218 Chapter 10: On-Demand Services
`
`Goals of On-Demand Services
`On-demand services share many of the same goals as interactive services (see Chapter 8).
`Successful on-demand services add value in some way for the programmer, cable operator,
`and customer. Added value can come from a number of sources:
`• Customer-generated revenue—If an on-demand service is useful or compelling, the
`customer will be prepared to pay for it. Specifically, video-on-demand has demonstrated
`significantly higher buy rates than impulse pay-per-view in a number of trials.
`• Advertiser-generated revenue—If an on-demand service can increase the
`effectiveness of existing commercials or allow the customer to access a company's
`product and service information, the advertiser will be prepared to pay the cable
`operator for that service.
`• Enhanced customer perception and retention—Increased competition from DBS,
`MMDS, and others make it even more important to cable operators that their
`customers are happy with their cable service. Availability of on-demand services
`enhances the customer's perception of the overall cable service.
`To be successful, on-demand services must provide a satisfying experience for the customer
`in the following areas:
`• They must be friendly and intuitive to use. The service needs to be easily accessible
`from the remote control with little or no customer training.
`• They must present an attractive, graphically rich interface. The customer is
`accustomed to the high production values of most television programming and will
`not tolerate an uninteresting or static user interface.
`• They must be reliable and highly available. If an interactive service becomes popular
`with customers, they will be frustrated if it is unreliable or unavailable.
`The business case for on-demand services is simple: An on-demand service is economical
`if the revenue generated by it can pay for the capital cost of the additional hardware in a
`reasonable period of time (for example, four years). Until recently, there were two
`fundamental barriers to on-demand services:
`• The cable system could not provide sufficient bandwidth to support revenue
`generating movies-on-demand service; that is, too few customers would be able to
`access the service simultaneously to make it successful.
`• The cost of the digital set-top was prohibitive and made the movies-on-demand
`business case uneconomic.
`The first problem has been solved by hybrid fiber coax upgrades, which provide a massive
`increase in the bandwidth available to each customer. The second problem has been solved
`because the digital set-top has been cost-reduced and is now justified economically for
`broadcast digital services. This chapter describes how the video-on-demand architecture is
`tailored to provide the best economic solution to these problems.
`
`DISH, Exh. 1011 p.0226
`
`

`

`On-Demand Services 219
`
`On-Demand Services
`There are many services that are enabled by on-demand technology. The following list of
`services provides some examples:
`• Movies-on-demand
`• Music-on-demand
`• Post-broadcast on-demand
`• Special interest programming
`• Distance learning
`• Library access
`• Video mail
`
`Movies-on-Demand
`Usually called video-on-demand, the idea of providing access to a movie library is very
`appealing. The movies-on-demand (MOD) service has been extensively tested in trials, and
`customer response has been enthusiastic.
`
`This service emulates the current video rental business, so it is easiest to justify from an
`economic point of view. The cable industry today has later windows (that is, the period of
`time at which particular movies are available) than the video rental market. However, if a
`movies-on-demand service can be successfully deployed, the cable industry will be well-
`positioned to negotiate better windows for MOD titles.
`
`Music-on-Demand
`The music-on-demand service provides access to a music library. Broadcast music services
`(for example, Music Choice and DMX) are effectively niche services, and it is unclear
`whether on-demand access to a music library will change the status of music services.
`
`This service emulates the CD player, but allows the customer to access any musical
`selection. For the same reasons that there are no CD rental stores, it is unlikely that there is
`much business potential in a music-on-demand service. However, music-on-demand can be
`used to promote CD sales in the same way as the Internet.
`
`Post-Broadcast On-Demand
`Post-broadcast on-demand uses video-on-demand technology to allow access to a library
`of titles that have been broadcast. It allows the customer to view shows they have missed
`and caters to the segment of the market that is prepared to pay for this convenience.
`
`DISH, Exh. 1011 p.0227
`
`

`

`220 Chapter 10: On-Demand Services
`
`This service emulates using a VCR to record broadcast programming for viewing at a later
`time. It provides convenience, because the customer does not have to remember to record
`the program, and flexibility, because many titles can be offered. Some early tests also
`showed that the success of this service is very programming type—dependent; for example,
`post-broadcast on-demand might work well for soap operas or hit sitcoms, but it might not
`be as successful for children's programming. The real question is whether this kind of
`service is commercially viable—will customers pay for this type of service?
`
`Special Interest Programming
`Video-on-demand can allow random access to special interest programs and provides an
`alternative to creating new special interest channels. New special interest channels are being
`created at an amazing rate; digital compression allows many more channels on a cable system,
`but at some point there is a limit to the number of channels. In addition, many special interest
`channels are hard-pressed to provide sufficient programming to fill their schedule. With the
`addition of each new channel, the potential audience is further subdivided, making advertiser
`sponsorship more difficult. Therefore on-demand delivery of special interest programming
`provides a solution to dedicated channel assignments and limited viewership.
`
`Distance Learning
`The capability of providing multimedia delivery of video, graphics, and text makes on-
`demand technology ideal for distance learning applications. Distance learning is an ideal
`candidate for on-demand services. In many ways, it is similar to special interest
`programming—appealing only to a niche audience and not justifying a dedicated channel.
`(Currently, distance learning channels aggregate a number of courses to justify a channel
`allocation, but this model can be very inconvenient for customers if the broadcast time of
`the course does not fit into their schedule.)
`
`Library Access
`The idea of a large online, multimedia library is not new, but the technology to retrieve titles
`in real-time makes this concept compelling. On-demand library access emulates the service
`currently provided (albeit poorly) by the WWW. This service allows retrieval of short clips
`of multimedia information and would primarily support research activity.
`
`Video Mail
`The store-and-forward approach to communications is often more convenient. This
`approach can be extended to a video mail service that allows one customer to send a video
`and audio message to another. Video mail is a communications service and is not really part
`of video-on-demand, but VOD technology could be used to provide the record and playback
`capabilities required for a multimedia e-mail service.
`
`DISH, Exh. 1011 p.0228
`
`

`

`On-Demand Reference Architecture
`
`221
`
`On-Demand Reference Architecture
`Figure 10-1 is a reference diagram for an on-demand system.
`
`Figure 10-1 On-Demand Reference Architecture
`
`Media
`Server
`
`Media
`Server 21
`
`Media
`Server
`
`Media
`Server
`
`•
`• •
`Media
`Server
`
`Switching
`Matrix
`
`3
`
`4
`
`N
`
`Provisioning
`Network
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Distribution Networks
`
`At this level of detail, the system is composed of the following major components:
`• Provisioning network—This allows new content or multimedia assets to be loaded
`onto the media servers.
`• Distribution networks—In a cable system, the distribution networks are formed by the
`hybrid fiber coax (HFC) plant. Each distribution network supplies a group of set-tops.
`• Media servers—The media servers provide the streaming content for distribution to
`the set-tops.
`• Switching matrix—The switching matrix provides a path from each media server to
`each distribution network.
`• Set-tops—The set-top provides termination of the streaming content and adapts it to
`the television set.
`
`Each of these components is described in more detail in the following sections.
`
`DISH, Exh. 1011 p.0229
`
`

`

`222 Chapter 10: On-Demand Services
`
`Provisioning Network
`The provisioning network provides secure delivery of content from the content provider to
`the video server location. Satellite or terrestrial links could provide this network. The
`important point to note is that the content provisioning can be modeled as a series of file
`transfers and, accordingly, the requirements for the content provisioning network are
`considerably relaxed compared to the distribution network. In fact, multimedia assets do
`not need real-time delivery, and the QoS of the provisioning network does not have to be
`guaranteed. Therefore, available bit rate services (on a satellite channel, for example) can
`be used to deliver multimedia assets.
`
`Distribution Network
`Figure 10-1 shows that in an on-demand cable system, there are actually multiple, parallel
`distribution networks. Each distribution network serves a group of set-tops so that sufficient
`capacity is allocated to meet the peak on-demand traffic generated by those set-tops (see
`Chapter 2, "Analog Cable Technologies"). Media servers are connected via a switching
`matrix to the distribution networks. Finally, each media server is connected to a
`provisioning network for content provisioning. Note that in Figure 10-1, for simplicity, the
`control and signaling network is not shown.
`
`Figure 10-2 shows the channel allocation to each set-top. The analog and digital broadcast
`channels are the same for all set-tops (and all customers), but the on-demand channels are
`narrowcast to a set-top group.
`
`Figure 10-2 Set-Top Channel Allocation
`
`
`
`
`
` Digital On-Demand Channels
`
`Digital Broadcast Channels
`
` Analog Broadcast Channels
`
`Out-of-Band Channels
`
`Set-Top Can Select Only
`One Channel at a Time
`
`Set-Top
`
`DISH, Exh. 1011 p.0230
`
`

`

`On-Demand Reference Architecture 223
`
`The size of the set-top group can be calculated as follows:
`• The aggregate capacity of the on-demand channels is equal to the number of channels
`multiplied by the capacity of each channel. The more channels the better, because this
`provides better statistical multiplexing. For example, if 10 channels are allocated
`(requiring 60 MHz of cable spectrum) and 256-QAM modulation is used, the
`aggregate bandwidth is 10 x 38.8 Mbps, or 388 Mbps.
`• The traffic generated by each set-top is determined by the on-demand session rate. For
`standard definition compressed digital television, 3.8 Mbps is a realistic number (and
`makes the math easier).
`• The next factor is the peak utilization of on-demand services. This is the percentage
`of all customers who simultaneously request on-demand services at the busiest time.
`On-demand services are usually engineered for a peak-utilization rate of 10% because
`this number seems to be validated by on-demand trials (see Chapter 1 I ).
`• The number of set-tops per distribution network is given by the following formula:
`aggregate capacity X peak utilization ÷ on-demand session rate
`Using the numbers in the example, each distribution network can serve 1,000
`set-tops. In a cable system serving 100,000 set-tops, 100 distribution
`networks would be required.
`
`The distribution network provides many areas for innovation in the cost-effective delivery
`of on-demand services. There are a number of technology choices to be made by the
`network architect:
`• Physical protocol—In a fiber-based network, the physical layer protocol can use a
`digital (for example, SONET) or analog physical layer, or a combination of both. In
`an HFC network, the digital payload must be modulated to traverse the analog part of
`the network.
`• Transport protocol—There are a number of alternative transport protocols to choose
`from for on-demand services. The leading alternatives are MPEG-2 transport, ATM,
`and IP.
`
`Physical Protocols
`In an HFC cable system, the distribution network is composed of a series of cascaded fiber
`links and coaxial links (see Chapter 2). The physical layer protocol over the fiber link can
`be digital (for example SONET) or analog, but the physical layer over the coaxial links is
`always analog. A modulator performs the conversion from digital to analog and may be
`placed anywhere in the signal path between the switching matrix and the first analog link.
`
`In practical terms, modulators are sensitive pieces of equipment that require a temperature-
`controlled environment; the possible locations are the headend or the distribution hub.
`
`DISH, Exh. 1011 p.0231
`
`

`

`224 Chapter 10: On-Demand Services
`
`Modulation at Headend
`Placing the modulators at the headend centralizes all the digital components of the system
`and uses analog fiber to transport the modulated carriers all the way to the set-tops.
`Centralizing the modulators might require considerable space—for example, a 100,000 set-
`top system would require 1,000 modulators at 10% peak utilization. However, located at
`the headend, the modulators can easily be reconfigured to best match the traffic pattern
`changes of the system as it grows.
`
`The use of analog fiber transport is familiar to most cable operators because it is already
`used for analog and digital broadcast channels; the only change is that instead of one
`channel line-up, a separate bank of on-demand channels is required for each set-top group.
`However, multiple fibers or wavelength division multiplexing can be used to carry each
`bank of on-demand channels to the set-top group.
`
`Modulation at Distribution Hub
`Modulation at the hub allows on-demand traffic to be integrated over a single transport with
`other digital signals, such as cable modem, telephony, out-of-band signaling, and network
`management traffic. In the distribution hub, the modulators are closer to the set-top (in
`terms of fiber miles), and there is less analog degradation of the modulated signal. Although
`digital transport provides more flexibility and can be extended over greater distances than
`analog transport, there is a cost premium to be paid for digital transport.
`
`Transport Protocols
`There are a number of alternative transport protocols for on-demand services. The leading
`alternatives are MPEG-2 transport, ATM, and IP:
`• MPEG-2 transport—MPEG-2 transport is optimized for compressed, streaming
`media.
`• Asynchronous transfer mode (ATM)—ATM transport is a compromise protocol
`designed to handle video, data, and telephony traffic.
`Internet Protocol (IP)—IP was designed for data traffic but is being extended to add
`features for streaming media.
`These transport protocols are discussed in more detail in Chapter 4, "Digital Technologies."
`There is no correct choice for all circumstances, but the following guidelines will help you
`select the best choice for your application:
`•
`
`•
`
`If a single traffic type dominates, select the best transport for that traffic type. For
`example, the best choice for a video-on-demand application is MPEG-2 transport.
`• The choice of transport type is closely related to the need for distributed switching and
`routing. MPEG-2 transport switches are not available commercially, and ATM or IP
`might be a better choice for flexibility if the additional cost can be justified.
`
`DISH, Exh. 1011 p.0232
`
`

`

`On-Demand Reference Architecture 225
`
`• What transport protocol is used by set-tops that are already deployed? Nearly all first-
`generation digital set-tops accept only MPEG-2 transport streams, and protocol
`conversion back to MPEG-2 in the distribution network will be required to support these
`set-tops if ATM or IP are selected. (Figure 10-5 provides an example of this case.)
`
`Media Servers
`The server has a number of functions and can be broken down into a number of functional
`elements to aid description (however, note that a specific implementation might not follow
`these lines):
`• Asset management—Each file stored on the media server is called an asset. Assets
`must be managed to allow them to be provisioned and retrieved by the streaming
`service.
`• Streaming service—The streaming service delivers an asset or group of assets
`according to the requirements of the on-demand service.
`• Directory service—The directory service provides a list of assets to the server
`application.
`• Server application—The server application provides the intelligence and control of
`the media server. Different applications are used to provide different on-demand
`services.
`
`Asset Management
`Asset management is the function responsible for making sure that assets are available to the
`streaming service when they are required. Asset management supports a number of functions:
`
`• Provisioning—Each asset must be downloaded onto the media server via the
`provisioning network. The asset is the movie file and related elements, such as the
`trailer video file, the poster image, the text description, the price, and so on.
`• Storage management—Each asset must be placed into the media server storage.
`Multiple copies of assets might be required based on server design and viewing demand.
`• Deletion—Assets must be deleted when they are no longer required.
`
`Streaming Service
`The streaming service delivers an asset or group of assets according to the requirements of
`the on-demand service. For example, to stream a movie requires that the video and audio
`streams are synchronized and delivered at the precise rate for the set-top decoder. The
`streaming service is often implemented with special hardware to satisfy the stringent real-
`time requirements for streaming MPEG-2 encoded assets.
`
`DISH, Exh. 1011 p.0233
`
`

`

`226 Chapter 10: On-Demand Services
`
`Directory Service
`The directory service provides a list of assets to the server application via an API. The
`directory service is responsible for keeping track of assets and liberates the application
`program from this task. A directory service is useful for providing asset search functions,
`for example, to locate a movie based on genre or actor.
`
`Server Application
`The server application allows the media server to be used for many different on-demand
`services
`typically, there is a different server application for each on-demand service. For
`example, a movies-on-demand application might support functions such as providing a list
`of available movies, movie previews, movie purchase, fast-forward, pause, and so on.
`
`Conditional Access
`Conditional access (CA) is required for the distribution of any streaming media that have
`value and must therefore be protected from signal theft. There are two approaches to
`providing conditional access to streaming media from the server:
`▪ Encrypt the signal at the output of the server. This approach follows the broadcast
`model and can conveniently use the same conditional access system that is already in
`place for digital broadcast signals. The conditional access system must support
`additional interfaces to support secure session establishment and key distribution to
`the on-demand client application.
`• Encrypt the stored assets on the server. This approach requires no stream encryption
`at the headend because the preencrypted assets are streamed directly from the server
`storage. Secure key distribution to the on-demand client could use the broadcast
`conditional access system or a separate conditional access system. The disadvantage
`to preencrypted assets is that the encryption is static and might be less resistant to
`pirate attacks. Preencryption also does not allow statistical multiplexing because the
`MPEG stream is encrypted and cannot be reencoded. Finally, updating preencrypted
`content to fix expired keys places an additional load on the server.
`
`Server Placement
`In theory, the server can be placed at any point in the distribution network between the
`headend and the set-top. However, the server requires a controlled environment, and the
`practical server locations are the headend, the distribution hub, or the customer premises.
`
`DISH, Exh. 1011 p.0234
`
`

`

`On-Demand Reference Architecture
`
`227
`
`Server at Headend
`Placing the media server at the headend allows it to serve the entire cable system. The
`distribution networks aggregate the traffic from all customers, and the law of large numbers
`allows efficient statistical multiplexing. In early deployment, when some distribution hubs
`support relatively few customers, centralizing the servers makes good sense.
`However, a centralized server farm might grow to require considerable resources in terms
`of space and power. A centralized facility is also vulnerable to disasters such as fire or flood.
`In addition, the distance-bandwidth product of the distribution networks is much larger than
`when the servers are placed nearer to the customer. Consider that every asset must be
`transferred across the distribution network for all customers individually—even if they live
`next door to each other.
`
`Despite these caveats, placing the media server at the headend is viable for cable systems
`with several hundred thousand subscribers. Distribution networks with sufficient capacity
`can be engineered using the almost unlimited bandwidth of fiber optic links, and the
`physical footprint and power requirements of media servers are being constantly reduced
`by Moore's Law.
`
`Server at Distribution Hub
`Placing the media server at the distribution hub reduces distribution network capacity at the
`expense of asset replication. The media server is closer to the set-top and effectively
`provides a cache function, holding all the assets that might be required by the set-top. This
`works for systems with relatively few assets, but the provisioning traffic and media server
`storage increases in direct proportion to the number of assets.
`The media server must be designed for unattended operation and must be managed
`remotely if it is located at the distribution hub. Any regular maintenance of the media server
`necessitates travel to and from the distribution hub.
`
`Server in Set-Top
`With the advent of multigigabyte storage at affordable prices, why not put the VOD server
`in the set-top? The set-top server is dedicated to a single customer, and therefore its cost
`cannot be amortized across a number of customers like a centralized server. However, the
`server complexity is considerably reduced because it plays only a single stream at a time.
`Moreover, server failure affects only a single customer and redundant hardware is not
`required.
`The main problem with placing the server in the set-top is the amount of storage and the
`provisioning bandwidth required. In a time-shifting application, where broadcast
`programming is cached by the server for subsequent playback, no incremental provisioning
`
`DISH, Exh. 1011 p.0235
`
`

`

`228 Chapter 10: On-Demand Services
`
`bandwidth is needed; but in an on-demand service, every asset that might be required by
`the customer must be provided, using one of two approaches:
`• Popular assets—for example, new movie releases—are broadcast to all set-tops to
`reduce provisioning bandwidth.
`• Random assets—for example, library video-clips—are delivered on-demand in real-
`time (or faster than real-time). Media servers are still required in the headend, and set-
`tops generate much the same traffic load as other on-demand approaches.
`In the first approach, considerable set-top storage is required to provide customer choice.
`In addition, because every set-top contains a copy of each new movie release, piracy is a
`concern; the set-top server assets must be protected by a bulletproof conditional access
`system. In the second approach, there is little or no advantage to the cable operator except
`when assets are downloaded and then played several times.
`In summary, placing the set-top in the server can provide only very limited on-demand
`selection unless servers in the network provide a second level of support.
`
`Switching Matrix
`With reference to Figure 10-1, the media server output is routed to the set-top via one of a
`number of distribution networks. (To understand why there are multiple distribution
`networks, please refer to the section Distribution Network, earlier in this chapter.)
`Therefore, a switching function is required to connect a media server to a set-top in a
`particular distribution network. In general, an M by N, nonblocking switching matrix is
`required, and Figure 10-1 illustrates the use of a such a switching matrix to allow any of the
`M media servers to be connected to any of the N delivery networks.
`There are a number of approaches to the requirement for a switching matrix (these are
`described in the next sections):
`• Content replication—In the content replication approach, each media server holds a
`copy of all the potential assets required by the set-top. This allows the media server to
`be directly connected to the distribution network.
`• Massively parallel server—Proponents of massively parallel media servers (notably,
`N-Cube) argue that the switching matrix can be best incorporated into a scalable
`media server. Such a server can output any asset on any output port and can be
`connected directly to the distribution network (at the headend or the distribution hub).
`• ATM switching—ATM is an effective switching technology for the switching matrix.
`A connection manager is required to set up ATM connections from the media server
`to the distribution network.
`
`DISH, Exh. 1011 p.0236
`
`

`

`On-Demand Reference Architecture 229
`
`•
`
`IP routing—Until recently, IP routing did not scale to meet the throughput demands
`required for the switching matrix. However, recent advances in multiple protocol label
`switching (MPLS) and IP QoS have made IP routing a potential technology for this
`application. The great advantage of IP is that it is self-routing—that is, no explicit
`connection management function is required.
`• QAM matrix—It is possible to build a matrix of QAM modulators in such a way as to
`form a space/frequency/time switch. The switching matrix is connection-oriented, so
`it requires a connection management function similar to an ATM network.
`
`Asset Replication
`Asset replication places a copy of all assets on every media server and allows the media
`server to be directly connected to the distribution network, as shown in Figure 10-3.
`
`Figure 10-3 Asset Replication
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Provisioning
`Network
`
`Media
`Server
`
`Media
`Server
`
`Media
`Server
`
`Set-Top
`
`- 411110 "
`
`Set-Top
`
`Set-Top
`
`Distribution Networks
`
`This approach is simple and eliminates the switching matrix but has a number of
`disadvantages that make it unsuitable for wide-scale deployment:
`• Asset replication increases the amount of storage required and directly increases
`storage cost. In a limited movies-on-demand application, where only 10 titles are
`offered at any given time, asset replication is acceptable, but it does not scale to larger
`systems. For example, in a time-shifting or library application, asset replication in
`
`DISH, Exh. 1011 p.0237
`
`

`

`230 Chapter 10: On-Demand Services
`
`each media server requires massive amounts of storage. Worse still, the provisioning
`bandwidth increases linearly with the number of media servers and with the number
`of new assets. Eventually, as system size increases, asset provisioning generates most
`of the activity in the system for assets that might never be retrieved.
`• The availability of such a system is the same as the availability of the media server;
`if it fails, no other path exists to the distribution network. This problem can be
`mitigated by using two media servers per distribution network, but this doubles the
`media server cost.
`• For efficiency, the demand of each distribution network must be accurately
`matched to the capacity of the media server. This is difficult to achieve as the
`customer base grows.
`
`Massively Parallel Server
`In a massively parallel server, the switching matrix is incorporated into a large, scalable
`media server, as shown in Figure 10-4.
`
`Figure 10-4 Massively Parallel Server
`
`Storage
`
`Storage
`
`Storage
`
`Storage
`
`Interface
`
`Storage
`
`Storage
`
`Storage
`
`Storage
`
`Stream
`Server
`
`Inter-
`connect
`
`
`
`Stream
`Server
`
`Stream
`Server
`
`Stream
`Server
`
`•
`
`Stream
`Server
`
`Provisioning
`Network
`
`Set-Top
`
`1
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Set-Top
`
`Massively Parallel
`Server
`
`N
`Distribution Networks
`
`Such a server can output any asset on any output port and can be connected directly to the
`distribution network. Massively parallel servers provide an excellent solution to the
`switching problem but must satisfy rigorous availability metrics because of their design:
`
`DISH, Exh. 1011 p.0238
`
`

`

`On-Demand Reference Architecture 231
`
`•
`
`• Massively parallel servers must be highly available, because if they fail, no service is
`available to all customers.
`It must be possible to increase server capacity by increments while the server is
`online.
`• The failure of individual disk drives, processors, memory banks, or power supplies
`cannot cause the server to fail (but might reduce total capacity).
`• Massively parallel servers must be cost-competitive with smaller servers over the life
`cycle of their operation. In other words, they cannot be prohibitively expensive for
`early, lower-demand operation and cost-effective only in a fully subscribed system.
`If massively parallel servers are built to satisfy these requirements, this technology will be
`very successful for the provision of on-demand services.
`
`ATM Switching
`Figure 10-5 illustrates the use of an ATM switch as a switching matrix. The ATM switching
`function can be distributed by building a network of switches located at the headend and
`distribution hubs.
`
`Figure 10-5 ATM Switching
`
`Media
`Server
`
`Media
`Server
`
`Media
`Server
`
`•
`
`Media
`Server
`
`ATM
`Switch
`
`1
`
`QAM
`
`QAM
`
`QAM
`
`ICG
`
`•
`
`Set-Top
`ya- Group 1
`
`Set-Top
`Group 2
`
`)
`
`Set-Top
`• Group 3
`
`QAN7
`
`
`
`• Set-Top
`Group n
`
`Provisioning
`Network
`
`ATM is a connection-oriented protocol, and a connection manager is required to set up
`ATM connections from the media server to the distribution network.
`
`DISH, Exh. 1011 p.0239
`
`

`

`232 Chapter 10: On-Demand Services
`
`Incremental cost of ATM switching includes the adaptation of MPEG-2 packets to ATM
`cells and reassembly hack into MPEG-2 packets before they are fed into the MPEG-2
`decoder. In Figure I 0-5, AM adaptation is done by the output interface of the media server.
`MPEG-2 streams are reassembled by the Integrated Cable Gateway (1CG) before QAM
`modulation to support an MPEG-2 digital set-top. (See Chapter 1 1 for an ATM-to-the-home
`case study.)
`ATM switching provides flexibility and efficiency for multimedia transport but requires
`sophisticated connection management services that are directly accessible from the on-
`demand application. Some approaches, notably X-Bind, show considerable promise toward
`meeting these requirements. (See "A Programmable Transport Architecture with QoS
`Guarantees," by Jean-Francois Huard and Aurel A. Lazar, in IEEE Communications.)
`
`IP Routing
`Figure 10-6 illustrates a routed network approach to media server to distribution network
`interconnection.
`
`Figure 10-6 IP Rotnin,f;
`
`Media
`Server
`
`Media
`Server
`
`Media
`Server
`
`Media
`Server
`
`Media
`Server
`
`Provisioning
`Network
`
`QAM
`
`N
`
`QAM
`
`
`
`IP
`Router
`
`Set-Top
`Group 1
`
`Set-Top
`Group 2
`
`Set-Top
`Group 3
`
`SetJop
`Group n
`
`DISH, Exh. 1011 p.0240
`
`

`

`On-Demand Reference Architecture 233
`
`Until recently, IP routing did not scale to meet the throughput demands required for the
`switching matrix. However, recent advances in multiple protocol label switching (MPLS)
`and IP QoS have made IP routing a potential technol