the time-value of the content in the food chain. Although home networks will exist, it is Copyright protection is necessary to essential that the home network be ‘trusted’ augment conditional access. in the economic sense – and, in the technical realm, ‘robust’ in implementation – unlike Copy Control Modes the previous evolution from traps to scrambling, the stakes are much higher. Generally, there are but a few business model concerns that need protection: If three equal home networks were to evolve, content companies need the same No uncontrolled redistribution on the level of security – and already have Internet – this restriction is mandatory in experience with “trusted” cable networks. every copy ‘restriction’ state – and also a Content security is a “new” feature to the requirement for the ‘Retransmission Control “other two” home network developers. Descriptor’ (RCD) (also known as Thus, MSOs have an advantage because of “broadcast flag”) if the content is not their participation in the revenue chain – and otherwise copy controlled. (Note that the the attendant concerns to preserve security, RCD is currently under industry discussion and to evolve conditional access as they and is not necessarily ratified in standards have before. mentioned below.)

COPY PROTECTION BASICS Copy Once – content marked copy once is restricted to one copy (even on removable Access to Content and archival material), but all playback devices must mark the playback output as In the 1970’s ‘legacy’ consumer “copy no more” to ensure that no subsequent environment, where “on and off” were copies are made. A copy on a device such sufficient business conditions – and there as a Digital Video Recorder (DVR) might be were no recording or distribution issues of ‘moved’ to an archive under the rules of concern – conditional access worked fine at Copy Once. a very simplistic level. As noted, when consumer copying of PPV became a Copy Never – content marked copy never potential threat, copy protection was added. is prohibited from having any recorded copies other than copies that are of short- Today’s access to content, if the duration (convenience-type) copies. environment was devoid of redistribution, copying and there was no desire to increase These three states are communicated by revenue by exploitation of the time value of two binary “bits” in several control content, then simple conditional access messages that are either embedded in would be sufficient. content or sent via control streams. Copy control information (CCI) bits can occur in However, the home network implies some or all of the following forms: connectivity, copying for convenience and device portability – and the risk of • Setup in the configuration of set-top uncontrolled redistribution of high-quality boxes, control software and digital interface content beyond the subscriber is too drivers. In this way permanent modes could economically devastating to ignore. be set via programmer (content provider) direction. • Analog (Copy Generation Management System – Analog (CGMS-A)) BASIC COPY CONTROL STATES bits are sent via Line-20 (International use, standardized by IEC); US VBI Bits Control Mode RCD Redistribution 0 0 Copy Freely 0 Allowed standardization of CGMS-A is included in 1 Prohibited CEA-608-B via Line-21. Note: The intent is 0 1 Copy Once x Prohibited not to restrict analog recording but identify 1 0 Copy Never x Prohibited copyrighted content to downstream Analog 1 1 Copy No More* x Prohibited to Digital (A/D) devices or encoders to give (* sometimes copy no more is flagged as copy never) instructions regarding redistribution and x = “don’t care” copying.

• Other signaling such as Program Map Table (PMT) is under discussion and is HBO / Cinemax linear content is proposed in some technology licensing designated “Copy Once” and HBO / agreements. Cinemax on-demand content is designated “Copy Never.” It is important to note that declaration of copy control mode, the mechanism that sets Fourth Step – Digital Rights Management the mode (whether that control is embedded in the content or ‘authoritative’ via secure The next step in the evolution that channel such as an encryption system) and optimizes the trend in time-value content the type (category) of content that is copy- revenue optimization, secure distribution of controlled is governed by the DMCA, FCC content, copy management, portable device rules, and technology licensing agreements management and takes the role of (so-called “Encoding Rules”). conditional access to its, arguably, ultimate level – is to include digital rights Encoding Rules as proposed by the cable management functionality into a ‘traditional’ and consumer electronics industries (and conditional access system. adopted by the FCC) propose the following ceiling levels (less restrictive levels are DRM will be the tool by which MVPDs allowed for a particular content type, but not and content owners will be able to use more restrictive) for copy control: copyright protection technology to create new consumer offerings rather than simply employing the technology in attempts to • Copy Never: PPV, VOD (on- preserve the status quo. demand PPV), SVOD.

Conditional Access systems, generally, • Copy Once: Linear premium and are ‘trusted’ for their security and have basic television evolved from quite simple on/off gates to

addressable, VOD-enabled, two-way • Copy Freely: Over-the-air broadcast session-based, transactional systems. television (however, the RCD may be applied). Digital Rights Management (DRM) systems, now deploying for Internet-related content delivery, provide basic security (encryption) technology – but have very complex on/off rules – sometimes embedded The integration by a MVPD of an within the content (not necessarily requiring ‘trusted’ conditional access / DRM system a two-way connection). On/Off rules can be into a whole-home-network environment soft descriptions of business cases (such as that exerts authoritative control over content subscription, copying permission, quantity within the network, provides all protected of viewing (e.g. x-views), duration of outputs from the network under that control viewing (self-destruction of copy after y- umbrella, is the genesis of a ‘Trusted days), etc.). Domain.’ Such a secure environment will be necessary to optimize the time-value Signaling the DRM control states is an revenue stream – that has served cable so ongoing industry discussion (“Extended well since its very first paying subscriber. CCI”) and is not yet embodied in cable But, that was a few hundred billion dollars standards or other rules. ago, and there are plenty more to go around. COPY PROTECTION STRATEGIES FOR THE DIGITAL HOME

Carmi Bogot NDS Technologies

Abstract transmission, downloaded from the Internet via broadband or uploaded from a digital Digital entertainment content is the camera or other recording device. driving force in the “digital home” revolution. Availability of content drives What makes the digital home appealing is demand for consumer electronics devices the high level of picture and sound quality and for the ability to interconnect these and the ease with which consumers can devices using a home network. At the same access content on several devices, sharing time, the new availability of digital content and transferring high quality content on a wider range of connected devices throughout a home network. It is this same makes content protection an ever-more level of quality and flexibility which gives pressing issue for content developers and rise to worries about illegal reuse and rights owners. redistribution of copyrighted content.

Content protection is a highly emotive Today, it is technically possible for issue, which impacts all the players in the digital content to move freely between digital content distribution value chain. This storage and access devices, and to reside in paper defines both the technical and many locations from the time it is business terms critical to content protection distributed until the time it is finally in the digital home. It proposes the Secure rendered and consumed by consumers. In Video Processor (SVP) — created by adding fact it is so easy to move digital content, that a small number of gates into an existing systems that try and prohibit the movement video processor chip design, creating a of content are easily hacked. Secure Video Processor or SVP. The SVP Alliance is a group of interested parties Rights owners and content providers, as committed to advancing the development of well as networks and platform operators, secure content distribution and the wide who generate revenues from reselling their adoption of SVP technology. The paper also content, do not care if an un-viewable / presents the business benefits that SVP unusable copy of content is made. They are brings to all the principal players in the concerned with the uncontrolled viewing of content protection industry: operators, content without proper content protection studios, chipset manufacturers, CE vendors mechanisms in place. and consumers. The SVP Alliance proposes a solution INTRODUCTION which balances the consumers’ demand for fair and flexible use of content throughout Digital entertainment content can take their home network — their “domain” — any of a number of formats — it can be with the rights owners’ need to protect their broadcast TV content, digital music, or business interests. SVP does this by creating movies. The content can be delivered to the an end-to-end chain of control over content home in any of a number of ways — from distribution until rendering. broadcast via terrestrial, satellite or cable Defining the Digital Home DVR vendors offer consumers the ability to stream recorded content from a central DVR A digital home features any combination to client devices over a home network. of TV sets — analog and digital — DVD and MP3 players, DVRs, media centers, Implementing home networks allows PCs, PDAs, cell phones and home video viewers greater convenience and flexibility servers, all interconnected via a home since content can be consumed on multiple network. The digital home consumer wants TV sets, DVRs, and DVD players within flexible access to high quality digital their homes as well as on portable devices content, on demand, on any of these devices. such as PDAs.

Digital TV content is playing an SECURE CONTENT IS THE KEY increasingly important role in the digital home. According to Strategy Analytics, Appealing, timely, and high quality there were around 100 million digital TV digital content is the foundation of the households worldwide at the end of 2003. digital home revolution. Without the By 2008, this number is projected to pass content, consumers obviously don’t have the the 300 million mark. incentive to purchase devices which enable them to use and enjoy digital content. But The massive appeal of digital TV content the availability of content depends on more creates an incentive for consumers to add than the simple existence of consumer another digital platform to their home — the demand for access to it — it depends on the digital set-top-box (STB). New STBs willingness of content owners to provide facilitate a two-way connection between the content and the service providers to offer it home and an external service. An increasing as part of a secure pay TV service. number of STBs also feature local storage (known as a PVR/DVR), allowing time- One of the most significant issues content shifted viewing and archiving of digital providers face in the digital age is illegal entertainment. In addition, the existence of a use, distribution, and redistribution of digital STB stimulates consumers to use and content. Unprotected digital content can be look for convergence between the STB and copied, redistributed, and consumed at its other devices in the home, further driving original quality by anyone. Clearly, content the development of the home network owners and service providers whose . business is based on selling content seek to Convergence, fair use, and flexible use prevent this by using content protection are all terms used when talking about mechanisms. Without the means to control consumers’ desire to access the same content rights in the digital home, content content easily on different electronics owners will limit their content offering for devices. this environment. Furthermore, service operators are obligated to protect the rights For example, most U.S. households of their content owners. Operators will include multiple TV sets. Owners of hesitate to enable their digital STBs to advanced DVRs and STBs with local interface with other devices in the home via storage express strong interest in having high speed ports if they cannot control time-shifted viewing capabilities on every access to the content once it leaves the STB. TV in their household. As a result, some For these reasons, balancing the rights of device. Upgrading any existing video the content providers with the needs and processing chip into an SVP is a simple demands of consumers is critical to the process that typically increases the existing future development of the digital home gate count by less than 2 percent. network. Flexible, low cost, user friendly, content protection mechanisms are required What does the SVP Protect? to satisfy the needs of consumers who should be able to enjoy greater access to and The SVP is designed to protect any “fair use” of high-value content. Such scrambled digital content, associated with a mechanisms provide benefits to consumers valid license. The content is typically SDTV as well, enabling dozens of new ways to or HDTV MPEG-2 / 4) and any form of purchase content, such as content rentals, digital audio. Content can be delivered to the video-on-demand, and other models made home via any existing method in use today possible by content protection. — cable, satellite, DSL, or terrestrial — and protected by any existing conditional access A FLEXIBLE, LOW COST SOLUTION technology. The scrambling algorithms used on the content are market-dependent, and The SVP-enabled chip ensures that could be, for example, DVB-CSA, DES, content is under control from the beginning 3xDES, AES, DVS-042, Multi-2 or CSS. of distribution until it reaches its final destination and is rendered. The content is Domain and Fair Use always encrypted, and the rights are defined in a separate, standard, tamper-resistant To facilitate the fair use of content across license. multiple devices within a consumer’s home, SVP enables definition of two types of Every device that consumes digital video domains, within which content may be must already have a digital video processing consumed. A domain is essentially a chip. SVP technology is embedded within consumer household, made up of a number this chip, turning the existing video of interconnected consumer electronics processor into a “Secure Video Processor” devices. Domains may be externally chip. The SVP handles both the content and managed or autonomous. the license. The SVP is unconcerned with the physical distribution method and it is • An externally managed domain is independent of the location or locations managed using a gateway connected to where content is stored and the networking an external network. An STB, or communications technology used to move connected to a service operator’s content between locations. network and using the operator’s conditional access, would be Designing the SVP as a hardware-based considered an externally managed solution enhances security while making it domain. easier to standardize — and therefore easier to produce inexpensively in mass quantities. The • An autonomous domain is a network SVP-enabled chip can be implemented in any of consumer electronics devices where digital device, including STBs, TVs, DVD one of the devices on the network players and recorders, DVRs, PDAs and other manages the domain and the network. portable devices — without requiring special customization for each type or model of Any device in the domain that wants to SVP provides a standard messaging access controlled content must have an SVP format for defining content rights. Each enabled chip. Under the proposed SVP movie should have one Content Segment solution, it is possible to limit the number License (CSL) and a set of Base Line ECM and type (STB, mobile device, TV, DVD messages (BL-ECM). A new BL-ECM is player, etc.) of devices within a domain. In required every time the control word addition, SVP can control access rights and (descrambling key) changes. specify how consumers may use content. Use of content can include any combination SVP FUNCTIONALITY of: • Rendering (accessing the content) The SVP-enabled chip is capable of receiving clear or scrambled content, and • Rendering content plus permission to can descramble and re-scramble the content record and reuse the content for a using various ciphers including an SVP short, defined time period, perhaps 90 standard cipher NSA that uses 128-bit minutes control words. The SVP can also receive and transmit standard format licenses, consisting • Permission to make and store of a CSL and any number of BL-ECMs. All recordings of the content processing of content, whether in scrambled and compressed format or descrambled and • Permission to copy the content and use decompressed format, and all of the it within the domain processing of the license occurs in hardware within the confines of a single SVP-enabled • Permission to move the content, which chip. entails creating a copy and deleting or disabling access to the original, and Each SVP is associated with a public using the copy within the domain or certificate that uniquely identifies the SVP, outside the domain and the properties of the device in which it resides — device type, whether it supports • Exporting the content to a different watermark detection capability, which content protection system. Such as decryption algorithms it supports, which transfer to 5C or HDCP control. video formats, and other properties.

How SVP Protects Content At any given time, an SVP is linked to a specific authorized domain. The maximum All content on an SVP-protected network number of devices within the domain is is accompanied by a usage license. Content limited by the certificate. SVP-equipped is never transferred between SVP devices in devices within a domain will mutually the domain “in the clear” — it is always authenticate each other and establish a scrambled either using the AES-based, SVP secure channel prior to exchanging content. Native Scrambling Algorithm (NSA), or the original broadcast scrambling. In some cases both algorithms can be in use at the same time, resulting in super scrambling.

SVP and Conditional Access benefits from the adoption of SVP as a standard approach for digital content An SVP can work alongside and in protection. conjunction with a service provider’s conditional access system: Network Operators 1. Content arrives at the STB under conditional access system control. The growth of pay TV businesses demonstrates an increasing willingness of 2. Content is then transferred to the SVP so viewers to purchase the rights to view it can be used by other standard devices compelling content. However, along with within the domain. that willingness is an expectation of being able to use that content in increasingly 3. To transfer or share control, the varied and personalized ways. Many, or conditional access system must be able most, viewer households include multiple to generate a standard content license, TV sets, as well as a DVD, a VOD service, a including a standard Content Segment DVR, and, increasingly, a home network License (CSL)1 and a standard Base Line that connects several home entertainment Entitlement Control Message (BL- devices. ECM)2, for the content. These viewers want to be able to access Three conditional access systems are their pay TV subscription on all the TV sets available in the STB — embedded, smart in the house, they want to make backup card, CableCARD. Each can act as a copies on DVD-R media, store content on gateway between the conditional access removable USB keys, and use it anywhere system and the SVP system. In addition, the that they are. conditional access system should be used by the network operator to perform actions such Network operators know that their as setting up authorized domains, issuing continued success depends on their ability to and renewing certificates, and passing SVP provide viewers with top content and certificate revocation lists. flexible accessibility. At the same time, the operators won’t be able to obtain that BUSINESS BENEFITS OF SVP content if they cannot demonstrate to content providers and rights owners their The digital entertainment industry ability to protect the content and control comprises a range of players whose interests access to it. They need a solution that offers often seem to conflict. Content protection is viewers the fair use they demand, while an often-contentious issue which affects protecting content and expanding the each of the principal players — network operator’s business models. operators, content providers, chipset vendors, CE vendors, and consumers — Current solutions require a second or differently. SVP offers a unique win-win third STB for every location that content is situation where every player can realize used. This is not only an unattractive solution from the consumer’s viewpoint, cluttering up the house with numerous boxes 1 The attributes and usage rights for content are defined in a standard tamper-resistant message. and cables; it is expensive for the service 2 BL-ECM messages are used to securely transmit operator. In most cases the service operator decryption keys. pays for the development of the STB and equipped TV sets, PDAs, DVDs, etc. in the supplies the boxes to subscribers at minimal home or domain. This enables new business or no cost. Furthermore, subscription fees models based on secure home network for secondary devices are usually priced at distribution to multiple viewing devices. very low rates. The result is that secondary Operators can offer extended content rental STBs generate cost, not profit. Even if the packages, for example, charging extra for operator implements a solution with a single the rights to view across the home network high-end STB which communicates with or on multiple TVs. Pay-per-view and numerous secondary, relatively inexpensive Video-on-demand content is now truly STBs, the operator still faces an enormous available on demand to consumers, when expense in developing, deploying, and and where they want to see it. Operators can maintaining large numbers of STBs. determine which additional devices may be used to access content and for how long — Viewers would pay for off-the-shelf and charge accordingly for these access consumer electronic devices to access rights. The operators also control copying content within the home, as long as these and storage rights, providing even more devices were and not linked to a specific opportunities for licensing fees. They can service operator. charge a certain amount to permit saving content for a limited time period with on- SVP solves these problems. SVP allows demand viewing, including the ability to any network operator to deliver content in a pause, rewind, and fast-forward the content secure fashion to any SVP-enabled third- — and charge more for permission to make party device. With SVP-enabled chips a permanent copy. The SVP license is even included in each consumer electronics flexible enough to enable operators to device on a viewer’s home network, the charge for rights to view content in an operator can extend access control additional domain or to “transfer” rights — throughout the home network and even to all canceling the viewing permissions in the devices — including portable devices — and original domain when a copy is made to be realize dozens of new business models and viewed elsewhere. program packaging options. This is accomplished at virtually no cost to the Furthermore, SVP protection extends operator, since proprietary and conditional even to content which is archived onto D- access components are required only in one VHS or DVD devices. The content license “gateway” STB, not in all of the networked defines the access criteria as determined by devices. Thus operators can enable access to the network operator. Operators can then stored content or video-on-demand (VOD) enable viewers to record and keep content content on all TVs within a viewer indefinitely, confident that the content can household while deploying / subsidizing only be viewed within that viewer’s domain, only one STB per viewer household. not recopied and redistributed freely to friends and neighbors. On-demand content can be sent directly from the network to any device using SVP Content Providers content protection. Broadcast and On- demand content can also be sent via an Content providers have a clear interest in operator controlled SVP enabled STB protecting their content and determining enabling redistribution to other SVP- who may access it and when. They need to protect content from piracy and other Consumer Electronics Manufacturers unauthorized use and redistribution, and to enforce business models. Manufacturers of consumer electronics devices might initially be inclined to argue Until now, only a few models have been that content protection is not their concern. available for protecting content. One method However, it is clear that the ramifications of of controlling access to content has been unlimited illegal distribution of digital using limited packaging and distribution entertainment media will ultimately affect options, such as a CD for audio content or their industry, as solutions are either DVD for video content. These do not legislated or imposed by content providers. provide reliable content protection, and offer no flexibility in packaging and marketing. Proactive implementation of content protection using SVP creates a situation Another method of content protection, where all players in the digital home particularly for video content, is to work entertainment arena benefit. SVP enables with network operators who use conditional CE manufacturers and chipset vendors to access to protect content rights. This offer a standardized, low-cost content solution is being challenged by the protection solution which does not proliferation of home networks — and is not significantly increase their manufacturing satisfactorily meeting consumer demands for costs. The security is placed in chips that are fair use access to content on these networks. already an integral part of the CE device. The SVP solution offers CE vendors new Historically, discussions of content business opportunities as well. SVP protection have focused on protecting facilitates home networking while interfaces / pipes and not on protecting preventing unauthorized use and content. For exmple using 5C for 1394, redistribution of content. If a content using DFAST for CIM, and using HDCP for protection solution is in place, content DVI. These are all interface protection providers will be more likely to enable the mechanisms that attempt to protect the use of high value content across these transfer of content over specific media but networks. The availability of high value do not provide an end-to-end method for content will, in turn, drive the demand for controlling rights to content. In addition, newer and more networked devices, new forms of digital copying and storage enabling CE manufacturers to create a wide that are becoming easily available to range of SVP-compliant devices which can consumers make it easier for consumers to be used to consume content in increasingly evade restrictions on content. sophisticated home entertainment systems.

SVP offers reliable protection for For example, an new device and delivery content, wherever that content resides and mechanism can be created by simply however it is accessed. It can offer content combing; a broadband connection, SVP and providers the confidence to permit better any existing viewing device like a DVD viewing windows for new movie releases, player. This would enable the device to opening up new business opportunities both receive video-on-demand directly from for the content providers and the network content owners — increasing the value of operators who distribute the content. the device and providing room for additional revenue. Thus, SVP presents the opportunity for consumed — what types of devices, even CE manufacturers to participate in pay TV which specific devices — and for how long. distribution on home networks. Without a This ensures that rights owners get paid for standardized content protection solution, pay their content while enabling viewers greater TV operators could only permit their own freedom of access and use than any solution proprietary devices to be used to decrypt previously implemented. Once consumers their programming. However, the ability to have purchased the right to access content, use SVP certificates and licenses to enforce they can use it on any device in their home the pay TV operator’s entitlements means network, even in a second home, or on that CE manufacturers can now enter this portable wireless devices. The SVP is lucrative market with non-proprietary designed to be inherently flexible, giving “horizontal” retail devices. This provides content providers and distributors more consumers with increased choice, while options for packaging and marketing spurring development of and creating new content, further expanding consumer choice markets for CE devices. . A SOLUTION FOR EVERYONE Consumers The issue of content protection is not It often seems to consumers that the going away; in fact as the digital home entertainment industry is putting a lot of becomes more commonplace, content effort into making it difficult for them to protection is becoming an ever-more access content. If a viewer has paid for complex problem. As proposed, SVP offers access to content, why shouldn’t that viewer a win-win opportunity for all the players in be able to decide whether to view it in the the digital entertainment food chain. living room or the bedroom? Consumers are increasingly willing to pay for access to top The SVP solution addresses the issue of value content, but in return, expect to be end-to-end content protection in a wide able to access that content when they want range of devices, yet does not require a large to, and on a variety of consumer electronics financial investment. It creates new business devices. opportunities for CE manufacturers, content owners and content distributors, while SVP presents a way out of the confusion, making it possible for consumers to access with a solution that protects the rights of high quality digital entertainment on any of content owners while making it easier for their home network devices. Adoption of consumers to access content in more ways SVP is a crucial step on the road to realizing and in more locations. the full potential of the digital entertainment revolution. As a standard chip component which enforces the rules set by the pay TV operator For further details, visit the SVP website who sells the content, SVP can be used in at www.svp-cp.org, or email us at any type of consumer electronics device. [email protected]. The license defines where content may be

DEPLOYING ENHANCED IP SERVICES USING PACKETCABLETM MULTIMEDIA QOS CONTROL

Gerry White Motorola Broadband Network Infrastructure Solutions Group

Abstract need to be directly involved with the QoS signalling. This abstraction layer between the As operators seek to deploy enhanced application and the QoS infrastructure services, the PacketCable Multimedia simplifies the addition of new services. initiative allows them to deliver high-volume multimedia services with committed QoS PacketCable Multimedia offers significant levels. PacketCable Multimedia allows opportunities for enhanced services because operators to deliver QoS-enhanced operators can deliver QoS-enhanced multimedia services—including both multimedia services over DOCSIS access residential and business video telephony and infrastructure while using industry standards conferencing services over DOCSIS 1.1 to ensure QoS control across metro and core access networks. This paper will discuss how networks. the PacketCable Multimedia initiative builds on PacketCable 1.x to support enhanced IP Multimedia applications present special multimedia services. It will explain how challenges to broadband network operators, Applications Managers and Policy Servers and evolving multimedia services will on the network can act as proxies for clients continue to present unique bandwidth, QoS, to request QoS levels from the Cable Modem and latency requirements. The objective of Termination System (CMTS) for multimedia the PacketCable Multimedia initiative is to sessions, and it will provide examples of define the core framework required to several multimedia applications. support QoS-base multimedia applications. It leverages the foundations established by the INTRODUCTION DOCSIS 1.1 and PacketCable 1.x specifications to allow network operators to PacketCable Multimedia defines an efficiently deploy profitable IP multimedia architecture to deliver QoS-enabled services services. over an IP-based cable infrastructure. It enhances the original PacketCable telephony PACKETCABLE 1.X COMPARED TO centric architecture to allow flexible support PACKETCABLE MULTIMEDIA for a broad array of multimedia services. Thus operators can support a wide variety of The PacketCable specifications were applications and services beyond voice. originally developed for telephony services to provide a mechanism to establish IP flows The key aspect of this CableLabs over the HFC network with defined QoS specification is that it defines a multimedia levels. PacketCable Multimedia builds on the architecture that is application-agnostic. concepts of the PacketCable QoS architecture Application managers and policy servers act but rather than being targeted at a specific as proxies to request QoS on behalf of client application (telephony) it is deliberately applications so that the client itself does not application-agnostic. This enables features originally developed With the PacketCable Multimedia for VoIP (e.g. QoS authorization, admission approach, the client device may not be control, event messages for billing, security) intelligent and client-to-server signaling is to be used to deliver enhanced multimedia isolated from network QoS signalling. The services. client requests enhanced services from network-based servers, and services are Multimedia applications are bandwidth- delivered based on network policies. intensive and sensitive to transmission delay In the original PacketCable architecture the within the network. The PacketCable CMS both provided application-processing Multimedia architecture was developed to logic and implemented network policies. allow operators to deliver a variety of IP- With PacketCable Multimedia, these two based multimedia services and applications functions are separated and the interface that require committed QoS levels. between the components exposed and defined using the Common Open Policy This service delivery framework provides Service (COPS) protocol. general-purpose QoS, event-based accounting, authentication, and security PacketCable used the concept of event founded upon the PacketCable 1.x messages to track telephony-specific events. specifications. But PacketCable Multimedia The telephony specific messages have been generalizes and extends 1.x in order to replaced with generic policy events but the support a broader spectrum of applications overall structure of the event messages is the and services. It supports both peer-to-peer, same. In the PacketCable 1.x model the client-based services as well as server-based eMTA signaled directly to the CMTS to services. request QoS based on its understanding of the telephony application. In the PacketCable The PacketCable Multimedia signaling Multimedia framework, the Cable Modem framework provides applications with access Termination System creates the DOCSIS to the full range of QoS capabilities defined service flows associated with the client based in DOCSIS 1.1. This includes all scheduling on information received from the policy algorithms including best effort, constant and server. variable bit rate services, and both symmetric and asymmetric upstream/downstream UNDERSTANDING THE PACKETCABLE bandwidth characteristics. MULTIMEDIA ARCHITECTURE

PacketCable 1.x supports telephony The major network elements within the services using the Dynamic Quality of PacketCable Multimedia Architecture Service (DQoS) specification. The user has include: an intelligent endpoint that communicates • Application Manager (AM) with a Call Management Server (CMS) • Policy Server (PS) through an Embedded Multimedia Terminal • Cable Modem Termination System Adaptor (eMTA). (CMTS) • Record Keeping Server (RKS) • Client Application Manager The PS devices are located in the MSO’s managed network and are responsible for The Application Manager (AM) is making QoS-related policy decisions based responsible for application or session control. on rules defined by cable operators. Each It may be a general-purpose engine such as a operator can therefore establish its own SIP proxy server capable of supporting network policies. For example, an MSO multiple SIP-based application types or may could establish policies for the number of support a single application such as a gaming concurrent authorizations for a particular server accessed via a Web browser. In either subscriber or service. case, the client talks directly to the AM (over PacketCable Multimedia the IP network) and either explicitly requests Network Architecture the resources to initiate a QoS-enabled service or simply requests a service which has implicit QoS needs known to the AM. Application Manager

The AM then formats a request to the Policy Client

Server and attempts to reserve the QoS HFC Network MSO IP Network resources. Policy Server DOCSIS 1.1 Cable Modem CMTS Policy Server Record Keeping Server The PacketCable Multimedia architecture allows for and indeed expects multiple Cable Modem Termination System application managers to exist in a network. These managers must all use a pool of shared The Cable Modem Termination System resources e.g. HFC bandwidth which can be (CMTS) is responsible for enforcing QoS- made available by multiple devices. Thus any related policy decisions. It performs QoS requests are directed to a further admission control and manages HFC network intermediary device which mediates between resources through DOCSIS service flows. the AMs and the CMTS/routers. This device The CMTS is based at the edge of the HFC is termed a Policy Server (PS) and is network and communicates with the Policy responsible for coordinating the QoS requests Server using the COPS/TCP protocol and for multiple AMs and delivering the with the Record Keeping Server using commitments necessary to the multimedia RADIUS/UDP. applications. It supports a single-phase reservation In a typical network the AMs may be model where access network resources are under the control of multiple application simultaneously reserved and committed for service providers whereas the PS would be immediate use, as well as a two-phase under the control of the network operator. reservation model in which access network Thus the PS applies policy as defined by the resources are initially reserved and then network operator and manages the committed for use as they are required at a relationship between AMs and CMTS later time. The CMTS also provides the platforms at the edge of the network. The PS boundary between the access and implements operator-defined authorization metropolitan networks so operators can and resource management procedures. implement end-to-end QoS for applications and subscribers. Record Keeping Server application. The client will not request QoS explicity. Type 2 and 3 clients are QoS aware The Record Keeping Server (RKS) and request explicit QoS parameters. Type 2 performs a mediation function between the clients do this via the AM while Type 3 PacketCable Multimedia elements and the clients bypass the proxies and talk directly to back-office applications for billing purposes. the CMTS. Currently, only Type 1 clients are It receives event messages pertaining to specified. policy decisions from the PS, and receives notification of actual QoS grants from the SAMPLE PACKETCABLE MULTIMEDIA CMTS. MESSAGE SEQUENCE

Client It is helpful to view a simplified message sequence for requesting QoS for an A multimedia client is a logical entity that application. In this example, an application can send or receive flows. The client can be running on a PC client requests bandwidth any one of a diverse range of devices, such as from the AM (step 1), which checks with the a PC, gaming console, videophone, etc. It PS to ensure that business rules permit the communicates with the AM via an allocation (step 2). The PS confirms application-specific signaling protocol. In availability of resources (to the best of it’s many cases this will be a standard such as knowledge), authorizes the QoS grant and SIP but any protocol understood by both requests the CMTS to deliver the QoS client and AM (such as PacketCable NCS) required to support the application (step 3). may be used. This protocol is not defined in the specification so application creators and operators have maximum flexibility; the Sample Message Sequence Application Manager client does not directly request QoS, the AM does so on behalf of the client. 7 Š Service Activated PacketCable Multimedia discusses three 2 Š Policy Request for QoS types of clients: 3 Š Policy Request for QoS 4 Š DOCSIS 1.1

Policy Server

Client DOCSIS 1.1 Client Type 1 are endpoints that lack specific Cable Modem CMTS 6 Š Policy 1 Š Client Event QoS awareness or signaling capabilities Requests Message Service 5 Š QoS Event Message * Acknowledgements not shown Record Keeping Client Type 2 supports QOS signaling based Server on PacketCable DQoS. It is aware of The CMTS determines whether the PacketCable Multimedia QoS and bandwidth is available, and if so it sets up a communicates with an AM to request service QoS enabled flow across the access network and obtain permission to obtain services from to the client’s cable modem using DOCSIS the access network. 1.1 signalling (step 4). If the CMTS can also act as a MultiProtocol Label Switching Client Type 3 requests QoS using the (MPLS) Label Edge Router (LER) it may Resource Reservation Protocol (RSVP) also create a virtual path across the metro without AM interaction network to provide a QoS-enabled connection to centralized server resources. The AM will request QoS for Type 1 Thus resources can be reserved end-to-end clients based on its understanding of the between the client and the broadband network operator’s server. (If the CMTS is DATA PLANE not capable of acting as an LER, the PS may initiate setup of the virtual path in the Two distinct mechanisms are required to metropolitan network). The CMTS sends a provide QoS to multimedia traffic flows over QoS event message to the RKS (step 5), and large IP networks. At the edge of the the PS similarly sends a policy event network, bandwidth is typically scarce and message (step 6) so the operator can bill the devices deal with a limited number of traffic subscriber appropriately for its use of flows. In this region QoS can be provided network resources. Finally the data path is set most effectively on demand at a per-flow up and the service activated (step 7). level.

A FRAMEWORK FOR QOS CONTROL But on the metro network, bandwidth is relatively abundant but the infrastructure is The CMTS controls DOCSIS 1.1 QoS shared between tens-of-thousands of clients. service flow mechanisms and the operator Thus, core switches and routers must support can support time-based and volume-based hundreds-of-thousands of flows. If these network resource authorization. The CMTS devices were to operate at an individual flow informs the PS, which informs the AM when level the amount of data to be maintained a volume limit or time limit is reached. The would be massive and systems would simply event mechanism enables operators to not scale, so that an aggregation mechanism implement event-based network auditing. is needed. Current aggregated QoS mechanisms include IP-based differentiated Application-layer QoS requests are services and MPLS-based traffic engineering. translated into DOCSIS QoS parameters by The CMTS/edge router must act as the the CMTS. The CMTS creates DOCSIS transition point between the two QoS service flows and determines the cable domains. modem the client is connected to based upon the Subscriber ID (IP address). DOCSIS Data Plane Network Architecture service flows are established on-demand Internet based on policy and released upon session Servers LER LSR completion. VOD LSR Server LER Farm LSR LSR LSR With this framework, deploying new LER/CMTS MPLS-Based services and applications is simple and fast. Metro Network HFC Gigabit and 10 The CMTS and the cable modem are Gigabit Ethernet IP Set-Top Cable Gateway unaware of any specific application—they Box Modem simply provide the QoS. The PacketCable Multimedia specification therefore requires The data plane requires QoS control across both no QoS signaling capabilities in the client metro and access networks, with QoS on the access device. Policy—and the ability to deliver network delivered using DOCSIS 1.1 and QoS on the QoS—are pushed from AM-to-PS, PS-to- metro delivered using MPLS CMTS, and CMTS-to-CM. PacketCable Operators can implement QoS in the HFC Multimedia can co-exist with PacketCable access network using DOCSIS 1.1 and 1.x telephony so operators can support intelligent edge router/CMTS platforms that telephony as well as enhanced multimedia support per-flow queuing. If the edge router services over the same infrastructure. can also serve as an MPLS LER, then it can map packets from the per-flow QoS in the and deployed to support them all (per-flow at access network into aggregated traffic flows the edge, aggregated in the core). across the metropolitan network. However the end systems for the IP traffic MPLS defines a mechanism to set up streams are radically different in terms of Label Switched Paths (LSPs) between user interface, programmability, and endpoints at the edges of the MPLS network. sophistication. PacketCable Multimedia Packets entering the network are assigned to clients range from gaming devices to video a particular path and a label added to the phones to application software running on packet identifying the LSP to be used. This PCs. Therefore, multiple mechanisms will be label is used by the MPLS routers in the core required to provide the control path signaling of the MPLS network to forward the packet based on the capabilities of the client system. to its destination endpoint. The core network does not need to examine the packet headers Operators use the control plane to beyond the label and can therefore focus on establish the appropriate QoS between switching traffic to its destination as quickly devices, and it enables resource reservation. and efficiently as possible. Each LSP can be The client device requests the appropriate associated with a Forwarding Equivalency QoS level, and the AM authorizes or rejects Class (FEC), which defines specific QoS the request based on policies defined in the parameters so that the MPLS network can PS. Operators can therefore specify which provide a mesh of QoS-enabled paths. applications or customers have the priorities to request resources. When a client requests An intelligent edge router/CMTS platform resources, the AM will check with the PS to at the edge of the network can therefore see if the client is entitled to those resources leverage the PacketCable Multimedia based on network policies. If the PS specification to deliver QoS across the HFC authorizes the services—and if those infrastructure using DOCSIS 1.1, and support resources are indeed available—the QoS QoS across metro and core networks using level will be assigned to the application. MPLS. The CMTS/edge router/LER provides the transition points between the two QoS There are several ways to support domains so operators can deliver the QoS signalling between the client and the AM. control necessary for enhanced multimedia For example, a user could go to a Web site services. and request resources, or requests could be generated using the Session Initiation CONTROL PLANE Protocol (SIP) or the Network Control Signalling (NCS) protocol, defined by The network requires a control plane to PacketCable 1.x. signal QoS requirements and to setup and tear down QoS-enabled paths. All Signaling from the AM to the PS and from multimedia applications have similar the PS to the CMTS is conducted using requirements for QoS. They all generate IP COPS. The AM or PS can also use SNMP to packet flows (each packet of which requires check on the availability of resources within the same QoS) that can be identified using an the network. appropriate classifier. Thus a common data path forwarding mechanism can be designed The CMTS will use DOCSIS 1.1 Once the session is established, stream signalling to set up per-flow QoS control can be implemented by the Real- reservations on the access network. It will Time Streaming Protocol (RTSP), a client- aggregate multiple flows into the aggregated server multimedia presentation control QoS mechanism of the MAN and set up protocol designed to leverage existing IP aggregated QoS paths across the infrastructure. VOD signaling can be metropolitan network using MPLS Label provided by SIP and SDP, and QoS can be Distribution Protocol (LDP) or the Resource provided on the access network using Reservation Traffic Engineering (RSVP-TE) DOCSIS 1.1 and across metro and core protocol. networks using MPLS.

The key feature of the control plane as defined for PacketCable Multimedia is that it Multimedia and SIP provides a common QoS control plane within the network, which can support multiple SIP client-to-AM signaling options.

DOCSIS 1.1 Policy Server Application Manager Cable Modem ENABLING ROBUST MULTIMEDIA Client CMTS Authentication, Bandwidth Bandwidth APPLICATIONS request Control and QoS COPS/TCP

PacketCable Multimedia delivers the flexibility to efficiently support diverse Another potential SIP-based application is applications over access, metro, and core video telephony service that can establish networks. For example, MSOs can deploy peer-to-peer voice and video sessions across VOD to residential subscribers. Signaling for the cable infrastructure. VOD could be provided via SIP, with the Session Description Protocol (SDP) used for In this application the caller issues an describing the multimedia sessions, including invite using the SIP protocol, and the AM the QoS requirements. and PS combine to provide a proxy server function. Once the call is authorized and In this case, the client requests an connected, the data path is granted the application flow with committed QoS levels necessary QoS and the voice and video call is using the SIP protocol. The VOD AM will established. check video resources and may provide authentication of the client and ensure that Operators can also deploy 1:N multimedia business rules permit the flow. It will then applications between groups of users, such as pass on the QoS request to the PS, which in multiplayer gaming services. Game players turn may check with local policy prior to from leading vendors are increasingly requesting QoS from the CMTS. The CMTS network-enabled, and MSOs can leverage then provides the required connection levels PacketCable Multimedia to offer high- to the client’s cable modem. performance experiences with QoS guarantees to gaming enthusiasts seeking an interactive, real-time multimedia experience. SUMMARY ABOUT THE AUTHOR

PacketCable Multimedia allows MSOs to Gerry White is Senior Director of deploy enhanced multimedia services over a Advanced Technology for Motorola common infrastructure. It is application- Broadband’s Network Infrastructure independent, so operators can deliver Solutions business and he has presented at emerging multimedia services while retaining NCTA several times. He is an experienced the flexibility to support evolving demand for author and presenter, and has written papers new multimedia services not yet even and articles for many of the leading trade imagined. shows, industry exhibitions, and trade publications. They can take advantage of industry standards to enable end-to-end QoS across Previously, he was CTO of RiverDelta access, metro, and core networks, and they Networks, a provider of next-generation edge can swiftly develop enhanced multimedia routing/CMTS solutions. He also served as services on their own or in conjunction with Vice President and CTO of Arris Interactive, third-party application vendors. Cable a leading provider of voice and data operators can protect their investments in equipment within the CATV market. While at network infrastructure by policing traffic Arris, and its predecessor LanCity, White flows and ensuring they charge for the assumed a leading role in the industry's volume and time of consumed network efforts to define, develop, and deploy resources. The PacketCable Multimedia standards-based voice and data technology. specification allows broadband network operators to deploy enhanced IP services Prior to LanCity, White held a number of with the QoS control necessary to increase senior technical positions at Concord revenues and profits. Communications, Wang Laboratories, Hasler Ltd, and Cable and Wireless. White is the co- author of several patents and articles on data communications technology. He holds a BSc. (Honors) from University College in London. DISTRIBUTED RESOURCE MANAGEMENT FOR ON DEMAND SERVICES

Bruce Thompson Cisco Systems Inc

Abstract managers responsible for allocating those resources. There are two predominate Today’s video-on-demand (VoD) systems schools of thought on how the session are centralized in many respects – resource manager interacts with resource managers. management, session management, and storage. However, there are many benefits The first has the session manager from distributed architectures for some or all responsible for the final decision on all of the above. Many of the lessons that have resources. This model is a logical outgrowth been learned from widespread deployments of session resource management vendors of VoIP and other content-based networks who currently implement both session and can be applied to VoD networks. Today’s resource management in the same node. In video servers in many cases actually perform this model, the session manager would several logical functions, including session receive a session setup request, and then based resource management, storage, video send individual messages to various resource streaming, etc. Separating, at least at a managers to ask for the potential resources logical level, many of these components can that are available for that session. Then provide significant benefits in terms of having all of the information locally in one scalability, cost, and performance. place, the session manager decides which resources should be used for that session. One candidate for breaking out from the This model is similar to the centralized call components that actually store and stream manager model used in some VoIP networks. the video is session and resource management. And beyond simply breaking Alternatively, session resource allocation out the different logical functions from each can be distributed in more of a flow-through. other, each individual function can be also In this model, the session manager would done in either a centralized or distributed receive a session setup request, determine the way themselves. It’s the “all knowing classes of resources needed for the session database in the sky and dumb edge boxes” and then let the resource managers vs. “smarter edge devices that manage their themselves determine the actual resources to own capabilities” debate that has raged on be allocated for the session. Instead of the for years in networking, most recently session manager communicating directly manifesting itself in the VoIP debates. with each resource manager, it embeds a list of resource managers that must be used into To enable distributed resource a resource request and sends the resource management, network components request to the first resource manager on the responsible for allocating different classes of list. Resource managers then communicate session-based resources can be separated with each other to determine the optimal set from the session and resource manager. A of resources for the session. This threaded component called the session manager is model of signaling, takes advantage of the then responsible for determining the classes concept of a signaling proxy which is of resources required for a session request common in Internet protocols such as HTTP, and communicating with the resource RTSP, SIP, and others. Internet deployments of signaling proxies have shown that they are allocating QAM resources for a population of very adaptable to changes in services and set-top boxes connected to the HFC plant. that they scale to very large networks since Another resource manager called the functionality is distributed. encryption resource manager is responsible for allocating real time encryption resources This paper will go into detail on the pros for a population of set-top boxes. The on and cons of distributed versus centralized demand resource manager is responsible for control architectures for session and determining the best video server to use for resources management. It will cover specific streaming a particular asset to a set –top box. call flow examples, processing requirements, The session manager provides the scalability concerns, and where possible real interface between DSM-CC session world examples of how this has been done in components and resource allocation existing networks today. components. In the session space, the session manager interfaces to the on demand client OVERALL FUNCTIONAL BREAK DOWN running on a set-top box and to service application which is typically bundled in the The DSM-CC defines a general VoD server. In the DSM-CC architectural architectural model and signaling protocol model, the session manager appears as the for client server applications. One of the DSM-CC SRM, the on demand client architectural components the DSM-CC appears as the DSM-CC client, and the specification defines is the session and service application appears as the DSM-CC resource manager (SRM). The SRM is server. The on demand client initiates session responsible for allocating resources for requests on behalf of subscribers and receives sessions between a client and a server. To tuning information as the result of those enable distributed resource management, the requests. The service application implements function of the SRM can be broken into a service and business logic. It uses the service component that is responsible for session and business logic to specify the classes of management (the session manager) and resources required for a particular session. multiple components called resource The session manager takes the classes of managers. resources specified by the service application and interfaces with the resource managers to In this architecture, each resource obtain the resources required for the session. manager is responsible for allocating a class of resources for a portion of the video Figure 1 illustrates the classes of topology. For example, a resource manager components used to implement distributed called the edge resource manager is resource management for on demand responsible for services. Session Components multiple broadcast segments that are DSM-CC Server DSM-CC SRM DSM-CC Client combined with RF splitters and combiners. In Service Session On Demand the resulting network, the combination of the Application Manager Client RF segment that an edge device is connected to and the RFC frequency that the edge Resource Components device uses as a carrier define which set- top boxes can be reached by that edge device. Edge On Demand Encryption Resources allocated for on demand session Resource Resource Resource Manager Manager Manager requests generated by set-top boxes must take into account the connectivity of the HFC Streaming Encryption QAM network. In addition, the IP transport network Server Engine may not be fully connected or may have Data Plane Components paths between components that are not Figure 1: Distributed resource management suitable for carrying video. components HFC and IP network connectivity will Session & Resource Management Functional affect the selection of edge resources and Break Down may also affect the selection of encryption and streaming server resources. The session There are 3 major functions that need to and resource management architecture must be implemented as part of session and take the HFC and IP network topology into resource management for on demand account when determining which resource services. They are: managers as well as which resources should be used for a particular session. 1. Business Logic 3. Resource Allocation Business logic is used to tie business rules to a particular service and to a particular Once the classes of resources required for session/resource request. Rules associated a particular session are known and a set of with business logic can be translated into resource managers are chosen to allocate parameters of a resource request. those resources, the actual resources to be used for a session need to be chosen. For example, resource requests for a pay per view service may have a higher priority The resources selected for a particular associated with them than resource requests session need be optimized to take into for a free on demand service. This priority account individual resource availability as can be carried as a parameter within a well as the connectivity/availability of resource request. Rules associated with bandwidth in both the HFC and IP networks business logic can also be used to determine which resource managers should be used for In both of the resource allocation a particular session request. architectures described in this paper, the session manager implements both business 2. Video topology routing logic and video topology routing. The difference between the architectures is in The HFC network that set-top boxes and how the actual resources to be used for the edge devices are connected to is divided into session are negotiated between the session manager and the resource managers the the session manager. The session manager session manager has selected for the session. sends a resource request message to each resource manager. Each resource manager The paper will also briefly describe a then responds with a list of potential method by which the tables used for video resources (QAMs, Encryption Engines, etc.) topology routing can be populated that may be used for the session. The IP dynamically using protocols similar to those transport address of each potential resource is used to implement dynamic routing in IP included in the response. networks. Based on the responses, the session Centralized Resource Allocation manager picks a set of resources to be used for the session. In order to make an optimal In centralized model for resource choice of resources, the session manager allocation, the logic for deciding which should take connectivity and available resources should be used for a particular bandwidth in the IP network into account as session is homed on the session manager. In part of the algorithm for selecting the this architecture, the resource managers resources to use for the session. Once the themselves simply report available resources session manager has picked a set of resources to the session manager while the session to use for the session, it informs each manager looks at all resources from the resource manager which resource was resource managers and decides which selected for the session. This second message resources should be allocated for a session. exchange is also required to exchange IP transport addresses between data plane With centralized resource allocation, all components selected for the session. Figure 2 resource requests for a session originate from illustrates the resource signaling flow for centralized resource allocation.

Session Manager

R e ) R s 1 e o ( 4) s u t ( o s ) u rc e e r e s (5 R ce u n S e R q p ) e s R e

o (2) Response Resource

8 Select (3) Resource Request (1) Resource

Response (4) Response Select o e p u ( le esp q R x c u u s i e t r e e e s c s F R e o t rc R n e n ( e o s S se 1 u e c p e ) o c r p o le ( s r u s n c 2) e u o e t o s ( R s R e 3 s e t ( ) e R c 4 R ) le e S

On Demand Encryption Edge Resource Resource Resource Manager Manager Manager

Streaming Encryption QAM Server Engine

Figure 2: Signaling flow for centralized resource allocation Pros and Cons of Centralized Resource complex. If any request fails, it must abort all Allocation resource requests that have successfully completed along with all outstanding Since the session manager makes the resource requests for that session. This final decision for resource selection, the greatly increases the complexity of the centralized resource allocation architecture implementation of the session manager and provides the most flexibility in making may lead to software errors where the tradeoffs between between business logic and resources of sessions that have failed end up resource allocation logic. getting “orphaned”. Orphaned resources must be cleaned up by either rebooting the entire While centralized resource management system or by running special processes that allows flexibility, there are a number of check for resource consistency between the limitations associated with it. components of the system.

With centralized resource allocation, the There are a large number of messages that session manager must implement all of the must be exchanged between the session logic of resource allocation. In addition, it manager and resource managers in order to must coordinate all of the individual resource allocate resource for a session. At least 2 allocation transactions between each resource message exchanges (4 messages) are required manager. Because of this, the session between the session manager and each manager becomes the most complex resource manager to allocate resources for a component of the system to build. Also, since session. The first message exchange is most of the per session processing falls on needed to obtain the potential resources from the session manager, it will be the each resource manager while the second performance bottleneck of the system. message exchange specifies the selected resource and also provides IP transport With centralized resource allocation, both address information for the data plane session and resource state are shared between components to be used for the session. the session manager and resource managers. The amount of state that is shared between Distributed Resource Allocation components complicates fail over and redundancy methods for the session manager The distribution of functionality and as and resource managers. When either the well as the signaling model used for session manager or a resource manager fails distributed resource allocation is similar to over to a redundant system, both session and the model used for the Web as well as resource state must be resynchronized streaming and interactive audio and video between the session manager and resource services on the Internet. managers. With distributed resource management, Since the session manager must the session manager directs the resource coordinate transactions between multiple managers to communicate between resource managers, the process of cleaning themselves to select the optimal set of up after failed resource requests can be quite resources to use for the session. Delegating resource allocation to the resource managers The session manager includes an ordered off loads this function from the session list of resource managers to be contacted for manager and results in a more even a particular session in the initial resource distribution of functionality/processing load request. The session manager determines the between the session manager and resource order that resource managers must be managers. contacted based on the order that the MPEG stream must flow through the data plane Each resource manager appears as a components that each resource manager signaling proxy in the signaling model used controls. The order that resource managers for distributed resource allocation. Each must be contacted is opposite of the order resource manager examines the parameters of that the MPEG stream flows through the data an incoming resource allocation request and plane components. For example, the MPEG allocates resources based on the set of stream associated with a VoD session that parameters it understands. Each resource requires encryption will originate from a manager then modifies the parameters of the streaming server and pass through an request based on the resources that were encryption engine on the way to an edge allocated. Finally, each resource manager device. The resource managers associated passes the request to the next resource with these data plane components are the on manager based on a list of resource manager demand resource manager, encryption addresses that the session manager embeds in resource manager, and edge resource the initial resource allocation request. The manager. Because the video stream originates proxy signaling model enables resource from the streaming server, its associated managers to efficiently negotiate the optimal resource manager (the on demand resource set of resources to be used for a session using manager) must be contacted last. Likewise, a minimal number of message exchanges.

Session Manager S et up On Demand J1 Se ( tup 1) Resource Network Re Manager Se S spo e n tu tu Resource se p p J2 ( R ( Monitor 6) e 3 sp ) o n J3 se

(4 Edge ) Encryption up (2) Resource Set ) Resource se (5 Manager spon Manager p Re Setu

J4 J5 Streaming Encryption QAM Server Engine Figure 3: Signaling with distributed resource allocation since the edge device is the last component RTSP enables each resource manager to that the video stream passes through, its implement a 2 stage reserve/commit process associated resource manager (the edge for allocating resources. Figure 3 will be used resource manager) must be contacted first. to illustrate the 2 stage process. The first stage of the resource allocation process is This ordering results in resource managers triggered by a resource manager receiving an whose data plane components are adjacent incoming RTSP setup(1) request from its being able to exchange IP transport down stream neighbor. The resource manager parameters needed for the MPEG media examines the parameters of the RTSP setup stream very efficiently. message it received to determine which resources it needs to allocate for the session. IP transport network resources are also In Figure 3, when the RTSP setup(1) allocated in a distributed fashion. With message is received by the edge resource distributed resource allocation, network manager, it allocates resources from one or resource allocation is broken into two logical more QAMs that can be used to service the functions. They are: network resource session. Once the resource manager reserves monitoring and network resource allocation. resources it passes the data plane transport Network resource monitoring is a function parameters associated with the resources it that provides information on the available reserved to the next resource manager in an network bandwidth between data plane RTSP setup(2) message. components while network resource allocation reserves network bandwidth The next resource manager in the chain between data plane components. The allocates its resources based on the threaded architecture implements the network parameters of the RTSP setup(2) request. The resource monitoring function in a component resource manager may check to see if called the network resource monitor and transport bandwidth is available between the implements network resource allocation in potential data plane components it has the IP transport network itself. selected and the data plane transport parameters it received in the RTSP setup RTSP can be used as the resource message from its down stream neighbor. This allocation protocol used between the session check can be used to trim the list of resources manager and the resource managers. RTSP is that are reserved. In Figure 3, the encryption well suited for use in this architecture since resource manager reserves resources RTSP was designed to support proxies as associated with one or more encryption part of the signaling architecture. Figure 3 engines when it receives the incoming shows the order of signaling flow for setup(2) message from the edge resource distributed resource allocation using RTSP as manager. The encryption resource manager the signaling protocol. In Figure 3, interfaces may check to see if there is available network J1, J2, and J3 are interfaces the resource bandwidth between the rncryption rngines it managers use to request information on has selected and the transport addresses it available network bandwidth from the received from the edge resource manager in network resource monitor. Interfaces J3 and the setup(2) request. J4 are used to request resources from the transport network.

Once the resource managers have reserved resources to choose for a particular resource resources in response to the RTSP setup request based on available network bandwidth to message, the resource commit process starts. downstream components already selected for that The last resource manager in the thread session. commits a specific resource to be used for the session. This resource manager then requests The network resource monitor (NRM) the transport network to reserve network contains a map of the IP transport network bandwidth between itself and its down stream topology and the amount of available neighbor and passes the data plane transport bandwidth on every IP transport link in the parameters for the chosen resource in the VoD topology. The NRM is implemented as RTSP setup response(4). In Figure 3, the on an internal routing protocol (IRP) listener demand resource manager picks a specific (OSPF, ISIS, etc) and knows how much streaming server and output port to use for network bandwidth is currently available the session. The on demand resource between any 2 IP endpoints. To allow the manager passes the source IP address / port NRM to understand bandwidth availability, number that was selected for the session in the IRP must include bandwidth availability the RTSP setup response(4). as part of the link state attributes that are flooded. To obtain available bandwidth As the RTSP setup response is passed information on each link, the IRP must back through the thread, each resource interface to the Network Resource Allocation manager examines the resources that were subsystem in each router. committed by its up stream neighbor. It uses this information to commit its own resource. With distributed resource allocation, IP In the process of committing its own transport network bandwidth is allocated by resource, each resource manager requests the the routers in the IP transport network transport network to reserve network themselves. IP networks have traditionally bandwidth between itself and its down stream used a distributed approach to both path neighbor and passes the data plane transport selection (IP routing protocols) and parameters for the chosen resource in the managing network bandwidth (RSVP). RTSP setup response. In Figure 3, the edge resource manager commits a single QAM Distributed resource allocation uses a resource when it receives the RTSP setup combination of RSVP in the control plane response(5). It passes the HFC parameters and the DiffServ architecture in the data such as the modulation frequency and MPEG plane to allocate bandwidth and minimize program ID back to the session manager in jitter and latency for MPEG video in the IP the RSTP setup response(6). transport network.

Distributed IP Network Resource Allocation RSVP messages are passed between the data plane components responsible for With distributed resource allocation, each originating and terminating the MPEG video resource manager takes IP transport network stream. The components from Figure 3 that resources into account as part of the decision originate/terminate RSVP requests are the of which resource will be used for a session. streaming server, encryption engine, and Each resource manager may use the services edge device. RSVP messages pass through of a component called the network resource the routers of the transport network along the monitor to determine which components/ same path as the MPEG video stream. As RSVP messages pass through the transport that need to be involved in the session and network, they are examined by every router forwards a resource allocation request to the along the path. When a router receives an first resource manager in the list. The rest of RSVP request message, it first looks up the the resource management processing is done outbound interface that the packet will be by the resource management components routed to and then checks to see if that specified in the list. This distributed model of interface and its associated transport link signaling and synchronization more evenly have enough bandwidth to service the distributes the processing load between the request. If it does, available bandwidth for session manager and the resource managers that interface is decremented and the RSVP involved in a session. message is forwarded to the next router along the path. If any router along the path does not Since the resource managers negotiate and have enough bandwidth to service an RSVP communicate allocated resource information request, the router fails the request and the between them selves, the only state that is message is returned along the same path that shared between the resource managers and it was originated from, tearing down existing the session manager is active session state long the way. information. This simplifies failure / recovery mechanisms for the session manager and When RSVP and DiffServ are used resource managers since the session manager together in the network, RSVP is used to and resource managers only need to have a provide admission control functionality while consistent view of active session state. DiffServ provides the per packet scheduling characteristics required for MPEG video. With distributed resource allocation, the RSVP works together with DiffServ at the ordering of resource requests greatly edge of the network to ensure that MPEG simplifies error processing when a request video flows admitted by RSVP are marked fails for any reason. When a resource request with the proper DSCP value to ensure proper to a resource manager fails for any reason, it QoS treatment through the rest of the responds with an error to the resource network. manager that sent the request. Each resource manager then de-allocates the resources it Pros and Cons of Distributed Resource allocated on the resource allocation request Allocation and forwards the error to its neighbor. This synchronous behavior greatly simplifies error Distributed resource allocation resolves processing logic. many of the issues associated with centralized resource allocation. With distributed resource allocation, a single resource allocation protocol is used Many of the performance issues and between the session manager and the complexities associated with the session resource managers. This allows each resource manager in centralized resource allocation manager to be implemented using the same are dealt with by moving the function resource signaling state machine. It also selecting an optimal resource set of resources enables new classes of resource managers for an on demand session to the resource can be added without having to modify any managers themselves. With distributed of the existing resource managers or the resource allocation, the session manager signaling state machine of the session simply creates a list of resource managers manager. dynamic method for populating video routing The signaling model used for distributed tables allows the session manager to re route resource allocation uses a minimal number of video sessions when resource managers fail message exchanges to allocate resources for a or when new resource managers are added to session. This is because the proxy signaling a system. It also allows the session manager mdoel used for distributed resource to choose the best set of resource managers to allocation only exchanges messages between use for a particular session based on both the components that must exchange static and dynamic cost metrics associated information in order to set up the session. with individual resource managers. Most of the information that needs to be exchanged in order to setup a session needs When a DSM-CC session request is sent to be exchanged between data plane from an on demand client to the session components that are adjacent in the flow of manager, it typically includes 1 or more the media stream. .For example, to set up the identifiers that can be used by the session media stream between an encryption engine manager to determine where the clients Set and an edge device, the edge and encryption Top is located in the HFC plant. The resource managers must exchange the IP identifier may be statically provisioned in or source and destination addresses along with it may be dynamically learned. One method the UDP port numbers to be used for the to enable a set-top box to obtain topology session. The proxy signaling model uses a identifiers dynamically is to configure each single message exchange between adjacent QAM with a unique identifier and then have components to exchange transport the QAM send that identifier in the MPEG-2 information. With centralized resource transport stream it generates. The TSID field allocation, 2 message exchanges are required. in the MPEG-2 transport stream is commonly used for this purpose. To obtain these Dynamic Routing of Video Sessions identifiers, a set-top box periodically tunes to a set of pre defined “beacon” channels and When resource management functionality stores the TSID information it receives on is distributed to multiple resource managers, those channels. each resource manager is responsible for allocating a class of resources for a portion of When multiple QAMs are fed into an RF the video topology. One of the functions of combining network, their RF output is the session manager that is common to both broadcast to all of the set-top boxes the centralized and distributed resource connected to the same RF network. The allocation models is the ability for the session combination of QAMs and set-top boxes that manager to select a set of resource managers can be reached in this RF network is typically to use for aparticular session based on the referred to as a serving group or serving area. constraints of the HFC and IP topologies. edge resource managers manage QAM This function is called video topology resources for one or more serving groups. routing. When an on demand client sends a DSM- This section describes a method by which CC session request to the session manager, it the tables used for video topology routing includes the list of TSIDs it has learned in the can be populated dynamically using session request. When the session manager protocols similar to those used to implement receives the session request, it must map the dynamic routing in IP networks. Using a TSID information it receives in the session request to the IP address(s) of one or more RFC 3219[7] (TRIP) is an IETF protocol edge resource managers that can be used to that deals with the problem of translating service this request. telephone numbers into the session signaling address of a telephony gateway that can be The protocol described in this section used to reach that telephone number. While enables the session manager to build a map TRIP was originally developed for telephony that translates a set of TSIDs to the HFC services, a subset of the protocol can be used serving group and the IP addresses of one or to dynamically populate the tables the session more resource managers responsible for manager uses to perform video topology allocating resources of that HFC serving routing. group. This protocol is called the service registration protocol. TRIP allows a telephony gateway to advertise the telephone numbers it serves In addition to providing topological along with its session signaling protocol information, the service registration protocol address to other telephony gateways. A can be used to provide other operational telephony gateway that sends TRIP information about QAMs or other video advertisements is called a TRIP speaker and nodes such as encryption engine. For a telephony gateway that receives TRIP example, the service registration protocol advertisements is called a TRIP listener. should be capable of providing information about the capabilities of an encryption When TRIP is used distributed resource engine. Another important requirement for management, edge resource managers act as the service registration protocol is to support TRIP speakers and the session manager acts the ability to detect the failure of an QAM or as a TRIP listener. Edge resource managers an encryption engine in a timely manner. advertise the TSIDs and serving groups of This capability will allow the session the QAMs they manage as well as their IP manager to reroute video streams from a address. The session manager uses this failed QAM or encryption engine to QAMs information to develop a the video topology or encryption engines that are still routing map. operational. It will also allow the SM to route new video streams to the EQAMs/encryption When TRIP is used with other types of engines that are still operational. resource managers such as the encryption resource manager it will be used to advertise Finally, when more than one resource the IP address of the resource manager along manager has resources capable of satisfying a with other properties such as the conditional session request, the service registration access protocols supported by the encryption protocol can provide a static administrative engines. Since data plane components such cost associated with that resource manager as as encryption engines may not be directly well as a dynamically updated cost that can connected to the HFC network, TRIP can be be used to reflect information such as the used to advertise an administrative cost to the amount of resources currently available from different HFC serving groups that can be the resource manager. reached by that component. Figure 4 Two alternative architectures for illustrates the role of in a distributed resource implementing distributed resource allocation management environment. were presented. In the centralized resource allocation model, the Session Manager is responsible for coordinating resource Session Manager allocation between the resource managers

T selected for a particular session request. In TRIP T R R IP Listener IP the distributed resource management model, TRIP TRIP the the function of selecting an optimal set of Speaker Speaker resource for a particular session is Encryption Edge T distributed to the resource managers ResourceR IP Resource Manager Manager themselves. While the centralized resource allocation model is a natural fallout from existing SRM implementations, the On Streaming Encryption Edge Demand distributed resource allocation model allows Server Engine Device Client the implementation of the Session Manager to be greatly simplified and allows the Figure 4: Role of TRIP in distributed network to scale by using protocols and resource management architectural methods that have been widely deployed in the Internet. SUMMARY Finally, the application of a telephony This paper described a few alternative routing protocol (TRIP) for distributed methods for implementing distributed resource management in On Demand resource management in for On Demand networks is described. The use of dynamic Services. It described a model for breaking protocols such as TRIP in On Demand the Session and Resource Manager (SRM) networks will allow these networks to scale used in current On Demand architectures into as On Demand services become more components responsible for session popular. management and resource management. DOCSIS SET-TOP GATEWAY (DSG): NEXT GENERATION DIGITAL VIDEO OUT-OF-BAND TRANSPORT

Sanjay Dhar Cisco Systems, Inc

Abstract DOCSIS data IP services and a separate, The cable industry has found a perfect proprietary digital video network. weapon to create a sustainable competitive advantage against satellite operators and Set-Top Box (STB) converge services over a common, standards- DIGITAL VIDEO based network. Offering value-added services NETWORK such as on-demand video and gaming creates T much needed differentiation that satellite HF technology cannot offer. Due to the success of DATA its high-speed Internet business, cable NETWORK operators have built the largest IP networks Cable Modem PC ever assembled using DOCSIS technology. (CM) The industry is looking to extend the power of their DOCSIS data networks to the digital Figure1: Current Cable Network video part of their business. The current video network is based on a costly, proprietary architecture that does not scale to efficiently Given investments made on the video or cost-effectively support high-bandwidth, network side for U.S cable operators, the interactive services. It is essential that cable migration to a standards-based, next operators evolve their digital video networks generation architecture is expected to occur in to a high-bandwidth converged architecture phases. As shown in Figure 2, the three main that leverages prior investments in HFC cable plants and enables the video side of aspects of the video architecture are the video operations to benefit from open standards. services/application plane at the top that comprise the video application level DOCSIS Set-Top Gateway (DSG) supporting the infrastructure; i.e., application technology promises a major step towards a servers, broadcast video, VoD servers, converged data, voice, video network. DSG conditional access servers (CAS), and billing. enables standardization of set-top technology and offers the potential to reduce set-top box The middle level comprises the control plane (STB) costs by half. While Korea leads in the that involves sending control messages such adoption of this technology, major U.S. MSOs as entitlements, STB provisioning, control are examining the technology and making session and resource management, system plans to deploy DSG later this year. information and program guide information to the STB. These control messages are UDP/IP This paper covers constraints of the traditional digital video network transport encapsulated. The lowest level in the architecture and discusses benefits to migrate architecture is the network transport plane that to a next generation DSG-based network. carries video traffic and control messaging. The network architecture consists of separate CURRENT CABLE NETWORK transport for the video transport and messaging plane. Cable operators today maintain at least two distinct networks: one for standards-based

Video Services / Applications Out-Of-Band (OOB) Transport

Traditional STB technology relies on the use of a dedicated channel to transmit control Video Content OOB messaging from the headend to each STB. MPEG-2 (EMM/EPG/SI) Conditional access (CA), system information (SI), electronic program guide (EPG), emergency alert system (EAS) and other STB Video Transport OOB Transport command and control messages are sent via a DVB-ASI (MPEG-TS/ATM) downstream RF channel that is separate from the channels actually being watched. A low- bandwidth upstream reverse channel is used Figure 2: Digital Video Network Architecture for interactive messaging.

The video transport backbone is migrating The OOB carriers require separate, from a point-to-point DVS-ASI network that proprietary headend equipment such as out- has limited scalability to a switched Gigabit of-band modulators and return path Ethernet backbone to a routed IP network demodulators. The OOB channel is typically from video servers up to the edge modulator. located in a reserved portion of the A future phase of video transport involves downstream HFC spectrum at 75.25 MHz and migrating to a converged data/video uses quaternary-phase-shift-keying (QPSK) infrastructure. The control plane represents modulation with 1.8 MHz of RF bandwidth. the next immediate logical point to migrate to An OOB gateway in the headend system a converged network. The legacy video receives the content for the OOB channel over messaging control transport is referred to as an IP/Ethernet connection from an application out-of-band (OOB) since this class of server, terminates the IP/Ethernet connection, messaging is transported outside of the main and converts the content to ATM or MPEG- video transport. TS frames before passing the content down the OOB channel to the STB. Legacy Video Network Architecture Most of the signaling messages from the At the core of the traditional digital video application server are fed through another architecture is proprietary technology between software program called a carousel and the STB and the headend network broadcasted on the OOB channel. The architecture. Due to this proprietary carousel periodically sends out the messages technology, introducing innovations on the from each application in sequence. The return network side and in the STB have been slow. path in such a system is typically operated The network architecture can not be changed through a polling mechanism over the OOB. without changing the STB technology and vice versa. That has locked cable operators into older video architectures, hampered the introduction of high-bandwidth, interactive services and kept equipment and maintenance costs high.

Video Set-Top Box (STB) QAM

TV Video OOB Servers/ HFC Servers/ Mod/Dem EPG

CAS STB Controller Mgmt Systems Cable Modem PC

InternetISP CMTS

Headend Facility D-Hub/Headend

Figure 3: Legacy Digital Video Network Architecture

The upstream is a low-bandwidth NEXT GENERATION VIDEO transport used to poll STBs for pay-per-view ARCHITECTURE subscription information. The combination of the OOB downstream and the upstream As part of the CableLabs® OpenCableTM return path resembles a first generation initiative, a number of vendors and cable DOCSIS system. operators have been working to specify an open architecture for cable headend The OOB data is essentially broadcasted equipment, application servers, and digital to all STBs in the legacy network path with cable STBs. OpenCable, the standards the application level filtering the data that organization laying the groundwork for belongs to the STB. All data, regardless if it traditional digital TV and new interactive is Broadcast, Multicast or Unicast in nature, video services, has developed the DOCSIS is broadcasted. This results in inefficient use Set-Top Gateway (DSG) Interface of network bandwidth. Specification. This specification defines interface requirements for transport of out- of-band (OOB) messaging between a set-top network controller, application servers, and customer premises equipment such as residential gateways and STBs.

STB with CM

Video TV Servers/ Video EPG HFC QAM

CAS STB Controller

CMTS

ISP CM PC

Headend Facility Routed IP Network D-Hub/Headend HFC Plant

Figure 4: Next Generation Digital Video Network Architecture

The DSG specification moves away from possible with plain DOCSIS-based STBs. proprietary OOB transport to widespread, IP- The second key benefit that DSG offers is based technology, while at the same time, that it allows cable operators to leverage preserving the current control plane DOCSIS networks and efficiently support messaging. The control information is carried new STB implementations. DSG on an “always-on”channel that is separate consolidates control traffic from the video from the video delivery channel. DSG is network onto the DOCSIS network, thereby typically implemented on a CMTS which is reducing operational costs. OOB traffic can the edge device facing the HFC network. be sent using DSG over a dedicated downstream RF channel or piggy-backed on DSG Benefits an existing channel used for high speed data. Analysis shows that consolidating OOB The first key benefit that DSG offers is traffic with high speed data yields maximum that it provides a robust transport path over cost savings. DOCSIS. DSG delivers OOB messaging in one-way plants. DSG provides downstream The third key benefit that DSG offers is transport for OOB even in the presence of higher downstream and upstream bandwidth reverse channel impairments. In two-way that accelerates rollout of new interactive plants, DSG continues to deliver OOB video services. This in turn leads to higher messaging even in case of reverse channel cable operator revenues. A DSG- impairments, which otherwise would not be interoperable STB with an embedded modem makes use of the higher DOCSIS bandwidth SUMMARY to transport interactive on-demand traffic such as VoD requests, and gaming. DSG not only offers cable operators cost savings from consolidation of OOB video Delivering IP Multicast or Unicast OOB to and HSD networks, but it opens up a number Set-Top Boxes Efficiently of opportunities to generate additional revenue streams from rollout of high DSG allows integration of OOB traffic bandwidth-intensive services. DSG provides that is sent via IP Multicast or IP Unicast digital video networks a gateway to sustained from application servers on the network side innovations of DOCSIS, including advanced to be delivered on a one-way HFC capabilities in spectrum monitoring, analysis, downstream to STBs. Although a STB may and network management. have an IP address, DSG acts as a proxy device to facilitate delivery of IP/UDP OOB DOCSIS is the standard of compliance traffic to the STB. In case of Multicast and fundamental approach adopted traffic, DSG maps the incoming traffic on a worldwide to enable IP data, voice, and particular Multicast Class D address to a video services on an HFC network. DSG tunnel that is characterized by a well- OpenCable and DSG pave the way for known MAC address. This MAC address is similar advancements in video. Using DSG, unique per CAS vendor and is either cable operators can overlay operation of embedded on the STB SmartCard or legacy STB technology and migrate to open- OpenCable Point-of-Deployment (POD) standards-based STBs. Cable operators can module, enabling the STB to listen on the leverage their HFC infrastructure and right tunnel. Alternatively, MAC addresses DOCSIS CMTSs to consolidate cable are dynamically passed on to the STB by the modem and STB data traffic on a shared DSG client as defined in the DSG advanced DOCSIS channel. mode specification. The DSG advanced mode specification is part of the next Korea is spearheading deployment of revision of the CableLabs DSG specification. DSG. One of the main reasons Korea is Most of the OOB data, being Multicast in leading in adoption of DSG is due to the nature, requires DSG to support Multicast on greenfield cable video market and absence of the network interface and efficiently delivers legacy systems. The Korea Digital Cable data to a specific set of STBs. Since STBs Forum, along with TTA—the regulatory are required to operate in a one-way mode, organization in Korea, announced DSG STBs can not make IGMP JOIN requests. adoption last year. Broadband Systems Inc, a DSG acts as a proxy Multicast host and subsidiary of Powercomm Korea is the first creates a static IGMP JOIN of cable digital video provider to deploy DSG. Using interfaces. To ensure only authorized DSG, BSI hopes to lead in deploying high application servers are allowed to send OOB bandwidth interactive services like video-on- to the STBs, Source Specific Multicast can demand and popular video gaming services. be used. U.S. MSOs are not far behind. They are defining solutions for their legacy networks and plan to evaluate and deploy the technology later this year. DOCSIS® ACCESS NETWORK CONSIDERATIONS FOR CONVERGED IP SERVICES

Doug Jones YAS Broadband

Abstract service over DOCSIS, the DOCSIS STB Gateway (DSG) are planned for 2004. This paper explores DOCSIS parameters that should be managed when offering With DSG, the migration of the video multiple services over the access IP network, business to the common IP platform is DOCSIS being defined as the connection underway. Not only will cost savings be between a CM and CMTS. When offering realized by making use of the existing multiple services on a DOCSIS connection, infrastructure, there is more innovation and there will be resource contention that needs less operational expense associated with the to be accounted for. existing IP infrastructure.

These three services (and others that will INTRODUCTION be offered in the future) have different traffic characteristics and system resource needs that Business Case place unique demands on the Cable Modem Termination System (CMTS). Any single The cable industry is migrating from service is easily accommodated because the proprietary technologies to standard specific demand on the system is more easily technologies based on the Internet Protocol monitored. But when multiple services are (IP). DOCSIS is the method for moving IP offered simultaneously, the demands on the packets over a Hybrid-Fiber Coax (HFC) system develop more complex interactions. network and within DOCSIS, the Cable

Modem Termination System (CMTS) Said simply, Cable is getting to the point arbitrates bandwidth allocation. where the DOCSIS connection (CM-to-

CMTS) should be engineered. This paper The IP infrastructure has proven to be will describe several DOCSIS system both general purpose (supporting many parameters and resources that could be different applications from web and email to observed and managed as multiple services VoIP and streaming media) and low cost. IP are deployed. has stood up in the face of competition from

other networking technologies such as Specifically with HSI, a system generally AppleTalk and Novell Netware. IP ran out of DOCSIS bandwidth before other technology just works, is widely deployed DOCSIS system resources were exhausted. and available from multiple suppliers. But when offering multiple services on a

single DOCSIS connection, other DOCSIS The first service over DOCSIS was High resources may be exhausted before the total Speed Internet (HSI). In 2003, several bandwidth of the system. If the DOCSIS operators began offering telephone service network is not managed properly, services over DOCSIS as well. Trials for a third offered over that network may not get the Whereas the DOCSIS protocol by needed service assurance guarantees. definition exists only between a CMTS and a group of Cable Modems (CMs), the entire Areas of Discussion Cable IP network extends well beyond the This paper will address several types of DOCSIS network and includes both the CMTS resources that may exhaust before the regional and backbone networks. This paper available bandwidth on the system is fully focuses on the DOCSIS connection only. utilized. These are: 1. BANDWIDTH RESOURCES 1. Bandwidth Resources, specifically just the raw amount of bandwidth needed to Introduction offer services in addition to HSI on the Bandwidth is a finite resource; the DOCSIS connection. DOCSIS connection is no exception. As the 2. How DOCSIS Quality of Service (QoS) number of subscribers and services increases can be used to “carve up” bandwidth for they all have to share the available individual services such that bandwidth bandwidth. for a particular service can be exhausted while there is still bandwidth available on Figure 2 shows how bandwidth is the channel. allocated in a typical DOCSIS system. 3. DOCSIS Protocol Resources that could possibly exhaust before bandwidth runs out. 100 %

As the DOCSIS connection is used to carry more and different types of services, operators will need to implement systems to better manage the bandwidth and the services 100% of bandwidth delivered to customers over the IP allocated for HSI service connection.

OVERALL SYSTEM VIEW

available Bandwidth Total A Cable IP network view is shown in 0 % Figure 1.

Figure 2

CM PC As shown, all of the bandwidth on the

IP Home CMTS is allocated to HSI. As the number of Backbone Network Router CM CMTS HSI subscribers is still growing, an operator has to decide how to get additional

CM PC bandwidth in order to support additional DOCSIS services because adding new services will take bandwidth away from the growing Figure 1 – DOCSIS System number of HSI subscribers.

Figure 3 shows a hypothetical allocation • Migration to 256 QAM for a higher of bandwidth in a system where more than speed forward channel. just HSI service is offered. The choice should be made to fit the business 100 % needs of the services. 20% of bandwidth allocated for PCMM services 2. DOCSIS QOS RESOURCES 40% of bandwidth allocated for HSI service Introduction It is a fairly common experience to be on a network that is perceived to operate slowly. 40% of bandwidth allocated This is the result of having too much traffic for Telephone service Total Bandwidth available Bandwidth Total and too little bandwidth which can happen on any network. There are technologies that 0 % allow premium services to receive premium Figure 3 service even when the network is congested. These technologies are broadly known as In this example, the 40% of available Quality of Service (QoS). bandwidth is assigned to HSI service, another 40% to telephone service and the final 20% DOCSIS 1.0 provides tools to prioritize to new services managed with PacketCable traffic from particular CMs; however, these Multimedia. tools are not widely used. These tools provide only prioritization, not a guarantee of Looking back at Figure 2 where 100% of a certain amount of bandwidth. the bandwidth was available for HSI, either that amount of bandwidth is shrunk to 40% This is one reason why primary line voice (having an impact on the HSI service) or the deployment had to wait for DOCSIS 1.1 total amount of bandwidth available on the which included the tools necessary to offer a system is increased by 250% such that 40% guaranteed experience for a service such as of that new bandwidth is equal to the voice. DOCSIS 1.1 tools can guarantee a bandwidth assigned to HSI on the old system. level of service for voice traffic (or any type of traffic) on the HFC connection. These figures illustrate that as additional services to the DOCSIS connection, it may But QoS is not a panacea. On any be necessary to add additional bandwidth. network connection there is a finite amount Adding additional bandwidth can be done of bandwidth and it is not possible to several ways, including: guarantee service for all comers. This is similar to having too many cars in a High Occupancy Vehicle (HOV) lane of a • Splitting nodes highway. The HOV lane has a finite capacity • Adding additional upstream and but as more cars enter it, it too begins to slow downstream channels to nodes down. • Migrating to DOCSIS 2.0 for a higher speed return path Much like an HOV lane can slow down, if The DOCSIS QoS rules are very flexible DOCSIS QoS is overused the channel can allowing the operator many configuration still congest and the customer may no longer options. But QoS will not create bandwidth be receiving the service guarantee needed for out of nothing; QoS only allocates bandwidth a pleasing experience. QoS can be used to that already exists. The available DOCSIS ensure certain services get a service bandwidth needs to be managed to ensure all guarantee on a congested channel, but that services, from telephony to simple web only means another service on that channel surfing meet service guarantees. The may get less bandwidth. availability of QoS technology in DOCSIS is necessary but not sufficient to ensure all Figure 4 shows a DOCSIS connection that services will receive service guarantees. is allocated solely for HSI service. As is the business today, HSI subscribers are continually added to this channel until the available bandwidth on the channel no longer allows subscribers to receive their expected VoIP grade of service. HSI

On-line games PCMM

HSI

Figure 5

DOCSIS QoS Services

DOCSIS 1.0 is what started HSI and provides prioritized best-effort service with Figure 4 bandwidth limits. Bandwidth limits come into play when there is more available Similarly, Figure 5 show how QoS can be bandwidth on the system than traffic. When used to “carve up” the large pipe into a larger this occurs, a particular CM will be throttled number of virtual pipes dedicated to to its bandwidth limits. As more and more individual services. The individual services users are added to a channel and the DOCSIS can be constrained to the size of the virtual bandwidth becomes fully utilized, the CMTS pipe even though there may still be available will be fair to all users and it is possible that bandwidth on the channel. In this particular individual CMs will not be able to send and example, the HSI service is left to use receive data at their limits. For DOCSIS, it is whatever bandwidth is available on the the CMTS that decides which CMs get to connection. transmit and which will have to drop packets. There are many algorithms designed to drop packets according to either fairness or There should be some “wiggle room” on the business policy that can be implemented on a connection to allow best-effort services such CMTS. as web and email a chance to get through. These best-effort services can be degraded at The DOCSIS connection gracefully certain time in order to meet peak QoS loads; tolerates overload. HSI services such as email however, users of Best Effort services may and web are tolerant to packet loss, unlike notice service degradation. digital video or voice. In the case of email and web, if a packet gets dropped there are PacketCable™ Multimedia “higher layer protocols,” such as TCP, that Whereas DOCSIS provides a QoS toolkit, cause the dropped packet to be retransmitted. PacketCable Multimedia (PCMM) provides So while there may be a delay of hundreds of the tools to manage QoS based on a set of milliseconds, the packet will eventually get business rules. As shown in Figure 6, through. Many Internet services are designed PCMM exists as a layer that controls IP to be very forgiving and can recover network QoS (including DOCSIS) based on gracefully from packet loss. However, as information received from operator back- more services migrate from single purpose office systems. networks to IP networks, there may be challenges to meet service guarantees for all PCMM supports the deployment of services. general Multimedia services, online gaming, video-conferencing, streaming media, etc., DOCSIS 1.1 can provide QoS guarantees over DOCSIS. These services can be for bandwidth, latency, and jitter. However, enhanced by assigning QoS to them as if that DOCSIS connection is oversold, it opposed to services such as web browsing, e- may not be possible to guarantee the delivery mail and instant messaging where customers of all services intended to be sent over it, may tolerate some slowness. Note, even even if those services are assigned QoS. If services such as simple web surfing can be the DOCSIS 1.1 upstream channel is enhanced using PCMM; the possibilities are configured to be 5 Mbps, then only 5 Mbps limited only by what the operator chooses to can be carried. The services guaranteed for implement. delivery over that connection simply cannot total more than 5 Mbps; it’s not possible.

OSS Billing System Net Mgmt Back-office Plane

PacketCable Multimedia / QoS Policy Network Policy Plane

Metro Backbone CMTS Transport Plane Network Network

Figure 6

While telephony or voice-based services But DOCSIS is being used for more and are not specifically excluded from PCMM, more different types of services, with the PacketCable 1.x specifications provide telephony being just one example. The specific QoS management methods to this DOCSIS Set-Top Gateway (DSG) effort type of service. Therefore, those allows DOCSIS to be used in STBs. PCMM specifications should be consulted as will allow new uses for DOCSIS as well, appropriate. PCMM could also be used for including on-line games and video voice as the PCMM framework was founded conferencing. As the cost of DOCSIS upon the mechanisms defined in PacketCable continues to drop, there is also more of a 1.x. chance for lower bit rate applications such as monitoring and telemetry to come into play. Most importantly, PCMM provides the capabilities to manage the QoS tools So while it is not clear if a DOCSIS available with DOCSIS 1.1 allowing protocol parameter may ever exhaust, it is operators to establish business rules for clear that DOCSIS is being used for more offering QoS on the DOCSIS network. and more applications so it is worth a look at one such parameter that has recently been under discussion by engineers. Impact on HSI Service Service Identifier (SID) When services requiring service guarantees (QoS) are offered on the same The parameter discussed in this section, DOCSIS channel as HSI, tools will be the Service Identifier (SID), is an upstream needed to ensure the QoS services do not resource that has a defined range; there are starve the HSI service of bandwidth. A only so many SIDs that can be in use at one system such as PCMM can be used to time. That number is just over 8,000 and is effectively manage the overall bandwidth on inclusive of all the upstream channels the channel especially as operators are associated with a downstream channel. offering higher-speed HSI tiers. Therefore with a 1 x 4 CMTS card where there is one downstream and four upstream 3. PROTOCOL RESOURCES channels, there can be a combined 8,000 SIDs in use at one time across those four Introduction upstream channels. DOCSIS is a protocol which is a fancy way of saying DOCSIS uses defined formats SIDs are associated with a CM and a and parameters to move IP packets across the DOCSIS 1.1 CM can have one or more SIDs HFC. Some DOCSIS parameters have finite assigned to it. DOCSIS 1.0 CMs use only a amounts or ranges. It is possible that a single SID. The SID can be active regardless particular parameter may exhaust before the of if data is being sent through the CM or available bandwidth on the system runs out. not. This situation is expected to be extremely rare if it could happen at all. With HSI . With HSI, typically between 2,500 and service there has not yet been a case where a 4,000 SIDs will be consumed before the parameter other than bandwidth has been system becomes bandwidth limited and the exhausted. subscriber experience degrades to the point where the operator would consider methods to offer more bandwidth to those customers, SUMMARY such as splitting the node. More and more services are being moved But with a combination of lower to the common IP infrastructure. Managing bandwidth services (such as iTV and the DOCSIS bandwidth will require tools that telemetry) and high capacity return channels were not needed when just HSI was being such as DOCSIS 2.0, there is an outside offered. chance that SID depletion could occur before the available bandwidth is depleted. Several PacketCable Multimedia is a tool to factors would play into the situation and manage that bandwidth based on operator while theoretically possible, further study is business policies. But offering QoS-based needed to understand if this situation might services may require the addition of more actually happen. overall DOCSIS bandwidth.

PCMM Example As counter intuitive as it sounds, as additional low bandwidth services are added PCMM, as well as PacketCable 1.x, to a DOCSIS connection, it is possible that contain methods to dynamically assign SIDs DOCSIS resources other than bandwidth may to services that need them. For instance exhaust before the available bandwidth is when a PacketCable telephone call is placed exhausted. over DOCSIS, an additional SID will be

assigned for the duration of the call.

This dynamic mechanism provides a Author Contact method to ensure that additional SIDs are only used when needed, however, every Doug Jones active CM is consuming at least one SID. YAS Broadband Ventures 1877 Broadway Boulder, CO 80302 Telemetry Example voice: +1 303.415.9400 x303 Telemetry in this usage assumes a [email protected] persistent but low-bandwidth connection between a CM and CMTS. This connection could be used to signal the setting of a home thermostat or to poll a water meter. Such applications use only a small amount of bandwidth; however, as long as that CM is booted a SID is consumed even if there is no data being sent.

Even though these CMs send and receive minuscule amounts of data, they still consume SIDs. DVRS—DRIVING THE DVR EXPERIENCE HOME: SOFTWARE AT THE WHEEL

Neil Jones Pioneer Digital Technologies, Inc.

Abstract And most importantly, with the user interface software as the engine, DVRs will Perhaps the most powerful TV evolution has allow consumers to record and view their been the ability to let viewers take total favorite programs or series of programs on control of their viewing, particularly in their own schedules via time-shifting recording whatever they want, whenever features with a simple click of a button on they want. It’s called DVR. the remote control.

And software is at the wheel, as set-top A life changing experience is probably value is defined by its ability to support a stretch, but DVRs are allowing cable services and applications through its subscribers to do what they want with their software. viewing experiences. And it is user interface software that’s driving DVRs. This paper will discuss the DVR software solution and the impact it is having on the DVR models are in their infancy, but the cable industry and its customers. stage has been set for DVRs as a valuable feature to cable’s digital offerings. Since their introduction, Digital Video Recorders (DVRs) have changed a For example, data from Leichtman consumer’s TV viewing experience—from Research Group (LRG) reveals that as a pausing live TV to recording an entire series cable set-top box feature, DVRs make good for playback according to the viewer’s business sense, and have generated high schedule. The time shifting advantage of interest among customers. Four times as DVR is what makes it so valuable. Making many customers were interested in a DVR it easy and convenient for the consumer is when it is charged as a monthly service, what will continue to drive the category and especially among the crucial 18-34 year-old take it to new heights. age group.

DVRs were projected to be what some In addition, 31 percent of current digital experts said would be a “life changing cable subscribers expressed a strong experience” and would drive TV viewers to willingness to pay for DVRs as a monthly a new user experience of time shifted service, while experts predict anywhere programming that included controlling live from 15-37 million DVRs in consumers’ TV with slow-motion, instant replay, homes by the year 2007. pausing and re-winding, fast-forwarding, and recording from on-screen program And, according to a Forrester Research guides, and more. study, the combination of VOD and DVR is expected to reach half of US households by pleasing experience that helps make them 2007, with customers paying $6 billion for feel they are using an advanced, consumer on-demand content. Additionally, oriented, and easy to use product and consumers now spend $13 billion per year service. for home video sales and rentals. Most experts agree a significant portion of that The usefulness of the IPG is another money will be channeled into the cable key to a positive user experience. The UI business once DVRs and VOD gain traction. should allow the recording of programs directly from the IPG, along with just Getting there requires two crucial watching TV or setting up recordings of TV components to DVRs—usability and series’ making it a completely seamless, positive user experience. Those are the keys integrated part of the overall navigation to DVRs’ success, beginning with the process. Recording part of the overall UI question: What does the consumer expect and not a separate piece is essential. from this product? Timely, rich, and accurate program data By integrating software into set-top is also a key to a positive user experience. In boxes, the one box/user friendly approach to a typical cable TV headend system, the DVRs has the attention of cable operators as program data is updated each night. an appealing feature to their expanding Efficient and intuitive DVR software should digital service. We know DVRs can record, have the native intelligence to track this data pause live TV, accommodate a viewer’s and use it for series scheduling purposes, own schedule, allow broadband connectivity even if a series has been re-scheduled for a for digital music and more. And all with a different time, channel or day. simple click of the remote. We also know it’s the user interface software that will Now that you can record, and you have allow all of this to happen, and create the a cache of stored programs on your DVR’s user-friendly experience of time shifting, hard drive, how involved must the user be in recording and pausing live TV that DVR managing and maintaining the contents on delivers. the hard drive? Since there is a finite amount of space to store recorded shows, an The User Interface (UI) is a crucial important element to usability is the ability component to the DVR experience, of the software to manage hard drive space including how the UI looks on the TV. DVR for the user. By simply pressing a few is a new product and service and it runs on a buttons, users should be able to select and high-end set-top with additional capabilities see what’s been recorded and choose and over the standard digital set-tops we are all modify the way they record a scheduled used to. Among these additional capabilities recording or a series of recordings—by time, is increased graphics abilities and processing day, category, etc. There is essentially a power. The UI (software) should take full table of contents, where users can modify advantage of those increased capabilities. the parameters of the scheduled recordings Using high-resolution graphics, among other of deleted shows they aren’t interested in. UI components, gives the user a more The experience should be as easy and The result is that more DVR users go to effortless as the user wants. Make it simple see what is on their hard drive, not what is to record and to manage the hard drive. The on TV, evidenced by a Forrester study user should not have to resolve all conflicts, which revealed current DVR viewers watch but must be notified of conflicts if they arise from the hard drive about 75 percent of the while setting recordings. The software has to time. be smart enough to do most of the hard drive management. An efficient and effective Advancing developments are now default way to manage content on the hard making it easier to find what customers want drive is to maintain the content on a first-in to watch or in the DVR world, record. Yet first-out basis. The last show recorded is at there must be logic in the software to better the top of the list and the first one to go is at identify program content. It’s all in keeping the bottom. with simplifying and maximizing the user experience and managing the hard drive. Each show is stored chronologically, The better the metadata, and keeping the but in an advanced process the users would data fresh will help take DVRs to the next be allowed to move shows up and down in level. the chronological hierarchy by identifying which show is next to be deleted if space is It’s no slam dunk, however. In fact, it is needed. That show being at the bottom of very difficult, especially with the increased the list. The user can also choose to delete or demands on networks, particularly with the tag the show so that it will not be deleted added demands of high definition television. without the user manually deleting it. Each DVR hardware evolves as quickly as show has an estimated time that it will be the market demands. DVR-capable set-tops available before being deleted so that must be able to evolve and stay current with something else can be recorded. The other technologies and services offered by estimate is based on hard drive space and cable operators, like Video-On-Demand an scheduled programs and series recordings. High Definition TV. Some DVRs can record

high definition, but the effect on the hard This all goes back to the benefit of drive space can be significant. time-shifting. Even though the user may have several recordings scheduled and This may require increasingly larger several TV series set to record, the user has hard drives or the capability to have an expanded window in which to watch external, renewable hard disk drives. those programs. They can decide, which puts them in control. In the meantime, UI software can help the user optimize the use of the hard disk For example, a recent Forrester study drive by providing behind the scenes found DVR viewers increasingly watch TV defaults for hard drive management. This stored on their hard drives, and on-demand can also be augmented by providing viewing will expand from 3.5 percent of all additional capabilities to allow the user to viewing today to 28 percent in 2007. fine-tune their choices and take more control if they desire. Having high resolution graphics on the UI is using instant replay anytime, pre-scheduling even more important when the user has a an entire series of TV shows, creating a high definition television signal. Consumers personal library of favorite TV shows or have grown to expect more from the HD displaying two programs simultaneously on services and poor graphics in an HD the screen. And all with a simple click of the environment leave much to be desired. remote control. DVRs are about usability, user experience, user interface and giving From a business perspective, DVR is a the customer what they want. And user revenue generator that doesn’t require a interface software is leading the way. huge capital investment, like a plant upgrade would, prior to getting the first customer. AUTHOR CONTACT Operators can buy a DVR set-top and make their technology investment decision one Neil Jones box at a time. As in every business case, Senior VP Operations operators have to weigh the features and Pioneer Digital Technologies, Inc. functions and how they affect their business. 2210 West Olive Avenue, 2nd Floor At the end of the day, DVRs are all about Burbank, CA 91506 giving consumers options and control over T 818.295.6686 their viewing experiences—controlling live F 818.295.6831 TV and pausing a show for up to an hour; [email protected]

EMERGING SONET-BASED SOLUTIONS FOR CABLE NETWORK EVOLUTION

Thomas Darcie, Eric Manning, Dale Shpak University of Victoria Nick Tzonev Syscor R&D Inc

Abstract latency, delay, jitter, packet loss rate), and Wide-area or Metro-area Network (WAN or Approaches for the design of cable MAN) connectivity. In response, vendors transport networks to support emerging have created a variety of transport network services are compared. While SONET is approaches, each of which has strengths and established firmly for wide-area network weaknesses, with no one approach serving all (WAN) transport, a diverse set of requirements well. Each approach presents a requirements drives a correspondingly continually evolving progression of features. diverse set of choices for metro (MAN) and secondary/hub networks. We evaluate the A central point of debate within the performance of next-generation (NG) SONET networking community surrounds the efficacy and contemporary alternatives in meeting of SONET and next-generation (NG) SONET these requirements, identifying strengths and technology, in comparison to evolving native weaknesses with both. Requirements or Ethernet alternatives. Numerous business and opportunities for new capabilities are technical factors have propelled a recent wave identified and used as guidance for a cable- of enthusiasm for NG-SONET technologies. friendly third-generation SONET technology. Once an expensive and static technology supporting only TDM services, improvements INTRODUCTION in cost, proven product stability, a pervasive embedded base, and data-friendly Having made massive infrastructure enhancements (RPR, GFP, VC, LCAS investments in DOCSIS, 2-way plant, and (described later)) have made Next-Generation digital video technology, the cable industry SONET (NG-SONET) the vehicle of choice now holds an enviable position for the for major telecom service providers in both delivery of a broad range of emerging WAN and MAN. Perhaps the strongest services. Yet with a growing pool of cable- argument in its favour is the ability to modem and DTV subscribers, and the vast capitalize on the enormous existing embedded majority of homes passed or penetrated with base of SONET technology. broadband capability, impressively lucrative new-service revenues remain just out of reach. However, the simplicity, flexibility, and How can cable operators leverage the utility of native Ethernet connectivity have tremendous capability created by prior driven its migration from access aggregation investments to improve return? or local area network (LAN) to MAN. Special challenges like multicast video-on-demand Popular directions for new revenue have driven cable networks to transport- potential include various incarnations of overlay alternatives (RPR, POS) or native voice, data, and digital video services. Each of Ethernet (GbE) in some portions of the the many possibilities present specific, often network, but the resulting network may be stringent, requirements for QoS (bandwidth, less suitable for other high QoS services.

In this paper we discuss future directions SERVICE OPPORTUNITIES for SONET, and how new, emerging SONET capabilities can address transport-network To provide context for evaluating the challenges within the cable industry. When suitability of various network approaches, we we compare proposed networking approaches attempt to categorize voice, digital video, and to delivery of key service opportunities, it data services that are most likely to have becomes evident that NG-SONET plays a key major market impact. We do this by service role in providing transport for high-QoS and type, bandwidth, QoS requirements, wide-area connectivity, while overlay MAN/WAN requirements, and other positive techniques using other technologies are or negative attributes sometimes preferred for digital video delivery and local data aggregation. This unfortunate situation exacerbates network complexity, increasing both cost and technical risk.

Finally, we suggest enhancements to NG- SONET, defined collectively as third- generation (3G) SONET, that add additional networking capabilities within the transport network. An example is SONET Frame Switching (SFS) [Shpak], which provides dynamic provisioning, multicast distribution, statistical multiplexing, and various tiers of QoS support, while retaining compatibility with the existing SONET infrastructure and transparency to existing data formats. This capability can be added incrementally to legacy SONET rings by adding low-cost switch hardware and simple (transport) network control.

REFERENCE ARCHITECTURE Figure 1: Reference architecture of hierarchical cable network with distribution and aggregation through HFC (fiber nodes, DOCSIS, QAM, etc.), MAN Recognizing that there is a diverse connectivity through hub ring, and MAN/WAN assortment of network topographies in use in connectivity through head-end ring. Gateways to other modern cable systems, we base our discussion territories and networks are through the master head- on the reference architecture shown in Figure end. 1. This captures the general notion of a hierarchy of interconnected networks leading Voice from access aggregation (HFC) to a WAN network (regional, national, global) spanning Voice services fall into various classes, independent systems. depending on distribution (Voice-over-cable or DOCSIS) and network (PSTN, Internet, or managed IP network) methodologies. Voice- over-Cable (TDM-based voice modems) aggregate bandwidth in the hub network can provides complete alignment with legacy become very large, especially as servers are PSTN approaches, but overlooks the pushed to the head end, cost per unit opportunity to make good use of the features bandwidth becomes a key parameter. of DOCSIS and the PacketCable initiative. We consider here the continued requirement We assume that the tradeoff between to transport the resulting aggregated DSO/T1- server and transport cost pushes the servers level streams through MAN and MAN. This towards the hub for popular content (e.g. top is referred to as Voice-Circuit-PSTN (or 10 hits or channels) and towards the master Voice-CP). We do not consider the use of this head end for customized or archival content. circuit-based voice distribution approach [A large amount of research on content within Internet or managed IP networks. networking [Cameron] examines ways to optimize the placement of content based on Voice over DOCSIS can be supported by access patterns and frequency.] To cover this all three network approaches, with or without tradeoff, we consider three alternatives for the invoking DOCSIS 1.1 QoS features. We delivery of demand-based video. The first consider only the specific important case of a (Video Centralized, or Video-C) involves a high-QoS voice service that is desired through centralized (master head end) VOD server a managed IP network. This approach, delivering streams of video to individual referred to as Voice-DOCSIS-Managed IP customers over the IP-MPEG/QAM path from (Voice-DM), is a next step towards higher the hub. The second uses a VOD server quality that goes beyond the models of located in the hub (Video Distributed, or existing VoIP-over-cable providers (since Video D) to provide transport-efficient DOCSIS 1.1 QoS capabilities are used.) service to all subscribers sharing the hub. The third (Video Head-end, or Video H) takes the Video middle ground, placing servers in each head end to serve all subscribers sharing the hub Digital video services can be categorized network as broadcast or demand- based. For delivery of broadcast digital video, a high channel Data count (e.g., 200) is distributed from a master head end to each head end, as illustrated in Data services include a large set of Figure 1. This is then converted to QAM applications, bandwidths, service quality channels and broadcast over analog fiber to attributes, and MAN/WAN requirements. hubs and distribution. Since this is a) business DOCSIS-based consumer-oriented Internet as usual and b) does not affect digital traffic access (Data - Broadband Access, Data BA), on the hub ring, it is not considered further. represents the existing service as supported by most cable providers today. The industry Demand-based video can be defined by the seeks to exploit this base and expand into location of the server or cache (hub, head end, other opportunities. These include, in order of or master head end) and the transport or increasing sensitivity to delay and jitter: distribution method (MPEG/QAM or VLAN (including Ethernet Virtual Private DOCSIS/QAM). The analysis is not sensitive LAN - EVPLan), Network Attached Storage to the distribution method. However, since the (NAS), Ethernet Virtual Private Line (EVPL), Storage Area Network (SAN), Ethernet Data-NAS describes remote data storage Private Line (EPL), and traditional T1. on centralized or shared network-based EVPLan and EVPL represent the primary (in media. Peak transfer data rates can be high, terms of likely demand) services recently while latency is not a prime consideration. defined by the Metro Ethernet Forum (MEF) Excessive latency is often an irritant rather [Clavenna1, Clavenna2]. It is important to than a failure to deliver the contracted service. note that these new Ethernet services need not be delivered by an Ethernet network: Data-SAN describes high-performance customer ports need only have Ethernet networked storage, typically using a dedicated interfaces. network (e.g., Fiber Channel) for this purpose. Although these networks commonly exist in While numerous other definitions could be data centers, it is often desired to extend the explored, we feel that these seven types span a SAN for remote backups or for servicing representative set of likely requirements. remote locations. To retain high performance, These will be abbreviated to Data-XXX in the network must have low latency and what follows. negligible packet loss.

Many of the services discussed are In an attempt to define a data-centric supported on Layer 3 (L3) router-based evolution for T1 service, Data-EVPL networks. This adds another layer through specifies various service qualities, such as which performance must be maintained. For Committed Information Rate (CIR) and the purposes of this transport-layer study, we Excess Information Rate (EIR). This would recognize that these services require QoS from support a combination of real time (voice) and lower layers end-to-end, and do not consider data requirements across MAN and WAN. the additional complexities of router evolution. Also supported by EVPL is a wide class of interactive real-time applications (real-time Data-VLAN represents the class of Virtual interactive gaming or music, video LAN services offered typically across LAN or conferencing/telephony, etc.), in which only a MAN, based on the IEEE 802.1q portion of the bandwidth must be allocated specification. With DOCSIS as the access with high QoS. vehicle, such services are clearly an attractive complement to existing broadband services. Finally, Data-EPL specifies a committed Included here are Ethernet LAN Services (E- rate over a dedicated channel (essentially LAN) as defined recently by the MEF. Ethernet over a SONET channel), effectively providing TDM-like guarantees, but over a Data-T1 represents the vast majority of data network. Data-EPL service is much like business services provided by the ILECs to Data-T1 but has Ethernet interfaces. provide Internet and PBX connectivity, and contributing roughly $23 B in high-margin In summary, we have defined two classes revenue (In-Stat/MDR Research). A of voice, three video, and seven data services, DOCSIS–based end-to-end means to attack for a total of twelve spanning a wide variety this market would be a valuable capability. of network requirements. All appear to be

within the opportunity space available to LCAS provides a mechanism for on-the-fly cable service providers. We now explore the modification the capacity of these provisioned networking technologies that can be channels. considered in support of these services. Although Next-Gen SONET improves NETWORK TECHNOLOGY CHOICES SONET utilization, it is still a point-to-point technology, and therefore requires numerous A daunting assortment of evolving point-to-point channels when interconnecting network choices is available. We summarize multiple nodes. Each of these channels must the main approaches in Figure 2. Included in be provisioned to support the peak data rate. “other” are those approaches that have Since the channels are not shared, it does not specific but narrowly-defined applicability, provide any statistical multiplexing gain to (e.g., Ethernet over SONET, Fiber Channel). improve utilization when transporting bursty traffic. FR ATM PoS Other RPR GFP TDM Ethernet SONET/SDH Optical In recent years, Optical Ethernet (OE) has Ethernet been proposed as a transport technology. The WDM main argument supporting Optical Ethernet is that the large deployment of Ethernet in the Figure 2: Layering of transport and overlay layer-2 enterprise helps to drive down the price of networks on wavelength-division multiplexed (WDM) components. Ethernet has several attractive transmission. features, such as the ability to rapidly provision bandwidth with fine granularity and Transport Layer simple internetworking with enterprise SONET is the backbone of telecom networks. networks. SONET provides high reliability, However, there are several reasons why QoS, scalability, and mature OAM Ethernet has not made major inroads into capabilities. Accordingly, many carriers carrier networks. Firstly, carriers already prefer to extend the lifetime of their SONET have substantial investments in their infrastructure. SONET/SDH and ATM networks. They would often prefer to improve the Next-Generation SONET (NGS) is performance of these networks rather than cap currently defined as the combination of their existing networks and grow a completely Generic Framing Procedure (GFP), Virtual new transport network. Much of the Ethernet Concatenation (VCAT), and Link Capacity traffic that is carried over transport networks Adjustment Scheme (LCAS). These is not transported natively: it is encapsulated technologies enhance SONET/SDH by adding in SONET/SDH. There are other concerns flexibility to payload types and by supporting about using Ethernet in transport networks, finer-grained provisioning of SONET such as standardized support for end-to-end channels. GFP is a simple framing method QoS (perhaps using MPLS), rapid protection that can be used to encapsulate many different switching (non-standard proprietary types of data. VCAT improves channel approaches with interworking and/or utilization by enabling finer-grained performance issues), mature OAM provisioning of point-to-point channels. capabilities, and scalability [MEF].

Overlay Networks Asynchronous Transfer Mode (ATM) and SONET are the major transport technologies The most common methods for used by telcos. ATM can transport many transporting IP data over SONET/SDH are different types of packet- and circuit-oriented Packet over SONET (PoS), Asynchronous traffic types. Notably, nearly all DSL traffic Transfer Mode (ATM), Resilient Packet Ring is aggregated using ATM-based DSLAMs. (RPR), Frame Relay (FR), and Ethernet. All ATM uses a small, fixed cell size that results these technologies have strengths and in two important hardware advantages: the weaknesses that make them suitable for some resulting hardware has a simple and regular applications and less suitable for others. queue structure and ATM can have a low forwarding latency. However, ATM has a Packet over SONET (PoS) is a long- connection-oriented protocol that is not serving method for transporting packet data efficient for small data transfers that are over SONET. A major weakness of PoS is typical for some Internet communications. It that it uses provisioned point-to-point does not scale well and it is not well suited for channels between each of the network carrying transient or best-effort traffic. It is elements. Because of the point-to-point nature more appropriate for establishing network of PoS, the interconnection of multiple nodes services requiring higher levels of QoS requires a mesh consisting of a provisioned (mainly voice). Other problems with ATM are channel between every node. This reduces high cost and excessive signaling overhead. network flexibility as each of these SONET channels must be provisioned to meet peak Resilient Packet Ring (RPR) was designed bandwidth requirements and there is limited to address many of the shortcomings of PoS. opportunity to exploit statistical multiplexing RPR introduces oversubscription into to improve bandwidth utilization over the SONET. With RPR, the nodes are individual channels. addressable, can share the capacity on a SONET ring, and any node can insert traffic Frame Relay (FR) is a very important into the shared channel. This helps to commercial packet service that is somewhat overcome the poor channel capacity like X.25 with cut-through forwarding. The utilization in PoS that results from having to nodes in a FR network switch packets over use dedicated point-to-point channels. In provisioned paths in the network (Permanent addition, RPR supports differentiated QoS Virtual Circuits). A FR network may also which can be used to offer tiered Service support Switched Virtual Circuits that can be Level Agreements (SLA). RPR was designed established on demand. With FR, each to be media-independent and can currently be unidirectional link is established having a carried over SONET or Ethernet. Therefore, it guaranteed or Committed Information Rate requires its own protection switching (CIR) (which may be zero) and a Peak mechanism. RPR’s most serious weakness is Information Rate (PIR). Since FR uses shared that it cannot natively span multiple rings. channels, it has statistical multiplexing gain. Typically, a router is used to interconnect RPR rings. An advantage of FR over VPNs is that a FR connection is truly private: no TCP/IP ports are exposed to hackers.

Ethernet: In a typical overlay deployment, As mentioned above, there are issues with Ethernet switches are connected over point- the scalability of Ethernet, including the speed to-point SONET links by mapping Ethernet of network recovery when using spanning tree frames into the SONET payload. protocols.

Each Ethernet switch in the overlay TECHNOLOGY SUITABILITY network must be capable of determining the required output port for Ethernet frames In evaluating the alternative networking destined to any Ethernet address anywhere on approaches, we compare key attributes of the the Ethernet network. Unlike RPR nodes, transport-layer and overlay networks, for each which simply forward packets that are not of the 12 classes of service we have defined, destined for that node, each Ethernet switch and for applicability in WAN and MAN must resolve the destination port for every networks. Ethernet frame that arrives on any of its interfaces. The advantage of this additional First, each service class has different processing is that an Ethernet switch has the degrees of tolerance for network impairments. capability to forward traffic any output port, Table 1 illustrates the tolerance to network thereby giving it the capability to interconnect outages, delay and/or jitter, and packet loss. rings or meshes (unlike RPR). There is a wide variation in tolerance for each impairment, ranging from low for Data-SAN, Data-EPL, Voice CP, and Data T1, to high for best effort data service (Data- BA).

Table 1: Tolerance to Network Impairments Outages Delay/Jitter1 Loss Voice-CP Low Low Low Voice-DM Low Mod Low Video-C Low Mod Mod Video-H Low Mod Mod Video-D Mod Mod Mod Data-BA High High High2 Data-VLAN Mod High High2 Data-T1 Low Low Low Data-NAS Mod High Mod2 Data-SAN Low Low Low2 Data-EVPL Mod Mod Mod2 Data-EPL Low Low Low 1. Jitter is often accommodated by adding delay (jitter buffer) 2. Assumes higher-layer error recovery (TCP)

These network impairments are mapped scalability, characteristics that do not affect onto the three transport layer alternatives service directly, but are of direct concern to (SONET, NG SONET, and Ethernet) on Table the service provider. Scalability is the ability 2. Also included are transport efficiency and to migrate to large networks and high speed, particularly across WAN. SONET has the step in increasing this efficiency. Depending clear advantage in providing protected, low- on the particular approach, Ethernet delay, low-jitter, and low-loss service that is restoration time can be large or small, but scalable across WAN, but with low transport difficulties can be encountered in scalability. efficiency. NG SONET takes a substantial

Table 2: Transport Network Characteristics SONET NG SONET Ethernet Restoration Time Low Low Low-High1 Delay Low Low Mod Jitter Very Low Very Low High Loss Very Low Very Low Mod Transport Efficiency Low Mod Mod Scalability High High Low-Mod 1. Spanning Tree, Fast Spanning Tree, Proprietary Faster Methods

Overlay Networks high bandwidth (video or broadband data) broadcast or multicast services. RPR satisfies Table 3 shows the strengths and weakness a broad set of MAN requirements, but its of the various overlay network approaches inability to span rings limits applicability in within the MAN. The point-to-point and WAN. Ethernet has broad applicability in static use of transport bandwidth limits the MAN, but struggles or fails for delay or jitter applicability of POS. ATM excels for most sensitive services and services requiring high services, but has cost and scalability availability. challenges. FR is not generally applicable for

Table 3: Suitability of Overlay Technologies for MAN PoS ATM FR RPR Ethernet Voice-CP No1 Yes3 OK1 OK1,3 OK1,3 Voice-DM No2,4 Yes OK4 Yes Yes Video-C No4,5 No7,8 No5 No9 Yes Video-H No4,5 No7,8 No5 No9 Yes Video-D No4,5 No8 No5 Yes Yes Data-BA No2,4 Yes No4 Yes Yes Data-VLAN No2,4 Yes Yes Yes Yes Data-T1 No1 Yes3 Yes3 OK1,3,9 OK1.3 Data-NAS No4 OK8 OK4 OK9 Yes Data-SAN No1 OK8 OK1 OK1,9 No1 Data-EVPL NoA Yes Yes Yes Yes Data-EPL No2 Yes Yes OK1, OK1 1. Excessive latency or jitter 5. No multicast 9. Cannot span rings 2. Poor network utilization 6. Channel is not shared A. No QoS mechanism 3. Requires emulation 7. Scalability is an issue 4. Inherently point-to-point 8. Cost is an issue

A similar evaluation of overlay alternatives - Provide an efficient means (unlike PoS) to for WAN reflects a combination of the create multipoint connections across performance of the overlay approach in MAN and WAN. This requires increased aggregating traffic and the capabilities of the flexibility in defining the connectivity underlying transport network. The inherent between interconnected SONET nodes, point-to-point nature of PoS is again a preferably on a frame-by-frame basis. limitation, as is the inability of RPR to span rings. Ethernet’s inability to provide hard QoS - Provide a transport capability that and limited scalability are exacerbated in dramatically simplifies the challenge of WAN. QoS-controlled routing. This requires sufficient flexibility, in terms of both FUTURE DIRECTIONS bandwidth allocation and addressing, to be built into the SONET layer. Several conclusions are clear from the preceding discussion. - Enable monitoring, policing, and billing - No one solution fits all requirements. of rate- or usage-based service level - SONET and NG SONET, in conjunction agreements. with the appropriate overlay technology, are the means by which WAN An example of an approach to 3rd- requirements will be met now and in the Generation (3G) SONET that attempts to future. provide these features is SONET Frame - Ethernet (overlay and transport) has a Switching (SFS) [Shpak]. SFS introduces two strong place in MAN, but must invoke L3 main components: for WAN. 1) a novel frame-based protocol, and Unfortunately, QoS mechanisms for L3 2) a network switching architecture that uses WAN data services remain a key challenge. segmented, hierarchical MAC addresses. Complexity of various alternatives (e.g. MPLS, flow-based routing) continues to A frame-oriented protocol is utilized in plague economics, operations and general order to take advantage of SONET framing, acceptance of these solutions. rather than using an octet-stream-based protocol that requires additional Therefore, we seek future directions for synchronization to locate packet boundaries. SONET and NG SONET that accomplish the Rather than layering legacy protocols on top following: of SONET, SFS uses a connectionless MAC- - Increase the transport efficiency for layer protocol. Hierarchical addressing data, particularly in MAN (as per Table provides minimum byte overhead while 3), while concurrently supporting low- enabling straightforward switching and latency TDM traffic. This requires topology discovery. SFS is presented in the increased flexibility in mapping different context of a ring-based architecture but other services into large (OC48-192) and small topologies such as linear or mesh can be (OC3-12) streams, the ability to readily realized. overprovision best-effort traffic on top of guaranteed, and broadcast/multicast support. To L3 protocols, all nodes connected REFERENCES through SFS appear to be on a contiguous L2 network (the SFS “cloud”). This allows for [Cameron] C. Cameron. S. H. Lo, and D. Wei, routing decisions to be made only at the “A Server Placement Algorithm for Content ingress nodes. The traffic can then be Distribution”, Pro. IEEE Information switched between multiple SFS nodes without Theory Workshop, 2002, pp. 26-28 making further routing decisions. To a router connected to the SFS network, any other [Clavenna1] S. Clavenna, “Standardizing router connected to the SFS network appears Ethernet Services”, Light Reading, Jan 9, to be separated by a single routing hop. SFS 2004. can be readily connected to legacy networks [Clavenna2] S. Clavenna, “Future Ethernet by extracting the IP packets and forwarding Standards Unveiled”, Light Reading, Dec 8, them to an existing router or, if the SFS node 2003. has a router, by making the routing decision within the SFS node and forwarding packets [MEF] “Metro Ethernet Networks A through one or more legacy-compatible ports Technical Overview,” Metro Ethernet on the SFS node. Forum, 2003. [Shpak] D.J. Shpak, P.-C. Tang, and E.G. SUMMARY Manning, “Supporting everything-on- demand in cable services,” 19th National In summary, with its pervasive embedded Fiber Optics Engineers Conference, base, unrivaled timing precision, and Orlando, Sept. 2003. restorability, SONET provides the ideal foundation for transport of emerging cable services. NG-SONET and emerging 3rd- generation SONET capabilities like SONET Frame Switching (SFS) can provide the efficiency and flexibility previously derived from Ethernet and Layer-3 networks, but with the timing performance and reliability expected from SONET. It is anticipated that these capabilities will result in substantial simplification of cable networks and reduce dependence on complex layer-3 alternatives.

ENHANCED PLANT AND SYSTEM MONITORING USING OPEN-STANDARD DATA CHANNELS AND TOOLS

Aran Sadja, Eric Holcomb & Lee Pedlow Sony Electronics Inc.

Abstract—The complexity and scope Using only open-standards, Sony of present cable networks eclipse its was able to create a unique and analog predecessor. The demand for encompassing system monitoring tool unimpaired service availability and with unlimited expansion capabilities that network robustness has been elevated as can in real-time analyze plant data and customer acceptance and dependency has automatically report any instances where grown for new revenue opportunities the quality of service is below established such as telephony, broadband data thresholds or suspected failures were services, video-on-demand and extensive found. channel offerings, all enabled by the digital revolution in cable television. Index Terms—Digital Cable These new services depend upon Television, Systems Integration, Open networks with high fidelity. The barrier Cable, Network Management separating error-free reception and complete loss of service is often less than a decibel.

Operators faced with the I. INTRODUCTION requirement for improved service, reduced downtime and 24/7 network This paper presents an overview of availability are also facing the need to technology and tool sets developed by Sony reduce capital expenditure, truck rolls allowing operators to gather extensive and personnel costs. Current solutions metrics on their entire network, including for extensive network monitoring are transports and channels, all the way from the both cost prohibitive and have limited head end to each subscriber device. scope of information as well as limited Information is aggregated and presented in a observability, often only providing data web-based tool providing operations staff at for a node or trunk and not to each any level information with variable scope at subscriber tap or CPE. the network, node, tap or device level. The system includes assignable watermarks for Cable network infrastructures can warnings and alarms that in turn vary from plant-to-plant, node-to-node automatically activates email, text and trunk-to-trunk. From mode hopping messaging and paging systems. Trend data lasers and noisy amplifiers to ranging is obtained at selectable intervals and this cable modems, operators are challenged data is stored indefinitely for future auditing, by a diverse set of issues that may seem QA, analysis or other purposes. Because the impossible to troubleshoot with any one granularity of scope is user selectable, quick existing technology. determination of where a system fault has occurred can more easily be made, allowing

deterministic action to be taken before a call standard SNMP (Simple Network center is flooded with customer complaints. Management Protocol) and the open- It also facilitates reduction of service labor standard HTTP (Hyper Text Transfer costs and unnecessary truck rolls because Protocol) using XML (eXstensible Markup problems can be pinpointed. Problems Language) or HTML (HyperText Markup localized to wiring or other problems at a Language) both of which are also open- customer premises are determined without standards. having to dispatch representatives on-site. In many cases, these issues, the source of DOCSIS requires, and CableLabs which having been conclusively identified, ensures, that every cable modem is equipped may then be delegated to customer service with a simple SNMP agent and responds to a agents and handled over the telephone with standard set of requests. This provides a the subscriber. uniform interface to the cable modem, regardless of manufacturer. All this capability comes without the need to purchase and deploy costly and To our chagrin, the rest of a cable plant complex specialized test equipment and the is not as simple to monitor. Luckily, SNMP data server is hosted on an inexpensive PC. is the prevailing choice for remote The information is made available using diagnostics due to its explicit design as a existing communications within the standardized method of “Network subscriber device and extensions to standard Management”. Any SNMP enabled device protocols. found in the headend will respond to SNMP requests, including all network II. OPEN-STANDARD DATA infrastructures (CMTS, routers, etc) and GATHERING most servers (Linux & Windows based).

The first hurdle in creating a plant- Another common remote query wide monitoring system is gathering the method used by equipment manufactures is appropriate data. Unfortunately, a typical by performing an HTTP GET request and digital cable plant may ultimately have parsing the response data. The data is equipment from a number of different returned in either in XML or HTML format, vendors, each one possibly using their own which can easily be parsed and the proprietary remote monitoring system. In appropriate data extracted, or the data is such circumstances, a comprehensive returned in HTML, as a diagnostics web monitoring tool would require a lengthy and page from which the appropriate data can costly upgrade for each individual system, also be parsed and extracted. the difficulty of obtaining an API (Application Programming Interface) from In rare cases, the manufacturer may the vendor to communicate with their not provide a remote diagnostics interface, equipment not withstanding. in which case a simple network connectivity test can help infer whether the device has Fortunately for MSOs, most crashed, is hung-up or has lost power. manufacturers do not create their own remote data query system and protocols. The use of open standards by Instead, they use one of two major open- manufacturers allows headend equipment to standard systems. be remotely monitoring with a single tool that can then analyze and present the data. The two most frequently used methods for gathering data are currently the open-

III. OPEN-STANDARD DATA STORAGE Management Protocol) and HTTP (Hyper- Text Transfer Protocol) enabled devices can An important aspect of a large be polled on scheduled intervals for status monitoring system is the ability to gather updates. Information about video/data data over extended periods of time in order delivery, device statuses and real-time log to examine trend information and detect files can be stored on a centralized possible future failures based on that data. monitoring server thru the use of databases This data needs to be stored in a and the Linux syslog service. By conveniently accessible fashion. Enter the monitoring and analyzing the incoming data, open-standard database query language, based on MSO defined criteria, the system SQL (Structured Query Language). can automatically take scripted response actions and/or alert appropriate personal The use of SQL allows the monitoring through pagers, text messaging, email or tool to communicate, sans modification, other means. This allows efficient problem with any SQL standard adhering commercial notification and resolution, minimizes or freeware database, providing the tool with downtime, and reduces maintenance costs. a multitude of options for data storage. This data is also accessible via HTTP through any standard web browser for IV. HEADEND REDEFINED operators to examine at their convenience from anywhere in the private, or The advanced cable systems of today alternatively, in the public Internet. are vastly different from their analog ancestors. Advanced devices and Video services are protected by infrastructure that provide digital video, monitoring source devices such as satellite video-on-demand content, and data services receivers, remultiplexers, rate-shaping are replacing headends once solely groomers, modulators and upconverters. populated with simple legacy analog On-demand video systems benefit from the equipment. The replacement equipment ability to monitor items such as server CPU provides more than just a system upgrade to loads, memory usage, as well as application digital video services, but as an integral part specific metrics maximizing uptime and of a networked multimedia architecture, this system availability for these advanced equipment is also capable of reporting on services. It is also prudent to monitor the the status of the device and on the programming to ensure that the premium information flowing through it. This allows channels are encrypted, while the clear today’s system operator to monitor the channels remain in the clear. equipment and meet the challenges of managing such an environment with its Based upon the operators’ settings, the corresponding expansion of scope and appropriate support staff will be notified complexity. when there is an ASI input failure on a modulator, an oversubscription on a V. HEADEND EQUIPMENT transport passing through a grooming device, or excessive CPU load on a cluster Effective monitoring begins in the of VOD servers. Using MSO defined headend, ensuring the core systems of the thresholds, problems can be averted or cable operator are functioning as expected corrected with early warnings triggered by a and providing early detection of problems. known pattern of events leading to system The Sony system uses various failures. In many cases, problems can be communication paths to aggregate headend detected by effective application of systems status. All SNMP (Simple Network

thresholds before they become service With two-way technology, every affecting from a subscriber perspective and customer device can play a role in corrective action taken prior to noticeable monitoring the cable plant’s health. A cable impairment being reported through a flood modem can help ascertain the health of of calls to a call center. signal and assist in determining the physical location of the customer. Additionally, two- Data services can be protected by way capable STBs containing DOCSIS monitoring network infrastructure and modems can provide more than just video related provisioning servers. Everything and program guide services to the customer. from the core backend routers to edge The STB can keep the operator apprised of devices such as the CMTS are monitored every aspect of the STB in the customer’s and the data displayed in a network centric house, ranging from the current video PID to web interface. Detailed information about the CA security device ID. This enhanced the health of the HFC network as it relates to use of two-way communications allows a the data services is analyzed and presented revolution in the way cable operators as it is gathered via SNMP from a DOCSIS monitor their plant, with a seamless logical CMTS. extension of the plant past the tap to the actual devices in the subscriber’s home. Bandwidth utilization, modems per Thus, for the first time operators are node, metrics on upstream/downstream provided the opportunity to monitor the interfaces and modem RF power can be network through its endpoints, not just the stored and analyzed providing real-time infrastructure. analysis or historical trend data for the delivery system. Alerts can be targeted to VII. CABLE MODEMS specific personnel specializing in the type of trouble or equipment reporting the The cable modem is the customer’s malfunction, streamlining problem high-speed gateway to the internet. In order resolution. to stay online, a cable modem performs a continuous electronic repartee with the VI. CUSTOMER PREMISES REDEFINED CMTS and in the process creates a plethora of useful network information stored on both With one-way technologies, the cable modem and CMTS and available determining the quality of signal at the thru the open SNMP standard. customers’ premises required the customer to call in and complain. The operator then The DOCSIS spec requires all cable faced the challenge of determining whether modems to be SNMP capable and that the the issue was confined to the customer’s MIBs (Management Information Base) and location or if it was an area wide issue, OIDs (Object IDentifiers) comply with affecting a branch, node or an entire hub. If standardized values. Because of this, every no other customers complain, it could be CableLabs certified cable modem responds assumed that the issue was localized but in a uniform way to a standard SNMP without further information, it impossible to request. This provides the operator with the conclusively determine without a truck-roll. freedom to choose any (or multiple) cable A technician must then be dispatched to the modem manufacturer(s) and never have to location, costing the operator in both time change the method of acquiring data. and human resources, to determine the specific source and location of the problem. The type of data available on the cable modem ranges from simple to obscure. Whether the operator wants to know if a

cable modem is online, or a technician is interface it is communicating on. When concerned with the number of T3 Timeouts, linked to the billing system, the monitoring this information is accessible thru the tool can provide the operator with specific monitoring tool. customer information for each cable modem and STB. Of the data available, one of the most valuable pieces of information to the The data available on the cable modem operator is the ability to get live reports on can help an operator at any level determine a signal health, a key element of which is the problem with a single cable modem or with upstream power as sent by an individual a whole trunk. This information is not only cable modem. When a cable modem has to available remotely, but can proactively “shout” to communicate with the CMTS, the monitor, alert, and even maintain a cable monitoring tool can help differentiate plant’s health. whether the issue is indicative of a poor return path due to a problem at the VIII. TWO-WAY CAPABLE SET-TOP customer’s site or a node wide issue. BOXES

In some ways, both cable modems and The digital STB is seen as the digital STBs are more finicky about signal customer interface to the digital television levels and quality then their analog realm, providing the user with a crystal clear predecessors. In analog systems, picture image and an electronic program guide. quality degraded proportional to the However, with a two-way capable STB degradation of the incoming signal quality. implementing the open TCP/IP standard, the In contrast, the magic and curse of digital box can also become a cable provider’s most transmission is that it provides perfect, error reliable monitoring tool, irrespective of free picture quality until the signal is whether the television is on or off. degraded to a point and then the complete transmission is lost. In many cases, the Unlike cable modems, there is yet to transition between perfection and total loss be a standardized interface for remote may only be a fraction of a decibel. It is diagnostics on digital STBs, but the use of important to maintain proper levels across a SNMP on a network-attached STB can plant to provide consistent service and provide a wealth of information. The Sony margins. By polling cable modems at STB, which has an integrated DOCSIS cable regular intervals and following trends in modem and uses a proprietary MIB set was signal levels, it is possible to detect current used in the design of our tool, but a and even predict future equipment failures standardized set of MIBs can be created to on the delivery infrastructure. allow easy integration of any two-way STB onto a monitored cable plant. Thru SNMP it is also possible to track a customer’s cable modem should it make Whether the STB has an integrated its way to another part of the system. With DOCSIS cable modem or uses an Ethernet our system and a network diagram, it is connection, the data available from each possible to determine with a fair degree of subscriber’s STB is priceless to the operator. accuracy, where that cable modem is For example, thru the Sony system, a physically located, and from previous polls, region-wide STB software upgrade can be where it should have been located, based on scheduled and the tool can provide an the time tick value, the upstream and upgrade report as well as schedule downstream frequencies and which CMTS subsequent upgrade attempts for failed boxes. If a box repeatedly refuses the

upgrade, the tool will alert the operator, notification and response system that is whom can then take appropriate action. accessible, intuitive, customizable, and capable of filtering the massive quantity of Information about the digital video data accumulated while monitoring a cable signal is also available on the STB. A plant. digital cable STB is essentially a debug tool embedded in the customers’ home. While To make the data available to the decoding the video, the box is also widest audience possible it is advantageous monitoring various aspects of the signal and to use an existing open-standard protocol this information is available via SNMP. that is already ubiquitous, so HTTP or the Mismatched audio & video PIDs can easily “web” was a natural solution. The use of be detected by a poll of the system, QAM HTTP allows worldwide access to the data data, downstream power levels, bit error thru any browser-enabled device, ensuring rates, SNR, mislabeled channel descriptors availability to technicians with a computer, and any other information specified in the cell phone, PDA, etc. The data presented can MIB can be monitored. In the event of a be filtered depending on the device and user, customer call, this information can be used optimizing the time needed to diagnose a to assist the support technician in problem. determining the cause of the problem. One aspect that has proven both CA (Conditional Access) information attractive and effective for a major cable is also made available thru remote operator was to provide key field staff diagnostics. Theft of service can be inexpensive laptop PCs and cable modems. addressed through real-time monitoring of This allowed these personnel to access the CA using the monitoring tool. When tied monitoring data in the field from any into the billing system, the monitoring tool available cable network tap to evaluate, in can track PPV or VOD purchases, check real time, the effect of their corrective when the last account report-back occurred, actions on the aggregate and individual compare service tier to actual viewing, and downstream customers. Sometimes data in the case of a roaming or rogue access was monitored in this manner even while cards or devices, an alert can be sent out and climbing a pole! the offending card then be deactivated. A monitoring system is not effective Additionally, a cable operator may without alarm and warning notifications. want to take advantage of the fact that real- Gathering data for presentation is useful, but time viewing statistics and habits can be requires constant monitoring in order to gathered from the STB, within the privacy repair problems as soon as they arise. To be limits allowed by law. A system poll can a proactive monitoring system, definable tell the operator how many STBs are tuned alarms and subsequent notifications are to any particular channel, and stored data necessary. can be analyzed and used for marketing purposes. All the data is sorted and formatted for presentation; visual contrasts (colors, font, IX. PRESENTATION, NOTIFICATION & size etc.) are used to delineate sources of RESPONSE information and to indicate warnings and alarms. Additionally, personnel can have a Collecting data is half of a monitoring customized view of the data that addresses system; the final challenge in creating a their specific needs. versatile solution is providing an interface,

For security, the web interface can This comprehensive monitoring either be made publicly available, or a user solution saves time by freeing personnel from authentication system can be used to protect repetitive tasks that can easily be automated. the pages. Monitoring system administrators By analyzing data and sending out have a separate login and can adjust alarm notifications, repairs can be made before the watermarks, data sources, email recipients customers’ viewing experience is and other aspects of the tool thru the web. compromised. This not only saves money by accelerating repair it also keeps the Gathering data for presentation is useful, customer satisfied, reducing churn and but requires constant monitoring in order to disconnects. repair problems as soon as they arise. To be a proactive monitoring system, our system Monitoring of backend networks in allows definable alarms and notifications as addition to the statistics gathered from necessary. The use of the open-standard, advanced 2-way CPEs (Customer Premises SMTP (Simple Mail Transport Protocol) to Equipment) employing DOCSIS provides email notifications allows a multitude of metrics on the entire HFC network. When devices to receive the warnings. this monitoring end-to-end paradigm is applied to the MSO’s network, the 100% Email alerts are fully customizable and uptime required by today’s demanding can be sent to any number of recipients and customer can be better realized. different alarms can be sent to only the appropriate personnel. It is also possible to REFERENCES tailor the notifications to the receiving device; a terse version of the alert can be sent [1] Data-Over-Cable Interface to pagers, cell phones and other mobile Specifications: Radio Frequency Interface devices, while a fully detailed message is sent Specification (Cable Labs SP-RFIv1.1-I08- to standard email accounts. 020301:2002)

The tool can also be configured to [2] ANSI X3.135-1992, Database Language perform automated responses to failures that SQL have definable remedies, e.g. When a device fails, the monitoring tool can switch the [3] ISO/IEC 9075:1992, Information system over to a backup device or when a Technology-Database Language-SQL simple power-cycle is the accepted resolution to an issue, the monitoring tool [4] Jonathan B. Postel, SIMPLE MAIL will contact the network-attached power- TRANSFER PROTOCOL, RFC 821, switch to reset the equipment. Information Sciences Institute, University of Southern California X. CONCLUSION [5] D. Harrington, R. Presuhn , B. Wijnen, Cable Headends are complex and An Architecture for Describing SNMP dynamic environments that may be difficult Management Frameworks, RFC 2571, The and costly for personnel to keep track of, but Internet Society (1999) by incorporating open-standard monitoring solutions, we have created an automated [6] Standard Generalized Markup Language system capable of monitoring an HFC (SGML) ISO 8879[SGML] network from headend to customer node.

Aran Sadja is in Digital Platform of America (DPA) Tel: 858-942-1438 e-mail: [email protected]. Eric Holcomb is in Digital Platform of America (DPA) Tel: 858-942-6089 e-mail: [email protected] Lee Pedlow is in Digital Platform of America (DPA) Tel: 858-942-2538 e-mail: [email protected].

EVOLUTION OF VIDEO ON DEMAND ARCHITECTURES

Weidong Mao, Kip Compton Advanced Engineering, Comcast Cable

Abstract • Edge device management and configuration This paper describes the architectural • Shared infrastructure with other evolution of Video On Demand (VoD) services, including IP-based streaming infrastructures in supporting open interfaces media services and expandability to future on demand • Expansion to multiple services such as services. Key technology areas that are networked PVR and interactive essential for the next generation on demand advertising architecture are discussed. ISSUES WITH EXISTING SYSTEMS With the advent of new technologies in Gigabit Ethernet IP networking and high Cable operators currently offer Video On performance storage/streaming as well as Demand (VoD) services through the software, it becomes feasible to evolve the interactive video systems that feature tight current VoD architecture from a vertically integration and customization across several integrated and proprietary system to an open system components, such as: asset and modular architecture. Significant management, session and resource economic advantages are realized as a management, billing and entitlement, network flexible and open environment makes it transport, and set-top client applications. possible to deploy new technologies much more efficiently, facilitating the rapid While today’s monolithic system proliferation of new services while controlling architectures helped cable operators bring a the associated capital and operating compelling product to market very quickly, expenses. there are several issues in the current VoD

architecture that must be addressed going This paper presents a reference forward: architecture that describes logical components and interfaces. In addition, it provides a detailed discussion on several key • Open Interfaces: In many of today’s areas of architectural evolution. They VoD architectures, the interfaces and include: protocols between different components are poorly defined and proprietary. Without standardized and • Standardized asset propagation and open interfaces, significant effort is management interfaces required to integrate any new vendor • Better utilization of resources via into the architecture. Earlier dynamic session and resource standardization efforts such as Time management Warner’s ISA and CableLab’s VoD • Secure content delivery via session Metadata project have addressed some based encryption and/or pre- of these issues. However, the encryption techniques interfaces related to key components, such as Session and Resource Management and Edge QAM, remain support other on demand services such as largely un-addressed. Switched Broadcast Video or networked PVR. • Multiple Service Support: Today’s architectures are typically customized The architecture is partitioned functionally for a very limited set of services (e.g. into a number of logical components. Each Movie on Demand and Subscription component is defined in such a way that the VoD). Unfortunately, a significant re- interchangeable module implementing the engineering effort is required to common interfaces can be introduced to work support the addition of new services, with the rest of the system. It is possible that like Switched Broadcast Video, or implementations may integrate several networked PVR. An extensible, on- components into a single product or solution. demand platform that allows multiple services to share the same underlying Each logical entity described in the infrastructure will create significant reference architecture may represent one or cost efficiencies and make it possible many physical entities in an actual to provide new services more quickly implementation. For example, multiple and easily. servers may be utilized to provide load balancing and scalability for the Session • High Performance, Scalability, and Manager (SM) component. Redundancy: The existing VoD The On Demand Client is typically located system deployments are typically at the customer premises in a digital set-top designed for 10% simultaneous usage. box. Any gateway server that is Content is typically duplicated on each communicating with other headend video server system at each server components on behalf of the digital set-top location. In addition, the network and box will be considered part of the On Demand edge resources are typically Client. All other components are located at “hardwired” or associated with a cable operators’ master headend, secondary specific server. To support services headend, or remote hub, depending on the such as HDTV on demand and specific deployment configuration and networked PVR, which will produce network topology. higher bit rates and viewing concurrency, a high performance and Key logical components include: highly scalable distributed approach with fail-over capability is needed. Asset Distribution System (ADS) – distributes assets from content providers’ or The following sections provide a VoD aggregators’ premises to the network reference architecture and discussions around operators. how today’s VoD architecture might evolve to address these issues. Asset Management System (AMS) – validates and manages the life cycle of assets and their associated metadata. REFERENCE ARCHITECTURE Real Time Source – generates assets from Figure 1 provides a Next Generation VoD real time encoders and / or broadcast feeds. reference architecture that supports Video On Billing System – manages customer billing Demand services and can be expanded to and service subscriptions.

Entitlement System (ES) – manages Edge Device – performs re-multiplexing entitlements and transactions. and QAM modulation. Navigation Server – presents assets and Network Management System (NMS) – service offerings; manages navigation from provides network management for all the the subscribers. components in the headend. Purchase Server – receives purchase authorization check from the subscribers and SESSION AND RESOURCE validates via the Entitlement System. MANAGEMENT

On Demand Client – provides interfaces In order to achieve the goal of managing with the headend components and enables end and sharing resources dynamically across all user applications. services, the reference architecture separates Asset Propagation Manager – manages Session Management and Resource asset propagation across multiple streaming Management. The Session Manager (SM) is servers. responsible for managing the life cycle of individual sessions for each on demand On Demand Resource Manager – service. The Resource Managers (RM) manages resources required from the manage shared resources (streaming, network, Streaming Servers. encryption, and edge) on the system. The Streaming Server – stores and outputs Session and Resource Management (SRM) contents and manages stream control. reference architecture and interfaces are Session Manager (SM) – manages session described in Figure 2. life cycles for on demand video services requested by subscribers. Typically, the SM will perform the following functions: Conditional Access System (CAS) – performs Conditional Access for the on • Communicate with the subscriber demand video services. device regarding session setup, session Encryption Resource Manager – manages status, and session tear down. encryption configuration for each session. • Interface with the corresponding Purchase Server to authorize the Encryption Engine – performs encryption session requested by the subscriber. of video services associated with the session. • Allocate the resources required for the It can be located anywhere between video session by negotiating with the server and edge device. resource managers for appropriate Network Resource Manager – manages server and network components. resources required in the transport network for • Dynamically add, delete, or modify each session. the resources associated with the Transport Network – transports video session to support integration of services from streaming servers to the edge. multiple on demand services. • Manage the Quality of Service for the Edge Resource Manager – manages session. resources required at the edge for each • Manage the life cycle of the sessions. session.

One of the main functions of the SM is to SM. In this model, the SM collects the obtain the required resources for each session choices provided by each resource manager by negotiating with resource managers for the and selects an appropriate combination of relevant server and network components. resources to enable the session. Several These tasks typically include the following: optimizations of this model are possible to reduce latency and increase the throughput of • Interface with On Demand Resource session and resource management. Manager to determine the Streaming Server resources such as asset location, allocated streaming server and output MULTIPLE VoD SERVERS port. (Interface S3) • Interface with Encryption Resource Another key area of evolution in the VoD Manager to determine encryption architecture allows multiple video servers to resources required for the session. operate in the same headend environment. (Interface S4) This enables multiple video server vendors • Interface with Network Resource using the common open interfaces and helps Manager to determine the uni- facilitate the over-subscription of VoD directional path that will transport the storage and streaming resources within or requested video stream to the edge across service groups. A reference devices that cover the service group architecture and interface diagram using the in which the subscriber resides. multiple servers is shown in Figure 3. (Interface S5) • Interface with Edge Resource Manager The On Demand Resource Manager is to determine the resources used at the responsible for allocating and managing the edge devices such as bandwidth resources that are required from the Streaming required and MPEG tuning parameters. Servers. Upon the session setup request from Digital set-top boxes can then tune to the client, the Session Manager (SM) will the MPEG program that carries the request resources from the On Demand requested content. (Interface S6) Resource Manager (via Interface S3), in conjunction with the resources of other The session and resource management components in the overall system. The process may follow any of several possible resources allocated by the On Demand implementation models: In a centralized Resource Manager may include: model, the SM will retrieve and aggregate the topology and resource information from each • Asset location: This includes the resource manager. The SM then updates this locations of the requested asset that information as necessary, assigns resources, has been determined by the and communicates with each resource propagation service. This information manager to configure those resources. may be retrieved from the Asset Propagation Manager (via Interface In a distributed model, each resource R1). manager is responsible for maintaining and updating the topology and resource state of • Server resource: This includes the the devices it manages as well as allocating availability of the Streaming Server the resources for the session on behalf of the that contains the asset and covers the

service group in which the requested • Content duplication: determine subscriber resides (via Interface R2). whether the content needs to be duplicated in a distributed manner. • Network resource: This includes the network resource allocated at the The interface between Asset Propagation selected Streaming Server output port Manager and Streaming Server (Interface A3) (via Interface R2). They may include is defined so that Streaming Servers from the UDP port number and IP address multiple vendors can be introduced to work that carries MPEG SPTS. within the same propagation service framework. It is essential that this interface The Session Manager (SM) will need to hides the internal implementation of each negotiate with the On Demand Resource Streaming Server’s storage system. The Manager as well as the resource managers for interface may include parameters such as the any other components which allocate required storage capacity, the available resources to enable streaming video from any storage capacity, service group coverage, and server to any edge. For example, the asset whether to duplicate a content file. files may not be available in the streaming server that is connected to a selected network The On Demand Client interfaces with the path to the edge, creating the need for an Session Manager for session signaling alternate server and network path. In this (Interface S1) and with the Streaming Servers case, the SM will need to negotiate with the directly for stream control (Interface C1). On Demand Resource Manager and other resource managers to reconcile the differences. SHARED EDGE RESOURCES

Single or multiple Streaming Servers may Another key area of evolution of the VoD be deployed across the network. Streaming architecture is the availability of a common, Servers may be deployed at a centralized open interface that enables multiple services headend or at distributed remote hubs or both. to share multiple edge devices on the same The choice of deployment architecture will network at the same time. This can be ultimately be driven by a number of factors, achieved by introducing standardized such as: operational feasibility, network interfaces between the Edge Resource transport availability, scalability, content Manager, the Session Manager, and the Edge caching and propagation, and the overall cost. Devices.

The Asset Propagation Manager is The Edge Resource Manager is responsible responsible for propagating the assets coming for allocating and managing the resources that from the AMS to the appropriate Streaming are required from the Edge Devices. Servers. The policies in the Asset Typically, the Edge Resource Manager needs Propagation Manager may be determined by a to know the topology of the service groups number of factors. For example: that Edge Devices are serving. When the Session Manager requests resources for a • Storage capacity: determine if there is specific session, the Edge Resource Manager enough storage for content files. needs to determine which Edge Device to use, specify input UDP port and IP address to be

utilized, define output MPEG program required to scramble the packets. EMMs are parameters, and set the output RF frequency. used to enable specific users to retrieve the ECMs that are required to decode the control Other functionalities of the Edge Resource words and de-scramble the packets. Manager may also include bandwidth management and quality of service. For In the case of pre-encryption, example, additional edge bandwidth may be ECMs/EMMs are generated a manner that required from a QAM to dynamically add enables a group of digital set-top boxes to content to an existing session. Quality of access content that has been pre-encrypted Service can also be provided by using and stored at the server ahead of time. In case techniques like MPEG bit rate reduction. of session based encryption, ECMs/EMMs are generated and assigned to a particular session. The main function of the Edge Devices is The content has to be scrambled on the fly at to receive the multiple MPEG Single Program the Encryption Engine. Transport Streams carried over UDP/IP from the IP transport network, multiplex them into Whether the content needs to be encrypted MPEG Multiple Program Transport Streams, may be determined by a number of factors. and generate QAM modulated signals. The Content providers can require the asset to be other features of Edge Devices may include: encrypted by enabling the “Encryption” field in the corresponding asset metadata file as • MPEG PID (Packet ID) and / or TSID defined by Content Specification 1.1. (Transport Stream ID) remapping Network operators can also require the • PCR (Program Clock Reference) re- specific service to be encrypted. In addition, stamping the system shall be able to identify which CA • Statistical multiplexing system to encrypt the content in case of • Bit Rate Reduction multiple CAS headend.

The Encryption Resource Manager is SESSION BASED ENCRYPTION responsible for managing the Encryption Engines and provisioning the encryption Session based encryption is the ability to resources required by sessions. dynamically encrypt on demand content in real time using a Conditional Access System The Encryption Engine performs real time (CAS) that the set-top box supports. This encryption of the MPEG-2 packets carrying makes it possible to enable multiple on demand content. It can be located conditional access systems to support legacy anywhere between Streaming Servers and as well as next generation set-top boxes. edge devices. For example, the encryption engine may be embedded in a multiplexer or The encryption of digital services can be edge QAM device. In order to perform the achieved by using the Entitlement Control session based encryption, the Encryption Messages (ECM) and Entitlement Engine needs to retrieve the appropriate Management Messages (EMM). ECMs are parameters such as the ECMs from the used to secure the control words that are corresponding CA system.

MULTIPLE SERVICES for time-shifted content, just as any video on demand service. In order to enable multiple services over the same on demand platform, a service • For live broadcast, the subscriber will architectural model is proposed (see Figure automatically trigger a Networked 4). In this model, each service will be able to PVR session by issuing a command use one type of the Session Manager. Several such as Pause or Rewind. different types for Session Managers can be defined to further optimize the protocols and The Networked PVR services impose actual implementation for a variety of significant challenges on the performance of applications. the overall system. For example, the Streaming Server will be required to handle Networked PVR large amount of real time stream ingest. The Session Manager and Resource Managers will Networked PVR (nPVR) services allow be required to manage a large number of the subscriber to watch broadcast simultaneous sessions in cases such as a programming on demand and interact with a popular live broadcast. The architecture and live broadcast programming (e.g. pause or interface design must take these issues into rewind). To effectively achieve this goal, the consideration. network operator must record, manage, and stream broadcast programming content in Interactive Advertising real-time based on each subscriber’s requests. The Next Generation On Demand Video The reference architecture and associated Architecture opens new opportunities for interfaces can be extended to support these providing innovative interactive advertising. features by adding a few more capabilities. In For example, advertisement can be inserted at particular, the beginning of a VoD session. The advertisement can be either determined • Real time asset ingest must be statically based on the asset metadata or supported to the Streaming Servers. In dynamically targeted to a particular subscriber particular, this includes the ability to based on a set of business rules. import metadata, like programming schedule information, into the Asset From an architectural perspective, there Management System. are several areas of interest surrounding interactive program insertion services. They • The segmentation of the digital video include: where the digital insertion will programming (start and end of the happen, how it can be done, and what programming segments) needs to be determines the advertising play list. addressed. An operationally friendly scheme is also required to address A digital program can be inserted either at programs that start late or overrun the Streaming Server location or at the Edge. their original schedule. Insertion at the Streaming Server provides an integrated approach and can leverage the • The subscriber will be able to perform existing storage and streaming infrastructure. asset query, purchase authorization Insertion at the Edge will allow separately request, and session setup / teardown managed Ad content servers to interface with

Edge Devices, eliminating the need for a sessions. For each channel change, the given Ad to be stored on the same server as subscriber will send a message to the Session the content it is being inserted in. Manager, which, in turn, determines if the requested channel is already being sent to the In both cases, CableLabs Digital Program subscriber’s corresponding service group. If Insertion standards can be used to define the the requested channel is already being sent to cueing messages that are required for splicing the service group, the session manager will MPEG-2 streams. In addition, digital instruct the subscriber to join the existing program insertion into encrypted streams broadcast session. A new broadcast session is should be handled properly. assigned if the requested channel is not already being sent to the service group. The Switched Broadcast Video Session Manager will negotiate with the

The architecture can be extended to Resource Managers to allocate resources support Switched Broadcast Video services. A required for the session. The Edge Device switched broadcast system only sends the needs to dynamically retrieve the MPEG digital broadcast video streams that are being single program transport stream that carries watched to their corresponding service the requested broadcast program (likely via IP groups. In addition, a subscriber can join an multicast) and generate the MPEG multiple existing multicast that is available to its program transport stream. As part of the corresponding service group. session setup response message, video tuning parameters, such as frequency and MPEG In more precise terms, Switched Broadcast program number, are sent back to the Video is a tool to save bandwidth rather than subscriber to access the requested broadcast a new service. From the subscriber’s channel. perspective, he or she still receives the same broadcast video service when switched Switched Broadcast Video imposes broadcast is used. In fact, an ideal switched specific requirements on the performance of broadcast implementation would make it the overall system. For example, the impossible to tell that the stream was switched broadcast session setup/channel change at all. If each one of the digital broadcast latency needs to be minimized to achieve channels is being watched by a subscriber in desired channel change response time. In the same service group, the Switched addition, frequent channel hopping in peak Broadcast Video approach does not yield any time can cause significant upstream traffic bandwidth savings. However, the more likely required to carry session messages. It is situation is that several channels in each necessary that the architecture and interface service group would go ‘unwatched’ at any design take these issues into consideration. given time. This “concentration ratio,” (i.e. the percentage of all available channels that Streaming Media to PC are watched by each service group) is the It is desirable to share the on demand video primary driver of any bandwidth efficiency service infrastructure among multiple services that can be realized through a Switched and multiple devices, including the Streaming Broadcast Video solution. Media services to PC and other video enabled One way to extend this architecture to devices. support Switched Broadcast Video is to utilize the Session Manager to manage broadcast

There are several aspects which should be REFERENCES considered when expanding the architecture to support shared Streaming Media platform. [1] CableLabs Video On Demand Content Specification Version 1.1, April 15, • Shared asset distribution and asset 2003 management system. • Shared session and resource [2] CableLabs Asset Distribution Interface management. Specification Version 1.1, April 15, • Shared streaming servers. 2003 • Shared entitlement and billing system. [3] Pegasus Session Setup Protocol, Version Future digital set-top boxes may support 1.0 and Version 2.1, Time Warner Cable advanced codecs, such as MPEG-4/AVC or Window Media 9 in addition to MPEG-2. [4] Pegasus Lightweight Stream Control They might also be able to receive video Protocol, Version 1.0, Time Warner content over IP/DOCSIS. The VoD Cable architecture evolutions discussed in this paper can be extended to support these capabilities [5] Pegasus Interactive Service Architecture in the next generation all digital platform. Version 1.3, Time Warner Cable

[6] ISO/IEC 13818-6: MPEG-2 Digital SUMMARY Storage Media – Command and Control (DSM-CC) This paper describes a next generation VoD architecture model and its associated [7] ISO/IEC 13818-1: MPEG-2 System interfaces that will enable multiple services to be provided through multiple servers, multiple [8] ISO/IEC 13818-2: MPEG-2 Video edge devices, and multiple customer premise equipments using the same underlying [9] IETF RFC 2326, Real Time Streaming infrastructure. Open interfaces among the Protocol (RTSP) various components are the key to achieving this goal. Work is on-going to finalize the [10] Insertion Cueing Message for Cable interface specifications and creating a Television System deployment migration strategy from today’s VoD architectures to the next generation [11] ANSI/SCTE 30 2001, Digital Program framework. Insertion Splicing API

HIGH AVAILABILITY CONSIDERATIONS FOR AN END-END CABLE IP NETWORK

Navin R. Thadani Cisco Systems, Inc.

Abstract highly available and reliable services such as voice over IP and digital broadcast video. As Cable MSOs start deploying PSTN It outlines the evolutionary path of the replacement voice services (over IP), current High Speed Data IP networks to availability and reliability become important highly available service delivery platforms considerations. However, there is a lot more in the future. to high availability than the number of nines on a certain network device. PSTN AVAILABILITY MYTHS

In addition to looking at device level The issue of availability is surrounded by availability, it is also important to several myths and misconceptions. Three of understand the availability of the end-end the popular myths are mentioned below … network taking into consideration the 1. The PSTN provides 99.999% (five 9s) various system level inter-dependencies. of availability and reliability end-end. Further, it is even more important to 2. One needs fives 9s level availability on understand, evaluate and design a network every platform in order to achieve keeping in mind “service availability”. In PSTN equivalence. the case of a voice service, two popular 3. Every failure in the network is service availability metrics are the number recovered in less than 50 milliseconds. of calls dropped and the number of ineffective attempts. THE FIVE NINES MYTH.

This paper reviews the availability PacketCable (VoIP Availability and requirements of a “primary line” (PSTN Reliability Model for the PacketCable™ replacement) voice service, dispels some of Architecture - PKT-TR-VoIPAR-V01-001128) the popular five 9s myths about the PSTN; does an excellent job in dispelling some of and then establishes a framework by which these myths. It notes that the idea of PSTN to analyze and design a network to achieve reliability being FIVE 9s is incorrect. It the required level of service availability. clearly breaks down the different subsections of the PSTN network and draws The paper also outlines some of the a direct analogy to an equivalent IP network. modifications in terms of redundancy as well As per these requirements, the end-end as routing optimization that may be required availability of a VoIP network should be on the edge, in the regional networks as well greater than 99.94% to achieve equivalence as the backbone networks in order to with the PSTN. support PSTN equivalent voice over IP networks. In other words, it reviews some of In addition, PacketCable also specifies the changes that need to be made for some “service availability” metrics. These transport at the edges and between include the number of calls dropped and the distribution networks in order to support number of ineffective attempts. As per the report, there should not be Defects Per Million DPM(Calls Dropped) or more than 1 in 8000 calls dropped (or cutoff DPM(CD). calls), and no more than 5 in 10,000 ineffective attempts. These are exactly the Ineffective attempts arise due to failures same as the PSTN requirements on in the signaling path of a voice call. As per availability and service availability as set the PacketCable definition, “an ineffective forth in Bellcore GR series specifications. attempt occurs when any valid bid for service does not complete because of a fault Cutoff calls arise due to failures in the condition (e.g., hardware or software bearer path of the voice call. At the two end failure)”. That means, if a user is trying to points of the bearer path (that is the CMTS make a call but cannot due to a signaling facing the customer and the PSTN gateway path failure, it is counted as an ineffective facing the PSTN in the case of an on-net to attempt. However, here again we have to off-net call) a cutoff call may occur due to a define a threshold. The popular ineffective failure on a line card and a failure to copy attempts threshold that exists in the industry call state information to the standby line today is 30 seconds. Hence if a user is trying card (in the event that there is redundancy). to make a call, it doesn’t get through, he/she However, in the rest of the network, there is tries again and the call is completed the no concept of call state (being IP). Let’s say second time, as long as the whole process there is a failure, in a core router in the completes in less than 30 seconds, it is not network, and it takes 40 seconds to reroute counted as an ineffective attempt. traffic to the alternate path. One has to Ineffective Attempts could also be expressed imagine that an end-user would get as Defects Per Million or DPM (Ineffective frustrated and hang up the phone after a Attempts) or DPM(IA). So 5 in 10,000 certain period of time. This should also be ineffective attempts could be stated as 500 considered to be a cutoff call. DPM(IA).

Hence in the realm of IP, a cutoff call THE 50 MSEC MYTH could occur due to two reasons. Inability to maintain call state at the end-points in the Originally the 50 msec threshold was event of a failure, and/or, inability to recover established in the 1980s because the voice traffic within a certain cutoff call threshold channel banks that were used in carrier in the event of a failure at the end-points or networks could not tolerate failures that in the core of the network. lasted more than 200 msec. When failures exceeded that threshold, a Carrier Group In reality, the cutoff call threshold is user Alarm (CGA) would be activated causing dependent, but most IP telephony providers the channel bank to perform a “trunking are settling on 3 seconds as that threshold. condition” procedure that would terminate That is, if there is a failure in the network, all connections on that particular T3 line. and the user experiences “dead air” for more Since the outage budget had to be less than than 3 seconds, they would hang up and it 200-300 msec, 50 msec emerged as the de would be counted as a cutoff call. Cutoff facto standard. This decision was ironic calls are sometimes referred to as “Calls because by the time the de facto standard Dropped” and can also be measured as was actually adopted, newer technology defects per million. For example, 1 in 8000 allowed a CGA timer of 2 secs. We must cutoff calls could also be referred to as 125 bear in mind that this is for circuit switched technology. In the case of IP, all signaling is dropped call, but exceed the downtime message based. Hence there is no hard and budget. At the same time, there could be fast requirement for 50 mesc recovery. repeated failures in the network, each of 3.1 Rather, a more practical “user perceived seconds in duration. This would allow us to threshold” of 3 secs can be adopted for a have 6096 Dropped Calls but still be within “dropped call”. our 315 minute downtime budget.

Given the findings in PacketCable, it is (315,6096) DPM (Calls X clear that … Dropped) 1. The PSTN does not offer 99.999% end- 125 Acceptable end. Although certain components in the Range PSTN network may be five 9s (as is also X (500,1) 315 the case with IP equipment) the end-end Mins of downtime

network meets a specification of > 99.94%. Similar logic can be used to see why we 2. It follows that all devices in the network need the third metric – Ineffective Attempts. do not have to be five 9s, rather, the Let’s say for example, that the call control end-end network should be > 99.94%. server is down, but the data path is up. This 3. All failures do not have to be recovered would mean 100% availability as per our in less than 50 milliseconds. Failures downtime definition but the user will still should be recovered within the calls not be able to make any calls. cutoff and ineffective attempts threshold and the end-end network should cause CALCULATING AVAILABILITY AND no more than 1 in 8000 calls cutoff and SERVICE AVAILABILITY METRICS no more than 5 in 10,000 ineffective attempts. The industry accepted In this section, we will cover some basic practical thresholds for calls cutoff is 3 theory around calculation of these metrics, seconds and that of ineffective attempts but will not get into too many details around is 30 seconds. the math. The main focus of this paper as mentioned earlier is to establish a These three metrics together framework by which to analyze and design (availability/downtime, cutoff calls and networks for High Availability. ineffective attempts) are required to define an operating range. We can’t use only Availability: availability to understand end-user service experience. The end-end availability budget Availability is commonly defined as as specified by PacketCable is 99.94% and MTBF/(MTBF+MTTR). Such a definition this translates to 315 minutes of downtime for availability is good for a simplex system per year. (a system comprising one box). However, in a network that consists of a number of Now, consider for example, a network trunks and routers, most failures are partial that has one major failure and a user sees an failures. As a result of a partial failure some outage for 500 mins. That means he/she customers will not receive service, while cannot make a call for more than 8 hours. Is others have un-interrupted service. Also, this acceptable? If they were on the phone at even within a router or a switch only one the time of failure, it would constitute only 1 line card may go down, and users connected to other line cards may not see any Based on the above equations, once we disruption in service. Hence availability is calculate the availability of each and every defined with respect to a customer of the component along the path of a voice call we network. To compute availability, we only take a product of these individual need to consider the components along the availability numbers to get the availability of path needed to provide service to a single the overall system or network. customer and then average this over all customers. It is important to note that this gives us just the availability and downtime of the In addition to partial failures described system or network and does not provide us above, we also have to take into any insight in terms of whether the service consideration redundancy. For example, (in this case voice) is available or not. For certain components such as line cards may that we need to examine two other “service be configured in terms of 1:N Active availability” metrics; calls dropped and standby or 1:N Load sharing. ineffective attempts.

In order correctly calculate the CUTOFF CALLS/ CALLS DROPPED / availability of a single part such as a line DPM(CD) card or route processor; one has to take into account several factors such as … From a high level, based on the definition a. Switchover time – the amount of time of dropped calls, it is easy to see that this taken to switchover from the active metric is a function of the MTBF. component to the standby component. b. Active Coverage Factor – the The Calls Dropped contribution by each probability that a failure is successfully component (line cards, route processor, detected and switched over chassis, power supplies etc) along the path c. Standby Coverage Factor – the of a single user needs to be calculated. probability that the standby is in working condition and can successfully For each component, we calculate the take over. DPM(CD) as shown in Figure 3 (all parameters are assumed to be in hours) … We can use a Markov State definition for each component like a route processor, line For an average 3 minute call. This can card etc within a router or a switch. This is also be expressed more generically as shown illustrated in Figure 1 for 1:N redundancy. in equation A in figure 3…

Given the value of the parameters in the Where for each failure, the switchover legend, the above Markov chains can be time (in case of a redundant part) or the solved giving the probabilities in all the repair time (for a non redundant part) is different states in the chain. greater than the calls dropped threshold of 3 seconds. For a given type of redundant part, the combined part availability, combined part MTBF and combined part MTTR are calculated as shown in the equations in Figure 2. INEFFECTIVE ATTEMPTS / DPM(IA) call rate at 3am to the mean call rate, to arrive at terms that are comparable and We follow a similar process to calculate additive. the Ineffective Attempts contribution per component along the path of a single user. The “Night” Factor The “Night” Factor

Again, we only count each failure where Daily Call Rate the switchover time or repair time is greater than the ineffective attempts threshold of 30 This is a very conservative seconds. assumption Unplanned Call Rate (mean) TREATMENT OF PLANNED 50%

DOWNTIME 6 am 3 am 6 am From the derivation of the equation of Calls Dropped, we see that the term “in- coming call rate” appears, but is cancelled In fact an analysis of the average call from the numerator and denominator. This volume to the call volume at 3 am means that the DPM(CD) (and DPM(IA) for (maintenance window) shows a 10-15% that matter) is for a uniform call rate. Since night factor. Please see figure 4. in the previous section we were calculating EXAMPLE OF USING THE OUTLINED unplanned DPM(CD) and DPM(IA) the FRAMEWORK TO ESTIMATE THE implicit assumption is that it was for the AVAILABILITY AND SERVICE mean call rate over a 24 hour period. AVAILABILITY OF A CERTAIN However, when equipment is upgraded NETWORK DESIGN or we have to perform any kind of scheduled We will now use the theory outlined or planned maintenance, there is also a above and work through an example of how certain amount of downtime and associated to estimate the availability and service loss of service in terms of calls dropped and availability of a certain network design. In ineffective attempts. To calculate the effect working through this example we also of scheduled outages on service availability, highlight the importance of a systems view, we can use a similar method as above with and the system level interdependencies that the exception that instead of a summation come into play. Further, we also stress the across a number of random failures (as is the role that the layer 3 routing architecture case with unplanned outages), we look at plays in high availability in a network. only calls dropped or ineffective attempts

that are caused due to the outage time during Consider the network shown in Figure 5. the upgrade (say twice a year). It consists of a non redundant CMTS Most often, scheduled maintenance is connected to a non redundant aggregation done late at night (say 3am) to minimize the router. This then connect to a pair of impact of downtime on service availability. redundant core routers which then connect Now, the incoming call rate at 3 am is to a pair of redundant switches in the data significantly lower than the mean call rate. center behind which are the voice Hence we would have to factor down the components like the Softswitch, PSTN DPM(CD) and DPM(IA) by the ratio of the gateway, provisioning servers etc. Step 1: Reliability Block Diagram and core router route processors have to be modeled in parallel as they are set up in a The first step in understanding the redundant fashion. availability and service availability characteristics of a network is to lay out a Step 2: Failure scenarios – Estimating MTTR detailed reliability block diagram of all the and switchover time. components involved in the path of the voice call. Please refer to figure 6 for more The second step in this process is to details. evaluate in detail the potential outage for a voice call in the event of failures for each In our basic example, the CMTS consists component along the path of the voice call. of 3 components; the RF line card, the route In order to do that, we have to evaluate the processor and the FE uplink card. This is a upstream and downstream outage for each simplistic scenario because in reality there failure. Further, we need to take into account are a lot more components including power system level dependencies such as routing. supplies, timing cards, software (on the For example if a line card fails, the outage route processor and line cards) etc. So our 3 time may be dependent on how fast layer 3 components are shown in series in the can detect the failure and route around it, reliability block diagram below, because both in the upstream and downstream they are single points of failure. directions.

The aggregation router is also assumed to In addition to the unplanned failures have 3 components; the FE line card, the described above, we also need to estimate route processor and the GE line card. The the outage caused due to planned upgrades. FE line card is shown in series because it is a single point of failure, but the route The table shown in Figure 7 outlines the processor and the GE line card are shown in possible outage times due to various parallel because they are assumed to be possible failures in the network. As intra-chassis redundant in our hypothetical mentioned above, this takes into configuration. consideration (where applicable) the routing system interdependencies. Further, at this Similarly, one has to define the detailed stage, we also need to estimate the outage RBD for the rest of the network as well. time due to software upgrades. For example, This includes the core router (in which all in the case of the CMTS it may take from 5- components will be in parallel), the Data 11 minutes (including reboot time and Center switch, the PSTN gateway and the IP routing table set up time) for a CM/MTA to Softswitch. register with the provisioning system and then start passing traffic. In the case of the Figure 6 represents the hardware aggregation router, this time will also be components. However, we also need to dependent on the route table establishment model software as series or parallel time. This can be anywhere from a few components. We assume the CMTS has seconds to a few minutes depending on the software on the route processor and line card. complexity of the routing setup. In this case since they are non redundant, they are modeled as serial components. Similarly software on the aggregation switch Step 3: Estimating MTBF and failure rates route processor and WAN card redundancy. Having laid out the reliability block • Inter-chassis redundancy on the diagram and the failure scenarios, we need aggregation routers. The CMTS is to estimate the MTBF of each component then dual homed to the redundant (hardware and software). Hardware MTBF pair of aggregation routers. is usually obtained from manufacturers databases. Software MTBF typically is Please refer to figure 9 for more details collected from network operations be on the network topology for case 2. measuring the unplanned software reboots or other failures over a period of time for a In addition to the topology modifications sample set of devices. from Case 1, in Case 2 we are also assuming some form of routing optimization for High With each of these assumptions in place Availability. Since this is a simple directly we now use the theory described in the connected Ethernet network, we would need previous section to calculate the availability, to reduce the SPF computation hold timer to calls dropped and ineffective attempts about 1 second (default value being 5 contribution per element (such as CMTS RF seconds) so as to reduce some of the failures line card, route processor etc). in case 1 (which were about 6 seconds) to <3 seconds so as to avoid any dropped calls. The idea is to reduce the above diagram to its serial equivalent components by In cases where the network is more calculating the combined MTBF and complex, with say Layer 2 SONET (or any combined MTTR of each of the components multi-access) connectivity between the involved. This is done using the Markov aggregation routers and a distribution layer, States described in the previous section. the routing optimization can get significantly more complex. In this case we The overall availability, DPM(CD) and may need to reduce the OSPF hello and dead DPM(IA) for the entire network are timers in addition to the SPF computation calculated using the formulas shown in timer. The default values of hello and dead figure 8. timers in OSPF are 10 seconds and 40 seconds respectively; and were set more The availability/downtime results for than 10 years ago keeping in mind data Case 1 are represented by the first bar in applications that did not need fast figure 10. The DPM(CD) and DPM(IA) convergence. As a result, without results are shown in figure 11. optimization it is possible to see some failures causing outages in the region of 40- Now, in Case 2, we make some 45 seconds. However, reducing the dead modifications to the network design and add timer to say about 1.5 seconds and setting some High Availability features to the the hello timer to 0.5 seconds with three network devices. hellos per dead timer will get the detection time down to about 1.5 seconds. Hence it Some of these are listed below. would be possible to see total outages in the • Redundancy at the edge of the region of less than 3 seconds. In certain network (on the CMTS). Line card, cases depending on the complexity of routing in the network, static routing in certain parts of the network and reduced In addition, implementation of advanced timers in others can lead to even sub second monitoring and early warning (especially on convergence. the RF side) HA software on devices will help further reduce downtime and improve There is however a tradeoff in this case service availability. where if the timers are set too low, it may cause instability in the network. This can The results for case 3 of the network are cause serious problems as a failure on one also shown in figures 10 and 11. line card which would have otherwise affected service only to a set of customers, COMPARISON OF RESULTS has now propagated to the rest of the network taking down service for a far grater As can be seen from figures 10 and 11, number of subscribers. the base case network can be expected to be down for about 102 minutes per year for the In addition to the routing architecture average customer. This consists of about 56 modifications, we also assume certain basic minutes of unplanned failures (that can HA features on the network elements. For happen at any time of the day) and about 46 example in the CMTS, if the route processor minutes of planned downtime (that occurs at fails, the cable line cards are not reset and 3 am in the morning, when the likely-hood the CMs are not dropped. We also assume of a call in progress is extremely low). that call state in maintained during switchover to the standby route processor The PacketCable availability budget for (whether it is within the same chassis or this portion of the network is 71 mins of implemented in an inter-chassis fashion). downtime per year. Hence the base case network is about 71% over the PacketCable Given the discussed enhancements from budget. case 1 to case 2, we have to re-define the Reliability Block Diagrams and the failure By introducing redundancy on the CMTS scenarios and redo the mathematical and the aggregation routers, optimizing the analysis. routing architecture and implementing HA features that maintain calls sate even in the The results for Case 2 are shown in event of RP failure, enabling the line cards figures 10 and 11. to continue forwarding traffic we can reduce these numbers by 60%. This brings us to a In case 3, we assume no additional total of about 41 minutes of downtime per network topology changes. However, we year, which is significantly better than the implement advanced HA features which PacketCable guidelines. enable us to upgrade software or hardware with interruption to service. For example, By further adding more advanced HA we can switch off one route processor, have features like the ability to upgrade software all CMs/MTAs be serviced by the standby without service interruption and advanced route processor (without dropping calls or HA monitoring features, this number can be losing call state), upgrade software on the driven down to 15.5 minutes of downtime primary RP, and then switch back over again. per year.

In the case of Calls Dropped, we see that When analyzing and designing a network the base case network will have about 26 for availability, there are complex system calls dropped per million and by making the level dependencies and interaction between recommended enhancements, this can be devices that need to be considered. Routing reduced to about 9 DPM(CD). Similarly plays a critical role in highly available DPM(IA) can be reduced to about 20 from a networks. current 128. In addition, for an IP network to be It is important to note that PacketCable equivalent to a PSTN in terms of availability, does not break down the DPM(CD) or it does not have to be five 9s end-end; rather, DPM(IA) budget by network component, so it needs to be greater than 99.94%. It should it is difficult to derive a ‘budget’ for this also meet the portion of the network. However, it is end-end service availability metrics of < 125 important to note, that this portion of the DPM(CD) and <500 DPM(IA). network (the CMTS, Local IP network and voice components) can contribute only It follows that all failures do not need to be about 5-7% of the end-end DPM(CD) recovered in less than 50 msec as long as the budget. This means that even if the CM, the number of dropped calls and ineffective HFC plant, and the IP backbone (or PSTN attempts do not exceed the above network depending on the mode of requirements. transport) contribute about 90% of the 125 DPM(CD) budget, the network will still The framework established in this paper meet the guidelines. can be used to evaluate the availability and service availability of an IP network, study Cable MSOs must conduct a similar the effect of redundancy at different points analysis for the CM and the HFC network to in the network, and make an economic based determine whether the end-end network decision in terms of the increase in meets the PacketCable guidelines for PSTN availability for a certain amount of capex. equivalent voice service. Lastly, it is possible for a well designed CONCLUSION IP network to meet and in certain cases exceed the availability of the PSTN. There is much more to availability than the number of nines on a box. In order to ACKNOWLEDGEMENTS design a highly available network, one has to keep in mind end-end network availability The author would like to thank the and more importantly the *service following people from Cisco Systems for availability*. In the case of voice, these are their invaluable guidance and contribution to calls dropped and ineffective attempts. this paper – John Chapman, Jim Forster, Madhav Marathe, Henry Zhu, Alvaro Retana, Paul Donner, and Latha Vishnubhotla.

1

(N+1) λ(1 −C) (N+1) λ c µ

2 β 3 ÷ µ (N+1) 2µ N λ

5 ÷(N+1) 4 ÷ N

State Descriptions Legend

1 All units working λ = Failure rate of individual unit 2 One of the units failed, detected. β = Rate of traffic switchover from the failed unit to other healthy unit(s) 3 N units have taken over the traffic from the failed unit. = Repair rate of part failures 4 Second unit failed before the first failed unit is repaired. µ C = Switchover coverage 5 One unit failed, undetected N = Number of units out of the total of N+1 loaload-sharingd units ÷(N+1) because each unit handles only 1/(N+1) traffic.

Figure 1 (1:N Active Standby – Markov State Diagram)

tyAvailabili combined = sum of probabilities that the redundant parts are in non failure markov states

1 MTBF = combined sum of transition rates fromnon failure states to failure states

1− Availabilitycombined MTTRcombined = × MTBFcombined Availabilitycombined

Figure 2: Availability Equations

∑ Existing calls __ dropped×10 6 all incidents>− 3sec Total Calls __ Attempted

∑ Existing calls at time of failure)_____( × ( ofnumber failuresof that scenario) ×10 6 all− failurescenarios>3sec Total incalls one____ year 1 ∑ (Incomin g Call __ rate× length of call ()__ 24365 ) ×××× 10 6 all failurescenarioss>3sec MTBF (Incomin g call rate)__ 365 ×× 24 3 1 ∑ ×× 10 6 all− failurescenarios>3sec 60 MTBF

Figure 3: Calls Dropped Equations

%% Ratio Ratio of of Call Call Rate Rate at at a a given given time time to to average average call call rate rate across across the the day day – – CalculatedCalculated for for weekdays weekdays (Monday (Monday through through Thursday) Thursday) based based on on sample sample market market datadata 200%

180% Field experience shows 160% that ratio of call rate at maintenance time to 140% average call rate is 15%- 17%. 120%

100% This means that there is a 15% impact of 80% downtime during a maintenance window.

60% PacketCable does not break down planned to unplanned budgets, but clearly there is 40% less of an impact.

20%

0% 0:00 1:00 2:00 3:00 4:00 5:00 6:00 7:00 8:00 9:00 10:00 11:00 12:00 13:00 14:00 15:00 16:00 17:00 18:00 19:00 20:00 21:00 22:00 23:00

Figure 4: Example “Night Factor” – average call rate across a 24 hr period

Hub site Master Head-end Switc h (A) PSTN Gateway / Softs witch and Router (A) CMTS Agg Router (A) other ser vers FE GE/OC-48 9 10

7 6 12 3 4 5 8 11

Switc h OC-192 910(B) Hub site

GE/OC-48

GE/OC-48 IP Bac kbone

Agg Router (B) Router (B)

Master Head-end

Figure 5 : Case 1: Network diagram

RELIABILITY BLOCK DIAGRAM (RBD)

Route Cable FE Interface Processor GE Interface Line Route WAN 1:1 Intra- 1:1 Intra- Card Processor Interface chassis chassis

Similar Reliability diagram for the rest of the network

CMTS AGG R OU TER (A)

MTBF, MTTR, Switchover time and coverage assumptions for every component in RBD

MTBF (hrs) 120,000 150,000 250,000 250,000 400,000 250,000

MTTR 3 hrs 3 hrs 3 hrs 3 hrs 3 hrs 3 hrs Similar Assumptions for Switchover NA NA NA 6 secs 6 secs 6 secs the rest of the Time network. Active NA NA NA 98% 98% 98% Coverage Factor

Figure 6: (Reliability Block Diagram and Markov Parameter Assumptions)

Example Network - BASE CASE Unplanned Failures Upstream Down- Upstream Notes Downstream notes Total Calls stream Outage Dropped ? CMTS 1 Route Processor 3 hrs 3 hrs This is a Single point of failure. Once this Same as upstream 3 hrs YES component fails, there is no path upstream and the outage time depends on someone physically going to change the card. According to PacketCable this time will be in the region of 4hrs 2 Cable Line Card 3 hrs 3 hrs Same as above Same as above 3 hrs YES 3 WAN Card 3 hrs 3 hrs Same as above Same as above 3 hrs YES

Aggregation Switch 4 Route Processor 2 secs 2 secs Lets say the switch has dual route Once the FE line card is reset, the CMTS 2 secs NO processors, but is not running any specific detects it within 1 sec and takes 1 sec to HA software. The route processor failure recompute its routes via the second resets the line cards. The directly aggregation switch connected core router detects the failure within 1 sec and takes 1 sec to recompute its routes via the second aggregation switch. The outage in this direction would be 2 sec. 5 FE Line Card 3 hrs 3 hrs This is a single point of failure This is a single point of failure 3 hrs YES 6 GE Line Card 0 secs 2 The CMTS continues forwarding traffic to The core router detects the failure of the 2 secs NO Agg router 1 because its is the only path GE card on Agg Router (A) almost available immediately, and then delets that route from its table. It then starts forwarding traffic intended for the CMTS through Agg Router (B) via the OC192 interface. Agg Router (B) still has a valid path for the CMTS because it is connected to a different line card on Agg Router (A)

Similar FAILURE SCENARIOS for the rest of the network

Figure 7: (Failure Scenario Examples for Base Case 1 Network)

Availabilitynetwork = ∏ Availabilitycomponent all series equivalentcomponents

DPM CD)( = ∑ DPM CD)( component all series equivalentcomponentsin pathbearer

DPM IA)( = ∑ DPM IA)( component all series equivalent components in signaling path

Figure 8: Formulas to calculate end-end network and service availability

Hub Site Agg Router (B) Master Head-end Switc h (A) Core PSTN Gateway CMTS Router (A) / Softs witch and FE oth er ser vers GE/OC-48 4 9 10 7 12 3 6 8

11

45 Agg Router (A) OC-192 Switc h (B) 910 Hub site

GE/OC-48

GE/OC-48 IP B ac kbo ne Ag g Router (C) Core Rout er (B)

Master Head-end

Figure 9: (Case 2 – Network enhanced for HA) DOWNTIME REDUCTION AND THE EFFECT OF REDUNDANCY

120 Planned downtime – D one 102.7 during mai ntenance Window 100 Unplanned downtime 14.9

80 46.1 PacketC able Budget for portion of the netw ork including CMTS, Local IP, Voice components

46.9 60 • Soft ware upgrades wit hout service interruption 40.9 • Advanc ed HA monitoring 40 • CMTS Redundanc y 56.6 25.4 • Agg Router redundanc y 31.2 • Routi ng Opti mization 15.5 20 • HA Features 5.7 9.7 9.7 0 Case 1 Results Reduction due to Case 2 Results Reduction due to Case 3 Results Case 2 Case 3 Enhancements Enhancements Base 60% Additional 62% Case Reduction Reduction

Figure: 10 (Downtime results for the end-end network – Case 1, 2 and 3)

DPM (CD) – Calls Dropped DPM (IA) – Ineffe ctive Attempts Reduction Reduction

30 140 128 25.6 120 25

100 20 80 15 60 10 8.9 40 6.1 31.7 5 19.6 20

0 0

Case 1 Case 2 Case 3 Case 1 Case 2 Case 3

Percentage Percentage of DPM(CD) 20% 7% 5% of DPM(IA) 26% 6% 4% Budget Budget

Figure: 11 (Service Availability results for Case 1, 2 and 3) IS MPEG-2 OUR NEW LEGACY?

Yasser Syed, Ph. D., Mukta Kar Ph.D. Cable Television Laboratories & Vinay Sathe, Ph.D. Multirate Systems

Abstract content. End users are now interested in “my 500 channels” rather than the standard “500 Cable operators have been looking at channels” on their cable networks. advanced video codecs in order to make more Proliferation of new digital services such as bandwidth available to meet ever-growing VOD, HDTV, DOCSIS, VOIP has increased user demands on cable networks, the need to be able to send more bits on the same pipe. This paper investigates the issues related to introduction of a new video compression Does the bandwidth crunch necessitate format in cable networks and how it can co- moving to new video compression exist with “legacy” MPEG-2 deployments. technologies? Yes and no. In this paper, we present the case that new compression formats should be looked at as complementing the INTRODUCTION MPEG-2 based deployments in the near term, not replacing them. The introduction of new The introduction of digital cable services video (and audio) codecs can act as a part of a using MPEG-2 over 10 years ago was done to bandwidth reclamation strategy instead of a meet several objectives: ability to provide complete overhaul of a cable network. hundreds of television programs, reduce capex and opex related to acquisition and In this paper we first list some bandwidth distribution, to name a few. The transition savings strategies in cable networks that can from analog to digital required plant upgrades, help reduce the bitrate needs without new encoders, and new STBs along with a abandoning MPEG-2 based deployments. revolutionary way of thinking. Today, MPEG- Next, we take a look at the issues involved in 2 set of standards forms the core of audio, introduction of new video compression video and transport protocols used in the formats. Unlike the analog to digital digital cable delivery system [1,2]. transition, only new encoders and STBs are required, but plant upgrades are not necessary. Over this same decade, user requirements This allows for new and old video have changed along with advances in video technologies to co-exist and many of the ideas compression technology. Recently, several originated in the earlier transition can be new video codecs have been introduced (e.g. leveraged again. WM-9 and MPEG-AVC/H.264). The powerful message of these codecs is hard to HOW FAR CAN MPEG-2 GO? ignore: 2-to-3 times the compression efficiency, Digital Rights Management MPEG-2 video standard specifies syntax support and increased interactivity with the for compressed video bitstream, but does not standardize an encoding algorithm to generate There are ways to extend the coding it. Over the years, better encoding schemes potential of MPEG-2 by adding new tools and have yielded superior picture quality at a creating new profiles. In the last few years, given bitrate using techniques such as researchers have been looking at compression improved pre and post processing, smoothing techniques beyond MPEG-2. By way of of motion vectors, better rate control (bit example, some of the potential new additions allocation) multi-pass encoding and so on. to the existing tool-kit of video compression Compared to the early days of MPEG2, the techniques include: typical bitrate required for broadcast quality • Multiple reference frames for prediction. video compression has gone down • Variable block size for motion estimation significantly through competitive processes • Special prediction modes for fades built into developing advances in video • 1/4th pixel motion estimation. encoding marketplace. However, it is • Multiple directions of prediction for I- interesting to note that certain applications macroblocks, such as VOD have associated business • Loop filter to control propagation of error, agreements that dictate the bitrate to be • Usage of arithmetic coding. constant (e.g. 3.75 Mbps) irregardless of the advances in encoding techniques. Adding these new tools, requires extensive modifications of the MPEG-2 standard and Video compression experts generally agree the creation of new products. In essence it that current state of the technology is such would be almost equivalent to creating a new that new encoding techniques to produce coding standard to handle these changes. In MPEG-2 video streams will probably not fact, some of the coding gain in advanced produce too much further improvement for codecs come from including these tools [3]. entertainment quality television programs (See Figure 1). Due to the maturity of the It should be noted that various alternatives MPEG-2 codec, most significant coding exist to reclaim bandwidth other than efficiencies at the main profile/main level changing the video compression codec, and (MP/ML) have already been explored through may be easier to deploy. A quick list includes: the competitive encoder development marketplace . 1. Higher QAM constellations (64 to 256: ~33%/QAM) 7 6 2. Shift analog PPV services to digital 5 (10X/QAM) 4 3. Switched broadcast Service (~1x/QAM)[4]

Mbps 3 4. Stat mux MPEG-2 VOD (~25%/QAM) 2 5. Convert all (most) analog to digital 1 (10X/QAM) 0 1995 1996 1997 1998 1999 2000 2001 2002 2003 Year STATE OF COMPETITION

Not surprisingly, the need for bandwidth is Figure 1: MPEG-2 Generational felt by DBS operators too. Due to the need to Encoder Improvements carry local programming, offer HD services and making bandwidth available for

personalized 2-way video services, satellite operators have been looking at various options including migration to higher algorithms used do not need to be constellation modulation, use of turbo codes standardized (the exception to this is targeting and adoption of advanced video codecs. to portable personal devices). Over time, as Satellite operators have very little margin left the average bitrate needed to deliver in reducing per-channel bandwidth used by entertainment quality video decreasesthe take their MPEG-2 video programming and it is rate on these services will rise. Whether or not inevitable that they have to commit to a new this is an opportunity or threat to cable compression standard in near future. operators remains to be seen [7]. However, to date, there are no announced plans of migration to a new video codec. Wireless network operators have also recognized the new market opportunities in With MPEG-2 technology, DSL providers being able to offer video services. Recently, have not been able to establish a firm foothold there is a renewed interest in MMDS in the video delivery market. The desire to technology (e.g. WiMAX consortium) that adopt an advanced video codec for video provides fixed wireless broadband access to delivery has also been reported in the DSL subscribers. At the bitrates of transmission industry because it can open up these markets. (1.5 Mbps), advanced video codec will be the Recently approved ADSL-2 standard enables logical choice for entertainment quality video transmission of data at 1.5 Mbps over existing services. It is too early to speculate if the existing phone networks [5]. This channel rate wireless operators will provide data and voice is not enough for entertainment quality video only or if they will provide data, voice and transmission if one is using MPEG-2 video with broadcast television quality video compression, but is plenty for advanced codec services. based transmission. However, High Definition programming still poses a problem because In terms of ability to offer converged 1.5 Mbps is not enough bandwidth for HD services (streaming, video conferencing, transmission, even for advanced video codecs. video cellphone, and traditional video using Clever attempts have been made to meet that same codec and protocols) MPEG-2 transport need by a very fancy arrangement of multiple and video technology is at a disadvantage. phone lines. For example, UK based Net-to- Migration to advanced codecs and new Net Technologies and Tandberg TV standards potentially gives a competitive edge experiment using two lines ADSL loop to the above network operators by enabling bonding, as referenced in an Internet report converged service offering. [6]. IMPLEMENTATION COSTS Another recent shift in business model due to availability of high bandwidth DOCSIS is Due to the broadcast nature of the video the content providers directly doing business signal, and number of decoder units; the cost with consumers by bypassing the need to have of equipment on the decode side far business relationship with network operators. outweighs the costs associated with upgrading Currently, there are only a couple of high encoding equipment. We therefore only talk profile (and many smaller) TV-over-the- about cost of implementation related to the Internet type services available. Two things to set-top box side. note here – decoders used for these services are typically software-based (WM9 or Real), In anticipation of a future move away from with DRM enhancements to the liking of each MPEG-2 technology to another video codec, content provider, and video compression several IC manufacturers have already started building their decoder ICs based on flexible and can be used to support carriage of both platform (e.g. TI, Equator, ST, Philips etc.) codecs. The concepts of modulation, Obeying Moore’s law, providing backward multiplexing, buffering, insertion, and compatibility becomes a less pricey option. caroseling can be reused and intermixed with the new video codec. As an example, MPEG decoder ICs are sold at approximately <$12 price range. The real analog legacy that still exists Current MPEG4 decoder pricing is very high today is not the plant and equipment (almost a $15 premium over MPEG2 upgrades; but the reluctance of a sizeable (MP@ML) only decoders), but the premium group of people (approx. 63 million paying in is expected to go down to less than 7 dollars 2003 [9]), who get basic cable service to all in 3-4 years [8]. the TVs in their household for one low fee, to switch to a digital service that requires a box H.264 w/MPEG- for each TV. The fear is that they may opt out 2 30 of cable altogether. Alternatively there is

25 constant pressure to shift from analog due to extensive piracy, bandwidth demands, and 20 less robust security mechanisms. This 15 ‘analog’ legacy is what in one part is creating 10 the bandwidth crunch on plant systems and 5 much study through the “all-digital efforts”, 0 including using new digital video codecs, are 2003 2004 2005 2006 2007 being considered for resolving this problem.

The important thing to realize is switching from an MPEG-2 Digital Video platform to a Figure 2 Average Sale Price of decoder IC new type of digital video codec does not (Courtsey: In-stat/MDR) create the same type of legacy and may, in New design and advances in hardware such as part, actually be a solution to this analog one. Media processors, VLIW processors are changing the traditional “hard-coded” nature The challenges created by switching to a of video decompression platform, making it new digital video codec are buying new more flexible to adapt to new video codecs. encoder equipment at headends or regional centers and replacing the millions of digital IS MPEG-2 VIDEO REALLY LEGACY? STBs already deployed. The most significant cost is replacing these STBs (approx 29 The transition to a new digital video codec million in 2003 [10]), but this could be argued is not the same as the earlier transition from as not insurmountable and actually somewhat carrying just an analog Cable TV signal to less of an effort than what has been launched using a new digital transmission platform. The in the past. There are a number of strategies earlier digital migration was much more that can mitigate and justify these costs, but physically involved than just replacing a much of this depends on being able to carry video codec; fortunately this time there is no both types of codecs in the same cable plant need to rebuild the digital platform. The and developing new boxes that can decode network infrastructure put in place for both MPEG-2 and the candidate digital video establishing MPEG-2 technology is still codec. With this and the inherent advantages useable; the cables, connectors, amplifiers, of the local distribution in cable systems, and transport mechanisms are already in place cable service providers can take several steps technology and services for a cost-effective to ease the burden of this transition: transition. Some factors to consider for this would be: • They can deploy new services on a region-by-region basis using an ROI to • New boxes need to bring in additional justify the deployment of new STBs. revenue • The type of video service is agnostic • Room for bandwidth needs to be to kind of digital codec used. This made. This can happen through means that MPEG-2 can initially be bandwidth reclamation strategies as used with a deployment of service to mentioned previously. build up demand to justify a box • Analog PPV services must be phased upgrade and then switch the codec out by replacing these boxes with used for the service. MPEG-2 STBs • Alternatively new codecs can be introduced in new tiers of services. A place on the cable plant where a total • Initial deployment of a few STBs to be replacement of MPEG-2 with the advanced used in a slow rollout can help codec could be beneficial, and less of a decrease the new STB costs for future hardship, is in the backend storage and larger rollouts. transport systems. With the projected • Existing MPEG-2 only STBs can be increases of Video-on-Demand (VOD) cable redeployed in other parts of the system services, larger amounts of content will need to assist in digital upgrades. to be stored, replenished and distributed in the backend plant. A content catalog of 10,000 New advanced codec products can be hours with 25% of it being replaced once a designed to decode both the new codec and week would require at least a continuous 56 MPEG-2 given an additional 10-15% extra mbps transport link to the headend along with performance in the chip. Though the new 17,000 GB of storage (this does not include codecs are not backwards compatible with storing multiple copies or transport to edge MPEG-2, the product can effectively support servers). To deal with this, more server farms both. The STB can be deployed to operate on will need to be added along with leasing of traditional MPEG-2 systems. As the new more bandwidth to transport content to one or codec gets deployed on the plant, the same many headends. These costs can be reduced box can be used to decode these new services. through the use of advanced codecs that will This strategy can allow MPEG-2 to be effectively halve the storage and transport complemented by the newer codec and can needs. To take advantage of these cost savings allow services to exist before a decision is real-time transcoding technologies between needed to switchover the codec. the new codec and MPEG-2 should be encouraged. There are already a large number digital MPEG-2 STBs out there, but this number will continue to grow due to a dramatic EXAMPLE USAGE SCENARIO: NEW increasefrom demand of MPEG-2 HD Boxes. VOD SERVICE TIER Given that new product deployments will be at least 1-2 years away, there will be a The introduction of a new video codec significant installed base of MPEG-2 boxes. needs to be part of an overall bandwidth Any new codec will have to work in reclamation strategy. A single RF QAM complement with the MPEG-2 existing Channel can have a tremendous range in video capacity depending on how it is being There are several new revenue sources to used (see Figure 3). Effective management of cable being captured by the on-demand this bandwidth can lead to new services as services: One is replacing the local video well as a cost-effective way to deploy an store (“watch a movie without the late fees”), advanced video codec. There are many ways the other is competing against the burgeoning to accomplish this, but this section will PVR market but with some extra features like attempt to describe one possible approach. an unlimited hard drive, access to unaired content, and never needing to record a show. The High Definition offering basically MPEG-2 Adv. Codec provides a “gotta have” service that can

35 increase customer loyalty (“I gotta have a feed

30 for my new HD set and cable has it”). The

25 problem is on the cable plant, the bandwidth

20 is disproportionately dished out. An HD

15 request can replace 4 or more SD requests, but the price is not that much different. The Video Programs 10 SVOD service can take up bandwidth that 5 could have been used by someone who didn’t 0 SD SD-VoD HD HD-VoD want to go to the video store to see the latest RF 256-QAM Channel movie releases. These situations can create either more contention or a disproportionate Figure 3: Codec Video Program increase in bandwidth demand on the node Capacity for Single 256 QAM Channel level to accommodate peak use. Some ways to deal with this is to limit the amount of HD Presently VoD services (projected to be a requests and constrict the size of the content 4-5 billion dollar market by 2007 [11]) catalog of an SVOD service to allow enough contain several different types of streams that take-up rates in the traditional MOD service. have radically different bandwidth behaviors. An alternative to this is to create a new tier of Two potentially bandwidth intensive services on-demand services. are High Definition (HD-VOD) deliveryand Subscription Video on Demand (SVOD). A This new tier would have a combined local High Definition program in MPEG-2 requires SVOD offering with an HD-VOD service that approximately 4 times the bit rate of the same requires a new STB capable of doing both channel in SD format. Besides needing a MPEG-2 and the new codec. The lower on- larger bit rate, HD-VOD creates a demand on demand tier for renting SD movies would be bandwidth resources by effectively blocking supported by MPEG-2 and the new tier would potentially 4 or more SD streams for each HD be supported by the new codec. With the stream in use. The SVOD service also creates doubling of bandwidth from the new codec, a demand on bandwidth resources by the SVOD service can have an expanded and effectively blocking pay-per-transaction SD localized content catalog with a predictable streams especially when ‘channel surfing’ monthly income source. Additionally enabling behaviors occur. As the take-up on these HD-VOD on this tier provides incremental services becomes more popular, a need to income other than the monthly fee as well as create a bandwidth resource policy based on freeing up bandwidth on the lower on-demand these different behaving streams will be tier for the SD MPEG-2 movie rental market. necessary.

By creating this tier, a new box with the REFERENCES advanced codec can be deployed with a trackable ROI instead of a pure capex [1] ITU-T H.222.0 (2000) | ISO/IEC 13818- expense. A tier becomes identified which 1:2000, Information technology – offers a value to the customer worth the Generic coding of moving pictures and additional monthly upgrade fee whether the associated audio information: Systems. subscribers has an HD set or just wants the [2] ITU-T H.262 (2000) | ISO/IEC 13818- additionally content on SVOD. In effect it ties 2:2000, Information technology – the deployment of the new boxes with a new Generic coding of moving pictures and revenue source while reducing the bandwidth associated audio information: Video. demands on the lower MPEG-2 tier. [3] “Overview of the H.264/AVC Video Consequentially, this frees up the lower Coding Standard,” T.Wiegard, MPEG-2 tier for capturing more of the SD G.Sullivan, G. Bjontegaard, and A. video rental market whose customer base will Luthra. IEEE Trans. On CSVT, pp.560- be largely those with existing MPEG-2 only 576. STBs. Since this is a node-based service, the [4] “Modeling Switched Broadcast Video new tier (and new replacement boxes) can be Services,” J.Weber, and J. Gong, NCTA targeted to specific areas or rolled out region- Show 2004, May 2-6, 2004. by-region. Furthermore it creates a type of bandwidth policy that can relieve some of the [5] ITU Standards G.992.2 and G.992.4 future pressure on a mixed on-demand tier. [6] http://www.rethinkresearch.biz Lastly, the deployment strategy addresses the [7] Modern Cable Television Technology: on-demand segment rather than the broadcast Video, Voice, and Data Communications, segment which for cable is more of an area for W. Ciciora, J. Farmer, and D. Large, future growth. Again this is only one type of Morgan Kaufman Publishers Inc, San usage scenario for initially deploying a new Francisco, 1999. codec, but there are others. [8] E-mail correspondence with Michelle Abraham, Senior Analyst, In-stat/MDR. CONCLUSION [9] Basic Cable Subs 2004 Estimate

Deutsche Bank Securities & Jupiter In this paper, we have discussed issues and Communications (2004 Outlook for challenges related to the introduction of Cable Jan 27, 2004). advanced video codecs in MPEG-2 based digital cable networks. One of many possible [10 Press Release for Digital Set-Top-Boxes deployment scenarios is discussed in details. Outlook for 2003 www.tracestrategies.com/ docs/ pr/ 10012003PR.html [11] Morgan Stanley & Co. 2003 revenue estimates cited in 9/8/03 Multichannel News article, "Demanding DVDs: Recordable Discs Could Pinch VOD Sales."

LARGE-SCALE ON-DEMAND DELIVERY ARCHITECTURES – TOWARDS AN EVERYTHING-ON-DEMAND FUTURE

Cliff Mercer, Ph.D., Director of Technology Satish Menon, Ph.D., Chief Technical Officer Kasenna, Inc.

Abstract • News-on-demand: news you want to watch, when you want to watch it – Pioneering technologists and business without being limited to rigid leaders have been driving our industry broadcast times at 6am, 6pm, and toward new ways of watching and interacting 10pm. with our televisions, that is, toward an • Music-on-demand: any tune ever “everything-on-demand” future. recorded from Hillary Duff’s latest to Accomplishing this lofty goal requires the Rachmaninoff. cost-effective implementation of “extreme • Education-on-demand: listen to narrowcasting” techniques, which in turn Feynmann’s physics lectures at your demand coordinated use of network convenience. technologies, powerful server architectures, • Sports-on-demand: re-live the and intelligent software design. memories of Pele’s most famous goal The early Video-On-Demand (VOD) in the 1962 World Cup. solutions of the past with their proprietary interconnection technologies and custom Realizing such a vision requires hardware designs fail to scale cost-effectively commercially viable implementation of with the new demands for flexible integration delivery techniques commonly referred to as with new services, delivery performance and “extreme narrowcasting” – the ability to storage capacity. Advances in commercial deliver a unique stream of rich media (such off-the-shelf computing and networking as audio or video) from the content source products along with smart software design (typically video servers) to an end user’s TV, meet these current and future challenges PC, mobile phone, or PDA. Extreme cost-effectively and without the obsolescence narrowcasting requires novel system designs, problem experienced with historical such as distributed network topologies, solutions. powerful server architectures, and intelligent software design, that can take advantage of INTRODUCTION “localities of reference,” predict future usage patterns, manage a variety of bandwidth Since the early 1990s, technologists and requirements and resource needs, etc. More visionary business leaders from a group of importantly, these designs must make for a pioneering companies in the content, cost-effective system, allowing operators to computer, networking, consumer electronic, deploy these services without breaking the broadcast and cable TV businesses have been bank and allowing them to realize returns on working to change the way we watch and their investment quickly. interact with our television. They have been working towards an everything-on-demand (EOD) future: In the past, proprietary servers, Modern Techniques for High Performance proprietary interconnection technologies, and custom designs were employed to solve these COTS cluster systems have become the problems of extreme scale in an initial most cost-effective way to satisfy high- application area: delivering movies over performance computing requirements. Real- cable plants to TVs. However, the advances time media streaming, VOD streaming, and in mainstream server technologies and on-demand delivery of rich media require the intelligent software designs can solve these best and most cost-effective high problems much more cost-effectively, while performance computing techniques available. avoiding obsolescence associated with According to Thomas Sterling of the proprietary technologies. Furthermore, these California Institute of Technology, “Cluster new technologies enable the delivery of a systems are exhibiting the greatest rate in plethora of additional video and rich media growth of any class of parallel computer and applications and services to a broader range may dominate high performance computing of end-user devices as the full vision of EOD in the near future.” [Sterling 2001] comes within reach. We define a COTS cluster as a collection In this paper, we highlight the technical of server-class computers built completely and economical challenges involved in the from commercial off-the-shelf components deployment of an extreme narrowcasting and which themselves are built using network. We begin with a discussion of commodity off-the-shelf chips and modern trends in high-performance components. The interconnecting network computing, which leverage commercial off- technology must use COTS components as the-shelf (COTS) hardware combined with well, the most cost-effective technology right intelligent software management to scale to now being Gigabit Ethernet. very high performance levels. Next, we relate those techniques for high-performance The latest COTS components have such computing to the particular application area increased price performance that using high- of concern: VOD and other video and rich performance COTS servers connected by media delivery applications. We discuss GigE interfaces yields a very high typical growth of a narrowcasting video performance computing system. For example, network from a centralized system to a a COTS cluster might consist of 10 potentially decentralized system as the commonly available dual-Xeon (> 2.2 GHz) number of subscribers grows and the amount servers, each with 2 GB of DRAM, up to 16 of content in the system increases. We disk drives at 146 GB each, and up to three introduce the concept of “hierarchical Gigabit Ethernet NICs. Such a cluster would storage” – an implementation of storage have: systems (from RAMs to disks) to address the requirements inherent in scaling up the • Processing capacity of 20 Xeon number of subscribers and amount of storage. processors, Intelligent software techniques help to solve • 20 GB of DRAM, these problems of extreme scale within the • 23 TB of disk storage and entire video network. Finally, we offer • 30 Gbps of networking capacity. quantitative comparison between common architectures for VOD and EOD deployment. Moore’s law has continually increased the COTS Clusters for Price Performance price performance of COTS computing Vendors that bring COTS servers to components for the past 40 years. Intel and market are under constant pressure to others spend billions of dollars on research leverage improvements of new components and development to drive this technology and in their server product lines. Vendors of the price performance forward, and COTS motherboards, NICs, and complete servers clusters benefit directly in terms of price spend billions of dollars in research and performance as a result. development to incorporate the latest underlying hardware technologies and The Fall of Proprietary Hardware provide standards-based products to the Hardware designed for specialized market as fast as possible. COTS NIC applications use the best currently available vendors are under similar constant pressure to technology at design time, but by the time the improve the price performance of their hardware design goes through prototyping, products. As a result, COTS servers and bug-fixing, revision and finally manufacture, network products track Moore’s law very the design is already outdated. Even before closely. the product ships, it is behind the technology curve. The specialized hardware design itself Commodity cluster techniques dominate is costly and negatively impacts the specialized hardware designs in terms of solution’s price performance. Furthermore, price/performance. In fact, a cluster of COTS maintaining a hardware design for a servers connected by fast Ethernet won the specialized market is extremely costly as Gordon Bell Prize for Price/Performance well. Bug fixes and upgrades required for any awarded in conjunction with the hardware design must be done by a small Supercomputing Conference in 1997 [Karp team employed by the single vendor. In 1998]. COTS clusters enable a “trickle up” contrast, COTS computing systems leverage effect, whereby technological improvements the design expertise of many competing and better price performance characteristics companies and benefit from the significant developed and perfected for standalone investment these companies make for the applications in the mainstream computer broader computer equipment market. industry are leveraged to provide significant advantages in the area of high-performance Over time the hardware design ages computing applications such as real-time, quickly. Within a year or two after initial high-bandwidth video, and audio streaming. product shipments, the hardware technology Figure 1 shows the COTS value chain, which is outdated, and the price performance is starts with processors, chipsets, and other significantly behind current technology and semiconductor products at the bottom. From products. Moore’s Law cannot be exploited those parts, components are developed and effectively to drive down costs over time for those move up into COTS servers, which can a proprietary hardware design targeted then be combined, to form COTS clusters. toward a specialized application. products that are themselves pre-packaged clusters with all the components priced together as a single bundle. CO TS Clus ters As an extension of the cluster concept, blade servers are the next, high-density incarnation of the commodity cluster. Blade Computer Industry servers do not yet have the same level of Driving Price Performance CO TS standardization that yields the benefits of Servers interoperability and competition for performance. Therefore, the price at this point is higher for blade server solutions. Commodity Chips With the benefits provided by COTS products and their application to high- Figure 1 COTS Price Performance Chain performance computing applications, the trend in the VOD market is toward open

Since COTS clusters are built from COTS hardware systems rather than expensive, servers, the component servers are flexible in difficult-to-maintain proprietary hardware terms of configuration. They can be systems. configured with more or less processing power, DRAM for main memory, and Applying COTS Clusters to VOD buffering and storage for content. Since the cluster itself is composed of smaller, high- VOD applications and more generally performance building blocks (the individual EOD applications have a structure that is servers), the clusters can scale from small well-suited to parallel processing and to capacity to large capacity requirements. COTS cluster techniques in particular

In addition to scaling conveniently to Mapping the VOD Application to the Cluster appropriate requirements for initial For example, if we first consider a installation and deployment of a particular centralized network architecture using COTS application, the clusters are flexible in that clusters for VOD, the cluster must satisfy additional capacity requirements that come in certain requirements including the maximum after the system has been in service for some simultaneous subscriber sessions, each with period of time can easily be accommodated an associated bitrate that may be 3.75 Mbps by the addition of new servers and storage for standard definition (SD) content or much capacity into the existing cluster. larger for high definition (HD) content. The cluster must also satisfy requirements for Clusters are becoming so popular that maximum content storage capacity. This some vendors are beginning to provide not application breaks down into the following only the individual servers and network primary compute or communication devices needed to construct a cluster, but also bandwidth intensive parts: • Business logic that allows a The upshot is that most of the subscriber to request content and pay computational load associated with VOD is for it. in the streaming of independent content to • Session resource management that independent viewers. Thus, computations do takes requests and identifies resources not interact with each other, so as a high- needed to fulfill them, including the performance computing task, relatively little content itself, processing capacity, network bandwidth is consumed in and network bandwidth available to coordinating within a cluster on the execution stream the rich media. of these tasks. The fact that tasks are • Computing capacity to actually independent makes them particularly well stream the high-bandwidth content. suited to the COTS cluster technique for • Low latency communication channels high-performance computing. from subscribers back to the server to control the streams (with pause/fast- Smart Cluster Management Software forward/rewind). To ensure that cluster resources are used most effectively, the cluster must look at The business logic and processing can be incoming requests and determine the best achieved using modern, flexible integration way to service the request. The cluster technologies that power the web’s still- management software must decide which burgeoning e-commerce activities. node in the cluster is best suited to stream a request for a particular piece of content. This Session resource management is a very software handles the load balancing, important piece where intelligence and availability and failover requirements for the accurate information about the current system cluster. This software uses distributed load and currently available resources are algorithms such that each server is capable of very important. This activity, however, does performing this management function, and not take a lot of processor resources or therefore no single point of failure exists in bandwidth - merely timely information. the cluster management.

The bulk of the work, in terms of In order to make resource management processor cycles and network bandwidth decisions, the cluster management software consumed that is required to provide rich must collect information from each node in media on demand, is the job of streaming out the cluster describing the content available on the media. Streaming can be anywhere from the node, the performance characteristics of 30 seconds to 3 hours in duration, for a quick the node, and the current load on the node. video commercial or a long-running movie, With this information, the cluster manager respectively. has information about available capacity to service new requests for specific content on The latency required for responsive trick various nodes in the cluster. It can then make mode transitions places more requirements on the appropriate load balancing decisions and network bandwidth and availability between assign the request to a particular node. the subscriber and the server site without putting a significant additional computation The cluster management software also burden on the VOD server (unless fast- keeps track of server or node availability in forward/rewind content are not computed in the cluster. For example, if a particular node advance and must be computed on the fly). is taken out of service or suffers from some type of hardware failure, the cluster manager Content Management Within the Cluster detects this condition and makes the Part of the job of the cluster management appropriate adjustments to its resource software is to make sure the streaming management policies for the period when the resources for particular content expand as the server is not available. Likewise, when a new demand for that content expands. One server is added to a cluster, it automatically technique is to divide the storage available on participates in the cluster and coordinates each node into a unique content partition and with the other nodes in the cluster to make its a cached content partition. The cache then performance capabilities and availability temporarily holds content that has become known. popular.

Driving Price Performance for VOD To utilize the cache partition of storage, The performance advantages of using one technique is to identify content that is COTS clusters for high-performance becoming popular by monitoring the request computing described above drive price rate in real-time. Once a piece of content is performance for VOD as a specific becoming popular, the cluster manager can application. Scalability for the COTS cluster copy the content to another node to increase translates directly into scalability of the capacity of the cluster and create streams streaming capacity for a VOD cluster. VOD for that content. clusters can scale smoothly from small numbers of streams (a few hundred SD Another technique is to bring new content streams using single dual Xeon class servers) directly into the cache partition of storage on to much larger numbers of streams (tens of the assumption that fresh, newly released thousands of SD streams when these single content is more likely to be popular. The server building blocks are racked up in caching policy then keeps the content for as quantity in larger COTS clusters). long as it is popular and replaces content in the cache based on policies that take into Additionally, COTS clusters for VOD can consideration usage characteristics. On an accommodate different storage configurations ongoing basis, the number of content copies that may be required for different types of that are stored in the cluster can be adjusted VOD storage requirements, performance up and down based on usage characteristics requirements, and usage patterns. For of the content over time. example, a VOD deployment that must support large numbers simultaneous users The content management UI for the (greater than 10,000) may benefit from a cluster must allow the cluster to be managed shared storage model that enables each of the as a single entity with the information from nodes in the cluster to share access to content individual nodes aggregated into whole- on a single large SAN-type storage unit. A cluster totals. This includes information about flexible software solution for COTS the current number of streams being served, a clustering can accommodate appropriate content list as well as number of copies of COTS hardware configurations and reap the each piece of content, etc. benefits of lowering storage costs in cases where large amounts of storage accessible by each individual node is needed. EXPANDING THE VIDEO NETWORK Transport Network Architecture for VOD

High performance in the COTS cluster More sophisticated techniques for content and effective cluster management provide the placement and management extend the idea essential foundation for a complete, scalable of caching to a general storage hierarchy that VOD rich media delivery system, but overlays the network topology hierarchy as additional techniques concerned with the depicted in Figure 2. distribution of clusters, the effective communication of information and content between them, and the efficient use of network bandwidth are also very important for large deployments.

Master Headend

Local Headend Primary Hub Access Network Primary Ring Secondary Ring

Figure 2: Hierarchical Network Architecture

A typical network topology includes: • Primary hubs that are closer to the subscriber and allow for effective • A master headend, which is a central aggregation of transport functionality site where many of the entire and services for the subscriber, and network’s operational services such as • Nodes that reside closer still to the transaction processing and billing are subscriber. housed, • The headend that acts as a point of origin for various video, data and voice services, Network architectures vary in terms of Storage Hierarchy Mapping to Network how much bandwidth is available through the The storage hierarchy can be designed to backbone between regional headends and how mirror the network hierarchy for purposes of much bandwidth is available from headends cost-effective use of bandwidth in the down to primary hubs. These bandwidth- backbone and in the access network and for provisioning factors impact the degree to effective us of network equipment at which streaming can be centralized and how appropriate locations in the network. much motivation there is to decentralize illustrates how this storage hierarchy would streaming. map onto a typical network hierarchy.

Master Headend

Local Headend Primary Hub Access Network Primary Ring Secondary Ring

Caching Cluster

Headend Cluster

Library Cluster

Figure 3: Storage Hierarchy

This storage hierarchy consists of levels • Smaller hubs and nodes that would that include: not typically contain active streaming on-demand servers but might house • Library servers, which act as large elements of the transport network content repositories for a wide range associated with rich media delivery. of different content, located at the master headend. Content Propagation in a Video Network • Centrally located popular content The video network architecture described repository and source for content provides for a great deal of flexibility in flowing toward the subscriber on a terms of network configuration. A typical day-to-day basis, typically from the usage scenario for the video network would headends. have content coming into the network • Localized “caching” servers, located through a combination of content aggregators perhaps in the primary hubs. that distribute via satellite and real-time content capture and ingest. This new content Moving down one level in the hierarchy, is stored initially in the library server cluster the VOD clusters located in the primary hubs in the master headend, perhaps using a economize on network bandwidth in the shared storage hardware configuration to higher trunks of the hierarchy by allowing accommodate very large amounts of content for very popular content to be cached in the such as those generated through real-time closest possible location to the subscriber. ingest applications. Assuming an “80-20” rule where 80% of the content requests can be serviced by 20% of The library cluster provides archival the actual content available, these caching storage for content that has the appropriate clusters in the primary hubs serve a relatively licensing for long-term storage. The content large percentage of streams with a relatively for which a licensing window applies is also small percentage of available content stored stored during the appropriate window and the locally. Thus, the bandwidth required to copy of such content in the library clusters service all of those streams is only consumed serves as the master copy of that content for from the primary hub down to the individual the entire system. subscribers rather than the same amount of bandwidth being required (and the amount The library cluster makes the content compounded by adjacent primary hubs) in available, not for streaming directly to the higher-level trunks for the distribution subscribers, but for transfer to streaming network. servers in other parts of the hierarchical video network. Streaming Bandwidth in the Video Network The streaming bandwidth savings The closest set of streaming servers possible in the trunk links of the network belongs to the VOD clusters located at when VOD clusters are distributed deeper headend sites. These clusters are expected to into the network qualitatively justify the store only content that is expected to be content propagation techniques as described popular or that has been dynamically above. However, a more quantitative analysis discovered to be popular. The headend of those savings is even more convincing. cluster will pull content from the library The following analysis compares a cluster in cases where a request comes in for centralized VOD cluster approach and the the content but that content is not currently bandwidth required through the network to resident in the headend cluster. Latency stream against a standard set of requirements associated with copying the content is to a distributed VOD cluster approach and minimized through the use of streaming the network bandwidth required in that case. mechanisms that allow for content streams to The requirements of higher-bandwidth be played out from the beginning while the content such as HD video streams must be later parts of the content are still being considered as part of the quantitative transferred in to the headend cluster. modeling of the bandwidth requirements

looking toward changes we are likely to see Mechanisms for monitoring content in the near future. usage and making copies of popular content in clusters closer to the subscriber, are based on the algorithms for dynamically creating copies of content within a single cluster as described above. QUANTITATIVE ANALYSIS Storage: One copy of the entire content collection is required at the single central Let us consider the bandwidth location, so the total requirement is 20 TB. requirements for a group of systems that require a total of 40 Gbps of simultaneous Decentralized with Duplication streams (equivalent to more than 10,000 SD Bandwidth: Streaming for each headend streams at 3.75 Mbps) to be delivered to starts at that headend and traverses the subscribers during peak load time, a total of Secondary ring and the Access network. 20 TB in archival storage and a total of 4 TB Therefore, we have the total of 40 Gbps of unique content readily available to crossing two networks for a total aggregate subscribers. Suppose that these subscribers bandwidth requirement of 80 Gbps. are served by 5 systems which need to support 12 Gbps, 12 Gbps, 8 Gbps, 4 Gbps Storage: The content is fully duplicated and 4Gbps for System A, B, C, D and E, at each of the five headends. With a content respectively. To compare the aggregate requirement of 20 TB at five systems, the network bandwidth required in the case of total storage requirement is 100 TB. the centralized video network and the distributed video network, we will consider Distributed Video Network the typical network architecture described above and look at bandwidth required in each Bandwidth: Assuming minimal case for the 3 levels of the network: primary streaming traffic from the Library cluster ring, secondary ring, and access network. across the Primary Ring, we focus on the traffic across the Secondary ring and Access System Number Bandwidth Storage network. Assuming that 80% of the streams of Hubs (Gbps) (TB) can be addressed with 20% of the content A 4 12 20 from the Caching clusters in the primary B 4 12 20 hubs, the bandwidth required for that 80% is C 3 8 20 D 2 4 20 only in the access network and totals 80% of E 2 4 20 40 Gbps or 32 Gbps. The other 20% of the streams are served from the headends; 20% Centralized Video Network of 40 Gbps is 8 Gbps and those streams must traverse both the Secondary rings and the Bandwidth: Assuming that central site is Access network for a total of 16 Gbps of different from any of the five headends: network bandwidth required for the 20% of Requires full 40 Gbps across each of the streams. The grand total for all 100% of three networks (Primary ring, Secondary streams is therefore 32 Gbps + 16 Gbps = 48 ring, and Access network), which is 120 Gbps. Gbps. If the central site is co-located with one of the headends, say one of the largest Storage: Firstly, the Library cluster ones, which supports 12 Gbps streaming, the requires storage for the entire system at 20 result is a savings of 12 Gbps since TB. Assuming a working set of bandwidth for that headend does not need to approximately 400 titles at the headend and go over the Primary ring. Therefore, total assuming each title is approximately 1.5 bandwidth required is 108 Gbps. hours, the headend clusters require 600 hours

of content. At 2 GB/hour that would be 1.2 TB of storage required at each of the five headends; grand total is 6 TB at the required for the fast-approaching EOD vision headends. In the caching clusters, we need of coming services. Novel techniques for the 20% most popular content of the 400 high-performance computing platforms and titles in the working set, so the requirement is efficient video network architectures are for 80 titles or 120 hours (assuming 1.5 necessary to realize this vision. hours/title). At 2 GB/hour, we require 240 GB per primary hub. In the entire system, we The COTS clustering approach to high- have 15 primary hubs each requiring 240 GB performance computing has proved to be of storage, so the total hub storage is 3.6 TB. successful in virtually all high-performance The grand total for all storage in this scenario computing applications since the late 1990’s is therefore: 20TB + 6 TB + 3.6 TB = 29.6 TB. and addresses the needs of VOD delivery very cost-effectively. This approach SUMMARY leverages all of the research and development The summary of bandwidth and storage investment in the computer industry, which requirements for the three scenarios amounts to billions of dollars, for the benefit addressed in the analysis above appears in of driving price performance in the VOD and the following table. EOD application arena.

Smart content management in the cluster Centralized Bandwidth: 108 Gbps management software as well as in the larger Storage: 20 TB scale distributed video network context help Decentralized Bandwidth: 80 Gbps to reduce costs elsewhere in the complete with Storage: 100 TB end-to-end system by using the network Duplication bandwidth and storage resources associated Distributed Bandwidth: 48 Gbps with the video network most effectively. A Storage: 29.6 TB representative analysis of a typical multi-

Clearly, the Distributed Video Network system VOD deployment quantifies the with Library cluster for archival storage and savings in bandwidth and storage that is Caching cluster near the network edges possible with this approach. And as the yields substantial savings in both aggregate requirements of the future scale toward network bandwidth and in storage required. extreme narrowcasting of everything-on- In comparison, the Centralized Video demand, the cost optimization available Network leans heavily on network bandwidth through these techniques becomes essential. and the major expense is in that area. The Decentralized with Duplication architecture, REFERENCES with duplicate content at each headend, leans heavily on additional storage, which 1. Sterling, Thomas. “An Introduction to PC contributes greatly to the high cost of that Clusters for High Performance Computing” solution. The International Journal of High Performance Computing Applications, Volume 15, No. 2, Summer 2001, pp. 92-101. CONCLUSION 2. Karp, Alan H., Lusk, Ewing and Bailey, David H. We have summarized the requirements “1997 Gordon Bell Prize Winners.” IEEE for the large-scale delivery architectures Computer, January 1998.

MODELING SWITCHED BROADCAST VIDEO SERVICES

Joseph Weber, Jiong Gong Cable Television Laboratories

Abstract subscribers by taking advantage of the finite number of simultaneous programs actually Switched broadcast video services can be being watched by subscribers on an individual used to offer many more broadcast programs, network. using less bandwidth, than traditional broadcast services. A typical 750 MHz cable In this paper we develop a mathematical plant could theoretically offer over a model for predicting the network capacity thousand broadcast digital programs to necessary to deliver a given number of video subscribers, compared to a few hundred programs to a network of a given size. The programs using traditional broadcast. model could be used for predicting capacity requirements, or determining when it is We have developed a model to predict economically advantageous to use a switched average and peak usage of bandwidth for a broadcast service over a traditional broadcast switched broadcast system given service area service. size and expected popularity of various programs. This model can be used to plan for We also extend the model to identify the equipment and capacity needed to offer a which channels are eligible for adding to a switched broadcast service. switched broadcast tier, and which should remain on traditional broadcast. INTRODUCTION OVERVIEW OF SWITCHED BROADCAST Cable operators continue to increase the number of broadcast channels they offer their Traditional broadcast video services subscribers. In a broadcast network, the transmit all available programs to all number of channels that can be offered is subscribers on the network all of the time. limited, however, by the bandwidth capacity Given the variability in content across of the last mile of the hybrid fiber coax (HFC) channels, it is highly likely that during some plant to the home. Operators will encounter parts of the day many programs or channels this limitation as they begin broadcasting are not being watched by any subscriber. With more channels, and more high-definition digital services, the number of possible content. While more efficient codecs can programs carried by the network increases increase the number of channels offered to dramatically, and therefore, so does the subscribers, with traditional broadcast probability that programs are being broadcast services the number is ultimately limited by but no subscriber is actually watching. available network bandwidth. In a switched broadcast model, programs In reality, as the number of programs being are broadcast on the local network only when offered grows, fewer subscribers are actually requested by a subscriber. When a subscriber watching some of the more obscure programs. selects a program for viewing via their A switched broadcast network gives the interactive program guide, the application operator the ability to offer an almost determines if that program is currently being unlimited number of new programs to broadcast on the local network. If so, the Like VOD, a switched broadcast video consumer premises equipment (CPE) simply service requires a two-way cable plant. Using tunes into that broadcast. If the program is their remote controls, subscribers select a currently not on the network, the CPE makes program from an interactive program guide a request to the server application to begin (IPG). In switched broadcast services, only broadcasting that program on the network. those programs that other subscribers on the When the subscriber switches to another local network are currently watching are program or is no longer watching the selected present on the local network. Therefore, a program, the server application is again request for a new program not currently on the notified. When the server application network must make its way back to the cable determines that there are no more subscribers headend so it can be added to the local on the node viewing a particular program it network. At the headend, the new program is can remove it from the network and, thus, free loaded onto the local network, and the location the bandwidth for other uses. of the new program is added to the local channel map. Switched broadcast, therefore, has the potential to reduce the amount of bandwidth The Potential of Switched Broadcast Video required to support large numbers of broadcast programs. With normal broadcast The model developed in this paper can be service, the amount of bandwidth required used to estimate the bandwidth required to grows linearly with the number of programs. offer a given number of switched broadcast For analog services this requires 6 MHz of programs, given the network size. With spectrum for every program, while digital traditional broadcast service, a 750 MHz plant services require about 0.6 MHz of spectrum typically can broadcast about 250 programs (a per program. With switched broadcast the mix of analog and digital services). Even if all bandwidth grows much slower with the video services are broadcast in digital, the number of channels. maximum number of programs the operator could offer is about 700. With switched digital services, the model presented here predicts that an operator could offer over a thousand broadcast programs on that same network (See Table 1).

Tier Present Broadcast All-Digital Switched Broadcast Allocation Broadcast Allocation Allocation Programs 6 MHz Programs 6 MHz Programs 6 MHz Channels Channels Channels Basic 26 26 26 26 26 26 (always analog) Standard 50 50 130 13 140 14 Premium 8 8 200 20 500 15 Digital 120 12 340 34 1000 25 HD 8 4 14 7 40 20 Total 212 100 710 100 1706 100 Key: Digital HD Switched

Table 1. Example Spectrum Allocations for Present, All-Digital and Switched Broadcast Systems Video-on-Demand (VOD) and network A MODEL FOR SWITCHED BROADCAST Digital Video Recording (network DVR) are a form of switched broadcast sometimes called Switched broadcast takes advantage of the unicast or narrowcast. For these services, each fact that the popularity of various programs is subscriber can request a unique program non-uniform: some shows are more popular received only by the subscriber. But both than others. Therefore, in a network with a services require significant network finite number of active viewers at any bandwidth because a stream would be particular time, some of the more popular dedicated to only a single user. Switched programs will be watched by multiple broadcast, on the other hand, uses a multicast viewers, while some of the less popular approach. If more than one subscriber has programs will not be watched by any viewers. requested a particular program, they all share Thus, if the operator offered m different a single broadcast stream, therefore using the programs to viewers, most likely less than m same bandwidth for one viewer as for one of those programs are actually being viewed hundred. Only in the unlikely case that a large at any one time. With traditional broadcast, as number of subscribers on the network each the operator increases the number of programs request a unique program would capacity offered for viewing, the bandwidth limitations come into play. requirement grows linearly with the number of programs regardless of the number of Events such as Thursday night primetime subscribers on the network. With switched and the Super Bowl, which could cause broadcast, the bandwidth requirement grows concerns for a network DVR service because slowly with both the number of viewers on of the large number of users on the system, the network and the number of programs actually place a lower demand on switched being offered. As a trivial example, if the broadcast since most viewers can be served network has only one viewer, then only one with only a few multicast streams (although program needs to be broadcast. If there are users do not have the same pause, rewind, and two viewers, a maximum of two, but other recording features available with VOD sometimes only one program needs to be and network DVR services). broadcast at any one time. As the number of viewers grows, many will be watching the Network Requirements same popular programs, so the number of programs required will be less than the Implementing a switched broadcast system number of viewers. can require a significant addition of hardware into the network, as well as software in both We developed a model to estimate the the CPE and the network. In particular, with number of simultaneous programs that are traditional broadcast service a single QAM being viewed by at least one subscriber, and modulator can be used for the entire network. the amount of bandwidth needed for switched The same QAM-modulated signal is sent to broadcast delivery on a network as a function every service group on the network. With of the number of programs offered and the switched broadcast, each service group would number of subscribers on the network. The need its own QAM modulator since the model uses rating information to identify the services available on each service group’s relative popularity of each channel on the local network is different depending on usage. network. This is combined with the variation in the number of active viewers during the day to produce estimates of the number of • Share = number of households tuned unique programs being watched as a function to a particular channel / total number of the number of channels being offered, the of households watching TV number of subscribers on the network, and the time of day. Share numbers also range from 0 to 100%. They indicate what fraction of households Program Ratings actively watching TV are tuned to that particular channel. We designate the shares The popularity of various cable and for a series of m program channels broadcast channels clearly changes with time. by si for = ,...,2,1 mi . Note that at any particular Weekday primetime might show a large m ri number of viewers tuned to network broadcast time ∑ si =1and si = . th )( channels, while Saturday evenings might have i=1 a preponderance of viewers tuned to cable movie channels. Effective Channel Capacity

Nielsen uses the following terms for Given the ratings curve, ri , and HUT identifying the relative popularity of various values as a function of time, (th ) , we can channels: estimate the number of unique programs being viewed on a particular local network • HUT = number of households (service group). This model is independent of watching TV / total number of TV the method used to obtain the probabilities. households They can be either empirical or estimated using a Geometric distribution. Given a The HUT measure varies throughout the service group in a network with n′ homes day, with a peak around primetime and a low passed, the number of customers within early in the morning. HUT is always a number households actively watching television is between 0 and 100%. We designate the given by the product of n′ and the HUT, or variable HUT by th )( where t is the time of = ′ thnn )(* . We extend the definition of HUT day. beyond that used by Nielsen, since a household might have multiple televisions and • Rating = number of households tuned VCRs, resulting in a HUT above 1.0. to a particular channel / total number Typically, HUT remains well below 1.0 of TV households during the early morning hours.

Rating numbers also range from 0 to The n customers watching television on the 100%. Ratings give the relative popularity of network can choose any of the m switched individual channels. They can also be used to programs being offered. The key to switched estimate the number of households watching a broadcast video economics is the fact that at particular channel by multiplying by the total any one time, many programs are not being number of TV households. We designate the viewed by any subscriber. The number of ratings for a series of m program channels those customers watching a particular channel by ri for = ,...,2,1 mi . Note that at any depends on the popularity, or share, of that m channel at that time. In switched broadcast particular time ∑ i thr ≤= 1)( . systems, the exact number of customers i=1 watching a channel is not important, only if there is at least one customer watching. As a result, the actual number of channels Although a channel has to be provisioned in needed, which we call effective channels and real-time even when only one customer is denote it by q, is then a selected integer viewing a channel, the probability of such an number where the probability of l exceeding q event occurring is still affected by the is pre-specified for this normal distribution. popularity of a channel program. That probability is commonly known as the blocking probability p. Under these For our purpose, the state of a particular conditions, the effective channel bandwidth program channel i, which we denote as xi , required is as follows: can be characterized as either on or off, with “on” meaning at least one person is viewing += lVaralEq )()( (E2.3) that program channel (requiring a corresponding channel to be provisioned), and where “off” meaning otherwise. xi is then naturally a Bernoulli random variable that can be 1 a 2 (1) a is such that p 1 −= ∫ − y 2/ dye defined as taking one value [1, 0] with the 2π ∞− n n associated probabilities i −−− ss i ])1(,)1(1[ . (E2.4) m The first probability characterizes the event n (2) ∑ −−= smlE i )1()( (E2.5) that at least one customer is watching channel i=1 m m i, and the second probability characterizes the n 2n event that no one is watching channel i at all. (3) ∑ slVar i ∑ −−−= si )1()1()( (E2.6) i=1 i=1 As a result, the efficiency of turning the program channel i into a switched program The first part of the equation E2.3 above channel is entirely driven by the probability of gives the expected number of channels being its state being off, which is in turn determined watched. Because channel watching is a by the channel’s popularity, or share, si . The stochastic process, not only the expected total channel usage over all the m channels, value but also the variation in the number of which we denote as l, is then the sum of m channels being watched is important. Bernoulli random variables Knowing that total usage follows a Normal distribution allows us to calculate the upper m bound of that variation given a pre-specified = ∑ xl i (E2.2) i=1 blocking probability as defined in the second part of equation E2.3. For sufficiently large m, the Central Limit Theorem says that this summation series can In practice, (E2.4) can be calculated by be approximated by a Normal distribution.1 using a standard Normal distribution table.2 For example, setting a = 2 yields a blocking 1 For a comprehensive treatment of Central Limit probability p = %28.2 . Setting a = 3 yields a Theorem, see for example “Probability Theory”, by M. blocking probability p = %13.0 . Conversely, Loeve, p 268-383, D. Van Nostrand Company, Princeton, New Jersey, 1963. Central Limit Theorem given a p value, a can also be easily found applies to the mean of a series of random variables. It from the table. can also apply to the sum of a series of random variables, provided that the summation series is bounded. Since m, the number of program channels in 2 One commonly used table is from P.G. Hoel, our case, cannot go to infinity, the summation series is “Introduction to Mathematical Statistics, “ 4th ed., thus bounded. New York, Wiley, 1971. The capacity requirement equation can be As expected, the number of actively viewed further simplified if we make the share channels grows slowly as the number of uniformity assumption that all the program channels increases beyond the number of channel shares are the same. For example, subscribers. However, the concave nature of µ these curves means that capacity requirements suppose 21 L ssss mi ==== , where µ m do not increase linearly with program represents the total share of these m programs. channels, but at a much slower rate, especially Then the above capacity equation can be when the service group size is small. As a simplified to: result, the bandwidth efficiency gain increases as the number of switched program channels µ µ µ increases. Traditional broadcast bandwidth mq n ma n −−−+−−= n ])1(1[)1(])1(1[ m m m requirements are linear in the number of (E2.7) channels offered and are shown for comparison. The share uniformity assumption should be a good approximation for several reasons, Bandwidth Usage vs. Number of Channels aside from the benefit of simplified Offered calculation. First, the tail of the program share Switched Broadcast (250 subs) distribution curve, which most likely 250 Switched Broadcast (500 subs) represents the channels which are candidates 200 Traditional Broadcast for switching, is indeed close to forming a 150 uniform distribution in reality. Second, E2.7 gives the upper boundary of the expected 100 usage of channels regardless of the share 50 distribution of the viewing profile. Bandwidth Required 0 Consequently, this approximation also gives 0 50 100 150 200 250 the least amount of bandwidth savings in a Number of Channels Offered switched broadcasting environment. Or stated differently, this approximation provides the most conservative estimate to channel Figure 1 Number of Channels Broadcast capacity provisioning, especially useful in the on Switched and Traditional Broadcast Services as a Function of the Number absence of accurate data on program share of Channels Offered numbers. We would expect that the number of Expected Channel Usage channels being actively watched would grow along with the number of subscribers on the Expected channel usage plays a dominant network. This is because the probability of the role in the effective channel capacity state of a program channel being “on” requirement equation. Generally, as we increases exponentially as the size of the increase the number of switched channels, the service group increases. The number of bandwidth efficiency gain from switching independent channels being watched will also tends to increase. Figure 1 below shows the depend on the shape of the program ratings expected number of unique channels being curve. In Figure 2 we graph the expected viewed with 250 and 500 subscriber groups as number of channels as a function of the a function of the number of channels offered. number of subscribers on the network for the average ‘Primetime’ ratings curve, and a typical Monday night football curve. The available program channels. For example in number of channels actively being watched Figure 3 when 100 program channels are approaches the total number of 200 channels switched for a service group size of 500 as the service group size increases. customers, percentage of bandwidth savings reaches 36%, meaning that only 64 channels Total Number of Channels are needed to offer 100 programs. If the Average Primetime service group size is further reduced to 250 250 Monday Night Football customers, percentage of bandwidth savings

200 increases to 58%, meaning that only 42 channels are needed for offering 100 150 programs. Here we assume 1% blocking 100 probability, 25% total viewership share for 50 the switched channels, and HUT=0.6, which

0 is a typical weekday primetime HUT value. Number of Active Channels 100 300 500 700 900 1000 Service Group Size Percentage of Bandwidth Savings

Figure 2 Number of Channels Actively 100.0% Service Being Watched as a Function of Group Size Service Group Size during Primetime 80.0% 125 and Monday Night Football Nights 60.0% 250 40.0% 500 Bandwidth Savings 20.0% 1000 Bandwidth Savings 0.0% The primary benefit of switched 50 75 100 125 150 175 200 broadcasting video is saving bandwidth so No. of Switched Channels that more program channels can be provisioned. Figure 3 shows the extent of bandwidth savings based on our effective Figure 3 Bandwidth Savings Versus channel capacity requirement, as a function of Service Group Size service group size and the number of channels switched, respectively. The percentage of Figure 4 shows the extent of bandwidth bandwidth savings is defined as one minus the savings as a function of variations in the ratio between the provisioned effective probability of blocking, while assuming channel capacity and the number of actual service group size is 500. As can be seen, there is not a large effective channel capacity variation between 1% and 2% blocking probability. But achieving 0.1% blocking requires considerable more capacity. traditional broadcast, a single QAM device Percentage of Bandwidth Savings can serve an entire cable system, while with 70.0% Blocking switched broadcast each service group needs a 60.0% Probability 50.0% dedicated QAM device. In this section, we 40.0% 0.1% create a model for determining which 30.0% 1% channels are candidates for being included in 20.0% 2% a switched broadcast service. 10.0% 5% Bandwidth Savings 0.0% 50 75 100 125 150 175 200 Selection of Switched Channels No. of Switched Channels We assume adding switched broadcast

Figure 4 Bandwidth Savings Versus capability to a cable network incurs a fixed Blocking Probability cost in terms of additional QAM modulators, Adding a Time Element new software on existing digital set top boxes (STBs), and switching controllers within the The HUT value varies throughout the day, network. The total additional system cost for a peaking usually during primetime. We can service group is divided by the number of model the variation in HUT as a sinusoid with customers within that service group to obtain a period of 24 hours, and peaking around 10 a per-subscriber cost. We assign the pm. Adding the time-varying HUT to our variableC as the cost per subscriber of adding model, we can simulate a node throughout the a channel to the switched broadcast service. day. Moving a group of channels to a switched It is, therefore, possible to dynamically broadcast service frees up bandwidth that can predict the peak usage as a function of time be used for other services, including during the day, and dynamically allocate additional channels. We assign the bandwidth resources appropriately. This variable R to the additional per-subscriber would allow a switched-broadcast tier to use revenue that can be obtained by the additional even less bandwidth during off peak times free spectrum. This can also be seen as an freeing bandwidth for other uses. opportunity cost for not freeing up the spectrum by remaining on a traditional OPTIMAL DEPLOYMENT MODEL broadcast service.

Not all channels are candidates for The model states that the number of channels inclusion in a switched broadcast service. freed by using a switched broadcast service is: Deploying a switched broadcast service is m −=−−=− slVaralEmqm )1()()( n more expensive than traditional broadcast ∑ i i=1 because it requires additional QAM m m modulators and two-way CPE. With n 2n i ∑∑ −−−− ssa i)1()1( i=1 i=1 (E3.1) From this equation the incremental savings in be placed on the switched broadcast service, bandwidth of adding program channel k to a while channels 1 to K −1should remain on program portfolio consisting of k −1 switched traditional broadcast tier. Channel K is the /1 n program channels is: first channel where k −< RCs )/(1 .

k k n n 2n Note that since every sk must be between zero k sas i ∑∑ −−−−− si )1()1({)1( i=1 i=1 and 1, adding switched broadcast service is

k−1 k −1 only viable if RC < 1/ , that is if the n 2n si ∑∑ −−−− si })1()1( incremental revenue of the freed bandwidth i=1 i=1 exceeds the costs of creating the service. (E3.2) Partitioning Channels Example The first term above captures the first order difference in usage of moving from a As an example of how channels can be portfolio of k program channels to that partitioned, we take the model approximation consisting of k −1 program channels. That is of prime-time share values using a Geometric essentially the expected usage for program k . distribution and assume a 250-subscriber The second term in parenthesis above service group. QAM device costs are set at captures the second order difference. That is $600 for the service group, and hardware and the difference in standard deviation of usage software costs are set at $3 per subscriber. between k program channels and k −1 Consequently, the value ofC is $600/250 + $3 program channels. In reality, the second order = $5.40 per subscriber. These numbers are for difference is actually very small and can be illustrative purposes only and do not represent ignored. Under that assumption, the actual pricing. incremental savings in bandwidth of adding a program channel k can then be simplified as The present value of all future cash flows n (1 − sk ) . for a single channel worth of bandwidth is taken be to $8, which is equivalent to $0.80 When the incremental revenue, in terms of per year revenue stream discounted at 10%. saved bandwidth, exceeds the incremental The value of R is therefore $8. Again this costs of adding a channel to switched number if for illustrative purposes and does broadcast service, then that channel should be not represent any typical expected revenue for added to the switched broadcast service. This bandwidth. is equivalent to Given these values ofC and R , equation n n E3.3 predicts that channels 1 through 37 k )1( >− CRs or k >− /)1( RCs (E3.3) should remain on traditional broadcast, while channels 38 and higher should be put on the If the sk are rank ordered from largest to n switched broadcast service. Figure 5 smallest, − sk )1( is an increasing function illustrates the result. of k . Therefore, at some channel K , the value n of − sK )1( will become larger than / RC , meaning that all channels from K to m should 16.0% 14.0% Traditional Broadcast Service Switched Broadcast Service 12.0% 10.0% 8.0% 6.0% 4.0% 2.0% 0.0%

25 35 Ch 1 Ch 5 h Ch 15 Ch C

Figure 5 Identification of Which Channels Should be Migrated to Switched Broadcast Service Based on Economic Model

CONCLUSIONS Switched broadcast video can be a valid element of the next generation cable In this paper we developed a model for network architecture. estimating channel capacity requirements for a switched broadcasting video service. This Our model appears to provide an accurate model uses the expected household television estimate of the amount of bandwidth that can viewership and program share data from be saved by adding programs with low Nielsen to approximate the likelihood of cable viewership shares to switched broadcast customers tuning to a particular channel. service instead of traditional broadcast. Given a desired level of blocking probability, Because typical channel lineups have 75 to the total channel capacity requirement is then 80% of viewership share concentrated in the derived as a function of the service group size top 30 channels, the remaining channels can and the number of channels on the switched have very low viewership shares. As a result, service. This model can be further used to many of the channels with low share are not derive the optimal number of program being watched for extended periods of time. channels to be switched, given a valuation of Switched broadcast can be used to deliver the bandwidth savings as a result of moving many of those remaining channels using some channels from traditional broadcast to significantly less bandwidth than traditional switched broadcast tiers. We further validate broadcast methods. If the saved bandwidth our model against simulated empirical data can be used for other revenue generating from a real cable system. The data collected services, switched broadcast can economically from the live system indicates that the model offer many more channels to subscribers. appeared to predict an upper bound on the service usage, and the bandwidth required to Bandwidth savings can be at least 36%, and ensure minimal blocking of the service. go over 60% for a service group size of 500 digital subscribers The channel capacity model we developed The efficiency gain from switched provides a very conservative upper bound of broadcast video also increases as the number channel usage given a blocking probability. of switched program channels increases. The Even under conservative assumptions and more channels in the service, the greater the blocking probabilities at 1%, bandwidth probability that some of those channels are savings from a switched broadcast service can not being watched. However, efficiency gain be substantial. In a realistic example of a tends to diminish marginally as we move up service group size with 500 digital the rating curve to include more popular subscribers, bandwidth savings during peak channels into the switched channel lineup. hours can be 36% when 100 channels are placed on the switched tier, and go over 60% Switched broadcast video provides a strategic when 200 channels are switched. By reducing competitive advantage over DBS providers in the service group size, bandwidth savings can terms of adding program channels and be even higher. efficiently using bandwidth

Bandwidth savings is an increasing function Switched broadcast video is an ideal of the number of switched channels, and a architecture for broadcasting hundreds of new decreasing function of the service group size program channels if each of those channels has a small rating share. The switched Bandwidth savings in a switched broadcast broadcast architecture may be viewed as an environment is driven by the service group interim step towards full on-demand size, and the number of switched channels. In television. one example where 100 channels are switched, bandwidth savings jumps from 36% Switched broadcast is a competitive to 58% as the service group size is reduced advantage for cable, due to the interactive from 500 subscribers to 250 subscribers. nature of the cable network architecture. It is However, reducing service group size incurs a unlikely to be matched by DBS providers. cost in terms of more QAM equipment and possibly the need for node splitting. MOVING BEYOND THE STANDARD: CREATING ADDITIONAL BANDWIDTH THROUGH EXTENDING DOCSIS 2.0 Jack Moran, Distinguished Member of the Technical Staff Motorola Broadband Communications Sector

Abstract DOCSIS 2.0 (SP-RFIV2.0-I02-020617) was specified by CableLabs and approved Advanced spectrum management can by the International Telecommunications allow operators to understand the nature of Union (ITU) in December of 2002 as the inevitable impairments on HFC networks Recommendation J.122. DOCSIS 2.0-certified so they can compensate appropriately for equipment is now entering the market. these impairments and create maximum The primary advantage of DOCSIS 2.0 is billable bandwidth. This presentation will upstream performance. DOCSIS 1.X offers a discuss how operators can more carefully maximum 3.2 MHz upstream channel width monitor performance in a non-intrusive and a maximum of 16 QAM modulation. The manner. It will show how operators can new 2.0 cable modems will allow operators to automatically correct for impairments and migrate to 32 or 64 QAM across a channel deploy extensions to the DOCSIS 2.0 width of 6.4 MHz. With extensions to specification to create more bandwidth for DOCSIS 2.0, operators now have the ability DOCSIS 1.0, 1.1, and 2.0 cable modems. It to double the bandwidth available to legacy will also explain how post equalization modems while concurrently deploying techniques can allow operators to nearly DOCSIS 2.0 modems that triple the double the throughput of their installed basis bandwidth available today to 1.X modems. of DOCSIS 1.0 modems by enabling them to run 16 QAM modulation virtually anywhere that QPSK is currently running. By implementing noise cancellation simultaneously over diverse populations of cable modems, operators can provide throughput levels beyond that offered by the specifications to increase revenues while efficiently migrating to DOCSIS 2.0.

DOCSIS: A HISTORICAL PERSPECTIVE

Over just a few short years, the Data over DOCSIS 2.0 doubles the channel width and triples the Cable Service Interface Specification upstream capacity when compared with DOCSIS 1.X (DOCSIS) has evolved as the industry cable modems. standard for ensuring interoperability between cable modems at the subscriber location and Cable Modem Termination System (CMTS) platforms at the operator’s headend. This requires selecting the appropriate The asymmetric nature of DOCSIS 1.1 technologies and solutions for taking results in limited ability to support services advantage of the many benefits of DOCSIS that are more symmetric, such as video 2.0. But the rewards for Multiple System conferencing or packet telephony. But the Operators (MSOs) that successfully deploy new DOCSIS 2.0 specification focuses DOCSIS 2.0 while leveraging existing primarily on the upstream path from the network assets are highly appealing. They subscriber to the network. It provides will be able to immediately deploy higher- significantly increased capacity and improved performance services over existing robustness to the upstream path, thus helping infrastructure, and capture new revenues from operators make maximum use of their existing premium services. infrastructure.

At the same time, they will be able to Higher Modulation extend the life of legacy cable modems while increasing throughput. They will be able to Higher upstream bandwidth is enabled by more efficiently utilize existing infrastructure adding a higher symbol rate and higher-order and deliver higher-speed services over modulations. Operators can create a greater existing Hybrid Fiber Coax (HFC) networks. range of tiered services with graduated pricing plans, and thus have the ammunition to better compete with incumbent carriers for small- DOCSIS 2.0: INCREASING and medium-sized business accounts. PERFORMANCE, THROUGHPUT Providing modulation rates above 16 QAM does of course come at a price, because DOCSIS 1.1 provided the QoS control higher modulation rates require a higher SNR and enhanced security lacking in DOCSIS from the network. Fortunately, improved 1.0, so the main focus of DOCSIS 2.0 became signal processing technology can allow the improvement of performance and the more operators to avoid the need for expensive efficient use of network capacity. plant upgrades, and DOCSIS 2.0 includes the ability to capitalize on advanced signal Upstream Capacity processing technology.

DOCSIS 2.0 triples the maximum Impairment Protection upstream capacity when compared to DOCSIS 1.1. It enables transmission across a The new specification provides better 6.4 MHz channel and increases upstream protection from impairments on the CATV throughput to 30.72 Mbps by using 64 or 128 network. DOCSIS 2.0 enables operators to QAM and Trellis Coded Modulation (TCM). support up to 16 correctable symbols, rather than the ten symbols available in the previous DOCSIS 2.0 cable modems utilize fewer specifications. timeslots to transmit a given amount of bandwidth than DOCSIS 1.X cable modems. Operators can understand the various This frees timeslots that can then be shared by impairments present in their infrastructure other modems, which increases their through the use of advanced spectrum throughput. measurement. This is particularly critical because in real-world environments, most—if not all—of the following impairments are tap equalizer. With pre-equalization, the present at some level the majority of the time. CMTS receiver equalizer convergences on a Ingress Noise, which is significant periodic burst and then sends the equalizer everywhere in the return path. coefficients to the cable modems for implementation in their transmitters. This Impulse Noise, which is not significant enables increased modulation and faster above 18 MHz. performance.

Common Path Distortion (CPD), which Advanced PHY is significant everywhere in the return path. DOCSIS 2.0 includes advanced Physical Layer (PHY) modulation techniques that Micro-Reflection, which is a significant allow operators to run higher modulation impairment everywhere. levels. DOCSIS 2.0 enables enhanced Amplitude Distortion, which is significant management of RF spectrum so that operators only above 35 MHz in 42 MHz systems, can more efficiently cancel out or avoid noise 48 MHz in 55 MHz Systems and 55 MHz impulses. This allows increased throughput in 65 MHz systems. and more reliable service delivery. The DOCSIS 2.0 specifications includes two Group Delay Distortion, which is also separate technologies for achieving these significant only above 35 MHz in 42 MHz goals: systems, 48 MHz in 55 MHz systems and 55 MHz in 65 MHz systems. Advanced Time Division Multiplexing (ATDMA) DOCSIS 2.0 allows operators to improve Synchronous Code Division Multiple the Signal-to-Noise Ratio (SNR). However, Access (SCDMA) operators need the ability to more efficiently manage spectrum and improve noise cancellation simultaneously over diverse UNDERSTANDING DOCSIS 2.0 populations of DOCSIS 1.0, 1.1, and 2.0 PROTOCOLS modems. These protocols allow operators to This requires a CMTS that not only increase the channel size to 6.4 MHz and they supports DOCSIS 1.0, 1.1, and 2.0 but also support statistical multiplexing to optimize offers a system architecture designed to bandwidth utilization. They enable the use of improve the SNR of both legacy and new a higher symbol rate and can deliver up to DOCSIS modems. triple the capacity of a DOCSIS 1.X channel. Both protocols can coexist on the same Pre-Equalization channel because each logical channel type is assigned non-overlapping timeslots, and these The DOCSIS 2.0 specification offers timeslots are interleaved based on demand. increased support for transmit pre- equalization. It enhances micro-reflection DOCSIS 2.0 supports immunity to (multipath) protection by increasing the Ingress Noise for both protocols, though length of the equalizer to 24 taps—which is ATDMA is more robust against this three times longer than the DOCSIS 1.1 eight- impairment and it supports enhanced channel equalization for both protocols to protect Enhanced transmit pre-equalizer (24 taps) against system linear impairments. It enables improved immunity against Impulse Noise CDMA spreading to provide some through the use of an improved Forward Error immunity to impulse and ingress noise Correction (FEC) that includes the technique Interleaving of SCDMA frames to provide of Byte Interleaving for ATDMA and Frame impulse noise immunity similar to that Interleaving for SCDMA, though SCDMA is provided by byte interleaving on the more robust against this type of impairment. ATDMA protocol

Both protocols support extended DOCSIS 2.0 specifies that the CMTS modulation formats up to 64-QAM for receiver must support DOCSIS 1.1, ATDMA, ATDMA and 128-TCM for SCDMA. and SCDMA modulation technologies on the Operators can implement either or both same carrier frequency, which is referred to as protocols, depending on their requirements mixed mode operation. and preferences.

ATDMA Supports: INTEROPERABILITY WITH .X CABLE MODEMS

A maximum channel width of 6.4 MHz DOCSIS 2.0 is backward-compatible and a minimal channel width of 200 kHz with the earlier specifications, and since it A maximum modulation rate of 5120 allows more subscribers and services on a ksym/s and a minimum rate 160 ksym/s single channel, operators can increase revenues from existing infrastructure while Increased modulation orders, including supporting symmetrical applications and QPSK and 8, 16, 32, and 64 QAM leveraging the QoS features of DOCSIS 1.1.

Enhanced transmit pre-equalizer (24 taps) DOCSIS 1.X cable modems of course do Enhanced Reed-Solomon error correction not support DOCSIS 2.0, so the new with byte interleaving specification provides for mixed mode operation for supporting 1.X and 2.0 cable Ingress noise cancellation with both modems. Unfortunately, this results in DOCSIS 1.X and DOCSIS 2.0 cable additional overhead of roughly 5-15 percent modems is possible. for ATDMA mode and 15-35 percent for SCDMA. This means that the existing SCDMA Supports: customer base of installed DOCSIS 1.X cable modems will experience a degradation in A maximum channel width of 6.4 MHz throughput performance as 2.0 is deployed. and a minimal channel width of 1600 kHz

A maximum modulation rate of 5120 However, there is an innovative approach ksym/s and a minimum rate 1280 ksym/s to transitioning to DOCSIS 2.0 without incurring this performance overhead. Trellis Coded Modulation—QPSK and 8, Operators can implement ATDMA receiver 16, 32, 64 and 128 TCM technology that is by definition directly compatible with DOCSIS 1.X systems and is capable of operating in a true DOCSIS 1.X DOCSIS 1.X and 2.0 operation across the mode. same infrastructure.

Operators can transition to 2.0 without Cable operators can double the upstream providing a performance burden to legacy bandwidth for a large population of modems, subscribers because the ATDMA CMTS thus creating increased billable bandwidth can operate in DOCSIS 1.X mode. without further network buildout. They can create upstream bandwidth that supports The ATDMA receiver technology higher-speed services and enables new provides post-equalization support that broadband services that command premium can increase throughput for existing pricing. customers by at least 50 percent by using DOCSIS extensions that enable virtually all DOCSIS 1.0 cable modems to operate in 16 QAM mode where they could ADVANCED SPECTRUM MANAGEMENT previously operate in QPSK. TO EXTEND DOCSIS 2.0 When there is a significant number of DOCSIS 2.0 cable modems installed, the Efficient migration to DOCSIS 2.0 cable operator can begin the ATDMA requires the effective management of Logical Channel Operation in which the impairments on the HFC network Symbol Rate remains the same (2560 ksym/s) but the DOCSIS 2.0 cable Spectrum management is essential so that modems can begin to transmit in a pure operators can identify impairments and make ATDMA mode of operation, i.e. with the necessary adjustments to improve extended FEC correction, byte performance. The most problematic interleaving (if necessary) using higher impairments—ingress noise and impulse constellation rates such as 32 QAM or noise—can be measured using the established even 256 QAM. Fast Fourier Transform (FFT) measurement technique. When the number of 2.0 modems exceeds the number of 1.0 modems, then the full However, FFTs alone cannot accurately logical channel operation of using assess the total impact of noise on network ATDMA mode in a 5120 ksym/s performance. operation can be implemented while the remaining 1.0 modems operate at 2560 The fundamental challenge is that ksym/s. measurement time directly impacts throughput, because in the typical scenario The financial benefits of this migration operators cannot send data while taking approach are compelling. Operators can measurements. Operators trying to improve accelerate revenue from 2.0 services, and they performance are therefore hard-pressed to can implement gradual migration at the pace impose increased demands on the that makes the most economic sense for them. infrastructure by performing continuous They can continue to support legacy modems testing that degrades the bandwidth being while introducing new services to these tested. subscribers, and they can concurrently support

But the implementation of advanced performance on any one of the upstream ports spectrum management on an extra receiver without impacting performance. Operators avoids the performance burden and allows can therefore non-obtrusively gain access to operators to collect the performance all of the return nodes connected to one of the information they need to optimize the use of receiver ports and perform tests on any network infrastructure. available modem on any one of the receiver The addition of an extra receiver on the port’s supported nodes. CMTS allows operators to monitor

MSOs can use the extra receiver to It can access all of the mapping identify the diverse types of noise and then information as well as a full list of cable implement spectrum management modems available to whichever receiver port enhancements that can process this is currently being evaluated. Therefore, while information and take measures to cancel it out the receiver port being monitored is in real time. The extra receiver can be performing its function at full capacity, the effectively connected in parallel with a live ninth receiver has the luxury of time to receiver port so the operator can measure perform detailed, lengthy, and coherent SNR traffic and performance in real-time on any measurements. This architectural approach to given live receiver port. the CMTS enables continuous monitoring and adaptation so that cable operators can aggressively implement advanced noise Post-Equalization cancellation in environments where the types and degrees of impairments change Post-equalization capabilities offer the frequently. operator the ability to increase the throughput of DOCSIS 1.0 cable modems by allowing Sophisticated Noise Cancellation for ATDMA them to operate in 16 QAM mode virtually and SCDMA anywhere that it is possible to operate in QPSK. The CMTS performs per-burst Extensions to the DOCSIS 2.0 protocol equalization which enables the receiver to can leverage the DOCSIS ATDMA equalize—and thus correct for—the effects of specification and add value by including micro-reflections, amplitude distortion, and advanced noise cancellation techniques that group delay distortion. work with all DOCSIS 1.X and 2.0 cable modems to help operators increase These impairments have historically been throughput. Cable operators can double the the limiting factors in achieving QAM performance of legacy modems while modulation higher than 4 QAM (QPSK). The concurrently deploying DOCSIS 2.0 modems combination of post equalization and superior that enable new services and increased ingress noise cancellation capabilities results performance levels. Rich noise cancellation in a DOCSIS 1.X system today where 16 capabilities allow operators to optimize QAM, error-free operation is achievable performance while operating in DOCSIS virtually anywhere in the return path. 1.X/2.0 mixed mode.

For example, operators can benefit from POWERFUL MONITORING, upstream transmission speeds that exceed MEASUREMENT, AND MANAGEMENT DOCSIS 2.0 specifications by 10.24 Mbps using ATDMA technology and 256-QAM, Sophisticated algorithms can be which is enabled by proprietary extensions implemented on a CMTS to determine which beyond DOCSIS 2.0. Therefore, while cable modems are most representative of the DOCSIS 2.0 has opened an opportunity for a return path under evaluation can use these standard cable modem to operate up to an modems to make signal quality measurements information rate of 30.72 Mbps (64-QAM or for the cable plant. This detailed monitoring is 128-TCM QAM), operators can leverage non-intrusive to the subscriber while enabling extensions to offer significantly higher the operator to continuously monitor noise throughput beyond the capabilities of a and improve performance. standard implementation. Operators can give performance The advanced spectrum management guarantees and implement flexible and capabilities, superior monitoring and automated means of continuously minimizing measurement, and rich noise cancellation noise and increasing performance. For allow operators to achieve higher modulation example, a cable operator can monitor rates. Cable operators can achieve 128 and performance and implement frequency 256 QAM for ATDMA implementations. hopping to a carrier frequency that will These higher modulation schemes allow support guaranteed, error-free 16 QAM operators to achieve even greater throughput operation. than offered by the DOCSIS 2.0 specification. With advanced spectrum management ABOUT THE AUTHOR capabilities on a robust CMTS platform, operators can optimize performance across Jack Moran is a distinguished member of network infrastructure that consists of the technical staff for the Network DOCSIS 1.0, 1.1, and 2.0 cable modems. Infrastructure Solutions (NIS) business for They can monitor and manage the use of Motorola Broadband. He is responsible for spectrum in real-time without affecting DOCSIS Physical Layer performance over an performance so that cable operators can HFC RF network system. carefully measure the factors impacting For the past four years, Moran has been spectrum utilization and compensate in real modeling the return path for DOCSIS 1.0, 1.1, time to optimize throughput. Operators can and now 2.0 performance capabilities. This implement a smooth transition from DOCSIS modeling effort has included several live plant 1.X to 2.0 while gaining the ability to improve characterizations in an effort to simulate on a the performance of the existing installed base repeatable basis the type of real-world so they can increase revenue during the impairments that DOCSIS systems must transition. overcome. He is also a member of the DOCSIS 2.0 Technical Team PHY Layer Issue as well as a member of the former IEEE 802.14 Cable Modem Group. MOVING SET-TOP BOXES TO THE IP INFRASTRUCTURE

Doug Jones YAS Broadband

Abstract be build from scratch. Equipment is available from multiple suppliers. Moving the STB to the IP infrastructure is Technicians are already trained for the a whole lot more than just delivering video network equipment. over IP. In fact, of the five areas of consideration discussed in this paper, video What remains to happen is the migration over IP is probably the furthest from of the video business to the common IP happening. When migrating STBs to the IP platform used by these other Cable services. infrastructure, the first steps will include Not only will additional cost savings be solving embedded CM provisioning and realized by making use of the existing moving OOB transport and interactive infrastructure, there is more innovation and applications to the IP infrastructure. Video less operational expense associated with the over IP will follow when the economics existing IP infrastructure. The move does prove viable. not include Internet access; rather, the move will have the STB business running on IP INTRODUCTION networks.

Business Case The IP infrastructure has proven to be both general purpose (supporting many The near-term drivers for moving the STB different applications from web and email to to the IP infrastructure are neither Internet VoIP and streaming media) and low cost. IP access nor delivering video over DOCSIS. has stood up in the face of competition from Rather, moving the STB to the Cable IP other networking technologies such as infrastructure is about reducing both capital AppleTalk and Novell Netware. IP and operations costs by placing all services technology just works, is widely deployed (data, voice, and the video platform) on the and available from multiple suppliers. same network infrastructure. Areas of Discussion The cable industry is migrating from proprietary technologies to using standard Providing either HSI or VoIP service technologies based on the Internet Protocol through the STB is not discussed in this (IP). It started when Cable operators began paper. Rather this paper discusses five areas building an IP infrastructure to provide high- when considering moving the STB to shared speed access to the Internet. Most recently IP infrastructure. These are: this IP infrastructure will be used to provide telephone service using Voice over IP (VoIP) 1. Embedded CM Provisioning technology. Running a second service on top 2. Out-of-Band Transport of a single infrastructure provides many 3. Interactive Applications business advantages. Specifically, the 4. Video Delivery network is already there and does not have to 5. CMTS considerations The section on CMTS considerations is connections through the CMTS, a section is include because items 2, 3 and 4 place traffic included to discuss the traffic engineering on the CMTS which could also be carrying impacts of this device. HSI and Telephone service. 1. eCM PROVISIONING OVERALL SYSTEM VIEW Introduction A system view with the five areas of discussion is shown in Figure 1. Because the STB is being added to the IP network infrastructure, the STB will use a DOCSIS® embedded Cable Modem (eCM) STB & interactive servers for both one-way OOB traffic and interactive STB traffic over the HFC. The eCM does not controller have to be used for either Internet access or Interactive applications RF for video delivery, just for OOB and combiner interactive traffic. VOD QAM HFC STB eCM CMTS 1-way OOB traffic (DSG) Having an eCM in a STB provides an 2-way interactive traffic interesting choice for operators. The STB is traditionally provisioned by the STB headend

eCM controller. DOCSIS CMs are provisioned by Provisioning a set of HSI servers. These two worlds have not met before and as a result, the operator Figure 1 – IP STB system view has the following choices when it comes to provisioning. Note that in addition to the embedded CM (eCM) in the STB, a large portion of the 1. Both STB and eCM provision from the diagram deals with integrating the Cable IP STB controller, which means the eCM network with the headend STB systems. may follow STB-supplier proprietary provisioning instead of defined HSI The five areas of the paper are represented provisioning steps. in this view. eCM provisioning is shown by the connection through the CMTS to the 2. Provision the STB from the STB eCM provisioning servers. OOB transport is controller and the eCM from the HSI show by the connection from the STB provisioning servers controller, through the CMTS as a one-way stream of information to the STB. 3. Provision both the STB and the eCM Interactive applications are shown in the provisioning from the HSI complex headend via an IP connection to the STB (which means the STB provisioning through the CMTS. Video delivery is show migrates to a single complex that is also through a QAM modulator and not the provisioning DOCSIS, PacketCable and CMTS as this is the most cost effective CableHome devices. means to deliver digital video at this time. There is no technical reason why digital This paper will assume that option 2 is video could not be delivered through the used because this allows continued use of the CMTS, however at this time the economic deployed STB controller while using reasons are not there. Finally, with all the standard eCM provisioning. In the longer • To maximize compatibility with existing term, since both HSI and VoIP are back-office management and provisioned from the “data” side of the provisioning infrastructure so that new house, it is expected that STB provisioning services enabled by eDOCSIS devices will also migrate in this direction toward can be deployed rapidly. (option 3); however, this is a future topic. • To architect eDOCSIS devices in such a way as to scale to new services and eDOCSIS™ applications, and to take advantage of technology innovations to achieve low There is an industry specification for cost and high functionalities. Embedded DOCSIS, called eDOCSIS. An eCM has special requirements placed The first DOCSIS specifications were on it beyond a standalone CM (the kind used created for stand-alone cable modems that for HSI service). Specifically when the eCM provide high-speed Internet services. The is provisioned it includes information that emergence of a class of devices that embeds identifies the device it is embedded in. additional functionality with a Cable Modem, eDOCSIS currently specifies how to embed a such as packet-telephony, home networking CM inside of two devices, a PacketCable and video, has necessitated the creation of Media Terminal Adapter (MTA) and a this specification to define additional CableHome Residential Gateway. eDOCSIS requirements such as interfaces, management does not specify how to embed a CM inside a and provisioning models. This is necessary to STB; however, it seems reasonable to follow insure the eCM will function and interact the blueprint already devised for these other properly with both the network and the devices. device in which it is embedded. For the purpose of provisioning, Figure 2 The goals for eDOCSIS are described in is one possible logical diagram of a STB with the following list: an embedded DOCSIS CM and it shows three separate elements within the STB that • To preserve functional separation of the need to be provisioned: DOCSIS cable modem entity from the device in which it is embedded. This 1. eCM ensures existing DOCSIS cable plant 2. eSTB integrity, cable modem configuration, 3. CableCARD™ management and provisioning security are not compromised. The eCM will be provisioned by the HSI • To isolate DOCSIS cable modem servers as a separate entity from the rest of functionality so that specification the STB. The embedded STB (eSTB) is compliance can be tested for the eCM provisioned by the STB controller over the component independent of device in DSG connection (described in a later which it is embedded. section). CableCARD provisioning depends • To enable the service provider to enable on operator policies and is beyond the scope or disable forwarding traffic between of this document. each application and the eCM within the eDOCSIS Device. video A/V ports An industry specification, DOCSIS Set- eSTB App ports Top Gateway (DSG), provides a method to

RF migrate the legacy STB OOB channel to the RF CableCARD common IP infrastructure. DSG supports

2-way comm retaining the legacy OOB transport for eCM LAN ports 1-way DSG already deployed STBs but allows newly deployed boxes to operate on the Figure 2 infrastructure.

eDOCSIS provides guidelines for how the DOCSIS Set-Top Gateway (DSG) eCM should be provisioned, including DHCP options. In addition, eDOCSIS provides a The DOCSIS Set-top Gateway function blueprint for defining the logical connections resides in the CMTS and is called the DSG within the STB to the CableCARD and the Agent. The DSG Agent provides reliable eSTB for consistent headend management and transparent transport for Out-Of-Band practices. messaging that has traditionally been carried on dedicated legacy OOB channels, 2. OUT-OF-BAND TRANSPORT specifically those defined in SCTE 55.1 and SCTE 55.2, over a DOCSIS forward channel. Introduction A DSG configuration is shown in Figure 4 OOB Transport enables a control-plane reflecting how as part of DSG the STB connection to the STB, even when there is an controller is networked to the CMTS. outage on the return path (and only the forward path is available to the STB). Traditionally this connection is from the STB STB & interactive controller over a QPSK modulator on the servers forward cable plant; a network completely STB separate from the IP infrastructure as shown controller in Figure 3. RF combiner

HFC STB RF STB & interactive combiner eCM servers CMTS 1-way OOB traffic (DSG) STB OOB 2-way interactive traffic controller QPSK

STB HFC Interactive applications

1-way OOB traffic eCM CM Provisioning VOD QAM 2-way interactive traffic - data - STB Figure 4 CMTS

To support operation during return path outages, DSG uniquely modifies the eCM to eCM Provisioning continue receiving downstream information in the event of a return path outage. Figure 3 Receiving the OOB information through

DSG takes precedence over registering as a DSG Advanced Mode DSG two-way CM. DSG Advanced Mode provides several Legacy OOB messages from the STB benefits over Basic Mode. controller arrive at the CMTS where the DSG agent encapsulates them into Ethernet First, DSG Tunnel Addresses are learned frames which are delivered over the DOCSIS dynamically from the network; in this case forward path. DSG further describes how to the DOCSIS downstream. This provides the address these Ethernet frame to allow an operator with more flexibility when eCM in a STB to find them on the DOCSIS associating STB controller information to channel. The eCM passes the DSG DSG Tunnels. information to the CableCARD or eSTB depending on how the system architecture is Second, Advanced Mode provides defined. additional options for identifying DSG tunnels. In addition to Ethernet addresses, DSG supports a minimum of 8 tunnels per both CA_system_ID and Application_ID are STB Conditional Access system (CAS). In included. These two IDs still map to this way, information can be segmented onto Ethernet MAC addresses on the DOCSIS individual tunnels for easier management. forward path, but using the IDs allows a For instance, System Information (SI) and more seamless integration with existing STB Emergency Access System (EAS) operations. information could be placed on one tunnel while Electronic Program Guide (EPG) Third, Advanced Mode provides a keep- information could be placed on a second alive message on the DOCSIS forward path. tunnel and CAS information such as In the event of a backend network outage Entitlement Management Messages (EMMs) when STB controller messages may stop can be placed on a third tunnel. The operator appearing on the DOCSIS downstream, this has the flexibility to configure the DSG keep-alive message provides a mechanism to tunnels based on their needs and preferences. keep the embedded CM on that downstream channel so when the network is restored the DSG Basic Mode STB will begin receiving OOB information as quickly as possible. In DSG Basic Mode the destination Ethernet address of the DSG Tunnel may be Fourth, Advanced Mode provides either a multicast (group) MAC address or a additional security features achieved through unicast (individual) MAC address. The DSG a combination of techniques. First, the Client in the STB recognizes a DSG Tunnel destination MAC address of the DSG Tunnel solely by the uniqueness of the DSG Tunnel may be replaced dynamically. If the MAC Address. address of a tunnel ever becomes widely known, it may provide the opportunity for a The tunnel addresses are known by the PC to spoof that MAC address and snoop the STB ahead of time. The tunnel addresses are DSG Tunnel. DSG Advanced Mode also either included in the software image of the provides a downstream filter which will STB or they are learned from a further qualify the DSG Tunnel based upon CableCARD™ when it is inserted into the destination IP address, source IP address, and STB. destination UDP port. 3. INTERACTIVE APPLICATIONS Network Topology

Introduction The back-office IP network is generally very specific to the operator. There are Interactive applications require a method numerous protocols and connectivity choices for 2-way communication between the STB available that can be used to meet business and the application server on the back-office objectives. Important considerations include: network. Since the network has migrated to • address plans that incorporate STB IP, supporting the end-to-end connection application servers in the network with over IP is not unreasonable. And since the HSI and VoIP servers. method for Cable to offer IP over the HFC • prioritizing STB interactive application network is DOCSIS, having an embedded traffic relative to HSI and VoIP traffic. CM in the STB provides a means for • designing and sizing network connections communications between the STB and • routing protocols for a resilient network. application server to be over common infrastructure. Interconnection of the Video and HSI networks should use common protocols and Three areas relating to interactive links designed from a systems perspective. applications are discussed: • OCAP Security • network topology and services • security Figure 5 shows an additional connection not included in Figure 1. All relate to support for IP networking and connections. STB & interactive servers OCAP STB controller OCAP, the OpenCable™ Application Platform, provides Application Programming Interactive applications Interfaces (APIs) that support two-way IP RF networking. Specifically OCAP 1.0 provides combiner support for the basic forms of IP transport VOD QAM (unicast and multicast) as well as obtaining an IP address using the Dynamic Host Internet CMTS Configuration Protocol (DHCP).

While the existing APIs are sufficient for basic IP communication, additional APIs eCM could be used for more advanced forms of IP Provisioning communication. Figure 5

Specifically the CMTS connection to the Video Delivery will be discussed in two Internet is shown for HSI and VoIP services, segments: from the video server to the HFC providing an indirect connection between the edge and from the HFC edge to the STB. Internet and the STB platform. The HFC edge can either be a QAM modulator that delivers video over native This is not a cause for concern for a well- MPEG transport or the HFC edge can be a engineered network. The HSI and VoIP CMTS that delivers video over DOCSIS. service elements also have this indirect Technically both work, economically there connection and are kept safe through prudent are differences. networking practices. The operator needs to be aware of the threats and take appropriate Server To HFC Edge measures to secure the STB platform. Backbone and Metropolitan Area 4. VIDEO DELIVERY Network digital video transport (broadcast and on-demand) is already using MPEG-2 Introduction framing over User Datagram Protocol (UDP)/IP carried over Gigabit Ethernet. Both Video delivery over IP is not a question of Single Program Transport Streams (SPTS) when but how far to take it. Video is and Multi-Program Transport Streams delivered from the video server to the edge (MPTS) are carried this way. An example QAM modulator over Ethernet connections of SPTS is a VoD stream at the streaming (typically Gigabit Ethernet, GigE) as IP server’s Gigabit Ethernet output. An packets today. Therefore arguably video is example of MPTS is the multiplexed being delivered over IP today. broadcast streams from a multiplexer.

But it is also technically possible to Current industry practice is to encapsulate deliver video over IP all the way to the STB seven MPEG-2 transport frames into a single by delivering the video to a CMTS. Both UDP/IP packet. Both constant bit rate (CBR) scenarios are show in Figure 6. and variable bit rate (VBR) encodings can be supported. The IP-based video transport is

QAM directed to an edge termination that may be either a QAM modulator or a CMTS.

... QAM

QAM HFC Edge To The STB

CMTS

... The debate about video over MPEG or video over IP should be about video over the CMTS most cost effective HFC transport. MPEG and IP are both digital technologies that Gigabit HFC Video Servers Ethernet Edge deliver content in the form of 1’s and 0’s; Switch devices however the current costs of the technologies fabric are different. To determine which is the most

cost-effective, the cost of MPEG QAM Figure 6

modulators should be compared to the cost of All of these traffic types can be carried IP QAM modulators. An IP QAM modulator over IP; however, each traffic type may is the QAM modulator in a CMTS. Count demand different resources from the CMTS. the number of downstream ports in a CMTS (each is a QAM modulator) and divide that Simply mixing these traffic types on the number into the cost of the CMTS to get a DOCSIS channel may result in the jitter or rough estimate of the cost of an IP QAM delay of all packets or in the worst case modulator. Compare this to the cost of the dropping some packets. DOCSIS 1.1 QAM modulator being used for VOD. While provides tools that can guarantee the delivery there are other considerations, this provides a of traffic types, not all traffic, but some types rough estimate. The decision is a matter of of traffic. The operator will need to apply economics, not technology. these DOCSIS Quality of Service (QoS) tools, along with proper network 5. CMTS CONSIDERATIONS engineering, to meet service guarantees.

Introduction The PacketCable™ Multimedia project (PCMM) describes one method to apply and Two considerations will be discussed that manage DOCSIS QoS tools. A discussion of illustrate issues when carrying multiple PCMM is beyond the scope of this paper; services over a DOCSIS consideration. The however, it will provide a valuable method to first is ensuring each class of traffic receives manage QoS on the DOCSIS network. the proper delivery guarantees. The second is ensuring overall that the CMTS does not CMTS Resources run out of other resources before becoming bandwidth limited. The DOCSIS protocol has certain design These issues need to be solved when a limits that should be respected when second service is added to the CMTS, such engineering a network. With HSI service the as adding VoIP when already offering HSI. CMTS usually runs out of bandwidth before With DSG and possibly other traffic types on it runs out of other DOCSIS resources; the horizon, solving these issues becomes however, as different types of services are even more important. added to the DOCSIS network, it is possible that other DOCSIS resources will come into Traffic Classes play before available bandwidth is exhausted.

The CMTS may be carrying several types One such limit within the DOCSIS of traffic including: protocol is the pool of available Service • DSG traffic Identifiers (SIDs). Each upstream • STB interactive application traffic connection between a CM and CMTS • HSI traffic requires a SID. Within a DOCSIS domain, • voice traffic defined as the group of upstream channels • video traffic associated with a downstream channel, there • on-line game traffic, etc. are approximately 8,000 available SIDs.

Field experience shows that with HSI a SUMMARY DOCSIS domain will run out of available bandwidth before 8,000 CMs are present; Moving the STB to the IP infrastructure however, the addition of DSG to the network involves more than just delivering video over will increase the number of low bandwidth IP. Nor is Internet access from the STB a CM connections. DSG is assumed to be low prime concern. Rather, there are bandwidth at the start because current two- fundamental provisioning and traffic way STB connections are much more engineering issues to address as more concentrated than HSI connections. services are added to the IP infrastructure.

With the addition of low bandwidth connections to the already existing HSI AUTHOR CONTACT connections, it is possible the CMTS SID limit could be approached. Theoretical Doug Jones calculations show that even with multiple YAS Broadband Ventures low-bandwidth connections the industry has 1877 Broadway a low probability of reaching the SID limit; Boulder, CO 80302 however, further field experience is needed voice: +1 303.415.9400 x303 to fully understand the issues. [email protected]

MPEG-2 VIDEO STREAMS ON A DATA NETWORK (or What Happens when Video gets the Jitters)

William Garrison Motorola, Inc., Broadband Communications Sector

Abstract a data network, this paper will focus on the issue of jitter. Also this paper will focus on This paper will review the issues involved MPEG-2 Transport Streams as traditionally in using standard wired and wireless IP data carried on HFC plants and decoded by networks to carry MPEG-2 Transport standard settops. Streams. Transport within the headend and within the home will be covered. IEEE What is Jitter? 802.11a/b/g, HomePlug, 100BaseT Ethernet, Gigabit Ethernet and proprietary IP network For MPEG-2 Transport Streams, jitter is protocols will be covered. The difference the “short-term non-cumulative variations of between carrying Constant Bit Rate and the significant instants of a digital signal from Variable Bit Rate streams will be examined. their ideal positions in time.“[1] New revisions of data networking standards are including provisions for carrying video MPEG 1 MPEG 2 MPEG 3 MPEG 4 MPEG 5 traffic. Will these enhancements be enough? What can we do until they are here? Perfect World

INTRODUCTION MPEG 1 MPEG 2 MPEG 3 MPEG 4 MPEG 5 -1 -2 0+1 MPEG-2 Transport Streams as used in Real World broadcast networks were designed to be Figure 1 – Jitter Timeline delivered as a continuous flow of bits on a HFC (Hybrid Fiber Coax) plant. This is quite For the purposes of this paper, jitter is the different from what you find on a data time difference between when a MPEG-2 network, where bit streams are broken into packet should arrive and when it actually does packets and then sent in discrete bursts. arrive. A packet might arrive early or late. In the Real World Timeline in the figure above, There also is another fundamental packets 2 and 3 arrive late and packet 5 difference between data and video traffic. arrives early. Packets arriving early could Data delivery is focused on accuracy while overflow the MPEG decoder buffer, resulting video delivery is focused on timeliness. in lost information. Packets arriving late Video decoding is designed to conceal errors could result in MPEG decoder buffer while data transfers require perfection. Can underflow, resulting in a packet that was not both of these coexist on the same network? available when it was scheduled for presentation. Although there are other potential issues when carrying MPEG-2 Transport Streams on If you observe these arrival time variations An Alternate Jitter Requirement? over a long period, you can characterize the jitter. The greatest positive arrival time jitter Some people read the MPEG-2 Annex D minus the greatest negative arrival time jitter and get excited because it says “4 is called the "peak to peak jitter amplitude". milliseconds is intended to be the maximum amount of jitter in a well behaved system.” [3] It also may be of interest to see if the What happened to the ± 500 nanoseconds? distribution of the jitter is completely random The jitter they are describing in this Annex is or correlated with other events. Correlated something completely different from the PCR jitter might be due to a protocol that packs jitter we have been discussing. This jitter is multiple MPEG-2 packets into a single called multiplex jitter. Multiplex jitter is transport unit. Random jitter might be due to often caused when packets are moved around random noise that causes packet loss and to create or modify a MPEG-2 MPTS (Multi- therefore packet retransmission. While the Program Transport Stream). After you type of jitter does not matter to the MPEG-2 reposition a MPEG-2 packet, you are still Transport Stream specification, knowing the required to re-timestamp the PCRs in the type helps us select the best way to mitigate packets to within the original PCR jitter jitter. tolerance of ± 500 nanosecond.

MPEG-2 JITTER REQUIREMENTS Jitter Allocation

The MPEG-2 PCR (Program Clock Most currently deployed VOD (Video on Reference) jitter tolerance is ± 500 Demand) systems that use MPEG-2 Transport nanoseconds.[2] The MPEG-2 standard does Streams allocate: not define a jitter budget on a per-device • a small portion of the jitter budget to basis. The standard instead places an absolute the MPEG encoder (no device is limit on packet jitter as received by the perfect!) MPEG-2 decoder. Unfortunately, some • a portion to VOD servers device manufacturers see this number for the first time and conclude that they may add up • a slightly larger portion to the output to ± 500 nanoseconds of jitter. However, the multiplexing and QAM (Quadrature MPEG-2 PCR jitter tolerance is a system level Amplitude Modulation) processing specification, not a device level specification. • a bigger portion to possible in-line remultiplexers/re-encoders that may be By the time a MPEG stream reaches your present (there could be 2-3 of these) home, the devices along the way may have • a negligible amount to the HFC “spent” all but 50 nanoseconds of the MPEG- distribution system 2 jitter budget. And why not? Every device along the distribution chain is trying to Notice that there are quite a few devices in perform its function with the greatest the path which are “spending” from a very efficiency and at the lowest cost. The lowest small jitter budget. cost often results in the maximum allowable jitter. So, as a practical matter, any in-home redistribution system is limited to adding no more than ± 50 nanoseconds of jitter.

NOT ALL MPEG-2 PACKETS ARE EQUAL DATA NETWORKING JITTER

Previously, it was mentioned that packets A data network is normally shared among arriving: many devices and supports different types of traffic. There is no concept of a continuous • early could overflow the MPEG-2 stream of bits. Bits are packed into groups decoder buffer which are then sent in discrete bursts. The • late could underflow the MPEG-2 protocol used by the data network determines decoder buffer when these bursts can be sent.

This is true for all MPEG packets. No widely deployed data networking Because there is some buffering at the MPEG- technology adds less than 50 nanoseconds of 2 decoder, underflow and overflow are jitter when carrying a MPEG-2 Transport infrequent occurrences unless the jitter is Stream. Let me back up this rather sweeping quite severe. However, there is another effect assertion by an example. that is more likely to cause problems replaying a jittered MPEG-2 transport stream Let’s look at the time required to send – but only for certain MPEG-2 packets. MPEG-2 packets over a common 100BaseT

Ethernet. Most systems try to minimize IP MPEG-2 requires a PCR bearing packet at (Internet Protocol) overhead by packing seven least once every 100 milliseconds. This MPEG-2 packets into one IP frame. This is means that most packets will not have a PCR. the maximum number of MPEG-2 packets Excess jitter between PCR bearing packets that will fit into a single IP frame. To send will typically disrupt the MPEG-2 decoder’s one of these MPEG-2 bearing Ethernet frames timing recovery process. The maximum requires: tolerable PCR jitter is a function of the decoder’s PLL (Phase Lock Loop) design. • 12 Bytes for the InterFrame Gap The PLL design affects acquisition time. So, • 8 Bytes for the MAC Preamble there is a tradeoff that every vendor must • 20 Bytes for the IP header make between acquisition time and jitter • 8 Bytes for the UDP header tolerance. Different vendors have made • 1316 Bytes of data (7 MPEG-2 different choices. Packets) • 4 bytes for the CRC (Check Sequence) You may send a MPEG-2 Transport Stream with more than ± 500 nanoseconds of Total = 1368 Bytes PCR bearing packet jitter to a settop and it may play without errors. Some decoders will At 100 Mbps, each bit takes 10 operate properly when fed a MPEG-2 nanoseconds. Therefore: Transport Stream that does not meet all of the MPEG-2 specifications. However, you must • 1368 bytes * 8 bits/byte = 10,944 bits remember that ± 500 nanoseconds is the • 10,944 bits * 10 ns/bit = 109,440 ns MPEG-2 requirement. This is all any MPEG- 2 decoder is required to support and this needs to be your design requirement.

As you can see, the jitter caused by level QoS. While HomePlug 2.0 does not packing these MPEG packets together is far support entertainment level QoS, HomePlug more than ± 50 nanoseconds. Fifty AV (the next version of HomePlug) will nanoseconds represents only 5 bit times! support entertainment level QoS.

Most other widely deployed Ethernet So, will 802.11e or any of the other QoS based systems, such as IEEE 802.11a/b/g and initiatives meet the MPEG-2 Transport 10BaseT are slower and some of these have Stream requirements? No. In all fairness, even more overhead. Step up to Gigabit how could they? As shown in the previous Ethernet and the numbers get 10 times better example, the physical transport network just than 100BaseT, but still far from ± 50 does not have the timing resolution required nanoseconds. Even at Gigabit Ethernet to meet a ± 50 nanosecond jitter specification. speeds, one bit takes one nanosecond, so 50 nanoseconds represents 50 bit times. The But I See it Work All the Time! InterFrame Gap required by the Ethernet protocol between every transmission is 96 When a protocol claims to support bits, which equals 96 nanoseconds! entertainment quality video, they usually are referring to “Streaming Media”. Streaming In addition, many data networks provide Media is a different kind of MPEG-2 delivery reliable service by continuing to retransmit a system. The big difference is in Streaming packet until it is successfully received. This Media, the decoder works from a “pull” is the correct approach to use for delivering model. In a pull model, the decoder requests data but makes things worse for MPEG-2 blocks of MPEG-2 packets as soon as it has Transport Streams. If video data is not the buffer space to store them. Streaming delivered by its presentation time, it is better Media decoders usually have a much larger to skip this packet and move on to the next. buffer, so jitter is less of a problem. In But if a packet contains a PCR, discarding the addition, they also run their display clock packet may mean there is no PCR within the from a local source so that they don’t have to 100 milliseconds required by the MPEG-2 worry about timing recovery. specification. ATM (Asynchronous Transfer Mode) has an encapsulation protocol with a In contrast, a MPEG-2 broadcast decoder concept of “PCR Aware”, but I have not seen is at the end of a fire hose of bits that keep this proposed for any other system. coming at it, whether it is ready or not! A broadcast receiver may need to either drop a QUALITY OF SERVICE frame when its buffer overruns or repeat a frame when its buffer underruns. This is how IEEE 802.11e is currently being developed the broadcast model can use one source to as a standard QoS (Quality of Service) feed millions of decoders. In the Streaming mechanism for wireless systems. It promises Media model, every decoder is making its to provide a QoS which meets entertainment own requests. So, if you replaced a broadcast video requirements. IEEE 802.11p provides a system that served millions with a Streaming prioritized QoS for wired LANs and Media system, it would require a video server HomePNA 2.0 includes a prioritized QoS, but farm that supported millions of sessions. these schemes do not support entertainment RECOVERING FROM JITTER MPEG-2 packets to present the content at the specified quality – and no more. So, while a Archimedes said that he could lift the earth Variable Bit Rate stream may be specified as if you gave him a firm enough fulcrum and a having an “average” bit rate of 3 Mbps, there long enough lever. Similarly, given a large may be intervals of time where the short term enough buffer and a large enough delay bit rate peaks at 9 Mbps – or more! The allowance, you can remove any amount of MPEG-2 specification allows standard jitter. However, if you have a large jitter definition rates as high as 15 Mbps. buffer, you must wait for the buffer to fill before you can view the video. This requires Most MPTS (Multi-Program Transport a patient viewer. Infinitely long levers and Streams) delivered from a satellite feed are a infinitely patient viewers are equally unlikely. constant 27 Mbps, but the individual program streams that it contains are Variable Bit Rate. Most current PC Streaming Media players If you separate out a single program stream use this buffering approach. They normally from a MPTS and then send it over a data buffer 5-10 seconds of video prior to playing network, you must be prepared to serve the the first frame. However, entertainment video stream at its peak rate, not its average rate. is often interactive, so “solving” the jitter problem with a large buffer at the receiver What is the difference between Constant will result in a system that seems sluggish. Bit Rate and Variable Bit Rate streams when Channel changes would become quite it comes to de-jittering? With a Constant Bit lengthy. Again, this sluggish response is the Rate stream, you can buffer the incoming typical experience we have come to accept packets and then play them back at the same with streaming media on a PC. rate as the rate at which they were encoded. With Variable Bit Rate, you never quite know Even with a large buffer, video glitches are when to send them to the MPEG decoder. In common today in the streaming environment. this case, you need to time stamp every The proposed QoS initiatives will take care of MPEG-2 packet when it arrives in the these problems for Streaming Media, but network so you will know when to play it out. won’t help our case of broadcast style MPEG- This adds quite a bit of complexity to the 2 Transport Streams. protocol plus the added timestamp adds transmission overhead to every packet To use a standard data network to deliver processed. MPEG-2 Transport Streams, you need to use whatever network features are available (such PROPRIETARY SOLUTIONS as QoS) to minimize the jitter added by the network. Then you will need to add de- Two proposed architectures, one wired and jittering equipment at the edge of the data the other wireless, claim to have solved the network. MPEG-2 Transport Stream jitter problem. Air5 is the wireless solution and MoCA Constant and Variable Bit Rates (Multimedia over Coax Alliance) [4] is the wired solution. Both these solutions are chip There are two kinds of MPEG-2 Transport sets that accept and deliver MPEG-2 streams. Constant Bit Rate streams are, as Transport Streams. How do they do it? Both their name suggests, a steady stream of use a network protocol that guarantees there MPEG-2 packets delivered at an steady rate. will be no collisions plus they have some de- Variable Bit Rate streams deliver just enough jittering built into the receiving chip. Unfortunately, as of the time of this [1] American National Standard Dictionary writing the Air5 developer, Magis of Information Technology (ANSDIT), Networks, has suspended operations and its available at future is uncertain. However, MoCA is going http://www.ncits.org/tc_home/k5htm/Ansdit.htm strong and MoCA products are expected to reach the market by the end of 2004. [2] ISO/IEC 13818-1 Generic Coding of Moving Pictures and Associated Audio: CONCLUSION Systems, 13 Nov 1994, section 2.4.2.2.

Traditional broadcast MPEG-2 Transport [3] ISO/IEC 13818-1 Generic Coding of Streams are not easily carried on a traditional Moving Pictures and Associated Audio: data network. The broadcast MPEG-2 Systems, 13 Nov 1994, Annex D.0.4 Transport Stream model is based on an SCR and PCR Jitter. unstoppable stream of bits directed at potentially millions of decoders. Strict timing [4] More information on MoCA is available rules are required to make this broadcast at www.mocalliance.org system work. If you follow the rules, the system works beautifully.

Streaming Media MPEG-2 systems were designed to be carried on traditional data networks. Therefore they work just fine on current data networks and the coming QoS initiatives will allow them to work beautifully. However, these systems put quite a load on the MPEG-2 source and the transport system because each decoder has a private session with the ongoing video source.

If you must carry broadcast MPEG-2 Transport Streams on a data network, use the best QoS that you can get. Then add a de- jittering device either at the edge of your data network or as a part of your MPEG-2 decoder.

OPTIMAL AVAILABILITY & SECURITY FOR VOICE OVER CABLE NETWORKS

Chun K. Chan, Andrew R. McGee, Martin J. Glapa, and Uma Chandrashekhar Bell Laboratories, Lucent Technologies

Abstract Now powerful, intelligent devices under end- user control are potentially able to access The Lucent Bell Labs Security model has signaling and network management now become the foundation of the newly information. Hobbyists have hacked into ratified ITU-T Recommendation X.805 cable modem hardware, tricking it to accept "Security Architecture for Systems Providing custom code. As a result, the end user has End-to-End Communications." The X.805 complete control of the cable modem and can standard was developed as the framework for surmount the bandwidth imposed by the the architecture and dimensions in achieving service provider [1]. In Voice over Cable end-to-end security of distributed network, it is conceivable that hackers can applications. In this paper, we introduce the also modify and gain control of the MTA. X.805 standard and describe how it can be Rolling out VoIP services over cable applied to the PacketCable™ Security automatically inherits some of the Specification (PKT-SP-SEC-107-021127) and vulnerabilities associated with VoIP. A recent the DOCSIS BPI+ specification (SP-BPI+- CERT advisory states that a number of I10-030730) for Voice over Cable (VoC) vulnerabilities have been discovered in networks. We identify areas of conformance various implementations of the H.323 and gaps that the current PacketCable protocol [2] [3]. Even though Voice over standards have with respect to the X.805 Cable does not use H.323, we expect similar Security Model and examine the effect on end- vulnerable implementations of relatively new to-end availability of VoC networks. The protocols. PacketCable™ reliability models (PKT-TR- VoIPAR-V01-001128) are generalized to Given that a Voice over Cable network is include downtime due to security vulnerable, cable operators can benefit from a vulnerabilities and attacks. Our analysis comprehensive, end-to-end security shows that the traditional reliability models framework to guide their network planning produce results that are optimistic if we do and the ongoing security assessments not consider both availability and security performed against their networks. The within a network dependability framework. recently ratified ITU-T Recommendation The X.805 standard can be used to augment X.805 "Security Architecture for Systems these models to provide optimal availability Providing End-to-End Communications" [4] and security for Voice over Cable networks. was developed to provide such a framework.

INTRODUCTION VOICE OVER CABLE NETWORK

The advent of next generation IP networks Figure 1 represents a simplified functional carrying converged voice and data traffic view of the PacketCable™ based VoIP obliterates the inherent, built-in security of network architecture. VoIP builds on a traditional telecommunications networks with DOCSIS high-speed data infrastructure. The their unintelligent end-user devices and out- CMTS (Cable Modem Termination System) of-band signaling and management networks. provides DOCSIS IP connectivity on the RF based cable network with a managed data across multiple MSOs. The network can be network or the Internet. For VoIP, DOCSIS distributed where softswitches and gateways is used to transport IP packets containing and other elements are scattered across signaling and bearer (voice) packets between multiple locations, or centralized where the an MTA device at the subscriber location and softswitch, gateways and other elements are various network elements. The MTA collocated together. Vendor products can provides the telephony implement single discrete functions of the termination/origination point. The MTA can reference architecture or can integrate several either be embedded with a cable modem (E- functions into a single product. MTA) or a standalone device that connects to a cable modem (S-MTA). E-MTAs can take Given all these subscriber, network and advantage of a Dynamic Quality of Service product variables, coupled with vulnerabilities (DQoS) feature to provide priority for voice mentioned earlier, security/reliability in this traffic, whereas an S-MTA cannot. network presents challenges to be addressed by a comprehensive, end-to-end security PacketCable Reference Architecture - Functional framework such as ITU-T Recommendation

SS7 Net wor k PSTN X.805. Ca ll Management Provisioning Server Media GW Server Controller Signaling Domain Name GW Media GW Server ITU-T RECOMMENDATION X.805 Managed IP Net work DHCP Server Media OSS TFTP Servers Server

SNMP Server The advent of next generation cable IP

CMTS CMTS Record Keeping Server networks carrying converged voice and data HFC HFC SYSLOG Net wor k/ Net wor k/ Server DOCSIS DOCSIS Key Distribution traffic obliterates the inherent, built-in Center CM/MTA CM/MTA • security provided by traditional • • telecommunications networks. In modern IP/cable networks we now have the situation Figure 1: Functional PacketCableTM where numerous, powerful, intelligent end- Reference Architecture user devices that can be used to launch network attacks are attached to cable A Call Management Server (CMS) networks. The signaling/control and provides subscriber side call processing management information is carried in-band functions such as origination, tear-down and with user information thereby making it Class 5 switch features. A Media Gateway susceptible to attack as well. Network Controller (MGC) is a softswitch function that operations or management security is often provides PSTN or trunk side call processing neglected when MSO network security is control of the Media Gateway (MG). The being considered and frequently provides a MG provides circuit-packet conversions for back-door entry into MSO networks. Since connectivity to the PSTN. Other various the insider threat represents a potential for servers provide SS7 interfaces, significant financial loss, this situation is a announcements, voice mail and back office recipe for disaster. The X.805 Security systems. Architecture provides a structured framework that forces the consideration of all these This type of network can be implemented factors to provide comprehensive, end-to-end in a single cable serving area or it can be network security. implemented to span multiple cable serving areas across geographic regions as well as The X.805 Security Architecture defines • Corruption or modification of the framework for the architecture and information, dimensions in achieving end-to-end security • Removal, theft, or loss of information of distributed applications. The general and/or other resources, principles and definitions apply to all • Disclosure of information, and applications, even though details such as • Interruption of services. threats and vulnerabilities and the measures to counter or prevent them vary based upon the Figure 2 depicts the X.805 Security needs of the application [5]. How each Architecture, which provides a systematic standard fits together in the end-to-end way of countering these five threats for large, security picture emanates from X.805. ITU-T complex networks such as today's MSO Recommendation X.805 also forms the networks. X.805 provides a comprehensive, foundation for the proposed ISO/IEC 18028 multi-layered, end-to-end view of network standard "Information technology - Security security across eight security dimensions. techniques - Network Security - Part 2: The X.805 standard defines a hierarchy of Network Security Architecture," which has network equipment and facility groupings into recently completed Committee Draft balloting three Security Layers: (1) the Infrastructure in preparation to becoming an international Security Layer, (2) the Services Security standard. Layer, and (3) the Applications Security Layer. We provide a brief description of the X.805 Security Architecture before • The Infrastructure Security Layer consists demonstrating how it can be applied to Voice of the basic building blocks used to build over Cable networks. telecommunications networks, services and applications, and consists of Security Layers Applications Security individual communication links and THREATS Destruction network elements including their Corruption Services Security VULNERABILITIES Removal underlying hardware and software

Privacy Disclosure Availability Data Integrity

Authentication platforms. Examples include the cable Access Control Non-repudiation Interruption Infrastructure Security Confidentiality Data Communication Security ATTACKS modem, CMTS, CMS, Signaling/Media

End User Plane Gateways, Media Gateway Controllers, Security Control Plane 8 Security Dimensions Planes Management Plane and Media Servers depicted in Figure 1. • The Services Security Layer consists of Figure 2: ITU-T Recommendation X.805 services that customers/end-users receive "Security Architecture for Systems from networks. Example services range Providing End-to End Communications" from basic connectivity and transport (e.g., Internet access) to service enablers The X.805 Security Architecture was (e.g., authentication, authorization, and developed as part of the ITU-T X.800 series accounting – AAA services) to value- of recommendations [6] to provide a added services such as VPN, VoIP, and methodical, organized way of addressing the Voice over Cable services. five threats to telecommunications networks. • The Applications Security Layer focuses The X.800 series identifies these threats as: on network-based applications that are • Destruction of information and/or other accessed by customers/end-users. These resources, applications are enabled by network services and are characterized by the end- The eight Security Dimensions contained user interacting with remote hardware or in recommendation X.805 represent classes of software in order to access information or actions that can be taken, or technologies that perform a transaction. Example network- can be deployed, to counter the unique threats based applications include basic and potential attacks present at each Security applications such as file transport (e.g., Layer and Plane: FTP) and web browsing, fundamental • Access Control is concerned with applications such as directory assistance providing authorized access to network (e.g., 411) and e-mail, as well as high-end resources. applications such as e-commerce, • Authentication is concerned with network-based training, and video confirming the identity of communicating collaboration. parties. • Non-repudiation is concerned with These Security Layers provide maintaining an audit trail, so that the comprehensive, end-to-end security solutions origin of data or the cause of an event or and identify where security must be addressed action cannot be denied. in products and solutions because each layer • Data Confidentiality is concerned with may be exposed to different types of threats protecting data from unauthorized and attacks. For example, a Denial of Service disclosure. (DoS) attack can be performed at the • Communication Security is concerned Infrastructure Layer by flooding a router's with ensuring that information only flows physical port with bogus packets, thus between authorized end-points without preventing or impeding the transmission of being diverted or intercepted. legitimate traffic. A DoS attack can also be • Data Integrity is concerned with performed at the Services or Applications maintaining the correctness or accuracy of Layer by deleting user account information, data and protecting against unauthorized thus preventing legitimate users from modification, deletion, creation, and accessing the service or application. One can replication. readily see that components of Infrastructure • Availability is concerned with ensuring Security, Services Security, and Applications that there is no denial of authorized access Security must be addressed in order to provide to network elements, stored information, a comprehensive, end-to-end network security information flows, services, and solution, and that different counter-measures applications. must be applied at each Security Layer. • Privacy is concerned with protecting

information that might be derived from Three types of activities are performed on the observation of network activities. any network, which are represented by the three Security Planes: (1) the Management Table 1 indicates how the Security Plane, (2) the Control Plane, and (3) the End- Dimensions relate to the X.800 threats User Plane. Different security vulnerabilities described previously; the cells marked with may exist in each of these planes – in fact 'Y' indicate the Security Dimensions that are each of these planes might be implemented by applicable to each of the five threats. In separate networks for a given network or particular, through this mapping, we can service architecture. Each Security Plane begin to identify the right security along with the three layers must be secured in mechanisms needed to thwart potential order to provide an effective security posture. threats. APPLYING X.805 TO VOICE OVER CABLE

CableLabs has developed a security specification for providing security to VoIP communications over the PacketCable reference architecture described above. The PacketCable Security Specification [7] was defined to provide confidentiality to user information flows (voice and data) across the PacketCable network and to protect cable Table 1: Applying Security Dimensions to MSOs against theft of service. The Security Threats PacketCable Security Specification defines The X.805 Security Architecture can also be the security architecture, protocols, addressed in a modular form, as illustrated in algorithms, functional requirements and Figure 3, to provide a systematic, methodical technological requirements that force any user approach to network security. Figure 3 shows with the intent to steal or disrupt network the intersection of a Security Layer with a services to spend an unreasonable amount of Security Plane. This represents a unique money or time to do so. perspective for consideration of the eight

Security Dimensions and can be considered a MSO TFTP PROV Remote OSS KDC SERVER SERV Telephony component, or module, of end-to-end network OSS KDC CR PK 14: security. Each of the nine modules in Figure 3 pkt-s

) IT ts TELEPHONY

S e

N k

y M KDC t c P i KI

C r P Ti

combines the eight Security Dimensions that are ( M : u

2 s

o Tickets

N c MPC P 1 r e

e pkt-s17:

S s T

- Kerberos t b

: S r F

k e Issue Kerberos Issue

T 0 3 p T K I ts v

applied to each security perspective. The : s e IN ke - 1 P u K c

t s P Ti s : k M s

- I 3 os t 1 r p

- N -s e k kt rb EBP E S e S

p p S K Tickets

Security Dimensions of different modules have K e CM I

u pkt-s17: :

s Kerberos 6 / s 1 s I t-s ero

Issue Kerberos Issue k

c p erb s

e K o / S

c r

S e S different objectives and consequently comprise PS e

P I

b

I

M

r

pkt-s20: TGCP pkt-s20: e pkt-s5: NCS C

CMS CMS :

K

MTA 6

IPSec / Kerberos + PKINIT /

p 1

different comprehensive sets of security kt- c

s s 1 e

us IPSe 6 - i : t

ad C S

R s c M k

6: / SS P I s pk Ke p kt- E- -) g M rb MP p IK s K t e measures. The tabular form gives a convenient c / E R -s ros Se K M d S 1 RKS IP I e V 8 / n s P : M c o a I In ti n t CMS K e b t S a - e S g e way of describing the objectives of the Security n S r r d IP i it v pk ( M y + e ) d O R s r C M I S b P t o / S - P K je V ) r KM s ba S o c O o P - 7 C t t M e d d Dimensions for each module. : C S a c P e e : K R c - te i M 8 s s / t D a a E d s n a a I

C - pk U K d K e t G Edge e

b b I I : / i k - - s E : h

P us t / t 1 p 5

S a - - u Router

S C

1 2 c s

b a e

s e

M MS - s A 2

- s -

c P

t +

t S C C S 5

u I

M i

k /

/ : Infrastructure Applications / P E

/ M d

H p I

P R

I pk

K

Services Layer c a K

C I

t

+ C a A

e

Layer Layer R /

/ E

c

d

+ r C

S A P

c e -

c

i

e j

T v

V

P u e

M

e r

I : h b

( s S

CMTS S e

H 9 p S

i P

Management O s

R

S P I

P

+ -

Module One Module Four Module Seven y I f

C t ( f

( C p d r it

k i Plane k r

T e t e

P p

P - t g s D h

R I 2 T e

C P a : 2 t : p S

i g : 9

R T 2 e n : U

c I e

: 1 1 C R D

4 ( r 1 /

Module Two Module Five Module Eight 3 P s

I

:

Control Plane P

s . K -

s - g J t : E P M

- t V

- g t

T 4 k

k K S

k : I s A

p p

R 2

- P R p t s : - B k DF 3 t /

Module Three Module Six Module Nine p k p User Plane s + k - p I t t -s P 2 k 4 B I : p PS U e DP Edge c / P IK Router :UD E Remote CM 23 + Access Control Communication Security t-s E- DF pk / IK MTA ec IPS Authentication Data Integrity

MGMG pkt-s10: TGCP MGC pkt-s11: ISTP SG Non-repudiation Availability IPSec / IKE- IPSec / IKE- Data Confidentiality Privacy Eight Security Dimensions Figure 4: PacketCable Security Architecture Figure 3: Modular form of X.805 Security Architecture

Figure 4 depicts the security architecture achieved by bridging unprotected user specified by the PacketCable Security networks (e.g., WiFi networks) into the Specification, which provides device PacketCable environment. authentication and authorization as well as Protecting the MSO network against encryption for the PacketCable network. The Denial of Service attacks is another area PacketCable Security Specification relies on where X.805 augments the PacketCable DOCSIS 1.1 [8] and BPI+ [9] to secure the Security Specification. Denial of Service information flow across the HFC portion of attacks are attacks on the Availability security the PacketCable network. The PacketCable dimension and are probably the most widely Security Specification extends encryption to publicized type of security vulnerability. the Call Management Server (CMS), Media Denial of Service can be achieved in many Gateway (MG), Signaling Gateway (SG), and different ways including: (1) an unauthorized remote Multimedia Terminal Adapter (MTA). user logging in as a system administrator causing a critical network element to crash, In order to provide confidentiality for user (2) deletion of user account information information flows and protect against theft of thereby preventing authorized users from service, the PacketCable Security accessing a service, (3) flood attacks like Specification contains detailed requirements "smurf" that consume network resources to for encryption algorithms as well as the point that no one can access it, (4) viruses authentication algorithms to be used in the and worms such as "Code Red" and PacketCable network. In summary, the "NIMDA" that exploit system vulnerabilities PacketCable Security Specification provides: to gain access to vulnerable machines and • Confidentiality of the user voice/data then propagate themselves to other vulnerable streams across the PacketCable network, hosts, which also results in the consumption • Secure bearer, signaling and management of network resources. channels across the PacketCable network, • Protection against CPE The emphasis placed on Access Control cloning/tampering/uncapping, and Authentication by the PacketCable • Protection against identity theft. Security Specification protects against Denial of Service attacks accomplished via The X.805 security framework can be used unauthorized access to network elements, to augment the Packet Cable Security with the exception of End-User devices, Specification to provide comprehensive end- which are considered out of scope. to-end security by including additional portions of the PacketCable network Cable MSOs can use the X.805 security architecture and the X.805 Security architecture to identify mechanisms that can Dimensions, Layers and Planes. When the be used to augment the PacketCable Security scope of the security analysis is extended to Specification to protect against the additional include the end-to-end PacketCable network types of DoS attacks. As evidenced by the architecture and the entire X.805 security recent Code Red attack's ability to cripple framework, vulnerabilities are identified in CMTS devices throughout the world [10], every layer, plane and dimension, even when every PacketCable network element (CMTS, the analysis is limited to the transport of MG, SG, CMS) as well as the Operations packetized voice across the MSO network. Support System servers, the back-office For example, cable MSO networks must be servers, etc. are potentially vulnerable to flood protected against unauthorized access attacks and network worms. X.805 also indicates that cable MSOs must also develop Privacy Security Dimension mechanisms to address Denial of Service X.805 Security X.805 Security Layer attacks achieved by attacking the user Plane Infrastructure Services information, etc., which is critical to the End-User Incomplete Incomplete Voice over Cable service. Control Not Applicable Missing Management Missing Missing Availability Security Dimension Table 3: PacketCable Coverage for X.805 Security X.805 Security Layer Privacy Security Dimension Plane Infrastructure Services

End-User Not Applicable Missing This section has provided some key results Control Not Applicable Missing Management Incomplete Missing of applying X.805 to a portion of the PacketCable network architecture in order to Table 2: PacketCable Coverage for demonstrate how the PacketCable Security Availability Security Dimension Specification can be augmented to achieve optimum security for the Voice over Cable The Privacy Security Dimension is another service. A complete analysis of the end-to- example of how X.805 augments the end Voice over Cable network architecture PacketCable Security Specification to provide utilizing X.805 has produced comparable comprehensive end-to-end security. This results for the remainder of the VoC network dimension is concerned with protecting architecture and remaining X.805 Security information about activities that take place on Layers, Planes and Dimensions. the network. For the Voice over Cable service, the source and destination of a IMPACT OF SECURITY communication flow would be an example of VULNERABILITIES & ATTACKS ON this type of information. For example, it may END-TO-END AVAILABILITY be important to protect the fact that two parties are communicating with each other The X.805 analysis in the preceding over and above the actual contents of the section briefly addresses the key security communication. The PacketCable Security challenges that need to be addressed when Specification utilizes the IPSec ESP protocol looking at the eight dimensions of network in transport mode [11], which does not security. When we look at supporting key encrypt the original IP packet header. applications on a cable infrastructure such as Therefore, the Privacy Security Dimension is voice we need to think about availability. In not addressed by IPSec ESP transport mode this section, we focus on the availability per se. The NCS messages that contain dialed dimension. The standard reference for Voice numbers and other customer information are over Cable availability is the Cable Labs carried as IP packet payloads which are specification, PKT-TR-VoIPAR-V01-001128 encrypted via IPSec ESP transport mode; [12]. We begin with a brief review of the however, once the call is established, the IP PacketCable™ VoIP availability allocation addresses of the communicating end-points process. are visible. The tunnel mode for the IPSec ESP protocol would provide complete The PacketCable™ VoIP models allocate coverage of the Privacy Security Dimension availability budgets to network elements so for these messages. that the end-to-end availability of a voice- over-packet network is the same as that of the PSTN. For instance, the MTA is allocated an availability budget of 99.9975%, which is Active CMTS equivalent to an average annual downtime of 13 minutes. An MTA is expected to be built Standby CMTS to this specification using standard reliability engineering practices such as thermal Figure 5: Active-Standby CMTS System management and component derating [13]. If all the network elements in the PacketCable™ based VoIP network architecture are built to meet their respective availability budgets, Active CMTS then the end-to-end VoIP availability is Denial of Service expected to be the same as that of the PSTN. Standby CMTS Even if all these budgets are met, we argue that it is unlikely that the end-to-end goal of Figure 6: Denial of Service as a Common PSTN availability will be met because the Mode Failure PacketCable™ VoIP availability/reliability models do not include downtime due to To estimate the downtime due to DoS, we security vulnerabilities and threats. Theft of use the measurements from [15]. According services such as MTA tampering does not to [15], 12,805 attacks were observed in one have a direct impact of end-to-end week on 224 Internet protocol (IP) addresses. availability, so such vulnerabilities and threats Since these 224 addresses represent a are not modeled here. theoretical maximum, we assume the number of active IP addresses to be two orders of Denial-of-service attacks such as Code Red magnitude below 224. Then, the mean attack and NIMDA have brought down CMTS rate per IP address is estimated to be 5×10-4 devices [10]. According to PKT-TR- per hour. This implies an attack frequency of VoIPAR-V01-001128, the CMTS is allocated about 5 per year. Combining this with the a downtime of 10 minutes per year. To meet average attack duration of 10 minutes [15], this downtime budget requires redundant this simple model shows that DoS adds an hardware, which is represented by a generic annual downtime of 50 minutes to the CMTS parallel system shown in Figure 5. This system. Figure 7 shows simplified Markov redundant system is fault tolerant with respect models to illustrate the impact of DoS on to hardware and software faults; when the system downtime. The state transition active fails, the standby takes over. However, diagram on the left shows the scenario with if security vulnerability is present, it will be in no DoS. The system fails where both the both the active and standby software. A active and standby units fail at the same time denial of service attack will bring down both (duplex failure) due to hardware and/or the active and the standby subsystems. This software faults. The duplex failure rate is λ common-mode failure [14] is pictorially and the system restoration rate is µ. For a represented by a DoS block in series with the CMTS system that is compliant with the redundant system, as shown in Figure 6. CableLabs specification, the system spends less than 10 minutes per year in the duplex failure state. If we include DoS, then there will be an additional failed state labeled as the analysis, all PacketCable network elements DoS state in the state transition diagram on (CMTS, MG, SG, CMS) as well as the right in Figure 7. It is seen the system is Operations Support System servers, back- expected to spend 50 minutes/year in the DoS office servers, etc. are potentially vulnerable state. The total system downtime is 60 to DoS attacks. In the remainder of this minutes/year. section, we show the impact on end-to-end availability if we add a DoS downtime of 100 Duplex Duplex minutes per year to the CMTS, MG, SG and Failure Failure CMS in the call path. (10 min/yr) (10 min/yr) Figure 8 shows the reliability block diagram λµ λµ for a local on-net call between two DOS subscribers served by the same Call (50 min/yr) Management Server (CMS). Each block µDoS Normal Normal contains the unavailability budget given in the λ DoS Cable Labs specification. If all the network elements meet their respective unavailability Figure 7: Simplified Markov Models Showing the Impact of DoS on System Downtime or downtime budgets, then the end-to-end availability is 99.97%, which is the same as A more detailed Markov model that takes its PSTN counterpart of 150 minutes of into account of attack frequency, security downtime per year. vulnerability arrival rate, security vulnerability removal rate, and system restoration rate is given in [16]. The detailed model in [16] is applicable to a softswitch CM/MTA HFC CMTS/ER CMS HFC CM/MTA (such as an MGC) that uses an off-the-shelf 0.0025% 0.01% 0.0025% 0.001% 0.01% 0.0025% server cluster for fault tolerance. Compared to a traditional circuit switch, which uses proprietary hardware for fault tolerance, a End-to-End Availability = 99.97% End-to-End Downtime = 150 min/yr softswitch relies mainly on software for fault tolerance. As a result, a softswitch is more Figure 8: Local On-Net Single-Zone Availability prone to security attacks [17] because it is virtually impossible to have bug-free If we include unavailability due to DoS, software. An attacker could exploit well- then the end-to-end availability is likely to known OS vulnerabilities to gain control of a degrade from 99.97% to 99.935% (Figure 9) softswitch. When the softswitch is in a resulting in a downtime of 340 minutes/year. compromised state, the attacker could erase It should be noted that unavailability due to critical system files so that a system re- CM/MTA power outage is not included in the installation is needed, then the mean time to end-to-end calculation. If we include the restore service could be hours. Based on the downtime due to power outages [18], the detailed Markov model in [16], we estimate degradation in end-to-end availability due to that the downtime due to this type of DoS is DoS is less pronounced (from 99.88% to of the order of 100 minutes per year. This is 99.84%), because power outage alone about two orders of magnitude higher than the contributes 240 minutes of downtime per year downtime allocated by the Cable Labs to each of the CM/MTA at both ends. specification. As pointed out in the X.805

CM/MTA HFC CMTS/ER CMS Local IP 0.0025% 0.01% 0.02% 0.02% 0.024% CM/MTA HFC CMTS/ER CMS HFC CM/MTA 0.0025% 0.01% 0.02% 0.02% 0.01% 0.0025% SG MGC CMS 0.02% 0.02% 0.02%

End-to-End Availability = 99.935% PSTN Gateway Circuit Switch Copper Loop End-to-End Downtime = 340 min/yr 0.02% 0.005% 0.01%

End-to-End Availability = 99.83% Figure 9: Local On-Net Single-Zone End-to-End Downtime = 900 min/yr Availability with DoS Figure 11: Local Off-Net Availability with DoS The impact of DoS is larger if we consider an off-net call path with more network The following table summarizes our results. elements that are vulnerable to DoS attacks. End to End Availability (Annual Downtime) An off-net call is defined as a call between an Scenario Baseline Baseline + Baseline + Baseline + DoS Power DoS + endpoint on a PacketCable network and an Outage Power endpoint on the PSTN. An example is given Outage Local 99.97% 99.935% 99.88% 99.84% in Figure 10. This scenario shows a call where On-Net (150 min) (340 min) (630 min) (840 min) the calling and called parties are served by a Local 99.96% 99.83% 99.91% 99.78% CMS and a Class 5 switch with a baseline of Off-Net (215 min) (900 min) (470 min) (1160 min) 215 minutes of downtime per year in its Table 4: Summary of the End-to-End Availability normal state (without DoS attack). (Downtime) Calculations The baseline calculations in Table 4 are based on Cable Labs allocations [12]. From the end user’s perspective, these numbers CM/MTA HFC CMTS/ER CMS Local IP 0.0025% 0.01% 0.0025% 0.001% 0.005% appear optimistic because they do not include downtimes due to DoS and power outages. SG MGC CMS 0.001% 0.001% 0.001% For a local on-net call, the impact of power outages is larger than that of DoS, because PSTN Gateway Circuit Switch Copper Loop 0.002% 0.005% 0.01% there are two CM/MTAs in the call path and they are both affected by power outages, End-to-End Availability = 99.96% whereas DoS only impacts the CMTS and the End-to-End Downtime = 215 min/yr CMS. For a local off-net call, the impact of Figure 10: Local Off-Net Availability DoS is larger than that of power outages, If we include unavailability due to DoS, because there is only one CM/MTA in the call then the end-to-end availability is likely to path and there are many network elements degrade from 99.96% to 99.83% (Figure 11). If that are vulnerable to DoS. we also include the impact of CM/MTA power outage, the degradation in end-to-end Since we expect that there are more off-net availability is from 99.91% to 99.78%. The calls than on-net calls, the overall (weighted) impact of CM/MTA power outage is less than impact of DoS is larger than that of power that of the on-net scenario because we have outages. Whereas extended power outages are only one CM/MTA in the call path, and the events that are out of the Cable MSOs’ control, call path has more vulnerable network the impact of DoS can be mitigated by elements. implementing software reliability engineering practices [19]. Following the downtime allocation process in [12], the MSOs should using a continuous, systematic, methodical, work with their equipment vendors to allocate end-to-end approach that has been missing downtime budgets for software in addition to until now. ITU-T Recommendation X.805 the existing hardware budgets. This will in turn provides such an approach by providing a drive the equipment vendors to improve their comprehensive, end-to-end, multi-layered software development process to reduce the view of network security across eight security number of bugs and patches. dimensions.

CONCLUSION ACRONYMS

In this paper, we introduced the X.805 AAA Authentication, Authorization & standard and showed how the current Accounting PacketCable™ Security Specification (PKT- BPI+ Baseline Privacy Plus Interface SP-SEC-107-021127) and the DOCSIS BPI+ CM Cable Modem specification (SP-BPI+-I10-030730) for CMS Call Management Server Voice over Cable (VoC) networks alone do CMTS Cable Modem Termination System not address end-to-end network security in a CPE Customer Premise Equipment manner that allows cable operators to have a DOCSI Data over Cable Service Interface S Specification secure and reliable network. From the DQoS Dynamic Quality of Service examples, we noted that in order to support DoS Denial of Service VoC and other value added services, the cable E-MTA Embedded Multimedia Terminal operators need to have their cable network Adapter designed, maintained, and able to support an ER Edge Router ongoing security program with controls to ESP Encapsulating Security Payload prevent, detect, and correct vulnerabilities FTP File Transfer Protocol resulting in maximum availability for the end- HFC Hybrid Fiber Coax users. In particular, the controls should IP Internet Protocol address the gaps noted in the security IPSec Internet Protocol Security dimensions - non-repudiation, privacy, MG Media Gateway communication security, data integrity, data MGC Media Gateway Controller confidentiality, access control, availability, MSO Multi-System Operator and authentication. By implementing these MTA Multimedia Terminal Adapter changes, and updating the models, the cable NCS Network Call Signaling network can be designed to support VoC and OS Operating System the next generation of services for the end- PSTN Public Switched Telephone Network user. RF Radio Frequency S-MTA Standalone Multimedia Terminal The global cost of cyber-attacks is Adapter estimated to be in the $145 billion range for SG Signaling Gateway 2003 alone, with 2003 also being regarded as SS7 Signaling System 7 the "worst year ever" for viruses and worms. VoC Voice over Cable Unfortunately, there is no end in sight to the VoIP Voice over IP continued onslaught of threats to network VPN Virtual Private Network security. Clearly in today's environment, network security can no longer be treated as an afterthought and must be implemented REFERENCES [10] J. T. McKelvey, “Combating Security Risks on the Cable IP Network,” IBC [1] K. Poulsen, “Cable Modem Hackers 2002 Conference, Conquer the Coax,” SecurityFocus, http://www.broadcastpapers.com/ February 5, 2004. ibc2002/ibc2002.htm [2] http://www.kb.cert.org/vuls/id/749342 [11] S. Kent and R. Atkinson, “Security [3] http://www.cisco.com/warp/public/707/ Architecture for the Internet Protocol,” cisco-sa-20040113-h323.shtml IETF RFC 2401, Internet Engineering [4] International Telecommunications Union, Task Force, Nov. 1998, Telecommunications Standardization http://www.ietf.org/rfc/rfc2401.txt. Sector, "Security Architecture for Systems [12] Cable Television Laboratories, “VoIP Providing End-to-End Communications," Availability and Reliability Model for ITU-T Rec. X.805, 2003, the PacketCableTM Architecture,” PKT- http://www.itu.int. TR-VoIPAR-V01-001128, 2000. [5] International Telecommunications Union, [13] P. D. T. O’Connor, Practical Reliability Telecommunications Standardization Engineering, John Wiley & Sons, 1985. Sector, "Security in Telecommunications [14] R. Billinton and R. N. Allan, Reliability and Information Technology," pg 1, Evaluation of Engineering Systems: December 2003, http://www.itu.int. Concepts and Techniques, Pitman [6] International Telecommunications Union, Publishing Inc., 1983. Telecommunications Standardization [15] D. Moore, G. Voelker, and S. Savage, Sector, "Security Architecture for Open “Inferring Internet Denial-of-Service Systems Interconnection (OSI) for Activities,” Proc. of the 10th USENIX CCITT Applications," ITU-T Rec. Security Symposium (Washington, DC, X.800, 1991, http://www.itu.int. 2001), http://www.usenix.org/events/ [7] Cable Television Laboratories, Inc., sec01/moore/moore.pdf. "PacketCable Security Specification," [16] C. K. Chan and H. Pant, “Reliability and PKT-SP-SEC-107-021127, 2002, Security Modeling in Upgrading http://www.packetcable.com. Wireless Backbone Networks,” Bell [8] Cable Television Laboratories, Inc., Labs Technical Journal, 8(4), 39-53, "Data-Over-Cable Service Interface 2004. Specifications DOCSIS 1.1, Radio [17] Security Vulnerability in Sun Cluster Frequency Interface Specification," SP- 2.2, May 20, 2003, doc id 51340. RFI-v1.1-I10-030730, 2003, [18] Allen L, Black, James L. Spencer and http://www.packetcable.com. Douglas S. Dorr, The Impact of [9] Cable Television Laboratories, Inc., " Commercial Power Quality on Fiber-in- Data-Over-Cable Service Interface the-Loop Service Availability, Ninth Specifications DOCSIS 1.1, Baseline Annual NFOEC Conference Privacy Plus Interface Specification," SP- Proceedings, June 1993. BPI+-I10-030730, 2003, [19] John Viega and Gary McGraw, Building http://www.packetcable.com. Secure Software: How to Avoid Security Problems the Right Way, Addison-Wesley, 2001.

REDUCING CAPEX AND OPEX THROUGH CONVERGED OPTICAL INFRASTRUCTURES

Duane Webber Cisco Systems, Inc.

Abstract PAPER

Today's Cable Operator optical This session will focus on the capabilities infrastructure designs are becoming more of Optical infrastructure options, when and important as customers are increasingly where to use them, and highlight real world demanding high-bandwidth services such as analysis derived from implementation by residential and commercial data and VoD. The several Cable Operators deploying deliver network must be able scale to increasing service voice, video, and data services to residential requirements and customers, be flexible to and commercial customers. deliver new services, and be cost effective to lower CapEx and OpEx. At the same time the To help ensure future profitability, solution must offer investment protection. To leading cable operators are exploring how accomplish this, optical equipment vendors are best to deliver new revenue-generating offering a wide array of solutions to allow services while lowering capital and Cable Operators flexible service offerings while operational expenses (CapEx and OpEx). reducing CapEx and OpEx through network convergence. Many operators looking to converged networks to improve efficiencies, especially New infrastructure platforms offer a breadth when delivering new services and of options for delivering an optical applications infrastructure based on two primary architectures. Transponders-based solutions Figure 1 shows an example of an integrate DWDM intelligence including auto unconverged network that comprises five topology discovery, wavelength provisioning, individual fiber networks. The network and auto power management. These platforms requires a total of 1960 fiber miles in order to can converge the network at multiple layers deliver digital broadcast video, voice over IP including DWDM, SONET/SDH, Ethernet, and (VoIP), high-speed data (HSD), SONET for IP. Passive-based solutions integrate optics analog video, and video on demand (VoD). into the switch/router via gigabit interface This approach incurs the high CapEx of connectors (GBICs) while using passive optical installing new fiber, especially in solutions to support native gigabit Ethernet metropolitan areas where cable operators are solutions. Both solutions enable the Cable reaching fiber capacity limits. Operator to offer new revenue generating services while optimizing network efficiency, lowering CapEx, and lowering OpEx.

Figure 1 An Unconverged Network Running Multiple Services

Convergence Benefits installation costs alone—all of the services could be carried over a single fiber pair, Convergence at different layers can bring ® requiring only 380 fiber miles. This is based significant savings. For example, the Cisco on an estimate that the cost to lay each fiber ONS 15454 with the ML-Series card allows mile is $3615, factoring in fiber, amplification convergence at layer 2, and further and regeneration equipment, and trenching convergence at the optical layer, layer costs. The $5.7 million figure is derived by 1,brings greater savings. With an optically multiplying the extra fiber of 1580 miles converged architecture, the network described required (1960 minus 380) by the per-mile above would save US$5.7 million in fiber fiber cost of $3615 (Figure 2).

Figure 2 Optical Convergence Saves Fiber

In addition, management becomes gracefully combines networks with different increasingly complex with multiple networks, quality-of-service (QoS) requirements, where each network has separate management bandwidth needs, protection, topology, and systems and processes that must be protocol requirements (Figure 3). A properly maintained. designed converged optical network is transparent for all higher-layer protocols that Rather than running each service transit the network, and provides simple application in a separate network silo, a manageability of the optical layer. flexible, adaptable converged infrastructure

Figure 3 A Fundamental Challenge is Migration to a Converged Network

Cisco Optical Architecture Options Cisco Systems® offers two primary native dense wavelength-division architectures for implementing converged multiplexing (DWDM) sources with the optical networks. The first architecture is an DWDM components of the Cisco ONS 15216 active, intelligent, auto-adjusting DWDM product line (filters, amps, and dispersion solution based on the Cisco ONS 15454 compensation units (DCUs), for example). Multiservice Transport Platform (MSTP). The The emergence of ITU-grid pluggable optics second architecture takes advantage of the for switches, routers, and optical transport fact that many client devices (multiservice platforms has made this architecture an provisioning platforms [MSPPs], switches, attractive and lower-priced alternative to the and routers, for example) can source ITU-grid MSTP-based design for less complex optics directly. This architecture couples these networks.

Figure 4 Optical Architecture Options

Intelligent Auto-Adjusting DWDM Solution accomplished with unidirectional Gigabit Ethernet (GE) or 10-GE wavelengths. Data The Cisco ONS 15454 MSTP delivers an networks can be connected with GE, 10-GE, intelligent, auto-adjusting DWDM solution. It SONET/SDH, or other signals. Voice network brings capital and operational efficiency by transport can be handled with the standard converging many disparate networks at DS-1 or DS-3 signals or with SONET/SDH multiple layers, including the optical layer. connectivity. Storage networking can be The Cisco ONS 15454 MSTP features all of delivered with 1-G and 2-G Fibre Channel the familiar capabilities of the Cisco ONS interfaces, Fibre Connection (FICON), 15454 MSPP, delivering time-division Enterprise Systems Connection (ESCON), or multiplexing (TDM), Ethernet, and SONET GE. Each wavelength can be customized as to services, but adds support for numerous bit rate, protocol, protection scheme, and standards-based, auto-adjusting wavelength direction (uni- or bidirectional). services within the same platform. The converged network can deliver several optical The Cisco ONS 15454 MSTP is the wavelength services, as well as sub-rate optimal choice when a separate, dedicated multiplexing. Video transport is typically optical platform is required to provide a demarcation between the optical and data domains, and is ideal for optical networks of • “Plug-and-play” card architecture for significant complexity in terms of topology, complete flexibility in configuring distance, protection schemes, and DWDM network elements––hub nodes, management requirements terminal nodes, and optical add/drop Technology Innovation nodes––within amplified or un-amplified networks. The Cisco ONS 15454 MSTP uses advanced photonics technologies, combined • Flexible 1- to 32-channel optical add/drop with innovative engineering, to address the multiplexer (OADM) detail, supporting unique requirements for both metro and both band and channel OADMs, for regional networks: greater ease in network planning and reduced reliance on service forecasting. • Scalable 1 to 32 wavelengths in a single network for superior cost-versus-growth • Integration of pre- and post-amplification. trade-off. • Multilevel service monitoring— • Transport of 150 Mbps to 10 Gbps per SONET/SDH, a G.709 digital wrapper, wavelength, as well as sub-rate and an optical service channel for multiplexing of TDM and data services, unparalleled service reliability. for maximum service flexibility. The • Network topology discovery, automatic Cisco ONS 15454 MSTP supports G.709 power control, automatic node setup, and standard encapsulation, allowing wavelength path provisioning to simplify wavelength transport independent of the DWDM network management. transport protocols embedded in the wavelength. Asymmetrical and Bidirectional DWDM Transmission • Transmission distances from tens to hundreds of kilometers (up to 600 km) The Cisco ONS 15454 MSTP supports without regeneration through the use of both asymmetrical and bidirectional advanced amplification, dispersion wavelength transmission. Wavelengths can be compensation, and Forward Error routed independently, allowing for Correction (FEC) technologies. Enhanced asymmetrical and bidirectional applications FEC on future products will extend this (Figure 5). range further without the need for regeneration.

Figure 5 Asymmetrical and Bidirectional Applications Over a Converged MSTP Network

Wavelength Protection client protection provides protection of all fibers and equipment in both the working and The MSTP also offers optical path protection path by allowing protection redundancy using three wavelength protection switching to be performed on the client schemes. Y-cable protection protects against device. Figure 6 highlights the protection both wavelength path and transponder schemes supported by the Cisco ONS 15454 equipment failures, splitter protection reduces MSTP. costs but only protects the optical path, and

Figure 6 Protection Schemes on the Cisco ONS 15454 MSTP

difficulty in predicting demand. Furthermore, New Advances for Managing Wavelengths in complexities are involved in managing metro the Metro DWDM network architectures. These include Traditional DWDM solutions have rigid dynamically adding and dropping network architectures and require wavelengths, controlling in real time the considerable manual interaction to manage, power across wavelengths on the fiber, and particularly when new sites are added or setting up and provisioning the DWDM network capacity is upgraded. Traditional network. The Cisco ONS 15454 MSTP deals solutions are optimized for low-cost, per-bit, with these complications through several fixed topologies that cannot efficiently innovations, including network topology address the operational constraints of metro discovery, wavelength path provisioning, and regional networks. Metro networks face automatic power control, and automatic unique challenges, such as the inherent network setup. Network topology discovery, based on the attenuation—yet at any given point in the industry-standard Open Shortest Path First network, particularly prior to amplification, it (OSPF) protocol, enables the auto discovery is crucial that the individual wavelengths in of nodes in a layered network without the the fiber be at a uniform power level. provisioning of neighboring nodes, and provides network topology information for Lambda Management amplifier power control and wavelength path In order to perform lambda management, provisioning. cable operators can make use of Cisco

Transport Controller, a powerful craft Wavelength path provisioning enables the interface tool resident on the Cisco ONS same “A-to-Z” provisioning that is widely 15454 platform, and Cisco Transport used on the Cisco ONS 15454 MSTP for Manager, a workstation-based element SONET/SDH circuits to be used for management system (EMS) capable of wavelength provisioning. The application managing thousands of elements. Cisco tunes transponders to the correct wavelengths Transport Controller and Cisco Transport and prevents the provisioning of unavailable Manager monitor the power of wavelengths wavelengths. as reported by the network elements.

Typically, individual wavelength powers are Automatic power control dynamically monitored at ingress to or egress from the monitors and controls optical power across all network, and composite signal power is wavelengths. This enables constant per- measured as the composite signal enters and channel power as channels are added and leaves active components. Monitoring of the dropped and constant per-channel power composite signal is an efficient and cost- under span loss degradation due to changing effective way to maintain the network— conditions, fiber aging, or laser aging. It also failure of an individual wavelength, once it is provides automatic provisioning of amplifier aggregated within a composite signal, is parameters (gain, for example) during extremely unlikely. Typically, impairments to network installation. the system at the composite signal level will

affect multiple wavelengths and be detectible Finally, automatic network setup equalizes via composite signal monitoring. Individual the optical power of all channels prior to wavelength monitoring of a composite signal amplification by controlling the ingress power at every point in the network clutters the level of each channel as it is added to the management interface, adds cost, and does network, further simplifying network not provide additional value. It is sufficient to management. monitor each individual wavelength as it

enters and leaves the composite signal, and to Equalization is necessary because monitor the health of the composite signal at individual channels will hit the ingress point all points between a given wavelength’s of the network at differing power levels, and ingress and egress from that composite signal. may take different paths through the network and thus experience different amounts of

Figure 7 Monitoring DWDM Power Levels

The Cisco ONS 15454 MSTP also has the Controller and Cisco Transport Manager, ability to set thresholds and triggers on including lambda layer and fiber layer views, monitored parameters at DWDM and and the ability access the port level of Cisco electrical levels. This enables the system to ONS 15454 equipment to isolate problems. proactively set off alarms or to take other pre- The lambda layer view provides a snapshot of programmed actions when signals cross the logical connectivity and the health pre-set thresholds. information between edge devices. The fiber layer view shows which lambdas are running To interface with other management on which fibers, filters, splitters, and systems, the Cisco ONS 15454 MSTP amplifiers. supports the following management protocols: Cisco ONS 15216 Product Line • TL1 from the network element with full The Cisco ONS 15216 product line fault, configuration, accounting, includes both unidirectional and bidirectional performance, and security (FCAPS) DWDM filters, as well as erbium-doped fiber features amplifiers (EDFAs) and DCU units. The • Simple Network Management Protocol Cisco ONS 15216 product line is designed for (SNMP), for fault management and solutions where the wavelength generation Remote Monitoring (RMON) statistics, function is separate, such as when a pluggable from the network element ITU optical module is integrated within the switch or router itself. • Common Object Request Broker

Architecture (CORBA) gateway with full A Cisco Catalyst® 4500 Series system, for FCAPS from Cisco Transport Manager example, configured with 1- or 10-G • TL1 gateway with full FCAPS from Cisco pluggable optical modules, enables 32 Transport Manager wavelengths to be delivered over the passive optical network. Voice, video (including • SNMP gateway with trap forwarding from VoD), and data services can be converged Cisco Transport Manager over this network with the switch or router

providing a scalable and flexible aggregation Finally, multiple network management point. views are enabled through Cisco Transport

In Figure 8, primarily passive building scalable, enabling as few as two channels to blocks are used to build unidirectional be deployed on day one and scalable up to 32 distribution trees for VoD. The solution is DWDM channels.

Figure 8 Example of an ITU DWDM Pluggable Optical Layer

using the Cisco ONS 15454 MSTP or an Integrated ITU DWDM Pluggable integrated ITU DWDM pluggable solution via Wavelength Management the Cisco ONS 15216 product family. Network administrators can manage the passive optical network through active Phase 2, targeted for the second half of components in the network, including the 2004, will provide increased network design EDFAs and the 100-GHz OADMs. For flexibility and management. It will include example, the EDFAs implement automatic improved aggregation for lower-cost transport gain control and variable optical attenuators with 10-Gigabit ITU pluggable optics moving that allow software control of power at the to switches and routers. These pluggable input of the amplifier. Information from these optics will provide optical monitoring to active devices flow up to both Cisco improve manageability of the integrated Transport Controller and Cisco Transport solution. Manager, allowing both management of and visibility into the network. Phase 2 will also focus on improved lambda distribution and flexibility through Optical Infrastructure Roadmap reconfigurable OADMs (ROADMs), allowing lambdas to be turned up remotely as traffic The Cisco optical infrastructure roadmap demands increase. As wavelengths may is being implemented in three phases. The change at the transport layer via ROADMs, current phase, described in this document, the EMS will maintain an association between includes the development of a robust network the client port and the wavelength for infrastructure. This includes an intelligent management visibility. auto-adjusting DWDM network solution Phase 3, targeted for the first half of 2005, maintaining only a single optical network to will provide sub-rate multiplexing of multiple meet all transport needs. Cisco is uniquely protocols (ESCON, IBM Fiber Connection positioned to offer all of the components of [FICON], Fibre Channel, and GE) onto 2.5-G the converged optical network. or 10-G lambdas, giving improved efficiencies and removing cost from the For more information on Cisco networking network. solutions for the cable services network, visit www.cisco.com/placeholder or view the Future of Optical Convergence datasheets of the products featured in this paper: The combination of rich new capabilities • Cisco ONS 15454 included in Cisco platforms such as the ONS http://www.cisco.com/en/US/products/hw 15454 MSTP and the ONS 15216, combined /optical/ps2006/index.html with the ability to converge at multiple layers, is laying the groundwork for converged • Cisco ONS 15216 optical networks in the cable market. These http://www.cisco.com/en/US/products/hw will offer increased revenues from new /optical/ps1996/index.html services as well as the reduced CapEx and OpEx that naturally results from building and SECURITY AND SAVINGS: GOING DIGITAL AND GETTING BOTH

Alec Main Cloakware Corporation

Abstract satisfaction from digital services, the challenge for cable operators is switching Achieving cost savings while converting the over the installed base of analog users. Those analog television ecosystem to digital reluctant to switch – who could represent the continues to be a difficult problem. The next potential wave of adopters – complain revenue opportunities available through the about too few digital channels, while some conversion of the analog bandwidth to digital subscribers have no intention of switching. In are well identified, but with the continued the meantime, the base of subscribers who do prevalence of analog legacy systems, the path upgrade from analog to digital can experience to conversion is a significant challenge. How reduced quality on analog channels due can the industry leverage the potential excessive conversion (i.e., analog to digital to revenues of digital bandwidth, while analog). continuing to provide service to those who may not want or need the digital experience? Consider the following broad categories of cable subscribers: This paper examines a cost effective path to conversion which proposes developing a 1) Premium service subscribers using basic service set-top box (STB) that would digital services on a set-top box with a include replacing hardware-based security smart card and two-way modem. solutions with a lower cost tamper-resistant software solution. This change would allow 2) Potential new digital subscribers subsidized digital STB roll-out and electronic considering upgrading to a digital provisioning of basic services, while STB. enforcing the rights of content providers and 3) Basic service subscribers using analog owners. services, who may never switch to digital. INTRODUCTION Until all of the basic subscribers switch The move from analog to digital networks over to the digital service, the MSO must allows cable operators to offer more services continue to provide analog services as well as to more customers than ever before. With digital. The fastest way to accomplish the digital cable, multiple service operators digital conversion – and to reap the benefits – (MSOs) can offer a host of new services such is to entice subscribers to make the leap by as video-on-demand, interactive television lowering the price of the STB to such a point and commercial-free CD-quality music, that it is easy for consumers to switch – say, giving subscribers greater choice, quality and $35 – or at least, to a point where an MSO can control. economically subsidize the box.

Taking the network all-digital frees up In order to achieve a cost-effective box, we bandwidth to offer these new, exciting propose boxes without smart cards or services. While subscribers report good CableCARDs™ that are targeted at the basic service subscriber. Security will be handled adding of the STB to the subscriber’s by secure software which is lower cost premises should be very simple. compared to hardware. The set-top box would have a unidirectional cable modem, analog The next issue is whether to scramble basic TV output (converted from digital input) and services – the goal being electronic use a standard remote control. There would be provisioning and eliminating truck rolls to no hard drive or personal video recorder activate or deactivate subscribers. To answer (PVR) capabilities and no additional outputs this question, we need to consider existing for home networking. It would be compatible piracy and subsidy strategies. with existing TVs and VCRs. A certain level of subscription fraud is This paper presents the challenges and tolerated today. Many subscribers have a benefits of going digital using secure splitter in their basement and feed multiple software: TVs on one subscription.

• Dealing with currently accepted piracy Going all-digital – whether the basic levels during the transition service is scrambled or not – requires a STB • Reducing the manufacturing and on- per TV. If the low-cost STB is not available going support costs through retail channels, then the MSO needs to consider piracy in conjunction with their • Realizing additional benefits of a subsidy strategy. software approach including electronic provisioning The low-end STB has a one-way cable • Making the software secure modem and an analog output. It provides only • Dealing with legacy CAS systems basic services and the keys for delivering premium services are never downloaded to It concludes that a software-based solution the box. The threat against such a box is low. is cost effective, while continuing to prevent However, by not addressing existing piracy, a subscription fraud. market for grey market hacked devices will develop. A strategy is needed to consider subsidizing one or even two low-cost boxes GOING DIGITAL per home, plus providing additional boxes for

sale. Subscriptions can charge for multiple Analog services tie up a significant TVs at a reasonable price – but not so high as percentage of bandwidth. Once the analog to stimulate a grey market. Most subscribers services are no longer required, bandwidth is will want to stay legitimate given reasonable freed up for additional premium service options. offerings.

Under this scenario, users would need to However, making the switch requires upgrade to another box for premium services. some planning. Many MSOs have already The basic STB could still support the Open moved some channels to digital, but the Cable Applications Platform (OCAP), but network infrastructure must be set to handle offer a limited set of capabilities (e.g. no all-digital. The switchover must be well interactive functions). The MSO may want to communicated and coordinated to ensure consider giving subscribers the option of the minimal disruption of service. The actual basic box, with a credit on a premium box There will, however, be additional costs when taking the network all-digital. for the CPU and memory, but it’s money better spent: this is the more practical place to Proper planning and a strategy to address add cost, since these processors support a existing piracy must be in place prior to wider variety of software and applications. making the transition to an all-digital network The CPU could also support Open Cable in order for subscribers to legitimately obtain Application Platform (OCAP) and this know- the services they want. how can be leveraged on premium boxes where more functionality is CPU intensive, such as DTCP-IP (Digital Transport Copy SAVINGS Protection mechanism for use on IP networks)

and PVR functions. Also, expect some How can we get the price down low additional cost for hardening the software enough to enable the transition to all-digital running on the box. input for the basic service subscriber? We propose cutting manufacturing costs by The last additional cost relates to the legacy replacing the security hardware with secure conditional access system (CAS) in place on software. Software can deliver identical the MSO’s network. This cost is discussed functionality to hardware with other added later in this paper. We believe that the savings benefits. will be greater by moving to a secure software

implementation, plus there are added benefits Cost savings are realized by eliminating to a secure software approach. the CableCARD, as well as the reader within the STB. A CableCARD is a PCMCIA-like card with a smart card slot or smart-card BENEFITS OF SOFTWARE SECURITY functionality embedded. While these prices are expected to drop, they are currently very In consumer electronics, cost reductions expensive. The cost of the card is covered by typically involve the replacement of soft parts the MSO, and may be recouped by a small with hard parts. Integration is usually the incremental monthly charge. If these cards are name of the game. hacked, which is likely if satellite TV is any indication, then they need to be replaced Security is an exception. periodically by the MSO. By eliminating these cards on a basic STB, the MSO saves Conditional access security is provided by initial and recurring costs, while minimizing an external component such as a smart card. the threat – it is very difficult for a basic The smart card is not integrated because it services box to be upgraded via hacks to full needs to be specific to an MSO, but also service. because it needs to be renewed periodically.

The additional major cost savings comes Software security can also be renewed, but from using a basic unidirectional cable at significantly lower cost. In addition, it is modem over a bidirectional DOCSIS® much harder to remove software from a modem. By running the secure software in a closed box, hack it and then insert it into other Linux® environment, there are no additional boxes. Smart cards arguably help create a operating system costs. pirate network because of the ease of removal and distribution.

Secure software – or more specifically a Do we even need to protect the box? secured software-oriented STB – has other Certainly the threat is not as severe as in a PC benefits: environment where the hostile user has complete control of the CPU and applications 1) New revenue opportunities – operators that are loaded. However, software protection can create new service bundles based steps are needed as the box will likely run a on the ability of subscribers to known operating system like Linux for cost download new technologies and savings. Regardless, tools exist to attack most features, even for low-end STB computer systems, so some level of protection owners. is prudent.

2) Increased flexibility – operators can Can we just encrypt the data? Data meet different standards as they confidentiality is only one component of the emerge. solution. The software also needs protection. Content protection standards, such as DTCP- 3) Increased renewability – new security IP, CPRM (Content Protection for Renewable countermeasures can be deployed Media) and HDCP (High-bandwidth Digital quickly and as frequently as required to Content Protection), all recognize the need for the entire existing installed base faster, software robustness.1 The requirement for reducing subscription fraud and piracy. software protection is mandated by these Software can be renewed selectively, standards. proactively, or reactively. In this case, the primary goal is to prevent 4) Ability to upgrade – subscribers don’t subscription fraud. Since this box is for basic have to buy a new set-top box to services only, the simplest mechanism is to benefit from new technologies and new make sure the box does not have the content features that can be downloaded. descrambling keys or functionality required for premium services. We assume some All of these benefits can be achieved scrambling is performed on the content and without the use of smart cards and the added the goal is to decrypt only the appropriate costs of replacing them. channels.

SECURITY Secondly, the attacker can always convert the analog output to digital (known as the The rewards of the secure software- “analog hole”), whereas our security goal is to oriented STB extend well beyond reductions prevent siphoning-off of the digital content. in cost to the MSO, but what kind of new Again, a basic service box reduces this risk risks are introduced to the MSO and how are since premium content is never descrambled they best mitigated? Any discussion on or available on the internal busses in the clear. software security should include a description of the threat model. The STB scenario is Lastly, we want to prevent the box from called a “hostile user threat”, where the being used for other purposes – a form of subscription fraud common with high-end legitimate user of the system may want to 2 hack it for the purposes of subscription fraud, media devices such as Xbox® . Since this is a piracy or theft of services. basic service box with limited outputs, such threat of service is also low risk. Since we have control of the operating that can only be resolved at runtime. There are system installation and software upgrades, we specific decompilation and disassembly can consider numerous techniques to harden prevention techniques that target these tools. the system. First let’s look at how software is Note that while very powerful disassembly attacked. A software attack follows this tools exist, most low-level code written in C general framework: or C++ is very difficult to decompile with only a few tools available. Software 1) Analysis – Classic reverse engineering protection is about using multiple layers of and analysis of the software and protocols to defense and all these techniques should be identify vulnerabilities. This can be static considered. analysis when the code is not running, such as disassembly and decompilation, or dynamic Runtime analysis of a system can be tracing of the executing code using debuggers prevented or made very expensive by the use and emulators. There are some more advanced of anti-debugger and anti-emulation and powerful forms of analysis such as techniques. A range of techniques unto collaborative and differential analysis, which themselves, these can be effective on we will discuss later. platforms where the operating system and applications are known in advance – such as 2) Tampering – Modifying the code and/or with our basic STB. In this case, the code can data such that it performs according to the be tied to the platform via node locking and attacker’s objectives. loading of new applications controlled by secure code signing techniques. Advanced 3) Automation – The creation of scripts or just-in-time decryption (or self-modifying code to apply the tampering attack to multiple code) techniques also raise the bar against copies of the application. These are also dynamic analysis. Authentication of known as “class attacks” or “global breaks”. components on the machine and encryption of In some cases, the tampered application must communication channels with protocol not be distributed, which is less desirable from an subject to replay attacks also prevent analysis. attacker’s perspective as it is more detectable In addition, data transformation techniques and prosecutable under legal measures. can be used to hide and randomize data values even when operated on within main memory. 4) Distribution – Once the automated White-box cryptography refers to specific attack is created, it must be distributed in an cryptographic implementations designed to effective, confidential means. Often bulletin prevent key extraction even when the boards, Internet Relay Chat (IRC) and peer- operation can be viewed by an attacker. to-peer networks (P2P) are used for this purpose. Static tampering is prevented with binary encryption techniques, as well as by The first goal is to make analysis and introducing data dependencies in the code to tampering difficult, time-consuming and/or change an easy branch jamming attack into expensive. The obvious approach to prevent tampering – increasing the effort required and static analysis is to encrypt the binary. involving multiple changes to the code. An However, there are many techniques to extract important technique to prevent tampering is these decrypted executables from memory. code signing, but the code signing mechanism However, there are also techniques that itself will be subject to attack and so must prevent static analysis such as control flow also be suitably hardened. Integrity flattening which introduces pointer aliasing verification of applications should be done The disadvantage of Sony Passage is that statically (on-disk) as well as in-memory to the MSO must make changes to their head prevent dynamic tampering attacks. end. The Sony Passage system consumes 2% to 10% additional bandwidth, but this will be Prevention of automated attacks is best amply compensated for by going all-digital. achieved by deploying code and data diversity There are a number of new players5 providing such that a successful attack will only work software-based CAS that can work with Sony for a sub-set of users. Diversity of code is a Passage or independently. result of most software protection techniques outlined above. It is similar to having CONCLUSION different keys (diversity of data) for different users. Diversity of code recognizes that This paper has described a secure and cost attacks will be on the software in addition to effective path to migrate analog cable users to the data. Automated attacks are also mitigated digital services. by software renewability, which can be made low cost – if designed in upfront. Conversely For the basic service subscriber, we can with hardware based security, renewability is achieve a cost-effective box without smart a major cost. Software can be renewed cards or CableCARDs. The set-top box would selectively, proactively, or reactively – have a unidirectional cable modem, analog depending on the strategy and the attacks to TV output (converted from digital input), and the specific system. would be compatible with existing TVs and VCRs. Diversity is a prerequisite for successful renewability; otherwise attackers will perform Secure software can effectively address the differential analysis. This is a powerful attack challenges of going digital and provide used to quickly determine the changes made additional benefits. These challenges include to software upgrades and shorten the time to legacy conditional access systems, dealing successful hack. with piracy and reducing overall cost.

Proper planning and a strategy to address LEGACY CONDITIONAL ACCESS existing piracy must both be in place prior to In North America, the vast majority of making the transition to an all-digital conditional access systems for cable are network. provided by Motorola or Scientific-Atlanta. In The other benefits of implementing a low order to integrate a software based low-end cost STB employing secure software include STB, there are three options: new revenue opportunities, increased

flexibility, increased renewability, and a 1) License the CAS system from cheaper upgrade path. Motorola or Scientific-Atlanta (e.g. 3 as done by Digeo and others) All of these benefits can be achieved

without the use of smart cards and the added 2) Utilize Sony Passage4 to run an additional costs of replacing them. CAS over top the legacy CAS. We conclude that a software-based 3) Roll-over to a new CAS during the solution is cost effective, while continuing to all-digital transition. prevent subscription fraud. TRADEMARKS REFERENCES

CableCARD is a trademark of Cable 1) Bar-Haim, Pam and Wald, Stephanie, Television Laboratories, Inc. NDS Ltd.; “The NDS Guide To Digital Set-Top Boxes: Third Edition”; 2002. DOCSIS (Data Over Cable Service See: http://www.broadcastpapers.com/ Interface Specification) is a registered data/NDSGuideSetTopBoxIndex.htm trademark of Cable Television Laboratories, Inc.

ABOUT THE AUTHOR Linux is a registered trademark of Linus

Torvalds. Alec Main ([email protected]) is

Cloakware’s Chief Technology Officer. Passage is a trademark of Sony

Corporation. He has spoken at key forums and

conferences including Copy Protection Xbox is a registered trademark of Technical Working Group (CPTWG), Media Microsoft Corporation. Summit, RSA, the Intel Developer Forum,

Information Highways, and Certicom – PKCS. END NOTES

1) See: Hitachi, Ltd., Intel Corporation, Matsushita Electric Industrial, Co., Ltd., Sony Corporation, Toshiba Corporation; “5C Digital Transmission Content Protection, White Paper”; July 14, 1998; http://www.dtcp.com/data/wp_spec.pdf

2) See: xbox hackz Website; http://www.xboxhackz.com/; 2002

3) See: Digeo Inc. Website ; http://www..digeo.com; 2004

4) See: Sony Passage Website; http://www.sonypassage.com; 2004

5) See: Latens Systems Ltd. Website; http://www.latens.co.uk/html/cable.html; 2004.

STORAGE: THE FUTURE OF CABLE

Bob Van Orden Scientific-Atlanta, Inc.

Abstract applications. These advances are evolutionary, as they continue to build on By definition, all new technologies existing technologies and trends. redefine what can be done in a given medium. Faster processors enable Other advances are more revolutionary. increasingly complex and sophisticated For example, Video-on-Demand (VOD) and computing activities to take place, at lower Digital Video Recording (DVR) and lower cost. Improvements in technologies give consumers unprecedented video/graphics technology enable more control over television viewing. Instead of colors and greater clarity. Certain key watching TV shows at scheduled times, technologies, however, do more than people can watch whatever they want, support evolution. They change everything. whenever they want, with the power to start, stop, pause, rewind and fast forward the This paper will look at the present and action. projected impact of one such revolutionary technology – storage – on the cable industry. Similarly, one must recognize the It will examine how storage enables new current trend toward increasingly capabilities beyond DVR. It will posit ways customized content. Correlative systems can in which the storage-based paradigm can make individualized recommendations offer operators opportunities to generate based on consumer viewing patterns. For new revenues, dramatically enhance example, if someone records every episode consumer relationships, create new leverage of Survivor, the system might recommend with broadcasters and content providers, other, similar shows that the viewer might ensure content and subscriber security and enjoy, enabling consumers to receive benefit from new advertising models. specialized programming based on their specific interests, likes and dislikes. These EVOLUTION AND REVOLUTION: advances are more than mere improvements ADVANCES IN CABLE TECHNOLOGY on what people already know and expect. They completely transform how people Without question, the cable industry is think of TV – and they are possible because undergoing tremendous change. We’ve of revolutionary changes in storage. advanced from a few channels to hundreds of channels. Content is becoming STORAGE CHANGES EVERYTHING increasingly specialized (e.g., sports channels, movie channels, home The ground-breaking effect of storage improvement/decorating channels, etc.). technology on consumer capabilities is not new. The computer industry saw a similar Image and sound quality continue to shift in what people could – and would improve as transmissions migrate from expect to – do with computers as storage analog to digital to HDTV. The speed of technology became less expensive and able broadband traffic is also on the rise, as we to fit on a smaller footprint. The industry move from dialup to high-speed cable – to evolved from mainframe computers that virtually unlimited bandwidth for a variety took up entire rooms to “dumb client” of data- and video-oriented cable environments in which multiple terminals

could tap into individual mainframes. The monthly fees for the service which are first PCs were limited by the amount of data typically $9.95 monthly in most cable that could fit on a floppy drive. When systems.2 More significantly, it appears that storage technology enabled PCs to the people who are embracing DVR services incorporate hard drives, however, the are not those who fit the “early adopter” capabilities of PCs increased dramatically – profile.3 DVR appeals to virtually everyone as did people’s interest in and need for who tries it. computers. BEYOND DVR: HOW STORAGE WILL First, the hard drive enabled increased TRANSFORM CABLE NEXT personalization of the software environment and the ability to save vast amounts of Beyond the DVR technology readily content. Then, when these storage- available today, storage in the set-top opens empowered computers were connected in a the door to even more services and distributed network paradigm via the capabilities. Internet, people could do even more. Email, the ability to get content and information via DVR Throughout the Home Internet downloads, and the ability to buffer content downloaded from the Internet for a Consumers are also making it clear that more seamless “streaming” experience once they have DVR service on one TV, became the norm. they want it throughout their home. With that in mind, operators will be able to The effect of storage on the cable benefit from services that enable consumers environment is quite similar. Storage in the to save content on one set-top, then share network enables video-on-demand services, that content with multiple televisions in the including Movies on-demand, Kids home. This is similar to the trend we have programming on-demand, Subscription on- witnessed in multi-PC homes with home demand (e.g., Starz/Encore on-demand, networking in a client-server like HBO on-demand, etc.) and Free on-demand architecture to share data. In this case, the (e.g., HGTV on-demand, DIY on-demand, DVR essentially turns into a home etc.). Currently available to an estimated entertainment server. 16.5 million U.S. homes1, these types of services give cable subscribers the ability to While content sharing brings up security access specific programs without regard to a issues, a viable system (such as Scientific- pre-defined programming schedule. Atlanta’s Explorer 8000 Multi-Room DVR system), should store only one copy of any Storage in the set-top, of course, enables given program recording and use existing DVR services, through which viewers can wiring in the home. The result simplifies watch what they want, when they want and installation and ensures that content can’t be create personalized libraries of preferred shared between subscribers within a given content. Likewise, DVR services enable cable service area. viewers to control live TV, just as if it were on a DVD recording. “Cable Anywhere”

Now widely deployed throughout North Just as storage in the PC evolved to America, DVR has proved to be a dramatic include CD and DVD recording capabilities, success. Consumers love DVR. Research we can expect similar developments in the shows that 75% of cable DVR subscribers cable environment to make it possible for rate DVR very highly and are willing to pay people to take the cable programming they

love with them wherever they go. This likely to want information about a given “Cable Anywhere” concept can create a new product or offering. distribution model for cable operators. Again, security would need to be ensured so The Complete Entertainment Experience. that DVD copies of programming could not A hard drive in the set-top can enable be re-copied or distributed. With secure storage of all kinds of content, in addition to DVD recording capability built into the cable video programming. Subscribers could DVR set-top, however, consumers can be “lease” a section of the hard drive to store offered the opportunity to own a DVD home videos, family photos, MP3s and recording of any program for, say, $9.99. games that could be played against other Likewise, it enables operators to expand the people in the home – or other subscribers in VOD paradigm by following any VOD the cable system. Ultimately, the set-top stream with the opportunity to own the hard drive could house all this, as well as program for only, a small surcharge. other downloadable applications and programs that have yet to be imagined. In The Personalized TV Environment this way, the cable operator enables the family room TV – or every TV – to become Customized Look/Feel. Advanced a complete, integrated entertainment center storage in the set-top enables subscribers to for the household. personalize their TV environment for a simplified, highly relevant entertainment All of these features and capabilities experience. Users will be able to build a would be able to be customized, configured unified, individual interface, based on and updated remotely by the subscriber. subscriber preferences for channels and content, as well as overall look and feel of THE NEW STORAGE PARADIGM interactive/information screens. Simply put, CAN HELP OPERATORS users gain the ability to group the functions they want within the interface where those Generate New Revenues from New Services functions are most convenient, much like a programmable remote controls do today for Clearly, operators can take advantage of linking disparate audio/visual components new revenues generated by these storage- together. based services and capabilities. DVR throughout the home is a prime candidate to Personalized Programming and be a fee-based service. “Cable Anywhere” Targeted Advertising. Advanced content services fit a per-program model, similar to search capabilities would allow a set-top to the iTunes, Rhapsody and Music Match automatically find programming based on a services for MP3 music downloads over the given genre, actor, director, subject, etc. Internet. Such a system would use individual subscriber recording habits and/or Enhance Relationships with Consumers information voluntarily given by the subscriber to deliver more relevant In the storage-based cable environment, programming recommendations than those the cable service becomes far more than a available on systems today. Similarly, the pipe for programming. If you are a cable system could use viewership and other operator, you become the must-have source subscriber information to create individual for trusted, relevant recommendations, as advertising profiles, in which advertising is well as all kinds of video entertainment. targeted directly at subscribers who are Moreover, you become the reason why consumers prefer to receive all their

entertainment – including home videos, information about subscriber viewership photos, music, etc. – on the TV. In short: the must be designed to protect subscriber relationship between the operator and identity while providing useful information. consumer goes from an impersonal Take Advantage of New/Different Kinds of consumer-vendor model to a more Advertising personalized relationship of greater depth and much greater value. Clearly, there are good reasons for the advertising community to be opposed to Create New Leverage with Broadcasters and technologies that enable consumers to Content Providers bypass commercials. At the same time, when people can control their programming, In the storage-based paradigm, even they expect to control it completely. The more than today, it is reasonable to assume ability to skip or fast-forward through that people watch programs more than they traditional commercials is hardly the death watch specific channels. When there is a of advertising. Programmers, broadcasters large, personalized library of programs from and cable operators have the opportunity to which to choose at any given time, the value sell program sponsorship, badging and of a broadcaster or content provider’s time bannering opportunities to advertisers. The slot to the consumer and/or advertiser is industry can expect to see an increase in diminished. At the same time, the cable more sophisticated product placement within operator’s value to broadcasters and content programming. Research shows that DVR providers as the direct connection to the users exercised the instant replay feature consumer – and that consumer’s preferences during the Superbowl to replay ads more – goes up. As a result, operators may be able than plays in the game. With that in mind, to leverage that consumer relationship and advertisers may choose to invest in more key information about how consumers creative and engaging ads to be shown interact with different programming to drive during a variety of events and shows. more favorable relationships with broadcasters and other content providers. Perhaps the most significant opportunity for cable operators to have new, more Ensure Content and Subscriber Security lucrative relationships with advertisers stems from the fact that storage-based capabilities As previously mentioned, services that and services give the operator the ability to involve distributed content (i.e., cable narrowly and directly target specific programming that can be accessed from subscribers. As such, cable operators can multiple set-tops in the home or the “Cable provide opportunities for advertisements that Anywhere” concept), raise security issues. also function as programming (e.g., the Indeed, such services essentially dictate that BMW ads by famous directors, featuring the cable operator become a major factor in famous actors). After all, when time-slots the protection of content and subscriber are no longer an issue, the length of a show information. Therefore, it will be essential is irrelevant. Will people actively choose to that operators employ systems that protect watch a three minute ad with adrenaline- this information. Systems that deliver pumping action and/or emotion-tearing content sharing in the home must also story-lines? Will people get hooked on ads protect content from being shared between that tell a serialized story over time? Will households on the cable network. Set-tops people want to watch performances of their with a built-in DVD recorders must ensure favorite actors or learn more about a product that content is paid for and cannot be re- for which they’ve expressed interest? Don’t copied or re-distributed. Engines that gather be surprised if the answer is yes.

STORAGE PROVIDES THE ENDNOTES OPPORTUNITY 1Kagan World Media, a PRIMEDIA It is important to remember one of the Company, Broadband Financial Databook key lessons taught by the computer industry: 2003, as cited in National Cable & For the most part, the companies who Telecommunications Association 2003 Year rapidly and eagerly embraced the complete End Industry Overview. combination of local storage and distributed connectivity have survived and thrived. 2Leichtman Research Group, Study of Those who ignored these opportunities have 900 Explorer 8000 households, November not. 2003. 3Business Week Online, New Analysis, It is easy to see that the trend toward “A Cable Lifeline for DVR Technology,” storage in a distributed environment is April 2, 2003. Also Canada Newswire, inevitable for cable. Likewise, this trend “DVD breakthrough underscores appetite indicates there are many opportunities for for interactive entertainment,” April 8, 2003. cable operators who embrace the storage- "Forward-looking statements," as defined in based paradigm. New services and the Private Securities Litigation Reform Act advertising models can generate new of 1995, may be included in this paper. A revenues. Equally important, however, is the variety of factors could cause Scientific- opportunity presented by the deeper Atlanta's actual results to differ from the relationship between the operator and the anticipated results expressed in such subscriber. forward-looking statements. Investors are referred to Scientific-Atlanta's Cautionary It has already been shown that DVR in Statements (Exhibit 99.1 to the Company's its most basic form is very “sticky.” Once most recent Form 10-Q), which statements subscribers have the service, they do not are incorporated into this paper by reference. want to give it up. Increase the amount of relevant content that the subscriber can watch, the degree to which different capabilities are integrated into the cable service, and the degree to which people become accustomed to getting “what they want/ when they want/where they want/the way they want it” and the service becomes even more deeply ingrained into the subscriber’s life.

SUPPORTING LARGE SCALE VOD THROUGH COMMON METADATA

Yasser Syed, Ph.D. Cable Television Laboratories

Abstract must consider whether the target audience of This paper explores the challenges in potential viewers is large enough to support video services and plant operations that the the costs of channel acquisition and resources operator has to undergo to support on- in the cable plant. Increasingly, this becomes demand services. In particular, the paper a hard decision because newer channels have describe how, a common set of metadata tags smaller target audiences or are repurposing will be needed to manage, describe, and content shown on other channels. present programs to the customer. Without a common set of metadata, there will be limits Given this, cable operators have been on the size of content catalogues, types of on- allocating more bandwidth for video-on- demand services, and search tools for these demand (VoD) and Subscriber Video-on- services. Demand (SVoD) services that provide customers a direct way to view content of This paper first describes what is their own choice. VoD customers, in turn, metadata and what are some of the types of need access to simple descriptions of the metadata (Application, Association, expanding video library content in order to Presentation, etc.) that are used today or select videos to view on demand. considered for use. The next section describes how metadata supports an on- This paper explores the changes in plant demand service in acquiring content, asset operations that the operator has to undergo to management, navigation, and user interfaces support on-demand services. In particular, we through the existing approaches enabled by describe how a common set of metadata tags CableLabs VOD Metadata specifications or will be necessary to manage, describe, and through new approaches. Lastly this article present programs to the customer. Without a describes how common metadata common set of metadata, there will be limits mechanisms can also be used to increase on the size of content catalogues, types of on- consumer demand for VoD content & demand services, and search tools for these services through guide-like interfaces and services. other approaches.

WHAT IS METADATA? INTRODUCTION Metadata is descriptive data associated The cable industry has traditionally been a with a content file or application. A content broadcast-oriented environment. With digital file in this case could be a Moving Pictures broadcast services, cable has over a hundred Experts Group (MPEG) video file, a still channels transmitted 24 hours a day for image file, or an audio file. The metadata viewing. In reality, very few of these may vary from merely identifying the channels are watched by a sizeable audience package title, to information for populating for more than a short period of time. Most of an electronic program guide (EPG), to the viewers do not pay for viewing content, providing a complete index of different but pay instead for the availability of a scenes in a movie, to supplying business variety of content that can be viewed. When rules detailing how the content package may a cable operator decides on a new channel, it be displayed, copied or sold. An asset is an it does not become necessary for the headend identifiable set of metadata plus its associated to fully understand the exact content or content file if it exists. metadata it is handling, but just the pertinent information for its asset management The information from metadata allows routines. An important concept at this level is cable operators to distribute, manage, track to be able to uniquely identify the asset such and present the described content or that updates to assets after distribution can be application to customers. The metadata itself feasible. Types of metadata for AMS are is just text that can be used or ignored. The Asset_ID, Provider_ID, Asset_Class, value in metadata is creating and Version, and Product_Offering. standardizing common and accepted text fields that can be used to build content. Application Specific Metadata  Metadata associated with applications like VoD/Subscription Video-on-Demand (SVoD) Categories of Metadata and can span more than one asset for information applicable to a collection or Metadata can be organized in following package. Application metadata is required in categories: order to put a collection of assets into a service. Examples of this metadata include Intrinsic Content Metadata  Metadata that association metadata, license metadata and is directly associated with only the content presentation metadata. A key item in this and does not change. This information allows category is the ability to perform operations for the content to be useable and routable, but (e.g., close) on the entire collection of related does not necessarily dictate how it can be assets. Specific types of metadata for this are played or presented. It could be encoding, license windows, sequence number, series file size, file type, genre, rating or other title (e.g., used to form groups of television information. Identifying what metadata episodes), and royalty_ID. belongs in this category allows for the content to be repurposed for different Association Metadata  Metadata on applications and services without how a group of assets are related to each retransmitting the entire collection assets. other. This can be expressed as explicit metadata information or implicit structural Non-Intrinsic Content Metadata  information through references or Metadata that is associated with only the associations. It can show that assets are content that does not usually change, but is dependents of other assets or shared among a not directly required for use of the content. group of assets. For instance, a VoD Title This is helpful but non-critical information asset can have as dependent assets: a movie that would assist in playing and asset, poster asset, license asset and preview understanding the content. Some types of asset. The concept of unique identifiers for metadata that fall in this category are chapter assets allows for relationships to be easily indexing, actors, language and studio. created. Asset Management System (AMS) License Metadata  Metadata of Metadata  Metadata needed to deliver, contractual nature for the application. Having distribute, identify and place assets within a license metadata asset for a content provider headend distribution system. With the proper to create one group of assets for many service metadata management wrapper for the asset, providers. It can then update the group of HOW METADATA SUPPORTS AN ON- assets with a unique license asset that dictates DEMAND SERVICE the contractual agreement for that particular provider. Some examples of license metadata Metadata can be used to support all key are license window, contract name, processes of an on-demand service and can display_as_new, movie preview_period, enhance performance as these services royalty_percent and billing_ID. become more heavily used. This section describes how broadcast services use these Presentation Metadata  Metadata that functions in the headend, and how this usage involves presentation of the video offering to changes as on-demand services become more the consumer through a user interface. This popular. Lastly, it shows how metadata could information can change often according to improve the performance of these functions display constraints and marketing strategies. for on-demand services and, in general, Some examples include summary, title_brief, headend operations. and category. For example, category is a marketing field for the assets that can change Acquiring Content over time (e.g., $1 movies, Christmas Movies, Weekend Blockbusters). This is In broadcast, content is usually a different from genre, which are permanent retransmission of a program channel from fields that can categorize the content and can either a local broadcast (ABC, NBC, CBS) or be used for navigation purposes. Other a premium-branded channel (HBO, aspects of presentation are display Showtime, TNN, MTV). Other types of characteristics. Common Metadata formats content that are directly acquired by the needs to also consider character limitations, headend are local advertisements and pay- Multilanguage aspects, scripts and symbols per-view movie content, which are placed in that can simplify displaying the information ad insertion and video servers. The arrival of on the user interfaces for all types of STBs. this type of content is negotiated in advance and delivered in batches (often by tape and User Metadata  Metadata that can be now more frequently by closed satellite link) used to target content to the right type of to the headend in a timely manner. consumer. In a passive way, metadata fields in the application and content assets can be In a VoD/SVoD system, content, which is used to allow the user to easily identify typically movies, is similarly negotiated content they want to see. It can also allow the beforehand and delivered in batches to the service provider to target ads for users headend in a timely manner. The catalog of watching a particular type of content (e.g., movies is often a magnitude larger than pay- advertising a blender for those watching a per-view selections, but still limited (~200- cooking show). In a more active way, 300 selections). Content is acquired via tape metadata can be created for a user that can delivery or over closed satellite link. indicate preference for particular content. In true on-demand services, the consumer This then can be used by the application or expects the equivalent of a personal program service to display content that the user channel. To acquire content in a scaleable prefers, and would be more likely to order fashion, content delivery will necessitate the (e.g., a personal barker channel for the user). following behaviors in content acquisition:

• It will incorporate different types of throughout the cable system, and then to content (Movies, TV episodes, Daily communicate with the pitcher in case a news, Sports, Instructional Video); “repitch” of the asset is needed.

• It will happen more often because the Internet/ VLAN size of the content catalogue will Studio(s): - Content Satellite(s) - Internal/External greatly expand and be actively Regional Data Center(s) Content Provider: Pitcher(s): Catcher: Asset Distribution Server: - MSO Contracts - Program - Program - Instantiating package modified (e.g., Sony’s content library - QC/ Packaging - Scheduling - Scheduling - QC/ Packaging - Metadata Authoring - Transport - Transport - Distribution contains over 6,500 movies and - Admission 1 35,000 TV episodes! ); Internet/VLAN

• It will be distributed in a manner where the complete package will be delivered over time starting with Shipping previews, continuing with asset Content Provider Package Transport Cable System updates, and finally the primary Figure 1: Cable Content Acquisition content file; Process

• It will allow for reuse and repurposing Metadata can assist in this process by of content for different types of uniquely identifying content assets and re- applications and services after initial associating them with intended applications offering. and product offerings. Additionally, it will allow for reassembling the collection of The collection of assets can be delivered distributed assets such that content (the most over a variety and combination of time-consuming element in the transfer mechanisms including satellite, tape, IP process) can be sent separately from network, and e-mail (see Figure 1). In most metadata and other assets.2 Furthermore, types of network delivery systems for assets, validation and corrections (another very there is a “pitcher” and “catcher” function time-consuming process) can be automated.3 that are aggregation points for transmitting Licensing details for a particular piece of and receiving the assets. The “pitcher” requested content can also be automatic once aggregates metadata and content from studios the overall contract negotiations for services and encoding houses, makes corrections to between the content provider and cable the assets when required, and usually acts as operator are established. As on-demand the business entity responsible for content negotiations and scheduled delivery. The “catcher” is the aggregated receiving point 2 Metadata is a small fraction of the size of the content file that validates the delivered asset, (KB vs. GB). Resending content with metadata whenever a metadata update is required can be a time consuming communicates with the pitcher, and process especially overlossy networks. Separation of content instantiates and distributes the assets and metadata can allow for easy updates of metadata while internally to multiple headends in the cable enabling content to be sent over other mechanisms that are more suitable and secure for transferring large files. network. An important function of the catcher is to validate the delivered assets 3 It can be very time consuming tracking and determining such that mistakes are not propagated mistakes in the metadata fields once the assets have been ingested into the system because corrections are usually manually inputted. Savings in time can happen if the asset 1 “Broadband and VoD Come of Age: Sony Exerts its’ can be validated before ingestion and rejected. This allows Enormous Influence”, Multichannel News, March 4th, 2002 for the content provider to correct the mistakes and repitch. services grow, the content catalogue also Standardized metadata is absolutely grows and becomes more active on a scale necessary to handle the multitude of asset that is magnitudes above what is currently management and business functions within a done. Automating and distributing some of headend system on a large scale (see Figure these processes to acquire content is 2). Metadata, such as asset class, can be used absolutely necessary at this scale. Having the to route assets to the appropriate application appropriate metadata fields allows for or content server. Resource allocation automation of these processes. algorithms can be developed using metadata fields like “year” to determine whether Asset Management content should reside at a central server or be allocated to local edge-caching systems. In broadcast, asset management was Licensing information can be used to limited to simple support of initial set-ups for determine availability of content and place retransmission and remultiplexing of this content in service at the appropriate time. broadcast channels, pay-per-view servers, Royalty information and user metadata and local ad-insertion servers. Broadcast combined with marketing offers would be asset management was also limited to needed to determine in real-time actual determining the program channel line-ups on validation and purchase price to the the plant’s physical spectrum. In VoD customer. Lastly, for the call admission, or systems, asset management functions become session set-up, bandwidth allocation more complicated because content is stored algorithms that know the runtime of metadata on distributed servers in the cable plant, and would be helpful in predicting near-term physical spectrum is a resource that is bandwidth capacities. There are numerous actively managed to service movie requests other examples. These are just some of the and peak-usage estimates. Furthermore, the ways a common set of metadata can assist in billing system in VoD moves more towards a asset management in the cable plant, real-time function that can handle single VoD especially on a scale that requires automation requests and subscription VoD models. in its processes. As on-demand services scale to larger Regional Data Center(s) volumes, there will be more Catcher: Asset Distribution Server: - Program - Package QC/ Storage - Scheduling - Instantiating package - Transport - Announcing Package application/content servers and distributed - Distribution - QC/ Packaging - Admission storage devices. An elaborate internal

Internet/ distribution system to continually redistribute VLAN content and assets via a combination of central Local Headend(s) AMS Server: VOD Application: - Package -Content distribution and edge-caching servers will - Product Presentation Database - Guide Internal Cable - Package ingestion HFC Network /lifecycle need to be supported. The billing and Network - Announcing Package To STBs QC/ Packaging reporting interfaces will need to validate and Headend Operations: VoD Server: - Asset Management - Video Streaming to STB --Billing - Support PPV & SVoD - - calculate bills in real-time, taking into account Reporting Asset Storage time of day, local marketing offers, subsidized Internet/ packages (e.g., on-demand with commercials), VLAN historical data, order reports, and customer Cable Company: profiles. Lastly, scheduling, determining - Contracts available plant bandwidth, and call admission Cable System strategies will need to be determined on-the- Figure 2: AMS within a Cable Network fly in accordance with the volume of requests. Navigation ability to query a database for a particular request becomes a useful feature as the size There is little search capability in the of the on-demand content catalogue becomes broadcast environment. The EPG serves as unmanageable for traditional navigation the navigator to the consumer. The customer interfaces. basically looks on the EPG, sees the possible content with some small descriptive information, and the time/channel it will be Presentation Interface shown. It is up to the customer to view in Printed media is an early form of an on- person, or to schedule a recording device, demand service. Books need to be presented when the content is played. in a manner to the customer such that he/she wants to buy it. If someone does not realize a The initial VoD systems are slightly better book exists, chances are that book will not be by allowing an interactive program guide read. But there are numerous presentation interface. The customer can search for a mechanisms that make the user aware and movie alphabetically and by predetermined want to read that book. This comes in the categories (e.g., just available, action, form of book lists, print and media comedy, HBO movies). The customer then advertising, and in word–of-mouth. can see a description of the movie, and maybe the preview, to determine if he likes Broadcast uses similar mechanisms to it. This type of approach can work for advertise its own content. Broadcast channels limited-size catalogues (under 200 are a good place to promote content since the selections) and becomes less friendly as the chances are a sizeable audience may watch a size of the content catalogue grows. promo for another show while seeing their favorite program. The drawback is there are As the size of the content catalogue and only a minimal number of spots with sizeable services grow, a better navigation system is exposure and some of those spots are needed necessary. An interactive query device to a to sell as advertising spots. As the amount of database that correlates metadata fields with content increases, there is less opportunity to content becomes necessary. Having the present each one. But in broadcast as a proper metadata fields and elements are program grows in popularity, it in turn can required to create a user-friendly search tool become a vehicle to advertise for other for the customer. content because there is a known viewer With common metadata tags in place, audience. This is hard to replicate in an on- different types of query needs can be easily demand model and could be one of the accommodated with the only restrictions reasons broadcast will still exist even though being the type of fields defined, and the way an on-demand service may become the the search tool is set-up. For instance, if preferred way to deliver most content in a someone wanted to watch Spanish-speaking cable network. shows to learn more Spanish, they could do a For present-day VoD, this presentation query that would look for country of origin or mechanism is not fully developed. Typically language option field. In another example, if exposure happens through the EPG, a barker a customer wanted to watch only Bob Villa channel showing movie previews and, rarely, instructional videos on building a porch, they a spot on the broadcast. The VoD or SVoD could do a combined query on actor, services (not content), on the other hand, may description, application title, and genre. This have lots of exposure via local ad insertion mannar by presenting the right content for on some of the cable broadcast channels. The selection to the user. This becomes more only constant mechanism that exposes valuable as the amount of content prevents customers to an unknown particular piece of the user from making a complete search on content is the barker channel. As the content his own. In this case, instead of searching, he catalogue increases, it would be increasingly is presented with a selection of content that hard to give the right exposure to content might be preferred. This is even more without targeting first the content to the user. efficient than creating another niche channel in a broadcast environment. Metadata can be used to present the content to the intended targeted audience. CREATING CONSUMER DEMAND Using the concept of barker channels, category barker channels could be created The mechanisms for creating consumer based upon genre or a designed market demand for content in Broadcast are well offering. For instance, there could be a barker established. It involves creating program channel for action movies [genre], $1 channel brand identity (e.g., American Movie offerings [category], or what’s the latest on Classics, HBO, Showtime). There are also HBO SVoD [display_as_new and advertising promo spots in other shows that product_offering]. Looking at more already have an audience share, as well as integrated solutions, metadata can be used to advertising in print and media. Lastly, one insert relevant previews for other content popular show can lead the audience into [based on genres] within a content that is another show by simply scheduling it right currently being viewed by the customer (this after (e.g., “Must See” Thursdays on NBC). is like trailers being shown in movie These various types of methods to create theatres). consumer demand capture an audience by:

Combining this with user metadata, there • Creating an interest in the content are even more dynamic ways of targeting through advertising spots that content to the viewer. For instance, user consumers are likely to see; statistics can be utilized to indicate those people who like this movie also liked these • Creating a brand for the program following movies (this approach is similar to channel that will draw an audience what Amazon.com or some of the Internet that believes any content on this audio-on-demand services do). Another channel will be interesting; approach is to use the customer’s history file, or inputs, to search through a content • Using the program order to lead an catalogue and present selected previews that audience into another piece of content. might suit the customer’s preference. This can then be presented on the customer’s Most of the devices to create consumer navigation or Graphical User Interface (GUI) demand for content in these cases are heavily screen as he is looking for something to reliant on the broadcast model. watch. For an on-demand service, new ways The key idea is that a common, agreed to create consumer demand needs to be upon set of metadata can enable targeting of developed since a schedule-driven content to the user in a passive or active programming approach does not exist. These new ways will be heavily reliant on making Marketing Offers  This could be similar to use of metadata for both the content and user. a lead-in and could be used for recurring Some possible new concepts are: series types of content. For instance, a viewer could order the most recent episode of an up- Video Magazine  This is like having a and-coming series and maybe get to view an video version of a National Geographic, Teen unaired episode of a popular series Like the Beat, Rolling Stone, MustSeeNBC. Each Supranos that the same studio is producing. magazine would contain pieces of content Creating these offers and personalizing them that would be of interest to users drawn would necessitate knowledge and use of the towards the brand. The brand of the metadata fields in lots of creative ways. magazine would create the consumer demand. Some of the brands or indexing in Availability Windows  The on-demand the brands would be assisted by metadata service can support previously made content, information. but could also support content just becoming available (e.g., 1st release of movies, current Personal GUI Interface  The user would episodes of TV series, content direct to on- have a personal interface that would suggest demand). To automate services based upon content of possible interest for viewing. This availability, this will require knowledge of content can be acquired by user metadata or the license metadata as well as metadata by user inputs on content metadata like related to types of assets. genre. Similar concepts are already being used in services like TiVO. There are several Broadcast Target-Insertion  The broadcast levels of complexity dealing with managing channel will still hold the largest audience for multiple users (family, individuals in family, a single viewing session. The promo spots etc.) as well as developing a proper user can be used to promote on-demand content or model (i.e., avoiding misinterpretations like services. In one approach, the entire audience my TiVO thinks I’m bilingual when I’m can view the same promo spot. In a second really not). Alternatively, the GUI can also approach, a different promo spot (but paid by be customized based on region/ethnic the same studio) can be sent to different information of content and user (e.g., show viewers based on user metadata and content option for a Yankees baseball game in NYC metadata. A product advertisement can also and a Giants Game in San Francisco). use a similar approach in both broadcast and on-demand environments. Spot-Insertion  For the right price, even viewers of on-demand content could tolerate Search Engines  The ability to search for promos for other pieces of content while content in as many intuitive ways as possible viewing something they ordered. These is highly desirable. To do this, developing promos can be selected based upon the user-friendly search engines based on metadata of the current content they are common metadata fields are necessary. This viewing. For instance, if a user was watching allows the user to tailor individual requests an action movie, they can be shown promos that can query a database (e.g., a Robert for other action movies. Promos can also be Redford fan can locate all movies he starred presented based upon the preferences of the in as well as directed). Without a search tool, user. the only other recourse is to develop a guide based upon general, anticipated types of requests (this will still also need metadata to REFERENCES keep the guide current). [1] Cablelabs Asset Distribution Interface Subscription On-Demand Services A Specification Version 1.1, subscription service can create demand by www.cablelabs.com/projects/metadata allowing users to sample different pieces of content without paying for each individual [2] Cablelabs VoD Content Specification item. The user feels that he is paying for a Version 1.1, service rather than an individual selection. www.cablelabs.com/projects/metadata This concept uses similar viewing habits as [3] Cablelabs VoD Content Encoding developed in broadcast where the user ‘surfs’ Profiles Specification, the channels for content, and can integrate www.cablelabs.com/projects/metadata the users towards trying on-demand services. This can also be a way to promote non- [4] “Wrestling with On-Demand Services: subscription content that needs to be How to Make VoD a Common purchased because the subscription service Household Name,” D. Defrees, CED, has only a limited choice (e.g., a Sci-Fi Nov.2002. subscription service can advertise for the season premier of “Quantum Leap” while [5] “Managing On-Demand: The Impending offering last year’s episodes on its menu). Challenges of Ramped Up Content,” Y. Gorden-Kanouff, Communications CONCLUSION Technology, Sept. 2002. In each of these cases, applications to create [6] A Heavenly Collision: when the HD and consumer demand can be developed around VoD worlds come together, will metadata. This process can come from a operators be ready?,” G Hardin and J. cable operator, content provider, or third Matarese, CED, Oct. 2003. party, but an agreed upon set of common metadata fields are required to develop consistent demand creation applications. An underlying substructure for maintaining constant customer demand is a sizeable content catalogue and service request availability. Common metadata can help in both these back office functions by creating powerful asset management systems that can maintain a large content collection, as well as management of bandwidth resources to minimize denial of service on VoD requests. Lastly, common metadata from both the user and licensing assets would be needed to automate billing functions and data to support these new types of demand-creation opportunities. TECHNOLOGY TO THE RESCUE – OPTICAL ARCHITECTURES FOR INCREASED BANDWIDTH PER USER

Oleh J. Sniezko, Scott Hunter, Richard D. White Aurora Networks, Inc., Rogers Cable, Cox Communications, Inc.

Abstract upstream paths without the need for upgrade in the fiber and RF coax part of the network. The requirements for segmentation levels Hence, the capacity per user can be (or segmentation granularity) have been increased without significant capital continuously evolving and have changed expenditures. In areas where nodes with a significantly since the first HFC networks large number of homes per node were were deployed. At that time, the main deployed earlier, or in new communities reasons for building HFC networks were growing outside of the existing boundaries, amplifier cascade reduction for improved the capacity per user can be readily reliability, better EOL performance and increased by adding new nodes without improved network level stability. In the adding new fibers on the existing fiber routes second half of the 1990s, a majority of HFC and without costly bandwidth upgrades in the networks were upgraded to 550 MHz and RF coax part of the network. Yet in other higher badnwdith. However, after the first areas when the network is newly built or fiber deployments, the demand for increased rebuilt for the reasons of inadequate cable, interactive bandwdith per user grew bandwidth or components, and in some areas significantly due the success of high- of extensive upgrades, a node architecture bandwidth interactive services. with superior capacity per user can be deployed. This paper describes recent progress in some areas of optical and digital This paper presents capital cost savings technologies that allow a significant increase realized by deploying these solutions with in interactive bandwidth per user. These real life examples. Finally, we describe some technologies enrich the toolbox available to feedback from field deployments of the HFC network engineers to increase available segmentation alternatives and the enabling bandwidth per user by providing several features of the node technology that allow segmentation alternatives that eliminate the easy alignment and maintenance. need for additional optical cable construction while reusing existing fibers. INTRODUCTION Further, we describe some of the optical components that permit significant reduction HFC Network Status in the cost of segmentation, the components having been developed for these applications Most older tree and branch networks have with emphasis on their robustness and been upgraded to HFC networks with 550 flexibility. MHz and higher bandwidth, with this upgrade process having accelerated over the This paper also presents segmentation last decade. The upper limit of bandwidth alternatives that take advantage of these new upgrades has been limited to 870 MHz. technologies. Several scenarios are possible. There is currently no pressure on or trend Nodes in areas previously upgraded can be toward expanding RF technology beyond 870 segmented to allow independent forward and MHz. HFC networks have proved their New Services and Increased Bandwidth per ability to support a multitude of services User Needs ranging from traditional broadcast entertainment to telephony services that With a change in behavior of high speed successfully compete with the services Internet access service subscribers, P2P provided by traditional voice service traffic explosion, introduction of VOD and providers. other one demand services (including the possibility of high definition video-on- Traditional Broadcast Entertainment Services demand), as well as additional symmetrical telecommunication services (voice), nodes After initial fiber deployments, however, larger than 1000 homes became a liability. the demand for increased capacity per user Even very coarse calculation shows that accelerated as well. Because the HFC VOD with 5% contention (in the number of networks were upgraded over a period of at households served) would result in 50 digital least 10 years, the node sizes in the networks streams in a 1000 HP node (five 64QAM upgraded in the mid-nineties were channels for SDTV with 10 streams per significantly larger (on the order of 3,000 to channel). After accounting for other 5,000 households per node) than the node interactive services, 6-8 QAM channels may sizes in the networks upgraded in more be needed for such a 1000 HP node. recent years (on the order of 500 and fewer Similarly, upstream bandwidth was not households per node). Moreover, many sufficient to support the myriad services, communities grew within their boundaries or especially when shared by more than 500 outside of their original boundaries. In many homes. cases, for today’s segmentation requirements, additional optical nodes would have to be All these factors contribute to continous added to the existing nodes. In outlying segmentation activities, especially in those areas, additional nodes would have to be HFC areas that were upgraded in the added to serve new communities. And in beginning of the HFC upgrade cycle with many situations, fibers on the existing optical large size nodes. Moreover, population cable routes have been depleted to serve the growth within the covered area as well as existing nodes or for other services. outside its boundaries also requires continous addition of bandwidth capacity per user. All these segmentation levels have proven adequate to support broadcast EMBRACING THE TECHNOLOGY services. Even after the introduction of digital TV, 550 MHz RF bandwidth (even in Fortunately, progress in several optical 5,000 HP nodes) was sufficient to support and digital technologies supported hundreds of video services. Moreover, even development of optical architecture design following initial deployments of high speed tools that result in lowering the capital cost Internet access services, the node sizes and of upgrades to the optical part of the HFC forward and upstream bandwidth per user network while taking advantage of the vast could support highly assymetrical high speed bandwidth provided by the RF coax segment data services (given very low P2P traffic and of recently upgraded HFC networks. typical customers’ behavior and applications characteristic to dial-up Internet connections). Optical Passive Components Table 1: Environmental Specification of Field-deployable Optical Components To fully utilize the optical fiber bandwidth capacity, many independent Parameter Specification streams are being transported over fiber deep Operating -40°C to +85°C into the plant. New passive optical modules Temperature Range enable selection of narrowcast information to Vibration Strand Mount Nodes be forwarded to a dedicated node. These Wind Load components are located as close as possible Humidity Up to 95% of relative to the final node location in order to preserve humidity fiber between their location and the signal origination (hub or headend) point. Only The importance of selecting and recently, a wide range components from: maintaining the required quality for these • DWDM filters components cannot be underestimated. The • CWDM filters plots in Figure 1 show how some • Dual-window optical passives components, not properly specified and have become available for outdoor screened, can behave in the field over the applications. The following environmental required temperature range. specification of these components permit their application in the most demanding environments.

Insertion Loss (including connectors) vs. Temperature

Brand X: 80 Brand X: 20 Brand X: 95 Brand X: 5 Brand Y: 85 Brand Y: 15 Brand Y: 95 Brand Y: 5

70

Note: 60 Connector splice joints for Brand X couplers are kept outside of the temperature chamber. Connector splice joints for Aurora couplers are kept inside of the temperature chamber. 50 Numbers at the legend represents the coupling ratio. I.e. 80 represents the insertion loss curve at 80% output 40 port

30

Insertion Loss [dB] Loss Insertion 20

10

0 -60 -40 -20 0 20 40 60 80 100

-10 Temperature [C]

Figure 1: Thermal Behavior of Optical Passives of Two Different Brands diagram and the relative sizes of some 2 ¼” integrated dual DWDM filters used to separate (or combine) discrete DWDM wavelengths with a shared wavelength carrying a broadcast signal (with dual filtration to eliminate ASE components and to achieve increased wavelength isolation).

Digital Technology

Digitization of the upstream bandwidth has allowed significant performance improvement in link CNR, dynamic range, Figure 2: CWDM Wavelength Filters thermal stability (refer to Figure 4), loss and 1310/CWDM Filters budget and link length (up to 200 km fiber length). Practically, digital links allow for

Pre-Split BC Input BC + NCn constant level and performance over the entire operational link loss budget and fiber DWDM INP NCn length. This significantly simplifies the link Out (BC +NC ) n alignment process. Drop NCn DWDM OUT Digital technology also allows for significant signal processing capability. Two Insertion Loss (max) or more independent signals, whether local BC to OUT 0.5 NC to OUT 1.6 or remote, can be digitized and time division NC pass-through 0.5 multiplexed to utilize fiber bandwidth more Operating Temp -40C to 85C efficiently and to share transport components (1310 nm and 1550 CWDM or DWDM lasers). Similarly, several digitized signals can be simply added together in a digital domain, thus simulating RF combining (or RF amplifier cascading). Moreoevr, manipulation of the bit stream allows for selective attenuation of the upstream path, thus improving problem troubleshooting granularity and enabling level-fine ingress mitigation without the intrusion of “wink- Figure 3: Dual Filtering NC switches”. These are examples of only the Wavelength Separator/BC-NC Combiner simplest types of signal processing capabilities that can be implemented within In parallel to the increased robustness of digitized upstream bandwidth. Additional components, innovative packaging allowed benefits include the capabilities of dedicating their miniaturization. Figure 2 shows the a communication channel for equipment size of some CWDM filters and monitoring and of adding traffic other than 1310/CWDM combining or separating the legacy RF upstream (e.g., native filters, while Figure 3 shows a functional Ethernet). 6

5

4

3

2

1

0 0511 2 2 3 3 4 4 5 Signal Strength --

- +25 +65

Figure 4: Thermal NPR Performance Stability of Cascaded Digital Links

(8 Links Cascaded)

Second 5 - 42/50 MHz

SFF SFF First Second First First 5 - 42/50 MHz 5 - 50 MHz 5 - 50 MHz 5 - 50 MHz Figure 5: An Example of Signal Processing Capability Inherent to Digitized Upstream Signals

Integration and Miniaturization Figure 7 depicts a module that integrates several modules previously available only as Packaging progress and equipment discrete parts. Moreover, it adds a integration also contribute to the capability significant functionality for level monitoring of unorthodox expansion of bandwidth and level management for remote single- capacity per user. Two examples of higher person level alignment and DWDM packaging density and integration are balancing. This integration eliminates the presented in Figures 6 and 7. need for at least three modules of significantly lower functionality, six optical jumpers and 12 optical connectors, thus lowering the cost and increasing the Miniaturization and integration trends, as reliability of the system. All of this is well as progress in optical components achieved under the assumption that level (environmemtal robustness), obviate costly monitoring and level management OTN installations by integrating entire OTN components are also eliminated from the functionality inside a single node enclosure. setup with discrete modules. To match The OTN-on-the-Strand shown in Figure 7 is completely the functionality of the integrated capable of feeding 16 nodes with dedicated module (including dual-filtering adavantages narrowcast signals while amplifying other discussed previously), significantly higher signals for more distant locations. numbers of discrete modules, jumpers and connetors would be needed.

DWDM INP

BROADCAST INP B1 #1 OUT Splitter #1

ADD

Tap#5/Pd #5 ATT#1

Tap#8/Pd #8 Tap#1/Pd #1

DROP #2 OUT

ADD BROADCAST INP A ATT#2 Splitter #2

Tap#9/Pd #9 Tap#2/Pd #2

DROP #3 OUT

Tap#6/Pd #6

ADD

ATT#3

Tap#10/Pd #10 Tap#3/Pd #3

BROADCAST INP B2 DROP #4OUT Splitter #3

ADD

ATT#4 Tap#7/Pd #7

Tap#11/Pd #11 Tap#4/Pd #4 DROP DWDM OUT

Figure 6: Integrated BC Splitting, DWDM Filtering and BC-NC Combining Module with Level Monitoring and Level Management Capabilities

+1.4 dBm

8.5 dB (cascadable) Digital 1x1 Node Digital Node 4

3.8 km/1 dB OP31S2D-55

-3.1

-3.8 OP31S2D-70

-2.1 OP31S2D-55 2.4 km/ -5.6 Hub on the Strand 1.6 km/ 0.65 dB 0.45 dB -3.1 OP31S2D-70

-2.1 -3.8 -5.6 Pad 0.5 km/ +1.5 +1.7 DWDM NC Signals accordingly 0.15 dB dBm 8.2 dB dBm from Remote FA4517S 8.4 dB -1.1 Headend OP4508-C -8.5 -7.1 Digital 1x1 Node Digital 1x1 Node (cascadable) (cascadable) FA4517S Node 2 Node 3 0.8 dB 0.8 2.9 km/ 2.9 OP31S2D-85

+1.25 dBm 1550 nm BC Signal +6.8 dBm from Remote FA4517S -10.2 8.65 dB (cascadable) Headend Digital 1x1 Node

-0.9 Node 1 0.4 km/ +7.5 dBm OP31S2D-90 DWDM Upstream 0.15 dB Signals from Remote FA4517S Nodes to Headend

OP31S2D-85

-1.1 To and From DWDM Upstream Remote Nodes Signals from Local -8.5 DX4505-31 DT4010-S Nodes to Headend

Figure 7: An Example of a Hub-on-the-Strand Configuration

TECHNOLOGY FACILITATES additional fiber or wavelength as long as the INCREASED CAPACITY PER USER node platform can be provisioned for separate upstream paths. If the platform does The technology capabilities so far not allow this, one alternative could be to described translate into capability for replace the platform while preserving the increasing bandwidth per user at significant investment in optical cable and RF coax capital and/or operational savings. The network. This translates easily into architectural implementations described significant capital savings. Figure 8 shows a below confirmed the saving estimates. conceptual diagram of node segmentation with application of digital technology. This Upstream Bandwidth Segmentation segmentation option becomes more complicated when node buses are not In areas of large nodes with reasonably balanced. In such cases, a different approach balanced buses, the simplest segmentation must be applied. One option is to segment option is to segment the upstream bandwidth. the buses with express coaxial cable. Digital technology supports two- to up to Another option is to push fiber deeper and four-play segmentation without the need for add a node. Figure 9. example ofsuchasystem ispresentedin excessive requirements onfibercount.An double theexistingnodecountandsoput count forinteractiveservicesmay easily outside theirboundaries,theadditionalnode nodes orinareasgrowingrapidlywithin in anareapreviouslyupgradedwithlarge problem ifonlyfewnodesaretobeadded, count. While fiberavailabilitymay notbea number offiberspernodeandhousehold guidelines includerulesonthepermitted is used.Mostoftheoperators’HFCdesign headend ifatraditional1310nm technology needed betweenthenodeandhubor with aseparatenode,additionalfibersare would requireaseparatenode.However, From FiberScarcitytoPlentiful Headend area BC from Number Near Table 2:UpgradeAreaStatistics One ofthenodesegmentation options BUS 1 + 2 + 3 2+ 1 + BUS + 4 Figure 8:UpstreamSegmentationwithDigitalTechnology Nodes Summary StatisticsfortheImplementation Area 2 0 4 . m(. B 19.4km (5.8dB) 0.4km (0.1dB) 34 30 22 of From Single Upstream Segment to Segment Upstream Single From Two Upstream Segments Number of Segments Future

L H L H L H L H

BUS BUS BUS BUS lowest loss (1550)from 4 3 2 1 Shortest distance /

BUS 1 + 2 BUS + 1 Keswick BUS 3 + 4 BUS + 3 available forotherservices. existing nodeswereactuallyrecoveredand segmentation. Thefibersusedtofeedthe dropped tobetween35withfullplanned number offibersonanytheroutes “D architecture (distributedDWDM, or route) wereredesignedwith“fiber-save” adjacent areaextendingonthesame cable one depictedinFigure9andasecond 1310 nm lasertechnology.Twoareas(the with atraditionalHFCarchitecturebasedon the number ofavailablefiberswhendesigned existing opticalcablesignificantlyexceeded fibers requiredinthemain routewiththe future growthofservices,thenumber of of newnodesaddedandprovisionedfor are presentedinTable2.With thisnumber 2 WDM” architecture).Therequired The statisticsfortheexample ofFigure9

MUX A/ A/ D D highest loss (1550) Longest distance / from Keswick L H L H L H L H BUS BUS BUS BUS 4 3 2 1

125298 New 19.2 km 125302 New 125294 22.4 km Future 125294 13.7 km Future 11.1 km 125296 125300 143076 New New Existing 125294 16.3 km 21.3 km 24.8 km Existing 125304 12.5 km Existing 24.3 km 125292 Existing 8.7 km

125290 Existing 5.6 km

25.2 km 125288 Existing 4.3 km

125280 New 1.2 km Expanded Area

125286 Existing 2.3 km Keswick Headend

125282 New 2.1 km

125284 Existing 0.5 km 125278 New Legend: 3 km 125276 125274 New headend New 4 km 1.7 km existing node

125270 abandoned existing node New 3.5 km future node

125268 new nodes New 3.9 km old cable 125272 125266 Future New 5.3 km 4.4 km new aerial cable

125260 125264 new underground cable Existing New 125258 5.6 km 4.2 km 125262 New Future 6.4 km 6 km splice enclosure (new to old cable)

125256 New 7.9 km 125254 New 8.5 km

139323 New 8.2 km

Figure 9: Upgrade Area Node and Fiber Route Configuration

125280 New 1.2 km

2 fibers new cable Hub-on-the-Strand 125282 New BC NC Rev 11 fibers 6 351 22413 231 old cable 2.1 km 90/10 85/15 80/20 Ch. 25 3 fibers

n n old cable BC +BC NC Drop NC 85/15

90/10 80/20 C+NC BC + NC Drop NC BC + NC Drop n n n n n n NC NC BC + Drop Ch. 27 Ch. 21 Ch. 23 70/30

Future second segment 70/30 of 125284 node 4 fibers old cable 13 fibers old cable 3 fibers new cable 3 fibers 50/50 new cable 50/50 BC + NCn Ch. 29 125284 Drop NC n 125278 Existing New 0.5 km 3 km

Ch. 31

nNC Drop

nNC BC + 3 fibers new cable Figure 10: Detailed Area Segment Configuration

The robust optical passive components with a strand mount enclosure to feed the already described in this paper have been upgraded area from a distant consolidated successfully deployed in the field for over a headend. Upon the completion of the year. The technology and configuration is project, substantial capital savings were illustrated in Figure 10, where a segment of realized. This project, initially budgetted at the upgraded area (encircled in Figure 9) is $1.2 million (for two adjacent communities), detailed. Moreover, with the miniaturization was completed for approximately $600 achieved for a Hub-on-the-Strand thousand, thus achieving a 50% capital cost configuration, the area hub could be replaced reduction.

Bandwidth Wealth Several implementations in North America and abroad have confirmed the The same technological wonders can be model numbers and expectations. An applied to new builds, rebuilds and extensive example of a system upgraded with this upgrades areas when the upgrade activity is architecture is presented in Figure 12, with a driven by the need to upgrade the coaxial detail design of a selected area presented in part of the network from sub-450 MHz to Figure 13. Table 3 details cost comparison 600+ MHz bandwidth. In this case, the between the deployed architecture and the digital and optical technologies permit HFC architecture after project completion. driving fiber-to-the-last-active (FTLA) at The final cost was within a few percent of lower cost than or at cost parity with the cost estimates. traditional HFC new-build, rebuild or upgrade, and several areas have been built with this approach in mind. An example of the optical design for FTLA architecture is presented in Figure 11.

1,276 Ft

2,007 Ft

1,277 Ft 1,261 Ft

Grp A 361 Ft s s Splice Closure 2,050 Ft s Grp B 483 Ft Legend 1,685 Ft Broadcast Fiber s Return Fiber s s Fusion Splice Optical Coupler Fiber Node

Figure 11: An Example of Optical Design for Fiber-Deep or FTLA Architecture

Figure 12: An Area Upgraded with FTLA Architecture

Figure 13: Detail Design of the Framed Segment

Table 3: Cost Comparison after Impelemntation Category HFC700 FD50 HFC$/Mi FD$/Mi HFC$/HP FD$/HP

Headend Optics $45,385 $135,773 $319 $955 $7 $21 Field Optics $39,050 $425,818 $275 $2,996 $6 $65 Fiber Cable $197,314 $410,813 $1,388 $2,890 $30 $62 Power Supplies $105,000 $89,554 $739 $630 $15 $14 RF Electronics $492,358 $14,450 $3,464 $102 $74 $2 Passives $93,323 $84,853 $657 $597 $14 $13 Coax $842,737 $864,260 $5,929 $6,080 $127 $131 Hardware $469,331 $422,770 $3,302 $2,974 $71 $64 Labor $418,410 $329,638 $2,944 $2,319 $63 $50 Subtotal: $2,702,908 $2,777,938 $19,015 $19,543 $407 $421

Variance $75,031 $528 $15

The following list summarizes the In this architecture, segmentation steps experience of the local crew during the first are practically limited only by the bus size of year of operation: tens of homes. Figure 14 shows the • capital expenses at or close to cost parity segmentation steps that allow segmentation with traditional HFC-1000 network of the node cluster down to the individual • remote monitoring from all nodes at no node, thus increasing the already expanded additional cost bandwidth per user by a factor of 5 to 8. • higher B/W per home passed than in Again, the technological advances presented HFC-700 network earlier in this paper are applicable. The • lower power consumption results in segementation can be staged in any order lower operating cost (downstream or upstream separately or in • lower number of actives results in parallel) and with any granularity, and can be increased reliability implemented on an as needed basis without • remote ingress mitigation on all any intrusion into the coaxial part of the individual nodes (approximately 100 network or network powering. homes granularity) • digital return from all nodes.

OP31S2D-EQ Forward 1550 BC + ITU OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx

1550 NC, or 1550 BC + -3.4 1310 NC or 1310 (BC+NC)

Reverse is either 1550 (ITU DWDM< or ITU CWDM or 1310 or 1550

digital Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable)

One spare fiber used for segmenting forward

OP31S2D-EQ Forward 1550 BC + ITU OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx

1550 NC, or 1550 BC + -3.4 1310 NC or 1310 (BC+NC)

Reverse is either 1550 (ITU DWDM< or ITU CWDM or 1310 or 1550

digital Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable)

OP31S2D-EQ Forward 1550 BC or 1310 OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx BC -3.4

Forward ITU 1550 NC (number of wavelengths C+NC + BC C N Drop C+NC BC + NC Drop C+N BC + NC Drop NC BC + N op Dr C+NC + BC NC p Dro C+NC N + BC C N op Dr C+NC BC + NC Drop C+NC BC + NC p Dro n n C C n n n n n n n n n n n n n increases with n segmentation)

Reverse is either 1550 (ITU DWDM< or ITU CWDM or 1310 or 1550

digital Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) OP31S2D-EQ Forward 1550 BC or 1310 OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx BC -3.4

Forward ITU 1550 NC (number of wavelengths C+NC + BC C N op Dr C+NC BC + r C N p Dro C+NC + BC C N op Dr BC NC Drop C+NC BC + C N Drop C+NC BC + C N Drop +NC BC NC Drop NC BC + NC Drop +NC n n n n n n n n n n n n n n n increase with n segmentation) OP31S2D-xx

Reverse is either 1550 (ITU DWDM< or ITU CWDM or 1310 or 1550 digital Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node (cascadable) with TDM module (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable)

OP31S2D-EQ Forward 1550 BC or 1310 OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx BC -3.4

Forward ITU 1550 NC (number of wavelengths C+NC BC + C N Drop C+NC BC + rpNC Drop BC NC Drop NC BC + NC Drop C+NC BC + C N op Dr C+NC BC + C N Drop +NC C+NC BC + NC Drop C+NC BC + C N p Dro n n n n n n n n n n n n n n n increase with n segmentation)

Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Digital 1x1 Node Reverse is either 1550 ITU (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) (cascadable) DWDM or ITU CWDM digital -3.4 OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-xx OP31S2D-EQ Figure 14: Staged Segmentation of Fiber Deep (FTLA) Node Cluster from Single Downstream and Upstream Segments to 8x8 Downstream and Upstream Segments

CONCLUSIONS the benefit of service providers. This paper described but a few examples of how today’s The optical and HFC designer toolbox is optical and digital technology can help in full of wonder tools. Although these tools increasing bandwidth per user (resulting in certainly amuse us, they have very practical increased customer satisfaction and the applications in broadband HFC networks continuing competitive advantage of HFC where the need for increased bandwidth per network operators) while lowering capital user is looming. Whether for significant and operational expenses and taking full capital expenditure avoidance or for advantage of the investment borne during significant operational benefit, these tools bandwith upgrades of the RF coaxial part of can be used by the designer to the benefit of HFC networks. broadband service customers and hence to

THE EFFECT OF OCAP MIDDLEWARE ON ITV APPLICATION DEVELOPMENT

Eric Miller Vidiom Systems Corporation

Abstract vehicle for the delivery of entertainment and information in our homes. Having a Cable networks have been active in laying nationwide standardized platform for the the groundwork for the interoperability of development and deployment of new services applications for interactive television. By promises to fundamentally alter the upgrading their physical plants, deploying economics of deploying those services to the more powerful set-tops, and pushing for both public, as much as the standardization of hardware and software standards through home personal computers created the basis for CableLabs, networks are poised to launch the internet to explode during the 1990s. OpenCable/OCAP systems in the US. This paper will address the historic and OCAP and the OpenCable standards will emerging forces that have shaped this have a powerful effect on the development of evolution, focusing on the OpenCable applications for these networks in addition to standards in general, and the OpenCable building a strong market for the creation of Application Platform (OCAP) specifically. application development tools. STANDARDS AND INTEROPERABILITY INTRODUCTION Historically speaking, the development and

adoption of technology standards are essential For the past 15 years, we have seen the for the successful deployment of consumer promise of Interactive Television looming on electronics. Open standards create the the horizon. Cable, Satellite, and Telecom conditions for many companies to develop companies have each, in turn, experimented products which are fundamentally with different transmission technologies, interoperable with each other. This hardware platforms, and software competition drives down the costs of environments. Each successive generation of manufacturing as well as the cost of trials has provided valuable insight and publishing software and multimedia content experience so that the next step down this for those products. road seems not quite so insurmountable. Although some aspects of cable television Lest we, as technologists become enamored networks are currently standardized, US cable with technology advancements for their own operators have, with only a few small sake, it is important to remember that exceptions, built their networks on proprietary consumers don't know or care anything about (non-interoperable) systems designed and the technology that we are deploying. They manufactured exclusively by two companies: care about the services that are offered to Motorola and Scientific-Atlanta. them and how using those services changes the way that they think of and interact with This is about to change due to the their television. OpenCable standards being developed by CableLabs. These publicly available standards That said, we stand on the brink of a specify an end-to-end architecture which will radical new transformation of television as a allow any number of companies to develop it cheaper to broadcast what might be truly interoperable products for the US cable considered a plethora of niche programming, networks. such as the History Channel or HGTV.

THE COMPONENTS OF Although there are different encryption INTEROPERABILITY schemes available for use, most aspects of the video broadcast are very highly standardized. There are several key components to the MPEG-2 transport streams using QAM interoperability of products and services in the modulation schemes combine several digital US cable network: transmission systems, set- broadcast channels into the same 6 MHz band top hardware platforms, and set-top software that previously carried only one analog environments (or middleware). channel.

From an infrastructure point of view, these Organizations such as the ATSC and components are intertwined, yet each has a SCTE have dozens of standards which govern fundamental dependency on the others. everything from the wireline protocols for Building a digital two-way network was delivery of data over that network, to the size necessary before advanced service set-tops and shape of the F-connector on the back of could be deployed. Similarly, set-tops with a the set-top box. While interactive services minimum, standardized feature set were such as those being deployed today would required before a consistent software have seemed unthinkable 20 years ago, the environment could be built which takes cable industry as a whole has done a good job advantage of those features. And finally, a of upgrading their physical plant to the point common software programming environment where bandwidth no longer seems to be a is required in order to support the wide scale critical barrier for new service deployments. deployment of applications throughout the country. To bring it full circle, the These improvements required a very large deployment of those applications is what will, capital investment over the last several years, in the end, pay for the network build-out but were fundamentally required in order to which was necessary to support them. support the tier of services which are even now being deployed. Transmission Systems Hardware Platforms When I first got CableTV in my house, in 1985, my cable company (TCI) broadcast The second critical element for analog video straight to my "cable-ready” TV. interoperability is the set-top box. In the mid- There was no set-top box involved and 1990s, Motorola and Scientific-Atlanta each ordering a premium channel involved sending selected a specific CPU for their respective out a technician to remove a filter from the set-top boxes and, in turn, designed several cable demarcation box set on the side of my custom processors for control of the graphics house. Those days seem quite remote. and video displays. Applications, such as the EPG, were written in C and compiled Now, we take for granted a digital, bi- specifically for those processors and their directional network. Premium tiers of service capabilities. are authorized remotely. Digital keys for decryption of those services are sent through This choice of processors and the the network, to target my specific set-top associated choice of operating systems and box(es). The increase in bandwidth has made API libraries has locked application developers (not to mention Motorola and S-A And even though the investment in themselves) into a compatibility straight- network infrastructure and hardware jacket. The vast majority of set-tops currently platforms is significant, it is fundamentally deployed are based on 8-10 year old designs, time limited. Once the initial investment is with a feature set that is deficient considering made to build a digital network, the costs of the types of applications that we want to maintaining that network are relatively minor. deploy today. And while each of those Similarly, the costs of buying a new set-top companies has begun to innovate based on will pale in comparison to the cost of writing those early designs, each incremental software for the box, especially when there improvement implies the need to rewrite all of are dozens, or even hundreds of applications the applications which run on those platforms, which will run on that box. while also maintaining all of the old versions that run on the installed base of legacy set- This comparison is illustrated by the chart tops. below.

It is this installed base which represents the next large infrastructural investment required of the network operators, but which, thus far, has proven insurmountable. The cable operators currently own all of the cable set- tops deployed in America, leasing them to Investment subscribers for a monthly fee, built into our cable bills. With tens of millions of set-tops deployed, it would cost billions of dollars to replace them all; an investment which has, so Time far, been deferred in preference for the upgrade in the physical plant. Network Infrastructure H/W Platforms Applications

This has created somewhat of a stalemate. Hardware capabilities, in general, have While this chart is not meant to connote evolved considerably over the last several specific dollar values for these activities, the years, reflecting continual incremental essential message is that investment in improvements in CPU speed and the cost of developing new applications for a given RAM, not to mention greater competition in platform continues long after the investment the business of developing powerful for building and deploying that platform. integrated multimedia components. But much of this power remains out of reach. One may ask why we don't see this investment taking place today. Even without Software Platforms a completely standardized environment, there are surely sufficient numbers of set-tops in the The last element of interoperability for market to justify the development of interactive television is the software applications for those platforms. application development environment, or middleware. The ability of the set-top box to One factor is the high cost of entry into the run sophisticated applications is the key to application business. Most network operators differentiation between broadcast television have a mix of different set-tops in their and the advanced services that we wish to networks, sometimes upwards of a dozen deploy. different models. And even though there are TVs have even more processing power than only two primary operating systems which many older set-tops). run on those boxes, older set-tops have not always been updated at the same rate as the At the heart of each of these OpenCable new ones are being deployed. This, combined hosts is the concept of removable/replaceable with new features available on the newer security. Any discussion of hardware boxes, makes for a non-trivial number of platforms would not be complete without a hardware/software combinations to target. reference to Conditional Access (CA) technology. CA is the basis for digital video A second barrier is that an ITV application encryption and creates the fundamental basis developer must install a very expensive by which Motorola and S-A maintain control headend in order to create and test their over the cable network architecture. The FCC applications. This initial investment can often has mandated that by July 2006 set-top approach a million dollars, an insurmountable manufacturers must move the video barrier to most companies. decryption circuitry off of the motherboard and onto a removable card. CableLabs calls To be fair, however, one could argue that this a Point-Of-Deployment module, or POD. we are, in fact, seeing an increase in the The POD would contain all of the network- deployment of applications. As the network dependent circuitry and software, thus build-out in support of two-way data traffic creating the possibility of competition in has been completed and more headends are building set-tops or other host devices. being upgraded with Video On Demand servers, a growing number of subscribers are OCAP being offered their first taste of interactive television. The limiting factor for wider The third component of the solution is, deployment of advanced services is naturally, a definition for a common software increasingly the base of set-tops on which environment that the mid- and high-tier those applications must run. devices are required to run. This environment is the OpenCable Application Platform OpenCable (OCAP).

The OpenCable standards being developed OCAP is based on the Java programming by CableLabs promises to break the language. This basis has two very strong interoperability gridlock by lowering some of advantages. First -- there are thousands of the key barriers to the development of a Java programmers who will come "pre- mature cable software environment. The most trained", as it were, without having to learn a basic element of OpenCable is the fundamentally new programming language standardization of the cable data transmission and environment. Second -- Java is machine- network. This includes both in-band (IB) and independent. A Java application will run on out-of-band (OOB) traffic, as well as refer- any set-top box, regardless of the CPU or ences to industry standards (ISO, DVB, operating system; a key to interoperability. SCTE, and ATSC). Other advantages of Java include the more The second element of this strategy was general efficiencies of Object Oriented the definition of standard hardware platforms. Programming. OOP is proven to reduce OpenCable defines 3 tiers of set-tops, or more software development and testing cycles, specifically "host devices" (since some newer especially in the second or third generation of applications, as developers build up the library of Objects that they can re-use and re- This is not meant to be an exhaustive list, purpose. Similarly, Java applications are but represents a number of extensions which much less likely to "crash" a set-top. Null are likely to be found in set-tops. pointers or other software errors are caught by the interpreter, before they cause catastrophic Beyond the simple equation of failure. This is especially important for competition, is the notion that set-tops will no consumer electronics, such as a television, longer be bound to maintaining CPU where users have a much higher expectation compatibility with older hardware designs. for reliability. They will be able to make processor decisions based on emerging technologies much more In short, OCAP combines the best features easily, instead of worrying about the cost of of having a powerful programming language, upgrading and maintaining different binary with the reliability demanded for use by versions of every application. This has a rip- unskilled users. ple effect through to the cable operators as well, who will need to maintain and distribute THE EFFECTS OF OPENCABLE only one version of each application.

Hardware Design Innovation Consumer-Driven Upgrades

Defining a standard set of OpenCable hosts One aspect of the retail market effect has several immediate effects, some of which which is often overlooked, is the fact that may, at first, seem contradictory. people are used to buying new consumer electronics at a fairly high rate. The average First, and most radically, separating the computer owner, especially the coveted high Conditional Access circuitry from the core end customers, will buy a new computer set-top box sets the groundwork for a highly every few years. The same is true of DVD competitive retail market in host devices. players and increasingly, even televisions. European, Japanese, and Korean consumer Ironically, they may have the same cable set- electronics companies will be able to top box that came with their cable service aggressively enter the US set-top business. when they moved in to their house. There has This will result in lower prices and more been no rationale to upgrade even if they had innovation in design. the opportunity to do so. The availability of advanced set tops at retail, and the attraction Although standardization and design of new services which require those set-tops innovation may seem incompatible, they for operation, will act to purge the networks really are not. There are still many features of the lowest end set-tops, without the which are only optional in either mid- or high- network operators having to spend the money tier OpenCable hosts, such as: directly to do so.

• High Definition Support Better Platforms for Applications • DVR • External I/O and storage devices In terms of software development, the • DVD players or writers primary benefit of OpenCable and OCAP is • Home Networking Interfaces that OpenCable hosts are, as a whole, far more capable than the vast majority of cable set-tops deployed today. Better hardware means that software can do more. There is so little RAM available in most set-tops that most optional or creative application features trapped by the underlying Virtual Machine, are, by definition, impossible to implement. long before they spiral out of control by crashing the machine. As well, the use of Creative Innovation higher level tools and data-driven application templates means that applications will be When the resources of a platform are “pre-tested”, again shortening the limited, much of a developer's time is spent development cycle tremendously. trying to figure out how to fit in to the constraints of the environment, not creating Niche Applications innovative new designs. Not only does this result in less interesting or compelling Next, having a larger user base will allow applications, it also drives up the cost of more niche programming. When there is a deploying an application. larger pool of users, even a small percentage of those users can support the deployment of a With a common, more powerful platform profitable service. This again, should act to for the deployment of services, the application increase the number of applications being developer community can start to build on the written, by broadening the scope of successes of each other instead of continually acceptably profitable services. This effect solving the same problems over and over. ultimately results in more subscribers electing This acts to drive down their costs, and thus to upgrade their set tops for access to services, increase the number of applications available. since they are more likely to want the services that are available. Better Application Development Tools CONCLUSION Developer tools become much less expensive to build and maintain when they The groundwork has been laid for the have only one target, instead of two or three. deployment of ITV applications. The network Knowing that the network operators are all infrastructure is in place and advanced set committed to the same platform will increase tops are being built for deployment. the confidence of tool developers to invest in this development. OCAP and the underlying OpenCable standards, coupled with consumer demand for Having an open network architecture and a more powerful set-tops and services (e.g. larger number of set top platforms will also VOD and PVR), are the crucial foundations drive the development of mini-headends for for building a dynamic application developer developers to use. Compared to the cost of a community. full Motorola or S-A headend today, a developer should be able to get by with The development of these applications will spending a small fraction of that amount. Key start slowly, but gain a lot of momentum as to this is having open interfaces between the more advanced set-tops are deployed. network and set-top, and increasing the More applications means more tools. More potential customer base for those mini- tools, in turn, drive down the cost of headends. producing applications, allowing for more companies to become involved. Reducing the Testing Cycle The bottom line is that application Java applications, as with all applications development for ITV will rapidly become written in high level, interpreted languages, much less expensive, which will lead to many are inherently more secure and less prone to new applications and services. catastrophic failure. Most real problems are THE TECHNICAL REQUIREMENTS FOR MULTICHANNEL QAM RF MODULATORS

By Ron D. Katznelson, Ph.D. CTO, Broadband Innovations, Inc., San Diego CA

ABSTRACT This paper addresses the technical requirements RF Interface specifications [1] (“DOCSIS and implications for multichannel QAM RF RFI”), these specifications will form a basis modulators and upconverters and the factors that for deriving proposed requirements for the affect their proposed RF specifications. RF output multichannel devices. The basic premise power levels per channel, adjacent channel noise underlying the construction of the proposed and broadband noise levels of such a class of noise and spurious specifications for N multichannel modulators will be discussed. A spectral noise specification mask that is equivalent contiguous channels is that the resulting to that obtained from a combination of multiple composite noise performance be no worse identical masks of a standard single channel noise than that obtained by combining the allowed specifications based on power addition will be noise degradations from N single channel presented. The power back-off required for devices that are compliant with the current maintaining the proposed distortion mask levels as DOCSIS RFI for a single channel. In the a function of the number of adjacent channels second part of this paper, the required power supported by these devices will be presented. A levels from such multichannel QAM proposed multichannel downstream RF modulators are addressed. specification standard based on these results will be presented for consideration for a new 2 Noise and Modulated Distortion DOCSIS™ class of high-density RF modulators. Specifications 1 Introduction The current DOCSIS RFI specifications for a The advent of new digital services such as single channel are contained in its Table 6-15 Cable IP Telephony, Video On Demand under various categories. In what follows, (VOD) and other Interactive TV applications general specification requirements that in this places further demands on HFC’s digital author’s view should be modified even for a spectrum and, in particular, the ability to single channel operation are addressed and the increase the number of Narrowcast channel appropriate changes to account for modulators deployed in head-ends and hubs. multichannel operation are subsequently In an effort to save scarce rack space and costs provided. per channel, the cable industry is well on its 2.1 In-band (In-channel) Noise and way on deploying downstream RF modulator Spurious devices that incorporate dual adjacent channel QAM signals at a single RF output port. DOCSIS RFI entries for this category are Prototypes having four contiguous channels given in Table 6-15 as follows: per RF output port have also been introduced. Total Discrete Spurious < -57dBc However, the technical requirements for such Inband (fC ±3 MHz) multichannel modulators have not received < -48 dBc; where channel spurious and noise includes all wide scrutiny or published analysis. The discrete spurious, noise, following sections will address these matters. Inband Spurious and Noise carrier leakage, clock lines, (fC ± 3 MHz) synthesizer products, and other undesired transmitter Since RF specifications for single QAM products. Noise within ± 50 channel devices are contained in the DOCSIS kHz of the carrier is excluded.

Unfortunately, both items specified above measures in the upstream channel cannot be measured in practice without specifications in its Section 6.2.21.3. actually disabling the QAM modulation. Because much of the in-band noise, (direct We therefore propose that regardless of the spurious, aliased spurious and distortion number of channels in the device, In-Channel components) is a result of the actual specifications in the DOCSIS RFI for modulated signal activity, a “clean bill of downstream channels be replaced by a single health” on these unmodulated measures is MER specification. The MER value to adopt mostly irrelevant as an indicator for actual for such specification is that which would performance. Rather, In-channel noise result from the currently permitted degradations are best measured with QAM degradation factors found in the DOCSIS RFI. modulation turned on. Moreover, in testing Factors we consider as affecting the MER that DOCSIS QAM sources for In-Channel are already specified directly or indirectly in downstream QAM compliance with ITU-T the DOCSIS RFI are: J.83 standard [2], CableLabs’ Acceptance Test Plan calls for evaluating and measuring I/Q (a) Inband Spurious and Noise. Phase Offset, I/Q Crosstalk, I/Q Amplitude (b) Phase Noise. Imbalance and I/Q Timing Skew [3]. It is (c) I/Q Phase Offset, I/Q Crosstalk, I/Q generally recognized that, in of themselves, Amplitude Imbalance and I/Q Timing none of these attributes including the two Skew. In-Channel DOCSIS RFI specifications in the table above, can predict alone the overall Item (a) above is specified at -48 dBc in the In-Channel performance of the QAM current DOCSIS RFI and as such, it is one transmitter. While careful characterization of component that would contribute to the each of these individual attributes may be of overall MER. Similarly, because the I and Q interest in diagnosing a particular deficient symbol components are assumed to be design that may have otherwise shown to fail statistically uncorrelated, the I/Q Crosstalk an overall QAM link performance, they do not limit of –50 dB specified by ITU-T-J.83 is provide an efficient overall single measure for another component that would contribute to satisfactory In-Channel performance. the overall MER. The other factors are Furthermore, the proof of compliance with addressed below: each (rather arbitrary) specification value of these attributes is time consuming and 2.1.1 Phase Noise Effects on MER expensive, particularly for multichannel We now need to account for the effect of the devices, wherein each channel in the group permitted DOCSIS RFI phase noise on MER. would have to be so tested. It is for that DOCSIS RFI specifies the permitted phase reason that the industry embraced the noise levels in its Table 6-15 as follows: Modulation Error Ratio (MER) measure as the operative criteria that encompasses all the In- Double Sided Channel attributes in one relevant measure Frequency Offset Band Integrated (see, for example, the tutorial in [4]). Noise Power Degradations in each of these In-Channel Band 1 1 kHz - 10 kHz: -33 dBc attributes will degrade the MER and the Band 2 10 kHz - 50 kHz: -51 dBc amount of such degradation in dB is Band 3 50 kHz - 3 MHz: -51 dBc essentially the ultimate relevant criteria for QAM source performance. In fact, the current By its very nature, the lack of a specific DOCSIS RFI already makes use of MER functional dependence on offset frequency

precludes this type of integrated phase noise specification from uniquely predicting the R )0( α (3) S )( ≅ω ϕϕ ; effect on MER. However, under certain ϕ ω−ω )( 2 assumptions discussed below, realistic 0 estimates can be obtained. The dependence where ω is the carrier frequency of MER on Phase Noise statistics is 0 analytically treated in detail elsewhere [5], and Rϕϕ ( )0 α is a constant that can be wherein results were derived for MER determined by integrating Equation 3 over the measuring systems that computationally frequency band of interest and equating it to a derive parameter estimates that account for a specified (or measured) integrated phase noise linear phase trajectory, as expected in ideal value in that band. It is shown in [5] that for modulation sources having no phase noise. It such Lorentzian approximation of the phase is shown in [5] that the contribution to noise spectral density, the approximate value measured MER from a source having a so derived is given by stationary phase noise ϕ t)( with zero mean is designated by MER and is given by 2 ϕ (4) MER −1 ≈ )0( αTR ϕ 15 ϕϕ (1) MER −1 = ϕ It is generally observed that in the DOCSIS 4 T phase noise Band 2 (10-50 kHz) the spectral R )0( −= TT 2 )()]()([ dRT τττ−τ−τ− ϕϕ 4 ∫ ϕϕ density is dominated by a decline in T 0 accordance with Equation (3). By integrating Equation 3 over the 10-50 kHz range and where T is the duration of the observation equating the dB result to –51 dBc, the value of period having consecutive symbols (often -1 called ‘Result Length’) assumed to be large Rϕϕ )0( α is calculated to be 1.96 sec . For a compared to the symbol time TS and typical MER measurement condition, a result length of 1024 symbols is selected and at a ϕϕ ϕτ+ϕ=τ ttER )]()([)( is the autocorrelation function of the phase noise that can be symbol rate of 5.3 Msps, the observation obtained from the two sided phase noise record time is 193 microseconds. Inserting these values in Equation 4 and expressing it in spectral density S (ω) by the inverse Fourier ϕϕ dB, we obtain transform: (5) +∞ i τω (2) Rϕϕ ϕϕ deS πωω=τ )2/()()( ∫ ∞− DOCSIS NoisePhase Limit MER =

2 It is shown that for practical purposes, a MER −≈ [log10)log(10 TR ])0( =α ϕ 10 15 ϕϕ relatively tight upper bound for MERϕ can be dB 43 =43 dB obtained by using values of the integrated phase noise over frequency offsets that exhibit an inverse square density decline in frequency. While this result was based only on the data of Such square law spectral density decline is DOCSIS’ phase noise Band 2, it includes exhibited by the Lorentzian density, for which effects from all other bands assuming the the single sided noise power density at the behavior depicted in Equation (3) is also offset frequencies of interest can be present in the other bands. In reality, approximated by however, at lower offset frequencies, the

phase noise profile is slightly steeper than a exist simultaneously, the worst-case square law, although such lower frequency composite MERD is calculated to be 40.73 dB. phase noise deviations are mostly tracked out and thus actual results are likely to be only 2.1.3 Other MER degradation Factors slightly worse than that derived in Equation 5. Other factors that generally may affect the Moreover, this limit must be used with observed MER on a given channel within a considerable judgment, as it depends on the multichannel setting containing DOCSIS observation period T. Nevertheless, for the compliant QAM channels are as follows: purposes of this paper, we shall use the 43 dB result. 2.1.3.1 Nonlinear modulated distortions 2.1.2 I/Q Phase Offset and Amplitude from adjacent channels Imbalance effects on MER We now account for the effect of the These permitted modulated distortion and permitted DOCSIS RFI I/Q Phase Offset and noise components that invade the test channel Amplitude Imbalance on MER. The arise out of Band I and II (treated in Section dependence of MER on such factors is 2.2). Because the test channel demodulator’s analytically treated in detail in [5], where Nyquist filter provides appreciable roll-off results were derived for a composite attenuation within the band-edge where the degradation of QAM modulators’ deviation power density of these components is the from ideal phase quadrature and I/Q gain highest, these weighted components from two balance. It was obtained by assuming that the adjacent channels are unlikely to exceed –60 MER measuring instrument finds the regular dBc and are therefore negligible for our undistorted reference constellation that analysis. However, this observation is by no minimizes the mean square errors of the means a reason to forgo MER measurements distorted received symbols over the entire in the presence of these adjacent channels, as reference constellation. For distorted signals their nonlinear interaction with other channels having small deviations from the ideal within the QAM device generates additional constellation, it is given by in-channel intermodulation components that are not accounted for above and that must be 2 included in the evaluation. −1 θδ−− )cos()21(1 (6) MERD ≈ 2 2.1.3.2 Linear modulator distortions where θ is the full axis-to-axis angular including frequency response and deviation from precise phase quadrature and group delay variations where the I/Q gain imbalance in dB is defined by δ as follows: To that end, ITU-T-J.83 specifies a maximum frequency response ripple in the Nyquist band 1+ δ  of 0.4 dB peak-to-peak and a group delay not (7) I/Q Gain Imbalance = log20 10 1 2 1−δ  to exceed 0.1TS . The DOCSIS ATP provides for an "expected" spectral flatness of

+0.3/-0.5 dB at the Nyquist flat-band edge, For the worst-case deviations permitted by ITU-T-J.83, one obtains for θ =10 an MER of D 1 See Figure A8 in J.83, wherein the ripple requirement 41.18 dB and for δ =0.00289 (a gain is set for Annex A, which is included in the Euro- imbalance of 0.05 dB) a value of 50.8 dB is DOCSIS. However, none of the ripple or group delay obtained for MERD . When both degradations requirements were incorporated in Annex B. 2 See Section 2.1.6 "CMTS Output Spectrum (PHY- 06.1)" of the DOCSIS ATP cited in [3]

although it is not an express requirement of 2.1.3.4 MER measurement instrument’s the DOCSIS RFI. Uncorrected, or linear and nonlinear degradations unequalized, these linear distortions can due to adjacent channels degrade the composite MER by a few dB. This factor is not insignificant because, as we However, unlike the other factors described in note in Section 2.1.3.1, multichannel QAM the previous sections that cannot be corrected devices are best evaluated with all their by adaptive equalization available in the channels ‘on’. Unfortunately, the adjacent decoders, the linear distortion effects of the channel filtering and the dynamic range magnitudes permitted above can be virtually limitation of the measurement receiver cause eliminated by the decoder’s adaptive equalizer some MER degradations. Experience with with negligible loss of noise margin multiple independent single channel associated with the equalizer’s deviation from modulators has shown that these degradations nominal response. It is for this reason that the exhibited by high-end MER measurement MER measures recommended herein must be devices can reach 0.5-0.7 dB when MER in performed with adaptive equalization the range of 43 dB are measured. engaged3. 2.1.4 Composite MER - Adding it all up 2.1.3.3 MER measurement instrument By accumulating the MER degradations residual MER and noise floor. accounted for in Sections 2.1.1 - 2.1.2 and including the permitted I/Q Crosstalk This factor relates to the dynamic range degradations discussed in Section 2.1 above, limitation of the measurement receiver and we obtain a composite MER of approximately demodulator itself. The residual MER is the 38 dB as given in Table 1 value measured by the instrument when . feeding it with an ideal QAM source. Although an ideal QAM source does not exist, Relative Degradation Factor dB MER instrument manufacturers specify their Power In-band Spurious and Noise -48.00 1.58E-05 product’s residual MER in the 40-43 dB Phase Noise -43.00 5.01E-05 range, depending on product. Actual values I/Q Phase error & Gain Imbalance -40.73 8.45E-05 normally encountered are in the range of 46- I/Q Crosstalk -50.00 1.00E-05 52 dB. Total Relative Power 1.60E-04 Composite MER 37.95 Table 1. Accumulation of worst case permitted DOCSIS degradations forming a composite MER.

3 The practice of characterizing QAM RF modulators This means that a single channel QAM source using unequalized MER has recently received some that just meets the current DOCSIS RFI industry acceptance because it has the advantage of requirements cannot exhibit an MER higher providing a single measure for the composite than 38 dB. degradations due to channel frequency response and group delay variations. However, this measure includes the test measurement receiver distortions that are However, it should be noted that in today’s difficult to separate from the overall result. Because the full digital implementation of quadrature frequency response specifications of MER test receivers modulation, I/Q Phase Offset, I/Q Crosstalk, are of the same magnitude as those permitted for QAM I/Q Amplitude Imbalance and I/Q Timing modulators and because instrument distortion that occur in opposite direction to that of the QAM device under Skew errors are virtually non-existent. On the test can result in actual measured improvements, these other hand, Table 1 does not include the MER measures are mostly non repeatable among different measurement device’s residual MER floor as instruments.

discussed in Sections 2.1.3.3 and 2.1.3.4. Figure 1 shows the actual measured power Therefore, it would be reasonable to expect spectrum of a single 256 QAM channel and its that digitally implemented QAM modulators related noise components impinging on its that otherwise just meet the current DOCSIS adjacent channels. The nearest adjacent RFI requirements and that are measured with a region (band I) contains the highest spectral typical laboratory MER instrument having a density which in most cases is dominated by 45 dB residual MER, would exhibit the MER the steep residual spillover of the linear Root- value shown in Table 2 below. Nyquist transmit filter at the band-edge. Digital transmit filters having longer impulse response (more taps and higher gate count) Relative Degradation Factor dB Power can produce lower side lobe levels. As a In-band Spurious and Noise -48.00 1.58E-05 consideration for the implementation realities Phase Noise -43.00 5.01E-05 and costs, as shown below, the DOCSIS RFI Meas. Instrmnt. Residual MER -45.00 3.16E-05 specifications for this band permits Total Relative Power 9.76E-05 appreciably higher spectral density than that Composite MER 40.11 over the next adjacent band, and it has been Table 2 Accumulation of residual MER and permitted shown that such allowance produces virtually DOCSIS degradations except I/Q degradations. no degradations to adjacent digital or analog It is for this reason that, in summary, we channels. propose a single Inband noise and spurious specification to replace those in the current Band II in Figure 1 is dominated by ‘spectral DOCSIS RFI as follows: regrowth’ due to nonlinear modulated distortion associated with third order RF gain In-channel compression. Band III is dominated by higher 40 dB MER, measured with Spurious order odd order distortions and flat RF noise. equalized demodulator and Noise DOCSIS RFI specifies the permitted noise and spurious levels for a single channel in each of 2.2 Adjacent Channels these three bands in its Table 6-15 as follows: In order to provide the DOCSIS RFI equivalent noise mask for multichannel QAM Adjacent channel < -58 dBc in 750 kHz (fC ± 3.0 MHz) to RF modulators, we now account for (fC ± 3.75 MHz) accumulation effect of the permitted noise Adjacent channel < -62 dBc, in 5.25 MHz, excluding up (fC ± 3.75 MHz) to to 3 spurs, each of which must be < levels from multiple contiguous channels. (fC ± 9 MHz) -60 dBc when measured in a 10 kHz band Next adjacent Less than the greater of -65 dBc or - Power Density (10dB/Div) 750 KHz, Band I channel 12dBmV in 6MHz, excluding up to three Linear filter residual skirt (fC ± 9 MHz) to discrete spurs. The total power in the spurs must be < -60 dBc when each is 5.25 (fC ± 15 MHz) MHz 6 MHz measured with 10 kHz bandwidth. Band II Band III

Third Order Figure 2 shows graphically a summary of these Regrowth Fifth Order Regrowth & specifications wherein, for clarity, the mask Noise does not include the –12 dBmV alternatives or the specification language on –60 dBc discrete -15 -9 -3 3 9 15 Frequency Offset from Center Channel (MHz) spurs.

Figure 1. Adjacent noise components of a single QAM channel and their DOCSIS mask boundaries.

Power −62 −65 −73 Spectral   25.5    Density 10 10 10 < 10 10log10 +    N −+ •10)2(10  6 Desired      channel   signal Power in dBc in 5.25 MHz. 6 MHz is the 0 dB Band I Band II Reference - 58 dBc 'Other Channels' - Band III - 62 dBc Broadband noise - 65 dBc -12 dBmV Band III: (Channel Group edge +(-) 6 MHz) (- 62 dBc to - 73 dBc) to (Channel Group Edge +(-) 12 MHz) :

− 10/65 − 10/73 Frequency < { N −+ • 10)1(10log10 } Figure 2. DOCSIS spectral mask for a single QAM dBc in 6 MHz. channel. It is also proposed that for practical purposes 2.2.1 Channels Adjacent to Multichannel related to testing and documentation, rounding Sources the results derived above to the nearest integer Adjacent channel noise specifications are best dBc value would be beneficial. An example provided in dBc, as argued below. For for the noise band superposition and multiple channel operation, we adopt an calculation results for two channels is shown underlying construction of the proposed noise in Figure 3 while calculated and rounded and spurious specifications for N contiguous results up to 6 channels are shown in Table 3. channels by combining the allowed noise degradations from N single channel devices Aggregating Band I Band II Band III Broadband that are compliant with the current DOCSIS - 58 - 60 - 64 - 70 RFI for a single channel.

As discussed below, we adopt the - 57.9 - 60.24 - 64.36 - 70 appropriately conservative requirement that - 62 - 58 - 65 - 65 ignores the absolute –12 dBmV relaxing - 73 alternatives and thus all specifications are in - 73 - 73 dBc including the Other Channels’ specification that is assumed to be –73 dBc Figure 3 A dual channel example of adjacent channel with an output power level of 61 dBmV for a noise mask addition and subsequent rounding. single channel. We denote by N the number of channels in the QAM group and obtain the N = 1246 following noise and spurious specifications in Computed dBc for N ≥ 2 in the three bands: Band I -58 -57.89 -57.86 -57.83 Band II -62 -60.42 -60.02 -59.65 Band I: (Channel Group edge) to (Channel Band III -65 -64.36 -63.31 -62.47 Group Edge +(-) 0.75 MHz): Rounded  −58  −65 −73  Band I -58 -58 -58 -58  10  75.0  10 10  < 10 10log10 +    N −+ •10)2(10  Band II -62 -60 -60 -60   6      Band III -65 -64 -63 -62 dBc in 750 kHz Table 3. Adjacent channel spurious and noise masks in dBc for multichannel QAM RF modulators having N Band II: (Channel Group edge +(-) 0.75 contiguous channels. MHz) to (Channel Group Edge +(-) 6 MHz) :

It is therefore proposed that these absolute levels of QAM signals that can specifications in the form of a table or the assume a range of levels and may be much equations provided above be adopted for lower than 61 dBmV (see comments in amending DOCSIS for multichannel QAM Section 3 below), adopting an arbitrary devices. The language allowing up to 3 absolute noise floor specification lacks any discrete spurs of –60 dBc may be kept as in basis. In fact, doing so would result in rather the existing DOCSIS specifications. punishing noise levels, particularly in conditions under which power level settings 2.2.2 Other Channels and the ‘spectrally below the maximum capability are non-aggregating’ noise exclusion encountered. For example, if a single channel In its Table 6-15, the DOCSIS RFI specifies QAM devices is used at a power level setting the permitted noise and spurious levels for a of 50 dBmV, the resulting allowed relative single channel in channels other than the three noise level will be at (–12) – (50) = -62 dBc. bands specified above. The single channel For QAM channels that are set 6 dB below specification is given by: NTSC peak sync, this corresponds to –68 dB relative to the NTSC channel. If one assumes Other < -12dBmV in each 6 MHz channel, that there are 50 such combined devices channels (47 excluding up to three discrete spurs. The total MHz to power in the spurs must be < -60 dBc when supporting 50 digital channels in frequencies 1,000 MHz) each is measured with 10kHz bandwidth. above the analog tier, the resulting aggregate broadband noise floor impinging on the Because the output power level required by analog channels will be approximately at DOCSIS RFI can be as low as 50 dBmV and -68+10log(50) = -68+17 = -51 dBc, - clearly as high as 61 dBmV, the above noise an unacceptable head-end noise aggregation specification for other channels corresponds to adversely affecting the NTSC channels. relative noise levels of –62 dBc down to -73 Clearly, a relative specification in accordance dBc. Even for single channel purposes, it is with -73 dBc improves the situation in this argued that the above specification is example by 11 dB. fundamentally flawed and requires amendment. As discussed below, while the Because relatively flat broadband noise -62 dBc noise specification would be adequate components typically aggregate from multiple for spectrally non-aggregating noise combined modulators or upconverters, the components (defined below), it is much too modulated distortion components can be at relaxed for spectrally aggregating noise substantially higher level but, by their discrete components. Furthermore, while the –73 dBc nature, they spectrally peak on frequencies was probably intended for spectrally proportional to the operating frequency and aggregating noise components, it is over- thus do not spectrally aggregate from multiple specified by more than 10 dB for spectrally upconverters tuned to distinct channels. Since non-aggregating noise components. no two modulators that are combined to feed a single RF transmission link operate on the First we note that maintaining the legacy same channel, by definition, their harmonics absolute –12 dBmV limit does not make much are spectrally non-aggregating. Noise or sense based on the recognition that even the spurious components that have power levels original DOCSIS specifications’ rationale that that do not vary much as the RF modulator gave rise to the absolute level specifications is device is tuned over frequency are defined as now unsustainable. Because network spectrally aggregating components. combining practices for broadcast and narrowcast services, result in requirements for

For ‘Other Channels’, the current DOCSIS at –6 dB with respect to NTSC peak sync level RFI unconditionally excludes up to three would have to exhibit corresponding discrete spurs in a 6 MHz channel wherein the aggregate noise levels in 6 MHz not exceeding total power in the spurs must be < -60 dBc. –65+6+10log(6/4) = -57.24 dBc. The Clearly, such exclusion can be devastating if rationale, the basis, the specification and the allowed components are produced on measurement methods for such 'Self frequencies that are independent of the device Aggregate' noise attributes in digital channels tuning frequency (i.e., spectrally aggregating) have been presented elsewhere [8]. For the because the use of identical modulators on DOCSIS RFI amendment, our proposal will distinct channels would cause these noise be based on a very common analog channel components to aggregate. Thus, these boundary of 550 MHz and on having 50 components cannot be excluded unless they digital QAM channels above that frequency. are spectrally non-aggregating. Similarly, it Thus, the aggregation specification for 50 can be shown that the current noise channels tuned above 550 MHz would specification of –62 dBc for ‘other channels’ become an additional requirement for ‘Other encountered in low power settings can be Channel’ noise specifications as proposed accepted for spectrally non-aggregating below: spurious and noise components. Other channels Less than -73+10٠log10(N) dBc in (47 MHz to each 6 MHz channel, where N is For aggregating spurious and noise 1,000 MHz): the number of QAM carriers in the components, a –73 dBc specification for a group, excluding second and third single channel would naturally translate to harmonics, the power in each excluded harmonic must be –62 -73+10٠log10(N) dBc for an N channel QAM dBc, and excluding up to three RF modulator device. Notwithstanding, this discrete spectrally non- specification might be argued to still be aggregating (carrier frequency related) spurs. The total power in marginal for protecting the analog NTSC the spurs must be <-60 dBc when service. While reducing (tightening) this each is measured with 10kHz specification level further on every channel bandwidth. Noise in such Other Channels self-aggregated from 50 would achieve the protection goal, it will, no equal power contiguous channels doubt, raise substantially the cost and size of situated above 550 MHz produced QAM RF modulators and upconverters. by the subject device on any 6 MHz channel below 550 MHz shall Rather, while maintaining the –73 dBc not exceed –58 dBc. requirement, broadband noise in ‘Other Channels’ should also be specified in the 3 RF Power Levels aggregate not to exceed certain levels when In adopting standards for the RF power levels aggregating multiple identical devices tuned to of QAM devices that can transmit up to N distinct frequencies. contiguous channels, we make the following assumptions: It has long been an industry practice to specify broadband noise of CATV modulators and (a) The unit may be configured to operate in upconverters in such a manner [6]. Most fewer than N channels and even in a single MSO’s have adopted head-end guidelines for channel mode. It should be power such aggregate noise floors at NTSC SNR of efficient in each of the channel setting 65 dB [7]. The NTSC SNR is defined with configurations. noise over a 4 MHz bandwidth per NCTA’s (b) The unit must comply with the recommended practice. To protect the analog multichannel specifications proposed channels accordingly, QAM channels inserted

herein including the adjacent channel In order to further evaluate the adequacy of modulated distortion noise mask. the resulting levels we propose, we will be (c) The output levels supplied by the unit at well advised to consider two factors relating to each channel number configuration should (i) actual products having multiple channels be sufficient for most head-end and hub and (ii) actual levels required in combining configurations. networks.

Thus, it is assumed that proper QAM RF Addressing the first factor above, an indicator modulator design would fully utilize the for a need of a product category or class is its available dynamic range subject to the prior vetting and adoption by MSO's through constraints imposed by the DC power product purchase decisions and deployment. consumption, and that such full utilization For example, many of the single channel leaves no excess margin for operating in the current RF specifications of DOCSIS were single channel mode. It follows that in order based on the then existing and deployed to be able to use the same hardware platform upconverters' capabilities. As currently under multiple channel settings, the required proposed the power limits for 1 and 2 power back-off from that used in a single channels appear to match that of generally channel would have to be that which preserves accepted dual channel QAM product the relative distortion as indicated above. specifications that have been recently approved and deployed by MSOs. The Unfortunately, the required power back-off as majority of these offer a maximum level not a function of the number of activated channels exceeding 56 dBmV per channel. It is argued is more than that which is loss-neutral when that there have been substantial investments in compared to practical binary passive design and optimization of such dual channel combining. Based on extensive experimental products that can benefit from such standard. results, it was found that the power back-off rate that preserves the relative distortion levels In order to address factor (ii) above, it will in most active RF stages, is up to 5 dB back- now be shown that the slightly lower power off in going from 1 to 2 channels and up to an levels we propose in our table would still additional 4 dB is required for another leave sufficient margin for all multiple doubling from 2 channels to 4 channels. channel configurations when applied in actual Thus, dual channel operation should be set at a head-ends and hubs. Multiple channel devices maximum level of 56 dBmV and 52 dBmV are predominantly deployed for Narrowcast for 4 channels, providing an optimal DC digital services and not Broadcast digital power consumption under both channel services. As such, unlike Broadcast RF number configurations as well as a single modulators, signals from RF modulators channel.. The proposed power levels are configured for Narrowcast applications shown in Table 4 below. traverse much less combining losses and are subject to very little splitting losses. Given # of QAM Carriers per Electrical RF Output that Narrowcast laser transmitter drive levels 1 2 3456 for digital QAM carriers are between 10 Output Power dBmV to 20 dBmV (see Table 5 below), most Per Channel 61 56 54 52 51 50 reasonably configured Narrowcast combining (dBmV) networks would require RF QAM levels that are no higher than 50 dBmV for single Table 4. Power level per QAM Carrier channel uncombined sources.

Input level (Per Input Return DFB Laser TX broadcast tiers [9] and much less to the NTSC Carrier) Loss Scientific-Atlanta 6473 Prisma 14 dBmV 16 dB treatment and loss budgets for the Narrowcast Harmonic PWL 4908/4910 15-22 dBmV 16 dB Harmonic PWL 4912/4914 21-25 dBmV 16 dB services. Motorola GX2-LM1000B 15 dBmV 16 dB C-COR TA3xxA-xx-3 15 dBmv 16 dB Narrowcast

Table 5. RF Interface Characteristics of commonly Narrowcast Narrowcast Typical used Nodal DFB Laser Transmitters. Note that for 40 - 50 Zone 1 Zone 1 Laser feed dBmV requirement digital 256 QAM channels, the drive level per channel To Other per QAM Narrowcast for all these laser devices is in the range of 10-20 carrier 16 Zones - 16 Combiner - 16 Combiner dBmV. Way - 4 10 - 20 - 3 dBmV - 3 To Zone For example, Figure 4 shows a Narrowcast 1 - 4 combining network serving service groups Broadcast Combiners Splitters - 28 - 3 split for two nodes (each driven by one DFB - 3 To Zone laser transmitter) wherein 16 Narrowcast RF 55-60 2 dBmV sources are combined per service group. In Other - 4 - 4 Broadcast Narrowcast Splitter this case, the required RF level from such All Zones Zones To Zone Narrowcast sources is in the range of 40-50 N dBmV, leaving significant margin if a 56 Figure 4 The RF Level Requirements for Narrowcast dBmV per channel device is employed. These QAM Downstream. Numbers shown next to passive margins will include a couple of dB for signal devices are their respective insertion loss in dB monitoring taps and/or general cable losses. The 16-way Narrowcast combiner that can Although we have occasionally encountered supply 32 channels (assuming a Dual Channel demands for levels exceeding 56 dBmV per device is feeding each port), provides ample channel in dual channel Narrowcast future expansion potential considering that applications, upon closer examination of the most Narrowcast operations today involve no purported applications, it was found that they more than 4-6 channels per service group. In were mostly due to the expediency of not systems having low Narrowcast penetrations, having to rewire cables and that there one can employ an 8-way Narrowcast appeared to be no sustainable system basis for combiner (a total number of 16 Narrowcast such requirements. While legacy wiring channels per service group) permitting preservation has some operational benefits, replacement of the two-way splitters serving Generally, our experience is that in virtually the zones by a 4-way splitter, thereby all stable Narrowcast applications, the total doubling the number of nodes per serving combining loss budget from the QAM zone, without changing the RF levels. Thus, it modulator to the laser transmitters' input does can be appreciated that certain invariance in not exceed 35 dB on a single channel basis. In total loss budgets exist over a wide variety of any event, the DC power savings and the Narrowcast applications. resultant density improvements associated with maintaining the power levels outlined in While other configurations for network Table 4 can be instrumental in sustaining very combining can be found that differ from that high density Edge QAM systems. illustrated in Figure 2, their deviations mainly pertain to the treatment of (multiple)

References

[1] “DOCSIS Radio Frequency Interface Specification”, SP-RFIv2.0. Available at http://www.cablemodem.com/downloads/specs/SP-RFIv2.0-I04-030730.pdf .

[2] ITU-T Recommendation J.83 Annex B. “Digital Multi-Programme Systems for Television Sound and Data Services for Cable Distribution”. (April 1997).

[3] See Section 2.1.4 CMTS ITU-T J.83 Annex B Compliance (Except Interleaving) (PHY-04.1) of “DOCSIS 1.1 & 2.0 RFI Acceptance Test Plan”, TP-RFI-ATP. Available at http://www.cablemodem.com/downloads/specs/TP-RFI-ATP-I04-030716.pdf .

[4] Gérard Terreault. “QAM Signal Impairments and Their Effects on MER and BER”. Sunrise Telecom, (2003). Available at http://www.sunrisetelecom.com/broadband/QAM_Impairment_Effects_on_MER_BER_104. pdf

[5] Ron D. Katznelson, “The dependence of MER on Phase Noise and other Modulator Impairments in QAM Channels”, Submitting to IEEE publication (2004). Available at http://www.broadbandinnovations.com/pubs/frm-pubs.html.

[6] See the 65 dB aggregate SNR on the front page of Motorola’s Commander® 8 modulator specifications at http://broadband.motorola.com/catalog/product_documents/C8M_Commander_Modulator_N TSC.pdf) and that of Scientific-Atlanta’s Continuum® modulator at http://www.sciatl.com/customers/Source/750885.pdf

[7] Oleh Sniezko, “Multi-Layer Headend Combining Network Design for Broadcast, Local and Targeted Services”. NCTA Technical Papers, pp. 300-315, (1997).

[8] S. I. Rubin, A. Basovich, B. Petrovic and R. D. Katznelson, "Specifying and Measuring Aggregate Broadband Noise of CATV Modulators and Upconverters" SCTE Cable-Tec Expo®, San Antonio, TX, June 5-7, (2002). Available at http://www.broadbandinnovations.com/pubs/frm-pubs.html.

[9] See, for example, the configurations described by: Jerry K. Thorne, “Using RF Amplifiers to Optimize Headend and Hub Site Isolation”. Quality RF Services, Inc. Available at http://www.atxnetworks.com/WhitePapers/RFAmp_OptIsolation.pdf

###

Author contact: For downloading any revisions or corrections of this paper go to: Ron D. Katznelson http://www.broadbandinnovations.com/pubs/frm-pubs.html Broadband Innovations, Inc. Tel: (858) 395-1440 Email: [email protected]

THE TRIPLE-PLAY NETWORK PLATFORM

Ran Oz BigBand Networks, Inc.

Abstract It is inevitable that operators will pursue growth by confronting each other and The convergence of cable services towards expanding into what had been each others’ IP has been an evolution in multiple steps legacy businesses. In cable’s case, operators building upon each other. IP technology is of are engaged in competition with satellite and course end-to-end for high-speed data services, telecommunications entities to sustain is increasingly driving network transport for all leadership in video and high-speed data services and media, and is keying the industry’s services, respectively. And, cable is expanding expansion of voice offerings. There are serious its competitive sphere with its voice-over-IP plans in place for increasing the relevance of efforts, while, at the same time, rural carrier IP to video as well, even in the home. experiments and fiber access planning by larger carriers indicate the coming The author proposes the emergence of a new intensification of video competitiveness by triple-play platform that consolidates all telecommunications operators. processing and communications functionality required for video, voice and data services, by As each type of operator pursues all growth combining the best elements of headend video opportunities, they evolve towards being services platforms and DOCSIS cable modem complete providers of video, voice and data termination systems (CMTSs). This platform triple-play services, which leads to further will be standards-based with particular advances by emerging innovative services that proficiency in IP technologies, but also support combine use of the different media. for legacy services and devices maintained in Subscribers have indicated a willingness to the network. The platform promises to key the accept all electronic services from a single emergence of far greater service functionality access network and service provider, and resource efficiency, and boost the cable especially if it holds promises of lower overall industry’s competitiveness, propelling it costs, functional enhancement and more towards being the leading provider of all convenient, consolidated billing and customer services and media to its subscribers. care.

TRIPLE-PLAY SERVICE NETWORKS Subscriber Diversification

Operator Competition The subscribers to access networks are ushering in the triple-play era by exhibiting The operators of different types of access interest in diversifying their media networks including cable, telecommunications, consumption experiences. The last few years wireless, and satellite, all face steep financial have seen music go from physical constraint pressures. They are expected to leverage the on CDs towards portable MP3 files used by investments they’ve made in their plants by the same person on an array of storage media increasing subscribership and expanding and personal devices; gaming on both PCs and revenue streams. dedicated consoles move towards wide area connectivity including voice chat among

1 players; and increasing numbers of viewers This puts cable in good stead to leverage liberating their television watching from already-made plant investments versus program schedules with time shifting achieved telecommunications handicaps in access by VOD and DVRs. bandwidth, and thus richness of media carried, and satellite challenges in two-way Subscriber empowerment trends are addressability, and thus the personalization proceeding to increasingly address the long- and interactivity of applications. considered question of whether the natural domain for all content and services is the Another key element positioning cable for television, the PC, or one for certain triple-play success is the industry’s effective applications and the other for others. The long- implementation of various standards to assure held stereotype of the TV for “lean back” interoperability among components and enable consumption and the PC for “lean forward” is the emergence of specialized best-of-breed now being challenged. Some subscribers might vendors. Several still-proprietary domains prefer one of the devices for all media, remain, as in video conditional access, but the whereas for others their preferences vary with embracing of DOCSIS, PacketCable, MPEG-2 time. video over MPEG-2 transport, and MPEG over Gigabit Ethernet for VoD, have spurred a PC video might be the right medium to more innovative industry providing more accompany routine PC-based tasks like choice and richer experiences to subscribers. checking email, whereas the TV suits family viewing. Likewise, while the PC is suited for INCREASING ROLE OF IP IN CABLE intensive data consumption, viewers are increasingly accustomed to accessing IP for All Media program-relevant data by EPG on TV, and other information, like full cast and crew IP’s role in cable is well established for listings for movies or statistics for sports is a cable modem Internet access. The industry logical extension, and is consistent with many developed DOCSIS to further standardize the people’s conduct on an Internet-connected PC offering, which led to ubiquitous availability, simultaneous with watching TV. Personal rapid subscriber growth, and open entry by mobility brings into play even more vendors of a range of innovative hardware, considerations of devices for subscribers, who software and service offerings . become increasingly empowered to drive their DOCSIS 1.1’s quality-of-service provisions own experiences. have sparked cable activity in voice over IP.

Cable Positioning for Triple-Play Subscriber interest in the service is also keen, given promise of reduced costs and expanded With video and high-speed data leadership functionality, while maintaining the quality and good empirical evidence on capabilities to and reliability expected of voice service. market voice, the cable industry has strong prospects to flourish in triple-play service To this point, video has been the one competition. Advantages include the technical medium over cable least palpably touched by combination of a high bandwidth access IP protocols. However, on closer inspection, medium along with the networking cable has begun to realize the benefits of IP for sophistication of the addressable headend-hub- video. Consolidation has expanded headend node hierarchy. service areas, in terms of both population and geography, and transport between facilities has

2 been key. This had been previously done by it still remains the case that different devices legacy protocols for video sensitivity are predominantly utilized to access particular considerations, but development of robust de- media and services. Broadcasters desire jittering algorithms has enabled UDP/IP video control over branding, advertising and access over Gigabit Ethernet to be used for transport and litigiously resist any attempts for their between facilities, even at long distances. content to move to the PC domain. Interactive television struggles that go back over a decade Gigabit Ethernet transport has multiple indicate difficulties that few are prepared to benefits for video, most especially the confront in order to bring data to the continuing steep downward sloping of the television, especially when emerging, technology’s cost curve, and the scalability primarily-video services like VOD and HDTV enabled by optical multiplexing and faster are providing new growth. Ethernet protocols. There is also an advantage gained in efficient plant management as voice Upstream of the client access devices, and data IP services are also commonly carried convergence also has yet to meaningfully over Gigabit Ethernet. And, Gigabit Ethernet’s impact the cable transmission plant either. IP functionality can be leveraged, as in Once content arrives at distribution hubs by gauging video at the transport destination, and converged transport networks, the different if there are problems with availability or services are still de-multiplexed from each quality, seamlessly switching by Gigabit other and each session is delivered to the Ethernet to an alternative source location. resources and spectrum permanently (and consequently, inefficiently) assigned to it. While IP has yet to move beyond the High speed data, video broadcasting, VOD and transport network for television consumption other services are assigned to their permanent of video, DOCSIS-driven video is also resource and frequency homes, and IP and promising. Major League Baseball has MPEG-2 content never co-mingle in the same demonstrated the viability of premium video portion of spectrum. offerings to the PC with its providing access to any game for a subscription. The Internet- This arrangement has ever-harsher accessed MovieLink offering from several consequences with increasing variety of major studios also commands attention of services available over cable. Finer and finer cable operators considering the availability of assignments are made for nichier services, and titles for their own TV-centric VOD services. there is a high likelihood that when traffic is high, there is excessive demand for some The diversification of cable data services to offerings (ex: popular, newly available VOD different speed tiers increases the imperative titles) at a time when it’s lower for others (ex: for more compelling video content available to little voice traffic late at night). PCs, in order to justify the subscription expenditure. IP video access also democratizes Such phenomena can create denials of video production, as people can post their service or compromised quality of service, personal clips online for friends and family, or concurrent with poor spectrum utilization. It even to try promoting more broadly. drives the ongoing consideration of more plant upgrades and other intensive capital Remaining Convergence Hurdles expenditure initiatives despite the competitive advantage cable should have in bandwidth While data-to-television and video-to-PC availability. More dynamic allocation of services are becoming increasingly available, spectrum to demand, perhaps through more

3 convergence of video, voice and data efficiencies by consolidating all tuning to increasingly utilizing IP, promises to remedy DOCSIS, receiving video compressed by this. advanced encoding, and transmitting upstream to the plant, both over IP. This will require Cable Initiatives headend equipment that delivers content to all video clients, beyond the scope of just the Cable’s long-standing innovative history is traditional set-top box. currently in a particularly rich period of advances across a range of services driven The proliferation of IP technologies from both by the industry as well as by subscribers. transport to transmission network and through Digital video recording is progressively to the home devices is beginning. Yet the altering how television is watched and legacy installation of MPEG-2 equipment enhancements to this service are coming throughout networks and homes, and the high along, such as multi-room provisioning investments made, necessitate that they and throughout the home from a single device. their associated techniques be maintained, DVDs, home video and digital photography even as IP spreads further. The carriage of drive multimedia PC purchases, which content and services MPEG-2 and IP formats themselves are increasingly engineered to act presents key management and efficiency as server for multiple devices, including challenges that must be addressed through televisions. And this is also a trend in which networks that are more flexible, intelligent and cable is involving itself, as a consideration of powerful across all media, while enabling the CableHome initiative. increasing leveraging of IP’s inherent advantages over time. PacketCable is expanding the industry’s device activities towards more involvement TRIPLE-PLAY NETWORK PLATFORM with telephones. Within this, PacketCable Multimedia expands the leveraging of Accruing IP Advantages DOCSIS 1.1’s quality of service for IP content from voice to greater overall richness and As cable operators expand their use of IP, interactivity, including implications for further potential advantages arise. For gaming. example, UDP/IP over Gigabit Ethernet is already commonly used for the scalable And the industry is also considering the use pumping of content from VOD servers, and of IP for video, even to televisions, by further IP evolution could lead to use of the leveraging the installed, expanding and protocol end-to-end for VOD, resulting in improving DOCSIS infrastructure. MPEG-4 more server and equipment options emerging and related encoding techniques may become from the open standards orientation. the next method of digital video broadcasting and other video services, with better device XML is an IP technology that determines economics, enhanced even more by the and utilizes profiles of subscribers, content and techniques’ bandwidth efficiency gains over other relevant entities. This could flexibly set MPEG-2. profiles on access rights to premium services or pay-per-view / VOD sessions. XML parsing Interest is increasing in the DOCSIS set-top and data transformation techniques spawned gateway, or DSG, which could further improve by its Internet origins can be carried into the economics and out-of-band spectrum

4 cable subscriber management space, allowing Such a platform can terminate distant or access, visibility and manipulation of local Gigabit Ethernet links feeding content, information in a way that was previously and when located at the edge, perform QAM cumbersome or unenvisioned. modulation for the accessing subscriber device. At first this could be done in separate In addition to operator cost savings, QAM outputs for MPEG-2 versus DOCSIS, subscribers also can benefit from more IP but these can be combined over the same utilization. Some might desire more high QAMs. Dynamic downstream spectrum performance gear that provides storage and allocation between MPEG-2 transport and advanced functionality locally while others DOCSIS, as well as with other service, will may desire least expensive, or most portable, allow the operator much more efficient use of customer premises equipment that leverages spectrum. Trends towards lower cost, lower the service functionality occurring upstream in power, higher density edge QAM technology the network. And subscribers might switch can be leveraged over DOCSIS and for all among methods including alternating services. The inefficiencies of fixed channel television and PC consumption of video allocation by service are overcome. depending on their situation at a moment. The resource allocation to achieve any This diversity of access methods could service is simplified on a network employing benefit operators too. As subscribers take more the triple-play platform. Communications, control and responsibility over what they use media processing and physical transmission to access services, they are increasingly likely elements are consolidated in this platform. to take retail ownership of their own devices They are programmable and incrementally for this access, and to alleviate operators from available as open pools of resources, to any current lease arrangements, which bear requiring service, so that those services are expensive amortization consequences. deployed with much less pre-set vertical integration of functionality. They can be Subscribers diversify the services they launched without necessity to precisely plan access, the media within those services, and for take rates, creating risks of wasting the devices they use. Doing so is facilitated by resources or exhausting capacity. expanding IP utilization and empowerment of the network. Overall benefits are accrued by A network deployed with all-IP equipment, both the cable network operators and the centralized by the triple-play platform, subscribers they serve. becomes much more capable of management by highly proven techniques such as SNMP. Responsive Network for IP Triple-Play The platform can also integrate openly into back office and customer relationship software Key to emergence of a triple-play network packages to simplify and automate for all media, all services and all devices is the management. Best-of-breed OSS tools that emergence of an intelligent network platform may currently provision cable modems will be that combines switching and routing with able to provision set-top boxes and other intensive media processing. This platform future client access devices. should be IP-based to take advantage of the protocol’s accruing advantages, but also More use of IP also advances better capable of supporting MPEG-2 and other resource management through to customer legacy techniques, to assure that transitions are premises equipment. The network can monitor economical and manageable.

5 devices and download necessary capabilities to the moment to dramatically boost plant maintain manageable subscriber environments. bandwidth efficiency. This can extend to having bandwidth available for commercial Improved Efficiency data and voice services during business hours and then for various sorts of home use during The triple-play platform will combine the evenings and weekends. IP’s power and best elements of headend digital video emerging DOCSIS controls can also platforms and CMTSs to maximize the dynamically balance the allocation of efficiency of a cable plant that drives upstream ports to downstream ports, which increasingly personalized and high quality can become very valuable in consideration of content. For example, agile transcoding trends like peer-to-peer traffic and multiplayer capabilities could allow all content to be gaming. accessed from a single source, and adapted on the platform for any subscribing device. Consolidation of switching and media processing for all services on the triple play Another technique which could conserve platform also can enhance the efficiencies bandwidth through the triple-play platform’s associated with enhancement of service support, is extending RateShaping techniques reliability. Powerful flexibility of the for bit rate adaptation of video streams to any platform’s components means that N+1 format or service. Currently this technique is redundancy schemes, available to any service, only economically applied to live broadcast can be achieved within the platform very content (in which one stream goes to many economically. One spare resource could apply thousands of subscribers), but with hardware to any service, rather than having spare and algorithmic improvements, it should be resources within each one. applicable to any video content, whether live or off of a server, broadcast or unicast, SDTV An intelligently designed triple-play or HDTV, MPEG-2 or IP. Bursty, best-effort platform can itself be a seat of complete data traffic is ideal to be statistically deployment efficiency, by disaggregating its multiplexed together with variable-bit video three basic stages of functionality: ingress streams in order to maximize capacity processing, Ethernet switching and egress utilization. processing. Each of these can be scaled in isolation so that the network’s resource A more IP-controlled plant can allocation conforms to exactly how it’s used. dramatically liberate conditional access to all And each can be modified in isolation as well, of these types of content. Established IP by software upgrade or modular expansion techniques like RSA key exchanges can be when possible, generally without disrupting efficiently performed as another boost to plant investments in the other elements. For performance. Authenticated firmware example, a non-blocking Ethernet switching downloads to end-user devices assure long fabric can be enhanced for continuing lives and reduce requirements for expensive improvements in speed, scale and reliability, field visits by the operator. Over time, while preserving investments already made in necessary functions like decoding schemes can well configured ingress and egress resources. be configured from the network. Or, at egress over time QAM technique could be modified from 256 to 1024 while More resource efficiency gains can be preserving the other two stages. Downstream, achieved by output QAMs that dynamically upstream and processing enhancement can be allocate their use to any service in demand at

6 by simple blade addition as required to better home. These can be dynamically scaled up if fit the specific areas’ traffic needs. popular, or down if not. Video conferencing, unified messaging and telemetry are examples Functional Enhancement of new services, driven through a triple-play platform infrastructure on which cable The triple-play platform as an anchor for operators could enhance subscriber loyalty and more IP deployment generally throughout the drive new revenues. network facilitates ongoing enhancement to the service functionality available. An early Overall, a more IP network driven through illustration, achieved differently, is in the a triple-play platform, vastly expands current collaboration between satellite and functional possibilities for subscribers. This is telephony operators that displays incoming largely, but not exclusively, driven by the caller ID to the television set. All-IP networks operator’s own services that it makes can do this even more robustly, able to available. There is also an attractive distribute such information to any connected democratization in that subscribers can device in the home. Viewers could use the leverage open, available protocols to set their open standards and rich functionality available own service environments, including retail- on their television and/or its associated devices purchased elements, and even embark on their to browse any information from the Internet own service provisioning towards other that’s relevant to what’s being watched. And subscribers, in partnership with their cable interaction no longer needs to involve the operator disruption of picking up a telephone – remote controls or other devices can drive ordering AUTHOR CONTACT something that’s advertised, or voting on a reality television show. From the network side, Ran Oz IP-related standards like IP MAC addressing Chief Technology Officer could better personalize distribution, as in a [email protected] demographically oriented, precisely targeted advertisement, with anonymity maintained. BigBand Networks, Inc. 475 Broadway New services can be rapidly launched and Redwood City, CA 94063 experimented with, either throughout the 650 995 5000 operator’s network, or by the subscriber in the http://www.bigbandnet.com

7 Yield Management: Turning Bandwidth into Bucks

Robert F. Cruickshank III Daniel J. Rice Stargus, Inc.

Abstract INTRODUCTION

This technical paper examines how Cable Operators (MSOs) around the world analyzing customer bandwidth consumption have built a far-reaching DOCSIS-based and optimizing MSO RF network capacity can infrastructure that increasingly supports voice, provide detailed insight and actionable video and data services. In the first quarter of information to increase the yield (i.e. utilized 2003 about 1.5 cable modems were installed traffic carrying capacity) and profitability of every second during business hours DOCSIS™ networks. worldwidei. By the end of 2003 there were over 30 million modems deployed This paper begins with a discussion of worldwide.ii As an industry, to maximize current bandwidth measurement approaches yield and profitability, we must understand and their challenges. An alternative is then the Customer Experience, what our discussed and results are presented—from a customers’ do to our networks, and how our software-based approach that uses the networks react to our customers’ actions. deployed cable modems and CMTSs as network 'sensors' to collect and analyze the CURRENT BANDWIDTH upstream and downstream usage (and many MEASUREMENT APPROACHES AND other variables) on a per-subscriber level in THEIR CHALLENGES hourly increments. This information is critical for enabling cable operators to assess the Consumption monitoring in DOCSIS impact of Peer-to-Peer (P2P) and other networks describes the process of measuring applications and overall network performance the amount of bandwidth resources consumed for Voice over Internet Protocol (VoIP). by each subscriber in the network, then processing those measurements in order to This paper then provides a detailed support operational, accounting or other overview of the process for optimizing business practices. There are four general DOCSIS networks to improve network steps common to any Internet-based capacity, performance, availability, and consumption monitoring technology: reliability by recommending and setting the appropriate configuration parameters for Collection: The process of accurately and each CMTS RF Interface. Given the dynamic reliably harvesting bi-directional consumption nature of HFC networks, deployment data from the network element layer on a per- experience demonstrates that a single channel host or per-subscriber basis. width, modulation and error correction configuration (i.e. a single modulation Processing: The application of algorithms profile) is not effective for all RF interfaces. to the raw collected data to ensure data- Field results show that network optimization integrity, remove measurement protocol has at least a doubling effect on the capacity overhead and compress element data into a and performance of DOCSIS networks. processed consumption detail record.

Persistence: The data storage required for layer visibility in pre- DOCSIS 1.1 and consumption information in order to support proprietary cable modem networks. business practices. This persistence timescale can range from the order of minutes to years. Vendor independent: Does not rely on any proprietary functionality in the CMTS or Presentation: The ability to present backbone network elements (switches, consumption information to various routers). Data can be collected without organizations or facilities within the OSS interfacing to network elements. infrastructure. This includes the automated distribution of consumption records in order Weaknesses: to enable billing mediation, traffic A lot of missed traffic: Although much of engineering, capacity planning, marketing DOCSIS network traffic is forwarded to/from research, abuse detection and other the Internet, a majority of traffic may remain accounting management practices. local to the CMTS, forwarded either within the same MAC domain, or to other MAC Although some technologies currently domains contained in the CMTS (e.g. on exist to address these requirements in the college campuses). This local traffic will only Internet today, almost all of them fail to meet increase as next generation DOCSIS CMTSs the unique technical challenges inherent in the evolve to higher densities (up to 100,000 monitoring of large-scale DOCSIS networks. CMs) and more applications, such as IP This section compares three methods for the telephony, peer-to-peer, and video monitoring of per-subscriber, bi-directional conferencing, run over the network. Because data consumption in order to determine the the intrusive probe typically captures traffic optimal method for DOCSIS networks. upstream of the CMTS, it cannot see this local intra-CMTS traffic. The three methods include: 1. Intrusive packet capture Scalability: Because intrusive capture is 2. Non-intrusive packet capture typically based on inserting a probe in all 3. Non-intrusive element polling possible paths to each CMTS, a 1:1 relationship between CMTSs and probes Intrusive Packet Capture results. This leads to a very high number of additional network elements making this an The intrusive packet capture method operationally complex and economically collects network usage data by use of passive unattractive method. "probes" inserted directly into the data path. All packets passing through the probes are Point of failure: Because the intrusive captured and processed in order to determine probe is inserted directly into the data path, per-host or per-user consumption. installation requires a scheduled network outage. After installation, network availability Strengths: becomes dependant on the availability of the DOCSIS protocol independent: The probe. Any redundancy or hot-standby DOCSIS 1.0 protocol does not include a solution for high availability would double the mechanism to furnish the capture of number of probes, resulting in two probes per application type (by port number). Intrusive CMTS. packet capture does not rely on the DOCSIS protocol and therefore provides application Dynamic addressing: Because the packets whether the probes’ I/O cards can handle the are captured by the probe beyond the DOCSIS faster line speeds upstream in the network segment, only a CPE's IP address can be used without turning into a bottleneck. to determine the packet's origin. Due to the nature of DOCSIS networks, CPE IP Non-Intrusive Packet Capture addresses are generally dynamic which does not create an authoritative relationship The non-intrusive packet capture method between packets and their CM origin assumes that all CMTS traffic is aggregated at (Hardware Address). In order to resolve this, an edge switch or router. The probe is integration with the provisioning system is attached to a port on the switch or router and required, adding additional cost to the promiscuously captures packets passing solution and creating an opportunity for poor through all interfaces (i.e. port spanning). usage-data origin integrity. Note that probes could be attached to the CMTS devices themselves, but this Physical co-location: This method requires configuration would result in a 1:1 that packet capture probes are co-located with relationship between probes and CMTSs, each CMTS. If the MSO supports a offering a solution that is not economically distributed HFC architecture (digital hubs, viable. micro-head ends) additional rack space, installation and maintenance costs are Strengths: incurred. Port unknown. Although application DOCSIS protocol independent: Has the layer visibility is provided through this capability to capture application layer traffic method, a large percentage of packets are not independent of DOCSIS protocol version. associated with well-known ports. In addition, popular peer-to-peer applications use No point of failure: Unlike the intrusive configurable ports making the application approach, this method does not introduce a type invisible. point of failure into the network.

DOCSIS unaware: Because the packet Weaknesses: capture is conducted upstream of the DOCSIS A lot of missed traffic: Like the intrusive segment at layer 3 and above, this method packet capture method, the non-intrusive does not provide any layer 2 (DOCSIS MAC) method is unable to view the traffic local to visibility. Both Service Identifiers (SID, the CMTS. As stated, a configuration capable DOCSIS 1.0) and Service Flows (SF, of capturing all packets would require a single DOCSIS 1.1) are invisible to the probe. As a probe provisioned for each CMTS. Again, this result, each packet cannot be associated with a would prove economically unfeasible. DOCSIS service profile or packet classifier in order to determine the quality of service it has Third party vendor dependent: Assuming been assigned. the probe is attached to the edge switch aggregating CMTS traffic for a headend, the While it is possible to move the intrusive probe depends on proprietary functionality in packet probes further upstream in the network the switch that forwards all CMTS traffic to to minimize the number of probes required, the capture probe. The extent to which all this drastically increases the amount of traffic CMTS, router and switch vendors implement that the probes miss and defeats the purpose and support a proprietary form of this of consumption monitoring. It is also not clear functionality is not certain and introduces the possible need for further development and order to determine the quality of service it has integration work. been assigned.

Network performance impact: The While it is possible to move the non- additional resource burden on the aggregation intrusive packet probes further upstream in switch to direct all inbound packets to the the network to minimize the number of probes capture port is not negligible. This effort may required, this drastically increases the amount impact the performance of the switch or of traffic that the probes miss and defeats the router's packet forwarding capability in large- purpose of consumption monitoring. It is also scale production networks. not clear whether the probes’ I/O cards can handle the faster line speeds upstream in the Scalability: Although this method requires network without turning into a bottleneck. fewer boxes than the intrusive approach, there are still a high number of probes required to Non-Intrusive Element Polling ensure that all inter-CMTS traffic is captured. Because of the limitations of both intrusive Dynamic addressing resolution: Like the and non-intrusive Packet Capture, we propose intrusive approach, the mapping of traffic to a new method known a non-intrusive element source CM requires integration with the polling. Non-intrusive element polling provisioning system and introduces potential leverages the network management data integrity issues. instrumentation already embedded in the DOCSIS network and the remote collection of Physical co-location: This method requires consumption data through the use of the that the packet capture probe is co-located Simple Network Management Protocol with the aggregation switch or first upstream (SNMP). router. If the MSO supports a distributed HFC architecture (digital hubs, micro-headends) Strengths: additional rack space, installation and No missed traffic: Unlike both intrusive maintenance costs are incurred. and non-intrusive packet capture methods, the polling method derives consumption data Port unknown: Although application layer directly from the entire DOCSIS network visibility is provided through this method, a down to the individual CPE device (CM, large percentage of packets are not associated MTA, etc.) level. As a result, there are no with well-known ports. In addition, popular missed intra-CMTS packets. peer-to-peer applications use configurable ports making the application type invisible. No point of failure: Unlike the intrusive approach, this method does not introduce DOCSIS unaware: Because the packet additional points of failure into the network. capture is conducted upstream of the DOCSIS segment at layer 3 and above, this method Scalability: Relative to the other methods, does not provide any layer 2 (DOCSIS MAC) a small number of probes are required to visibility. Both Service Identifiers (SIDs, collect data from the entire network. DOCSIS 1.0) and Service Flows (SF, DOCSIS 1.1) are invisible to the probe. As a No physical co-location: Remote polling is result, each packet cannot be associated with a conducted over IP allowing for the probes to DOCSIS service profile or packet classifier in exist at any location within the network infrastructure. This results in reduced Functionally extensible: Because of the configuration, operations and maintenance nature of SNMP and the extensibility costs. provided by MIBs, the polling method can be used to gather other operationally significant Vendor independent: Non-intrusive data from the DOCSIS network related to element polling is independent of vendor traffic consumption (interface utilization, infrastructure (CMTS, switches, routers) and etc.). does not rely on proprietary vendor functionality to monitor consumption. Weaknesses: DOCSIS protocol dependent: Application- DOCSIS aware: Unlike both packet layer visibility is introduced in DOCSIS 1.1 capture methods, the element polling method networks. Meanwhile, because DOCSIS 1.0 has been adopted by the cable industry for the limits traffic visibility to layer 2 (octets in/out monitoring of DOCSIS networks. As a result, by SIDs), polling provides visibility into layer 2 of the DOCSIS network and can monitor service Performance impact: If not engineered flows and quality of service in the DOCSIS properly, management traffic generated segment. With the adoption of DOCSIS 1.1, through polling the DOCSIS network can application layer visibility will be embedded negatively impact network performance. in the CMTS.

Figure 1: Bandwidth Utilization of Top 5 Most Congested of 88 Downstream Interfaces

Figure 2: Bandwidth Utilization of Top 5 Most Congested of 450 Upstream Interfaces CURRENT BANDWIDTH pipes can be “filled up” more often and yield MEASUREMENTS management increased. It would make business sense to find ways to “fill the pipes” Once Consumption data is collected with during off-peak periods, as Telephone non-intrusive element polling, analytics may Companies did years ago when they used to be applied to determine which downstream discount, incent and encourage off-peak usage and upstream interfaces are most congested. (in their case after 5 PM). Figures 1 & 2 show an MSO’s most congested channels; note how much of the OTHER IMPORTANT FINDINGS time the channels are way underutilized! Notice the daily peaks and valleys in Non-Intrusive element polling may be used utilization. During the hours after midnight to collect other DOCSIS parameters (in and before sunrise the DOCSIS network is addition to consumption) that further show way underutilized. Considering that Figures 1 how the network reacts to customers’ actions. & 2 show the most congested channels, every For example, every packet traveling upstream other channel is even less utilized than the from the customer typically has Forward ones shown here. In summary, there certainly Error Correction (FEC) redundant coding is a lot of “fallow” capacity throughout this applied which attempts ensure that bit-wise network (empty transmission opportunities packet errors (due to noise, etc.) can be that are not being used to transport customer “recovered” without retransmission. data). Surely the

Figure 3: Correctible Error Rate of Top 5 Most Errored of 450 Upstream Interfaces

Figure 4: Uncorrectable Error Rate of Top 5 Most Errored of 450 Upstream Interfaces The derived DOCSIS parameter known as error levels, these latent and worsening HFC Correctible Codeword Error Rate (CCER) plant problems are often undetected for shows the percentage of packets that have months until customers become exasperated been “recovered” from an errored state. with the performance of their applications. A Figure 3 shows the worst upstream CCER degraded subscriber experience results in interfaces. Notice that at times as many as operating inefficiencies such as heightened 40%–60% of all packets on certain channels customer care and network maintenance costs, were being recovered automatically by the as well as subscriber churn. DOCSIS network. By using the capabilities built into Another derived DOCSIS parameter known DOCSIS, the errors and inefficiencies can be as Uncorrectable Codeword Error Rate (CER) minimized through configuration shows the percentage of packets that have optimization, thereby allowing MSOs to defer been “unrecovered” and are forever lost in the capital expenditures, reduce operational VoIP case (or possibly retransmitted for other expenses, and ready their infrastructure for TCP/IP applications). Figure 4 shows the advanced IP-based services, particularly worst upstream CER interfaces. Any time that VoIP. any of the interfaces peaks above 3% CER, VoIP call opportunities are lostiii. Notice that Optimization Goals the interface indicated by the top (red) line in Figure 4 would not have been able to support When designing or operating digital data any VoIP calls for any subscribers during communication systems, there are several most of the first half of the week. goals that help drive system optimization:

OPTIMIZING DOCISIS NETWORKS Goal 1: Transmit as much data in the shortest amount of time possible through the system. Today many last mile networks are operating with impairments and inefficiencies Goal 2: Transmit this high rate of data using that are tolerated by email, web surfing, and as little of the physical resources (spectral other less data intensive or less time critical bandwidth and power) as possible. applications. Because of the FEC built into DOCSIS and the resilience of Transmission Goal 3: Transmit this data reliably at a much Control Protocol (TCP), these errors are lower rate of errors than will impact the largely unobserved by and unaffecting to performance or reliability of any of the broadband customers that are unaccustomed services. to the intermittent start-stop nature of web surfing and email. However, advanced, IP- Goal 4: Develop and operate this system with based services such as VoIP, video- as little expense and complexity as possible. conferencing and streaming audio/video are rendered inoperable by such errors and The challenge is that these four goals are inefficiencies. not completely independent. The error performance and the capacity of the network Error levels often get worse over time and are interdependent and must be managed are typically associated with transient noise together for a quality customer experience. and interference due to HFC plant problems. While email and web surfing mask these low Method for Optimization packet error protection, including many parameters for the downstream and upstream DOCSIS Cable Modems (CM), Multi- signal path. All these parameters are media Terminal Adaptors (MTA), Advanced dependent on one another and optimizing Set Top Boxes (ASTB), and Cable Modem them needs to be considered as a collective Termination Systems (CMTS) can all be task. For example, increasing the symbol rate utilized to detect and manage errors while without optimally setting the mini-slot size providing bandwidth intelligence data. In so and codeword structure will result in much doing, the DOCSIS network can be less capacity gain than would be expected. configured to "four wheel drive" through most Additionally, setting the mini-slot size service affecting errors – errors that otherwise incorrectly can make large PDUs impossible result in degraded or complete loss of service to transmit. to subscribers - while maintaining optimal bandwidth capacity. This can be This process is iterative and continual – accomplished while notifying operators of certain parameters such as FEC (“k” degraded network quality before it becomes codeword size and “t” correctable bytes), SLC service impacting, as it occurs, and also (Shortened Last Codeword), and max burst providing isolation and identification of the size should be adjusted and the results faults. analyzed over a period of days. After the results are known, further adjustments can be CMTS vendors ship their equipment with planned, including the analysis of modulation default modulation profiles that are extremely order, minislot size, and symbol rate. Over conservative, and significant opportunities time as RF levels fluctuate, periodic exist to reap additional capacity and error adjustments will ensure optimal network protection from DOCSIS networks based on performance. Using this technique of actual network conditions. The DOCSIS 1.0 methodical and measured change, an optimal – 1.1 – 2.0 specifications compromise a network configuration can be reasonably and progression of features that result in ever- cost-effectively attained and maintained increasing efficiency and capacity, but without undue risk of network instability, but configuring these networks’ elements has is necessary to constantly collect and analyze become increasingly complex. As a result, network data. capacity and error protection optimization techniques have become increasingly critical PROTECTED TRAFFIC AND NEWLY to successful deployments of VoIP services. AVAILABLE TRAFFIC CAPACITY

The technical expertise required to Adjusting the many DOCSIS configuration manually adjust DOCSIS parameters is “knobs” and “levers” in an automated way significant. Moreover, RF levels within HFC results in the right level of packet error networks are prone to fluctuations, both protection for all traffic and a huge gains in periodic and random, and continual use of an overall traffic capacity. Field experience automated system to monitor levels can shows that small portions of the HFC network reduce the operational expense of attaining require extreme error protection and low and maintaining optimal configurations. There transmission speeds – while the vast majority are many different “knobs” and “levers” that of the network can run at extremely high are available in DOCSIS networks that can be speeds. Figure 5 shows a relatively noise-free tuned to enable capacity optimization and interface whose configuration, over the course of a month, was changed from QPSK @ 1.6 CONCLUSION MHz to 16QAM @ 3.2 MHz. The upper jagged (pink) line represents the channel In order to maximize yield and utilization, indicated on the right hand vertical profitability, we must understand what our axis, which initially peaks at 80%-90%. With customers’ do to our networks, and how our the added capacity from optimization networks react to our customers’ actions. utilization is reduced to routinely below 25%. Non-intrusive element polling is the best way Moving downward, the flat (thin light blue) to measure the impact of customer traffic— line represents the unique CMs (subscribers) and the corresponding response of our on this upstream, the jagged (thick dark blue) networks. Measurements show that our line shows latency due to high utilization networks are often underutilized and that we (peaking every evening as expected), and as an industry have opportunities to increase lastly the lower oscillating (thick red) line yield management. shows the number of CMs (subscribers) active over time. Non-Intrusive polling can also collect DOCSIS performance metrics that can be Experience shows that virtually all utilized by an automated system that upstreams can be optimized to operate error- optimizes HFC packet error protection and free for high-quality, high availability VoIP network capacity so that VoIP works reliably service. In addition, as much as 2/3 of all HFC throughout the network. This optimization upstreams can be optimized to operate at results in overall double or better capacity much higher transmission speeds—which gains—resulting in even greater opportunities further incents MSOs to consider ways to for yield management. increase yield management.

i Source: Cable DataComm News Report, 2003 ii Source: In-Stat/MDR Report, 2003 iii "Network Tolerance of Highly Degraded Conditions for an H.323-based VoIP Service", Peter Holmes, Lars Aarhus, Eirik Maus, Norwegian Computing Center, P.O. Box 114, Blindern, Oslo, Norway

latency active-cms unique-cms util 500 100 450 90 400 Service 80 350 Peak Utilization 70 300 Reduced to 25% Slowdowns 60 250 Eliminated 50 200 40 150 30 100 20 50 10 0 0 % Bandwidth% Utilization 2/1 2/2 2/3 2/4 2/5 2/6 2/7 2/8 2/9 3/1 3/2 3/3 3/4 3/5 3/6 1/28 1/29 1/30 1/31 2/10 2/11 2/12 2/13 2/14 2/15 2/16 2/17 2/18 2/19 2/20 2/21 2/22 2/23 2/24 2/25 2/26 2/27 2/28 HTTP Response Time (msec) Time Response HTTP Opt Opt Opt Opt Opt Opt 1 2 3 4 5 6 Figure 5: Bandwidth Utilization (right), HTTP Response Time, Unique & Active CMs (left) of a Typical Optimized Upstream Interface Undergoing 6 Iterative Optimizations ZERO CONFIGURATION AND THE FUTURE HOME NETWORK

William Garrison Motorola, Inc., Broadband Communications Sector

Abstract of Windows PCs. However, these protocols were completely separate from any WAN This paper will review the state-of-the-art (Wide Area Network) Protocol and served a in Zero Configuration initiatives as it applies limited range of devices. They did not allow to an in-home network. The IETF Zeroconf, a wide variety of appliances, PCs and Macs to UPnP (Universal Plug-and-Play), Apple interoperate. Rendezvous and IPv6 Stateless Address Autoconfiguration initiatives will be covered. An additional Zero Configuration In addition, from a MSO’s perspective, does networking benefit will be the creation of new Zero Configuration represent a lost revenue kinds of networked products. These products opportunity or a lost headache opportunity? will become commercially viable only when How will the Zero Configuration home the inconvenience and support costs of network connect to the internet? traditional networking technologies are removed. INTRODUCTION This paper will cover the Zero Consumers do not want to own a home Configuration of IP (Internet Protocol) based network. Instead, they want applications and networks. IP was selected because it is the services to simply work, as if by magic. If native protocol for the ubiquitous Internet. this magic requires a network, then it should be easily invoked with, at most, a simple COVERAGE incantation. This is how the consumer sees it. This is how we need to see it to get consumers A whole home network should be to deploy applications that rely on home comprised of all of the data paths available networks. between devices. These data paths include wired and wireless technologies. Wired The purpose of a home network is to technologies include power line, phone line, enable new services by combining the coax and dedicated CAT5 wiring. Wireless capabilities of both new and existing includes 802.11a/b/g, Zigbee and UWB (Ultra elements. Why can’t I view a program that Wide Band). These technologies have happens to be stored on my PVR downstairs significantly different data rate, delay and on my TV upstairs? Why can’t I listen to the jitter characteristics. However, we need them music stored on my iPod on my home theater to interoperate seamlessly and with zero user without hooking anything up? Well, with a intervention. Zero Configuration home network, you will be able to do all these things and more! REQUIREMENTS

Zero Configuration is not a new idea. In As described in the internet draft entitled the past, AppleTalk handled Zero “Requirements for Automatic Configuration Configuration for Macs and NETBIOS of IP Hosts” by Aidan Williams [1], Zero provided similar features for small networks Configuration requires that we: • Distribute IP addresses without a to interact with a corresponding entity at the DHCP (Dynamic Host Configuration same layer in a remote host. Zero Protocol) server Configuration applies to the Application and • Provide name resolution without a Network layers, as outlined in Figure 1. DNS (Domain Name System) server • Find and list services SECURITY • Distribute multicast addresses Security is also a Zero Configuration In addition, while the system must operate requirement. In the wired world, there is in the absence of DHCP and DNS, it also some physical security in a local network. In must operate properly in their presence. Zero the wireless world, it is a challenge to tell the Configuration must not thwart their normal difference between your wireless device and function. your neighbor’s. Zero Configuration may bring tighter security requirements. You cannot depend on the user noticing an Application intrusion because the user is further removed Name Resolution / Find Services from what is actually going on. At the very least, the protocols used on a Zero Configuration network must be as secure as a Presentation non-Zero Configuration network.

Does Zero Configuration degrade security? Not really, a cracker can find out the same Session network information as offered by Zero Configuration using standard tools. There are plenty of tools floating around that let the unsophisticated “script kiddies” get into Transport lightly protected networks. Zero Configuration’s most likely security effect is to increase the number of networks that Network someone might try to intercept and increase IP unicast/multicast addresses the number of devices connected to those networks.

Data Link ZERO CONFIGURATION APPROACHES

Many people are currently trying to solve the Zero Configuration problem. Current Physical initiatives include: • Apple Rendezvous

• IETF Zeroconf Figure 1 – The OSI Reference Model • UPnP™ (Universal Plug-and-Play) The OSI Basic Reference Model divides a • IPv6 Stateless Address networking system into seven layers. This Autoconfiguration layering system enables an entity in one host Each of these will be covered in its own Vulnerability to denial of service attacks is section. probably unavoidable. Even a simple ploy, such as a rogue device responding to every IETF ZEROCONF ARP (Address Resolution Protocol) so as to claim all available IP addresses, could shut The Zeroconf Working Group of the IETF down a Zeroconf network. However, is this (Internet Engineering Task Force) was any worse than being unable to speak to the chartered in September, 1999. Its goal is to person next to you on the train due to an “enable networking in the absence of unruly child? Given local nature of a configuration and administration.” Their goal Zeroconf network, the universe of people who is so inclusive that it goes as far as to include could interfere with your network is small. allowing “impromptu networks as between Given the local nature of the network, in the the devices of strangers on a train.” [2] case of a network interruption you could simply search for the scoundrel. Zeroconf is a link-local technology. This means that the link-local addressing and Industry Support naming are only meaningful to devices directly connected to the local network. A standard needs industry support to have Because these addresses and names are not real-world relevance. Support for Zeroconf unique globally, Zeroconf only applies to has been announced by: small wired or wireless networks. Zeroconf is • Apple appropriate for: • Canon • Home and small office networks • Epson • Ad hoc networks at meetings and • HP conferences • Lexmark • Two devices needed to share • Philips information • Sybase Inappropriate applications of Zeroconf would • Xerox potentially result in serious networking • World-Book problems. Working Group Status Security The Zeroconf Working Group could not Zeroconf security is primarily based on the reach a consensus on security and service requirement that all included devices must be discovery issues. Therefore, it is not going to connected to a single link. Therefore, a produce a specification on those issues. The Zeroconf connection can only be hacked by a Working Group is producing a protocol device that is close by and easier to detect. specification, describing automatic generation However, if you are indeed going to and assignment of link-local IPv4 addresses in wirelessly network with the stranger on the environments lacking host configuration train, you will want some way to prevent (static or using DHCP). This document is in networking with the stranger in the next train draft form and will be submitted in the spring compartment. of 2004 for consideration as a Standards Track RFC. A Zeroconf network is relatively vulnerable to some fairly standard attacks. Further information on the IETF Zeroconf Working group can be found at http://www.zeroconf.org/.

APPLE RENDEZVOUS

Rendezvous is Apple’s name for IETF’s Zeroconf. You might want to think of it like Rendezvous is to Zeroconf as Firewire is to 1394. Rendezvous is an open protocol, which

Apple has submitted back to the IETF as a part Figure 2 – A Rendezvous Assigned IP Address of the ongoing standards creation process. Apple is using Rendezvous to transition from In the above figure, you can tell the IP AppleTalk to an all-IP network. address came from Rendezvous both by the 164.256/16 address and the annotation “Self- Rendezvous matches the Apple customer’s assigned”. expectations of a friendly, easy to use system. Apple added Rendezvous services in the Apple then uses mDNS (multicast DNS) to Jaguar release of the Mac OS X Operating handle DNS requests. In mDNS, each device System and is using those services in its own runs their own mDNS responder. The mDNS applications. responder provides traditional domain name services by having every device respond to Rendezvous uses three technologies the name queries that they know how to • Automatic IP addresses (IPv4) answer. Traditionally, devices contact a single • Name to Address Translation (DNS known DNS server for name lookups. In queries via IP Multicast) mDNS, when a host needs to look up a name, • Service Discovery it sends the query out to a local multicast group that includes all of devices that have A Rendezvous device first tries to obtain locally registered Rendezvous services. an IP address by a standard DHCP request. If there is no DHCP response, it: Rendezvous also uses a DNS-based • picks an address at random in the Service Discovery called mDNS-SD. 169.254/16 range Essentially, services are resolved to devices • Checks to see if this IP address is used similar to the way host names are resolved via an ARP (tries another IP if it is) using mDNS. • Periodically checks for DHCP server Rendezvous Applications The device periodically checks for a DHCP server because it wants to participate in the Apple uses Rendezvous in its iChat instant network with the widest possible reach. messaging application. Besides working with AOL Instant Messenger, it also works with Rendezvous-enabled Macs. So, if the stranger on the train has a Mac, you can easily chat. And if you change your status from (Simple Service Discovery Protocol) to find “Available” to “Away”, all Rendezvous an interesting device with an interesting clients are notified of the change service on offer. This is accomplished using a automatically as a part of the Rendezvous multicast search message (HTTP over UDP mDNS Service Discovery. over IP). Replies are unicast to the requestor. The multicast address, as well as the Further information is available at mechanism for advertising, searching, and http://developer.apple.com/macosx/rendezvous. revoking, are defined by the SSDP.

UPnP™ If a Control Point wants to know more about the services offered by a device, it UPnP is being developed by the UPnP requests an XML format description Forum. The Forum was formed in 1999 and document. The XML document describes the now consists of over 650 member companies. device and all its embedded devices. The The primary purpose of this Forum is to description includes services supported by the produce DCPs (Device Control Protocols) that device, manufacturer information, version of describe standard methods for device the device, device web site, serial numbers interaction. UPnP is based entirely on open and other relevant information. standards such as IP, TCP, UDP, HTTP and HTTPMU (a variant of HTTP that works on The Control Point can now access the top of UDP multicast). advertised services on the destination device What is universal about UPnP? UPnP uses via the SOAP (Simple Object Access common protocols rather than vendor-specific Protocol). For a control point to invoke an device drivers. UPnP is independent of the action on device, it must get the device physical media and can be implemented in address, discover the device, retrieve any programming language and on any descriptor, get URL for control and then send operating system. The basic foundation of actions. UPnP is somewhat unique in that UPnP is a client-server architecture, where the control is included as a part of its Zero client is called a “Control Point” and the Configuration standard. server is called a “Device.” UPnP also supports Eventing, where a Operation Control Point will be notified of device state changes. In order for a Control Point to UPnP covers the following device operations: register for Eventing, it must get the IP • Obtaining an IP address address of the device, discover the device, • Discovering other devices retrieve the device description, get the URL • Controlling other devices for Eventing and then subscribe to the events • Receiving state change notifications from device. The subscription must be for all (Eventing) events on the device. There is no way to • Presenting User Interface for other subscribe to just a single type of event. devices While Rendezvous does not explicitly A UPnP device obtains an IP address the support Eventing, similar features are same way as described for a Rendezvous provided through its mDNS-SD service. device in the previous section. Once it has an IP address, a device will use the IETF’s SSDP Security Windows XP and Windows ME provide various levels of UPnP support. More UPnP does not directly specify any information on UPnP is available at security measures in the basic protocol. The http://www.upnp.org. basic protocol relies on the security features in the standards-based protocols on which it is IPv6 Stateless Address Autoconfiguration based. In addition, UPnP is seen as relatively secure because it sends only data and keeps IPv6 is a redesign of the original Internet the implementation private. Because no protocols. Most often you hear about how executables are exchanged, there are fewer IPv6 supports a larger (128 bit addresses vs. security concerns. 32 bit) address space. This removes the need for NATs (Network Address Translations) UPnP has recently (Nov 2003) added a and private addresses, so you have end-to-end Device Security standard. UPnP security adds transparency. However, it also includes a Access Control Lists and a Security Console number of Zero Configuration features that which runs on Control Points that lets you edit are well suited for the home environment. In Access Control Lists. Security is controlled addition, IPv6/IPv4 translation mechanisms down to the Service level. So, a Control Point let us take advantage of these features before might be able to set a clock’s alarm but not its the whole world transitions to IPv6. time. While this adds security, it requires significant manual intervention and therefore IPv6 Autoconfiguration requires a does not qualify as part of Zero multicast-capable link and begins when a Configuration. multicast-capable interface is enabled. When a device interface is enabled, the host will Compatibility generate a link-local address. The link-local address is sufficient for communication UPnP and Apple Rendezvous use among nodes attached to the same link. The essentially the same link-local address Link-local address is constructed by specification. In both protocols, the IP appending the well known local prefix address 192.164/16 is understood to be a link- fe80:0000:0000:0000:0000: to the device’s 64 local address. The Rendezvous version is bit interface ID. For Ethernet, the interface based on a slightly newer version of the RFC ID is based on the 48 bit MAC address and than the one used by UPnP. Therefore, UPnP generated according to IEEE EUI-64 and Rendezvous devices can exist on the same (Extended Universal Identifier). network. The differences are Rendezvous can communicate with devices with routable Before any address can be assigned to an addresses and Rendezvous uses a packet TTL interface and used, however, a node must (Time to Live) of 255 while UPnP (as attempt to verify that this "tentative" address implemented by Microsoft Windows) uses is not already in use by another node on the 128. The routable address feature is just an link. To do this, it sends a Neighbor added benefit and the TTL can be handled by Solicitation message containing the tentative changing the default TTL value. address as the target. If another node is already using that address, it will reply with a Neighbor Advertisement. One unfortunate part of this specification is LOST REVENUE OR LOST HEADACHE? that if a duplicate address is found, the device must be configured manually. However, This is an easy answer. Zero given that the interface ID should be unique, Configuration is a lost headache opportunity. you should never have a duplicate address. A service provider can bill for services, but it is hard to bill for support of the backbone Once an interface has a link-local address, network. You could charge a per-hour or a it can use this address to obtain site-local per-incident fee, but in the end the call center and/or Global-scope addresses, if desired. To will not be a profit center. Zero Configuration get these addresses, the link first tries a DHCP is a revenue generator because the “lost request. If it does not get a response, the headache” of configuration will allow the roll- stateless mechanism allows the host to out of new and profitable services. generate its own addresses using a combination of its interface ID and the subnet CONNECTING TO THE INTERNET prefix advertised by a router. An address created this way will be a proper Site-local or Zero Configuration is designed for link- Global-scope address, depending on the router local connections. This means that you configuration. IPv6 is designed for interfaces connect to devices on the same wire (if a to have multiple IP addresses. wired network) or same channel (if a wireless network). In order to connect a IPv6 also supports the easy renumbering of Zeroconf/Rendezvous device to the Internet, an entire site. This means that if a home is you will need a DHCP server or a bridge suddenly connected to the Internet, there is a device to go from the link-local domain to the simple way to change all the device addresses Internet. UPnP explicitly allows a single from local addresses to global addresses. device interface to have multiple IP addresses. So if a DHCP server became available, a Security UPnP device could seamlessly connect to the Internet. IPv6 allows a device to get a Global So, a device can use a Link-local address address from DHCP or create one based on a that limits its traffic to inside the home or a router advertisement. link level address to limit communication to devices to which they directly attach. A clock radio might select a Site-local address to make CONCLUSION sure only people within the radio’s household can set the alarm. Or, it may select a global The biggest remaining question is “When address so you can set it from the office. will Zero Configuration become commonplace?” Zero Configuration is more All IPv6 nodes support the IP Security difficult to roll out than many other services protocols (IPSec) standard for cryptographic because it requires wide adoption before it has authentication and encryption. So, all devices significant value. Metcalfe's Law says that can send and receive packets with some the value of a network grows in proportion to confidence that the packets are from the the square of the number of users. By expected source and their contents have not analogy, the value of Zero Configuration will been modified. grow rapidly as the number of devices which . offer it grow. Connecting one device that supports Zero Configuration and one device that does not will not leave you with two Because of its support in Microsoft devices that are half configured! It Windows, and the widespread deployment of unfortunately leaves you with two devices to Windows, UPnP looks like it could be the configure, one of which may not have been winner. However, IPv6 is certainly a very intended to be user configured. vendor neutral option and it too is supported in the latest versions of Windows. The race is When will we get to the tipping point? far from over. When the market decides the When will people refuse to buy a device winner (perhaps in a matter of 2-3 years), the unless it has Zero Configuration? Well, first winner will rapidly become ubiquitous. one of the Zero Configuration approaches must become the clear winner. Or some combination of approaches will become the REFERENCES clear winner. Why not use Zeroconf for the link layer and then above that use an [1] Aidan Williams, “Requirements for evolution of UPnP which has evolved to Automatic Configuration of IP Hosts”, draft- support IPv6? This certainly sounds like a ietf-zeroconf-reqts-12.txt. practical and powerful combination. We will have to wait for the market to decide this one. [2] Erik Guttman, “Zeroconf Host Profile Applicability Statement”, draft-ietf-zeroconf- host-prof-01.txt. 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.