I

A Ksion for Residential Broadband Services: ATM-totheHome To achieve the grand vision of the residential broadband market, thousands of applications must be enabled on these networks. Instead of constantly searching for the single killer application through market studies and speculative guesses, the best insurance policy is to build a platform of thousands of applications such that, collectively, these applications all become killer applications. Tim Kwok

T I M W O K I S the chrefATM architectfor Microsoji

14

he convergence of the computer, communication, cable, and television industries has transformed from only a vision of the future to a reality in the past two years. Major network operators, both telcos and cable companies (MSOs, or multiple system operators), around the world are racing to deploy residential broadband services. Cable vendors are selling equipment to telcos, as well as MSOs, to build broadcast and interactive TV networks, while telco vendors are selling equipment to cable operators, in addition to the telcos, to build telephone and interactive TV networks. Meanwhile, all major computer companies (hardware and software) are developing servers, set-top boxes and software for both telcos and MSOs. This trend is spurred by three powerful forces: deregulation, consumer demand, and technology. For decades, most network operators have had monopolies in their geographical areas or markets. However, deregulation has forced these network operators to rethink their strategies and redefine their markets in order to survive and grow. Otherwise, competition from other network providers and entrepreneurs could significantly affect their market share. Most of them have invested very heavily (in billions of US$) in a new market: residential broadband services. Such services include video on demand, home shopping, video games, (high quality) video telephony, and eventually thousands of residential applications being envisioned or to be invented. At the same time, consumers have become more accustomed to using new technologies for entertainment, education, and work. One-third of all American homes already has a PC [l],and six out of ten Pentiumm-based PCs were sold to consumers and not businesses [2]. About 4.3 percent of all American households use on-line services today, and research has projected that this will grow to 17.6 percent by 1998 [3]. With

0890-8044/95/$04.001995 0 IEEE

the arrival of residential broadband networks at much higher data rates than current narrowband access, new residential broadband applications will explode in number. More importantly, technological advances in software, computing, video compression, and networking hzfye made deploying residential broadband services economically possible within the next few years. The arrival of asynchronous transfer mode (ATM) networks has enabled a wide range of new interactive multimedia applications for the residential market. The objective of this article is to present a vision for supporting universal residential broadband services based on an ATM-to-the-home (ATTH) network architecture [4]. This network architecture applies to the various residential access network ( R A N ) architectures being deployed today, such as hybrid fiber/coax (HFC), fiber-tothe-curb (FTTC), fiber-to-the-home (FTTH), and asymmetric digital subscriber loop (ADSL) technologies. First, this article addresses today’s residential networks and applications, to understand why a switched broadband residential network is required to support residential broadband services. After exploring residential broadband application requirements, a new class of service is proposed to support a very important class of residential broadband applications that has been not addressed. Then, the technical and strategic motivations for using the ATTH architecture are discussed in detail. A universal model for residential broadband network architecture based on ATTH is described, which is shown to apply to various RAN architectures. Finally, this article discusses the signaling requirements of residential broadband services and explain why the ATM multiconnection per-call model is much more efficient than the DSM-CC session control approach for the ATTH architecture.

IEEE Network

SeptembedOctober 1995

Drop

+-

Amplifer/ line extender

-

I

I

/

Feeder

Neighborhood area A

t

Trunk amplifer

Neighborhood area C

About 2000 homes passed ~~

iFigure

1 . Traditional cable Wtzenuork architecture

Turget Res iden tial Broadband Network Today’s residential networks can be broadly classified as into broadcast and switched networks. A broadcast network, such as the cable TV network. provides a one-way, one-to-many distribution of information. As shown in Fig. I, analog video received at the headend is broadcast, via the tree and branch architecture of the coaxial cable distribution plant, to each subscriber. A switched network, such as the telephone network, provides two-way, point-to-point communications between any two users on the network (Fig. 2). Unfortunately, neither the existing cable TV nor the current telco networks can support the envisioned residential broadband services (Table 1). Although the traditional cable network has the advantage of a broadband transmission medium to the home that can potentially deliver information at Gb/s, this is a broadcast network and has no switching capability. Hence, it cannot support point-to-point connectivity or switched interactive services. Another problem is that the “cable network” actually consists of disjointed islands of local cable systems. Communications between these islands of cable networks has not been necessary for their core application: broadcast TV application (except their shared common source of programming from national headends). Today‘s telephone network provides the key value of ubiquitous communication between any two phones on the network anywhere in the world, by providing a sophisticated signaling infrastructure across the network. Unfortunately, the current bandwidth to each home is very limited, typically at 64 kbps or less. To support fully interactive residential broadband applications, both telcos and cable networks must be upgraded t o a switched broadband residential network.

Today‘s Residential Applications Although today’s residential networks do not support fully switched broadband applications,

IEEE Network SeptemberiOctober 1995

Tandem circuit-l

I

I

I

circuit switch U

I

Phones

W Figure 2. The telephone network architectiire.

they d o support various degrees of switching (interactiveness) and broadband capabilities. Today’s telephony network supports a number of interactive applications such as telecommuting. on-line services, and Internet access (especially through the World Wide Web). However. these are all narrowband applications. The esplosion of interests in these interactive applications have generated significant demand for broadband access to make applications more responsive. Current cable networks also support a range of near-on-demand (NOD) applications. These NOD applications include both video based. such as impulse pay per view (IPPV) and near video o n de.mand (NVOD). and imagei’data based such as broadcast carousel distribution of video games. NVOD provides an illusion of video on demand by broadcasting each movie over multiple channels at staggered. start times (typically 10 to 30 minutes apart). so viewers can start watching the movie by waiting until the next available time when the movie begins on one.of these channels. Viewers can also .’pause“ and “rewind/forward” by tuning to the corresponding channel. However, the large number of

15

'

To support

fully interactive residential broadband applications, a quality of service based, two-way switched broadband residential network architecture is required.

Table 2 . Comparison of traditional telephone and cable networks with the switched broadbandrGidentia1

network. channels that must be dedicated to NVOD to support a single movie makes it very bandwidth inefficient, as spectrum is expensive. One way to improve the bandwidth efficiency (especially in low viewership scenarios) is a video juke box model, whereby each channel originally dedicated to NVOD for the same movie is used only when viewers request the movie for that time period. The broadcast carousel model is useful for game distribution. It continuously broadcasts a set of games in the cyclical fashion, so the viewers can download the desired game to the local game machine by waiting (on the order of minutes) until the desired game broadcast on the network. In general, NOD applications take advantage of the intrinsic sharing of information of the broadcast channel while giving viewers more choice for the start time (for video based applications) or the content (for image/data based applications). However, this incurs a waiting time for video o r images to appear, and the choice remains limited. Ideally, we would like to have true video on demand (VOD) as viewers are unique in their interest and viewing habit. Note that VOD applications include not only movie on demand, but also T V on demand (TVOD), which allows individual programs in the past or future to be viewed on demand.

Switched Broadband Residential Network To support fully interactive residential broadband applications, a quality of service (QoS) based, two-way switched broadband residential network architecture is required. A broadband network is required to support the high data rate, in addition to low bandwidth applications. A switched network is required to support independent connections to each viewer. A QoSbased network guarantees the performance of real-time delivery of both time-based information (such as video, audio, and animation) and non-time-based information (such as image, graphics and text). Ideally, the high bandwidth to the home should be symmetric in order to allow high bandwidth peer-to-peer communications. This will enable each home to be a source of multimedia information, without being limited to service providers as the information source. This will be important in the creation of a vibrant

16

residential broadband service market, as it will enable a large variety of sources of multimedia information from the general public. This could result in a similar;but potentially bigger, revolution than that of the desktop publishing industry.

Res iden tial Broadband Networking Req uiremenfs

1

his section explores residential broadband applications requirements in greater detail.

What is an Application? A communication application is defined as a task that requires communication of one o r more informatiqn streams between two or more parties that are geographically separated. More specifically, an application can be characterized by the following main attributes [SI. lnformation Types - In general, the information to be communicated can be classified as timebased or non-time-based. Time-based information must be presented at specific instants to convey its meaning; i.e., time is an integral part of the information communicated or the information has a time component. Typical timebased types of information are video, audio, and animation. Non-time-based information includes images, graphics, and text. An application can include both time-based and non-time-based information. When an application involves multiple streams of information (possibly of different information types), synchronization among them is an important issue [6]. Delivery Requirements - An application can also be classified according to its information delivery requirements as a real-time or a non-real-time application. A real-time application is one that requires information delivery for immediate consumption. In contrast, non-real-time application information is stored (perhaps temporarily) at the receiving points for later consumption. The former requires sufficient bandwidth, while the latter requires sufficient storage (and potentially bandwidth as well if delivered at high speed). For example, a telephone conversation is considered a real-time application, while sending elec-

IEEE Network

SeptembedOctober 1995

I

tronic mail is a non-real-time application. In the usaswho annmunicatevia a real-timeappliother cation must be present at the same time, whereas those who communicate via a non-real-time application can participate at different times. It is important to distinguish between the delivery requirement (real-time or non-realtime) of an application from the intrinsic time dependency of its information content (timebased or non-time-based). This is illustrated in Table 2. Video conferencing and image browsing are examples of real-time applications, while sending video clips and electronic mail are nonreal-time applications. In t h e case of image browsing, although the image is non-time-based information, a maximum response time constraint is imposed to ensure an interactive response experience for the user; therefore, it is considered a real-time application (with nontime-based information). O n the other hand, sending video clips is a non-real-time application because, even though the information content is time-based, the video clip can be treated as a single file transfer (similar to electronic mail), since it is not being displayed in real time during delivery. In general, the communications requirements for supporting an application depends on both t h e information type and the delivery requirement of the application. Since residential broadband applications encompass all combinations of information types and the delivery requirements, they must all be supported. Symmetry of Connection - In general, a communication application is bidirectional, and bidirectional applications can be classified as either symmetric or asymmetric. A broadcast application is an extreme example of an asymmetric application that is one-way only, while telephony is a symmetric application. Number of Parties - Depending on whether there are two or more parties involved, an application can be classified as point-to-point (two parties) or multipoint (more than two), respectively. Obviously, point-to-point applications require only point-to-point connections, while multipoint applications require multiparty connection types, such as point-to-multipoint, multipoint-to-point, o r multipoint-to-multipoint connections. All of these must be supported in the residential broadband network. This will be discussed in more detail later.

Application Characterization

, Timebased information

Video conferencing. telephony

Non-time-based information Image browsing --____._

Video clip transfer Electronic mail

2

by its traffic generation process. Since the traffic generation process (or traffic pattern) is basically a sequence of packets generated at arbitrary instants, each packet having an arbitrary length, it can be modeled as an on-off source. The traffic pattern can be characterized by two stochastic processes: a) the packet generation process (or packet arrival process); and b) packet length distribution [7]. Two important traffic patterns in residential broadband services are periodic and bursty traffic. If packet generation occurs at regular time intervals, it is a periodic traffic pattern. If these packet lengths are fixed in size, it is a constant bit rate (CBR) stream. Otherwise, it is a vari: able bit rate (VBR) stream. For example, compressed video streams (such as MPEG-2) are typically periodic traffic and can be CBR or VBR. Bursty traffic is characterized by packets of arbitrary length generated at random times, separated by gaps of silence of random duration. Typically, the period of silence is long compared with packet generation periods, leading to the distinctive high peak to average bandwidth ratios. Image browsing is a bursty application because images are transferred on demand by the user, and such requests are unpredictahle. Since this is a real-time application, performance guarantees (delay) are required. In general, residential broadband applications consist of both periodic and bursty applications. The need to support both types of traffic with real-time requirements becomes the most challenging traffic management problem. Communications Requirements - The key communications requirements of an application include bandwidth, delay, and error guarantees. The residential broadband network must support a minimum set of communications requirements to enable a large range of interactive multimedia applications. To support real-time (time-based and non-time-based) applications, the residential broadband network must provide bandwidth and QoS (delay and error) guarantees.

An application can be characterized by its traffic.

characteristics and the corresponding communication requirements. A communication application’s traffic characteristics can be described as one or more sequence of packets, of arbitrary length, generated at a certain time and destined for one or more locations. Each packet has an associated set of communications requirements. Traffic characteristics, together with the corresponding communications requirements, determine the network resources (bandwidth and buffer) required to support this application. Traffic Characteristics - The traffic characteristics of an application can be formally specified

IEEE Network

SeptembedOctober 1995

Bandwidth - The bandwidth requirements of an application (in each direction) are typically specified in terms of peak and average bandwidth. For CBR applications, the peak and average bandwidth are the same. To support broadcast quality movies, a CBR MPEG-2 compressed stream requires 3 to 4 Mbls. The bandwidth requirements can increase to 6 or even 9 Mbls for real-time compression of live sports events, due to the fast motion content. Since the amount of compressed information varies according to the content and instantaneous scene changes, compressed video is variable bit rate (VBR) in nature. The CBR MPEG-2 stream is created by

17

, I traffic shaping, using a rate buffer and real-time adjustment of compression ratio (quantization Absolute level) to maintain the buffer fullness without overflow or underflow. Real-time transfer of non-time-based informadelay, or tion generates bursty traffic patterns, which are VBR applications. The peak bandwidth of such latency, is real-time applications very often exceeds that of real-time transfer of time-based information, such as real-time video. For real-time transfer of one of the non-time-based information applications, there is an absolute delay constraint specified as part key QoS of the QoS requirements (usually about 100 ms end-to-end from the application level) to achieve pe dormance the perception of instantaneous interaction. For image browsing applications, a full screen photo image of 3 Mbytes (1000 x 1000 x 24 b), after parameters compression 300 Kbytes by JPEG compression [SI. This requires about 24 Mbls link (peak) that must be bandwidth to satisfy the response time requirements. However, this can be reduced with reduced image size, or high compression ratio, or satisfied relaxed response time to less than 10 Mbls. Specifying the average bandwidth requirement for a the residential bursty application is a challenge because it varies according to the duration for which the broadband average is taken. Furthermore, the values obtained vary widely across different users (such as with image browsing), even for the network. same applications, because everyone has a unique usage pattern. When residential broadband services are first deployed, each active subscriber may require about 10 Mbls peak downstream bandwidth to support at least 1 MPEG-2 compressed stream that requires 4 to 6 Mbls, plus additional bandwidth for downloading applications or graphics with acceptable latency (less than 100 ms). The minimum average bandwidth should be about 5 Mb/s to support at least 1video stream, while the other applications may share the bandwidth among multiple users. In the upstream direction, a peak bandwidth of about 1-2 Mb/s should be available to satisfy the latency requirement for a variety of interactive applications. Of course, this 1 to 2 Mbls bandwidth can be shared among multiple users. To support videotelephony, up to 384 kbls sustained bandwidth should be available for the active user. Again, symmetric bandwidth of 10 Mbls or higher should be available in the future to allow any user in the network to become an information source.

by

Delay - Absolute delay, or latency, is one of the key QoS performance parameters that must-be satisfied by the'residential broadband network. As mentioned above, to provide interactive response to viewers the response time between a user action and its effect should be less than 100 ms. T o support network-based video games, a response time of 50 ms or less is required to support twitch actions. This puts an upper bound on the transmission times in each direction, which imposes a minimum upstream and downstream bandwidth requirement. Delay variation is another important QoS parameter. Real-time video probably requires the most demanding delay variation constraint to minimize the buffering requirement for end-to-end synchronization.

18

A New Class of Service general, to support residential broadband I"supported applications, three classes of service must be [5]: Best-effort delivery. Real-time delivery of time-based information. Real-time delivery of non-time-based information. Although all three are important for residential broadband services, only the first two have been studied in detail. The first class is being addressed by the ATM Forum with the available bit rate (ABR) class of service. The second class is the CBR or VBR with timing requirements (bounds on delay variation). This can be supported by reserving peak bandwidth for each application over a QoS-based network. The third class of service includes image browsing, which is required for home shopping applications. This third class is probably the most challenging class of service to support because it is bursty and has an absolub delay requirement. Such unpredictability makes it very difficult to allocate bandwidth efficiently to support this class of service. Further research is required in this area.

ATM-to-the-Homefor Res iden tial Broadband Services

s3

nce the adoption of ATM as the solution for -1SDN in the late 'SOS, the focus on ATM has gone full circle. ATM was originally chosen by the telecommunications industry as the cornerstone of future public networking technologies. It was envisioned as the ultimate public networking technology to provide integrated services to a wide range of applications (especially those involving multimedia) for both residential and business services. In the early O OS, ATM was quickly adopted by the computer industry as an excellent local area network (LAN) solution to replace legacy LANs networks (such as ethernet and token ring) in supporting multimedia applications and the ever increasing bandwidth requirements of data applications as desktop computing power increases [9]. ATM LANs have become the key focus of the ATM Forum since its inception in late 1991. In 1994, ATM became important again for the public network operators deploying residential broadband services as it provides central office switching capability for residential services [lo]. ATM-to-the-home (ATTH) provides an endto-end pure ATM connectivity between headend or central office servers and intelligent home devices (such as PCs and set-top boxes). In other words, in an A l T H network, ATM terminates at the set-top box or the PC. A l T H architecture is being adopted by network operators deploying FTTC, FTTH, HFC, and ADSL [lo 1. In addition, the ATM Forum has just established a residential broadband subworking group to specify standard interfaces for the ATTH architecture for different residential access networks. Residential broadband services and applications have also been designed to support ATTH for different residential access networks [lo]. There are also alternative approaches to

IEEE Network SeptembedOctober 1995

5 t

deploying residential broadband services. For example, in the case of the HFC access architecture [ll](an improvement of the cable TV network by replacing coaxial trunks by fibers, as shown in t h e following section), t h e r e is a n MPEG-2-to-the-home hybrid network architecture approach using ATM protocols at the heade n d o r central office only, and MPEG-2 Transport Stream (TS) packets over a n H F C architecture between the headend o r central office and the home devices. ATM cells are converted to MPEG-2 TS packets, which are 188 bytes with a payload of 184 bytes, at the headend or central office. Such a hybrid architecture is optimized for carrying MPEG-2 compressed video and audio, while carrying other information types is an afterthought and requires further study. As explained below, A'ITH is a far superior architecture for supporting residential broadband applications for both technical and strategic reasons.

Technical Reasons Since the technical advantages of ATM are well understood [9], ATM has been generally adopted as the switching technology to be deployed at the headend and central office for supporting residential broadband services. However, the advantages of deploying a pure ATM network between the headendlcentral office servers and the home devices are not well recognized. Next, this article discusses the important technical advantages of deploying the pure ATM solution (A'ITH) for residential broadband services.

dential applications and services, it is essential that there is a universal residential broadband network model for developing those services. ATM provides the necessary layer of abstraction so that upper layer software can be residential access network independent; applications can use the same model for HFC, FTTC, FTTH, and ADSL, as all of them can b e based on ATTH. Therefore, by designing an end-to-end ATM network such as ATTH, a universal residential broadband network model can b e achieved, significantly reducing t h e cost of deployment of residential broadband services and applications across different network architectures. Rich Signaling Capabilities - The rich signaling protocols of ATM makes it possible to support the different connection types required by many multimedia applications, such a s point-tomultipoint, multipoint-to-point, and multipointto-muitipoint connections (in addition to the basic point-to-point connection type). Since the A?TH model supports such rich signaling capabilities a n d connection types t o all home devices, it enables a wide variety of residential broadband services. If, however, ATM is terminated at the headendkentral office as specified in the MPEG-2-to-the-home architecture, then most of these connection types will be unavailable the home devices, thus tremendously limiting the number of residential services that can be deployed. Enable large Numbers of Applications - Provid-

Guaranteed Performance - Since ATM is a connection-oriented protocol with signaling mechanisms for network resource reservation, ATM can guarantee end-to-end QoS and bandwidth. This is the key advantage of ATM because this is the fundamental requirement to support interactive multimedia applications. By deploying a pure ATM network solution, such performance guarantees can be achieved across the entire residential broadband network, all the way to the home devices. Each home device can specify and will have direct access to the ATM signaling protocols it requires to support each of its applications. Otherwise, if the above hybrid network model is used (i.e., ATM is terminated at the headendkentral office), another performance guarantee model must be invented (and standardized) between the home devices and the headendkentral office. Since ATM signaling protocol is not available to the home device in such a hybrid model, there is no standard way of. communicating such requirements over the nonATM network/link (such as an MPEG-2 TS link) from the home device to the headend/central office. Furthermore, the interoperability between the ATM and non-ATM models (such as QoS parameters mapping) needs to be designed and standardized, which will delay standardized solutions. ATTH prevents the nightmare of solving the interoperability of heterogeneous networks, yet provides a clean solution for guaranteeing QoS end-to-end between servers and home devices.

.

A Universal Residential Broadband Network Model - To create the biggest market for resi-

IEEE Network

SeptembedOctober 1995

Since ATM is a connectionoriented protocol with signaling mechanisms for network resource reservation,

ATM can guarantee end-to-end

QoS and bandwidth.

ing performance guarantees and rich connection types gives A T T H the flexibility t o suppbrt many different types of applications with diverse networking requirements. The ability to tailor the virtual connections for each application to the required bandwidth and QoS allows maximum flexibility to support a variety of different applications efficiently. This is extremely important to enable the residential broadband market because we need to support thousands of applications with diverse networking requirements on this platform. No one can predict which application will be the killer application. Since the ability to support thousands of applications means that, collectively, these applications can become a killer application, using ATTH is the best insurance policy for network operators deploying residential broadband services. In contrast, MPEG-2-to-the-home has been optimized for MPEG-2 compressed video and audio only; it is unclear that these applications alone can justify the investment for upgrading to such residential broadband networks. '

Single Platform for a//Applications - The flexibility of A m in supporting many different multimedia applications implies that a single platform can be used to support all residential broadband services and applications. This is an extremely important advantage of ATTH. Since the network architecture can remain the same as new services and applications are added to the platform, ATTH significantly reduces the incremental cost of deploying new residential services and applications.

19

r

-

N o Need for a Separate level

Many

management capabilities already exist as part of the management plane of the ATM protocols, providing

A77H with s tu ndardized management protocols.

I

Gateway

-

The function of a Level 1 gateway in the telco video dial tone paradigm is to provide connectivity through the Level 1 network and basic billing capability. It is unnecessary in the A'ITH architecture because the ATM switch by definition is to provide connectivity to all connected parties, controlled by the ATM switch's switch controller (part of the ATM switch). Also, public ATM switches provide billing records of all connections, including durations and resources usage. Bandwidth Efficiency - Since ATM is based on packet switching and asynchronous time division multiplexing, it is very efficient in multiplexing information in both upstream and downstream directions. Furthermore, the ABR service of ATM allows best effort traffic to fully utilize the slack left from guaranteed services. Hence, with ABR service, the A l T H architecture can achieve almost 100 percent network utilization in residential access networks. This is very important because access bandwidth is extremely precious for many access network architectures (which represent the last miles). There has been a myth that ATTH is inefficient because of the 10 percent header overhead of ATM cells. However, if ATM is not used for the links to the home, the access network utilization cannot be 100 percent in the first place; such a packet network needs to reduce the load (and utilization) to below 80 percent to prevent overload (switch buffer overflow) [7]. Thus, contrary to the myth, ATTH can achieve a higher efficiency than the MPEG-2-to-the-home solution. Furthermore, the ATM header should be viewed within the context of its capabilities for providing QoS guarantees, flexibility for multiplexing, and rich connection types, rather than as overhead only. Network Management - Since A'ITH is a pure ATM network model, it significantly improves network manageability because it is not necessary to manage multiple networks with different networking technologies within the residential broadband network. This removes the interoperability problem between heterogeneous networks, where many standard management interfaces must be defined and interoperable. In addition, many management capabilities already exist as part of the management plane of the ATM protocols, providing A?TH with standardized management protocols. Scalability - ATM switches have the highest switching capacity of any commercially available switches with standard interfaces. ATM switches running in the multi-gigabit range are commonplace today, and ATM switch capacity continues to grow to even higher capacity. The scalability to very high bandwidth is extremely important because the volume of information supported at t h e headend o r central office for full scale deployment can easily be hundreds of Gbls, due to the combination of multimedia applications bandwidth requirements and the large number of homes supported. Also, scalability of the link bandwidth is important for residential access

20

network architectures, such as FZTC, HFC, and ADSL, as their upstream and downstream transmission rates improve in the future. Since ATM is independent of the data rate of its underlying physical layer, ATTH supports all access network architectures independent of the data rate of those architectures and their future improvement.

Strategic Reasons

r

ere are also a number of important strategic reasons for deploying ATTH for residential broadband networks. These are explained below. Support More Services, Applications, and Revenue Streams - It has been recognized that the

VOD application by itself cannot justify the billions of dollars of investment in various residential broadband networks being deployed worldwide [12]. To achieve the grand vision of the residential broadband market, thousands of applications muSt be enabled on these networks. A'ITH provides bandwidth and QoS guarantees to support all interactive multimedia applications and provides a scaleable platform that can support as many applications as needed, while meeting the widely diverse networking requirements of these applications. Only with such a universal residfntial broadband networking platform can we support a large number of services and applications that will generate sufficient new revenue streams to justlfy the large investment. l o w Cost - With the large number of ATM vendors (over 100) already in the market and more going into the market, the ATM market is destined to become a commodity. ATM prices are already falling at a rapid rate; they have dropped by a factor of more than six within three years of first commercial introduction. It is extremely advantageous for network operators to ride t h e cost curve of ATM by deploying ATTH. The cost of ATM is already low enough for residential broadband deployment. Unfortunately, the cost of deploying ATM is probably the most misunderstood issue in the residential broadband industry. Many mistakenly believe that ATM is very expensive and that it will remain so for a long time. This is primarily because of the following misconceptions.

Myth # I : A l M Based Set-lop Boxes are Very Expensive - The worst cost impression of ATMto-the-home probably comes from the most publicized residential broadband trial: the Time Warner full-service network trial in Orlando, Florida. Time Warner is deploying ATTH for their HFC network. Because their set-top boxes each costs more t h a n US$5000, many who review the high costs assume the expense is due to the ATM components in the set-top box. In reality, the set-top boxes are unmodified SGI Indeo workstations. The high cost is not simply ATM in the set-top box Many think that because ATM adapters still cost hundreds of dollars, they will be expensive to put into the set-top box. However, if ATM is implemented in an integrated fashion with other components in the set-top box, its cost will not

IEEE Network SeptembedOctober 1995

T

come close the cost of implementing ATM as an adapter. In fact, the cost of ATM components is negligible as compared to other set-top box components such as memory, CPUs, graphics processing, and MPEG-2 decoders.

Myth #2: The Technology is Too New - Many think that since ATM is such a new technology, the cost will remain high for many years. However, they fail to realize that ATM technology is unique in the large number of networking vendors, both existing and startups, that are pursuing this market because they believe ATM will be the most important networking technology for years to come. The ATM Forum has grown from four to more than 700 member companies in only three years, and is still growing at the same rate. I n fact, since t h e introduction of ATM adapters on the market in 1992, the price of OC-3c adapters has dropped from $4500 to $695 (as of March 1995), a factor of six in less than three years! The trend is bound to continue, given the number of vendors and the availability of standards. ATM is destined to become a commodity market like ethemet.

Myth #3: ATM Switch Very Expensive - One of the biggest misconceptions about ATM prices is that of the ATM switch. This mainly arises from the huge discrepancy between the price of a public ATM switch and the price of a local ATM switch. Public ATM switches a r e those that are designed for the traditional telco market and meet the stringent fault tolerant capability and reliability requirements specified by Bellcore. Local ATM switches are those deployed for the LAN market; these do not need to meet the same Bellcore requirements, but are still reasonably reliable. For the same switching capacity, number of ports, and speed, the public ATM switch can be ten.times more expensive than the local ATM switch! When people claim that ATM switches are expensive, they are referring to public ATM switches. Unfortunately, they fail to realize that public ATM switches are not necessary for residential broadband services; there is no need to provide lifeline services for entertainment applications, which will be the primary residential broadband applications (at least initially). As local ATM switches continue to drop in price and increase in functionality and reliability, this will encourage network operators to gradually adopt local ATM switches for residential broadband services. Also, it is very likely that local ATM switch vendors will build versions of their switches that are slightly more expensive but have increased reliability to meet the residential market requirements. In fact, an analogy of public versus local ATM switches is t h e market relationship between mainframe computers versus personal computers (PCs); the force of the high volume commodity of PCs has brought on the demise of the mainframe market and the same might happen to public ATM switches. Myth #4: MPEG-2 TS-to-the-Homeis Cheaper than ATM-to-the-Home- It is unclear that ATTH is more expensive than its alternatives, such as MPEG-Zto-the-home in the case of HFC archi-

IEEE Network SeptembedOctober 1995

tecture. It is possible that deploying MPEG-2-tothe-home can be more expensive both in the short and long term. In the short term, supporting MPEG-Zto-thehome means that set-top boxes must terminate the 188 byte packets. This requires the set-top boxes to process the 188 byte packets, reassemble them from transport stream (TS), and convert them into PES format (the video and audio packetized elementary stream format). Moreover, by limiting the packet size of compressed video to only 188 bytes, software processing becomes unfeasible due to the resulting high interrupt frequency [ 131. Furthermore, ATM-to-MPEG conversion is required if the headend supports ATM switching. Such conversion processes are proprietary and expensive because no standard exists and very few vendors support them. (Currently, the equipment for converting a single OC-3c MPEG-2 TS over ATM to 5 MPEG-2 TS streams at 27 Mb/s can cost easily more than $30,000, more than the cost of a complete 2.4 Gb/s local ATM switch with 16 OC-3c ports !) In contrast, in the ATTH architecture, MPEG-2 compressed video can be carried over AAL5 either by Program Stream packet format, which can be up to 64 Kbytes long, o r k MPEG-2 TS packets (where k can be negotiated by signaling) [13], thus avoiding the above high interrupt frequency problem. In the long term, MPEG-Zto-the-home creates a “future legacy” problem as operators realize the limitation of MPEG-2-to-the-home for supporting additional applications. The upgrade to ATM is going to be expensive, not because of ATM itself, but because of the cost of redeR\oyment of new set-top boxes or set-top box interfaces and new headend equipment. Even if we just add ATM to other frequency channels in HFC for supporting newer applications, the new ATM set-top boxes for such applications will be more expensive due to the need for backward compatibility with MPEG-2 TS channels. Lastly, ATTH does not require a separate Level 1Gateway while MPEG-Zto-the-home does. Fundamentally, deploying a hybrid networking architecture like MPEG-Zto-the-home is more expensive because of the additional cost of managing multiple networking technologies, the need for (unnecessary) protocol conversions, and the development of standards for such conversions and the corresponding standards development delay. Furthermore, t h e intense competition in t h e ATM market is bound t o make it more cost effective than the MPEG-2 to-the-home architecture, which has far fewer vendors supporting it. Lastly, ATTH does not requires a separate Level 1 Gateway while MPEG-Zto-the-home does.

The intense competition

in the ATM market is bound to make it more cost effective than the

M PEG -2-tothe-home archi tecture, which has far fewer vendors supporting it.

True Multivendor Support and Standard Interface - At an early stage of ATM standardiza-

tion, it became widely recognized that, although ATM is the superior network architecture for the future, to make it successful in the market place multivendor support with true interoperability is essential. To avoid repeating ISDN’s mistakes of non-interoperable products, ATM vendors quickly realized they needed an extensive interoperability agreement across all ATM

21

For the first time, both the public networking

industry and the LAN

industry have adopted a common

Seamless Public and Private Network Architecture - ATM is the first high-speed network technolo-

networking technology

-

ATM

the

- as

interfaces in a rapid fashion. As a result, the ATM Forum was formed in late 1991 to expedite this process. The ATM Forum activities help tremendously in creating a multivendor ATM market and also help drive down the price of ATM to a commodity in a fast trajectory. Furthermore, since the key basic interfaces for ATM such as UN1 3.1 [14] already exist, they can be used for the residential broadband market to expedite deployment. In contrast, if a hybrid networking architecture such as MPEG-2to-the-home is employed, ATM to MPEG-2 conversion needs to be standardized, and it is not clear where and when this will occur. Since, in the cable industry, no standard network interfaces exist and they have been relying on proprietary technologies from few suppliers (which are intrinsically more expensive), creating standards for the cable industry becomes more critical. This is yet another reason for not using proprietary technologies such as MPEG-2-to-thehome.

networking technology to support a// future multimedia applications.

gy that has strong support across multiple industries. For t h e first time, both t h e public networking industry and the LAN industry have adopted a common networking technology ATM - as the networking technology to support all future multimedia applications. In the public networking arena, ATM is the only networking technology being pursued for supporting broadband multimedia services. For private networking, ATM has been viewed as the premiere next-generation networking technology to deploy for campus backbones and to support multimedia applications in corporate networks. Hence, it is clear that ATM technologies will be pervasive across many different areas and will become the networking technology to invest in. As the public network evolves to an ATMbased network in the backbone, access to public ATM services will become critical. There are already a number of telcos and IXCs that are providing direct ATM access to businesses. By using ATTH, residences with broadband access can take advantage of the public ATM services as they become available. Furthermore, we can assure a compatible public network architecture that will seamlessly connect residential broadband networks. We can provide end-to-end ATM connectivity between homes across the country and beyond; an end-to-end guaranteed QoS networking platform will appear. This is extremely important in supporting ATM endto-end services across a wide area (beyond the residential broadband access networks). (This is actually the original vision behind adopting ATM for B-ISDN.) Moreover, by deploying ATTH, we can minimize the future risk of supporting alternative technologies that become islands of proprietary technologies (such as MPEG-2-to-the-home). Future ProoF - By deploying ATTH at the outset, we will put in place the networking technologies that are future proof. ATM is capable of supporting a wide range of applications and is data rate independent, so it will not be necessary

22

to deploy alternative networking technologies as we add new services o r transmission bandwidth to the home improves.

Broadband Res iden tiaI Network Architectures

A Universal Model To enable a residential broadband market with upwards of thousands of applications and services, there must be a large number of content and applications developers for this platform. Since it is very expensive for content developers to develop for multiple platforms, it is critical that they have a single model for content creation that works over the different network architectures being deployed. In other words, we need to design a universal model for residential broadband network architectures to abstract a single model for all content and applications developers. As discussed in the previous section, the universal model for residential network architecture should be based on an ATM-to-the-home (A7TH) network architecture. The fundamental objective of the residential broadband network is to provide a communication network among service providers and home devices. The sgrvice providers typically provide their service (content and applications) using a collection of servers located in the headend, central office, a remote location, or the provider’s premise. Currently, in the United States, the service providers can also be the cable network operators but .not the telco network operators (except through a separately owned subsidiary). Home devices can be PCs, set-top boxes, and other intelligent devices. Note that although there is a debate over whether PCs or set-top boxes are the right home devices to support interactive residential services, it is clear that both are important because they serve different (but overlapping) classes of interactive applications. For example, PCs are more oriented to the individual due to usage distance from the user to the PC, whereas the set-top box is more family or group oriented. In the following discussion, the term set-top box is used to denote a generic intelligent home device; any occurrence of set-top box can be replaced by other intelligent home devices, such as the PC. It is also very important to provide connectivity from residences to other services beyond the local service providers. These include access to the Internet, the PSTN, the emerging public ATM network, and the corporate networks. Since telephony (plain old telephony service, or POTS) remains as the core residential communication application, access to the PSTN is required in the universal model. Also, access to corporate networks is important for telecommuting. To address the general problem of supporting residential broadband services, a functional model is required to understand all the networking requirements. Figure 3 shows a high level functional model for providing residential broadband services (including narrowband) over a universal residential broadband network model.

IEEE Network

September/October 1995

r

The universal model divides the network into three portions: private, public, and service provider networks. The private residential network refers to the in-home network connectivity. The set-top boxes can communicate either directly to the external public network or to a public network termination (NT) device or even a home network. NT allows multiple set-top boxes connected to the public residential network, while the home network provides direct communication between set-top boxes without going through the public network. If the set-top box connects directly to the external network, it can also be viewed as having a passive N T in between. The public residential broadband network can be divided into two subnetworks: the residential access network (RAN) and the service distribution network (SDN), both of which belong to the network operator. The RAN provides broadband connectivity from each home to the headend or central office. The SDN provides connectivity for different service providers to all the homes through the RAN. The SDN is also responsible for providing connectivity to the Internet, PSTN, and emerging public ATM networks for both the residential customers and service providers. The term SDN is used to distinguish this network from a Level 1 network because a Level 1 network applies to telcos only, in the video dial tone model. Furthermore, the SDN is responsible for connectivity to entities other than service providers, and thus is a superset of the Level 1 network definition. The service provider network (SPN) belongs to the individual service provider. The SPN is responsible for connecting the various servers of the service provider and connecting them to the SDN. In the simplest case. where the service provider has only a single server, there is no SPN and the server is connected directly to the SDN. In the ATTH architecture, all networks between the service provider servers and the intelligent home devices are pure ATM networks. These ATM networks include the RAN, SDN, SPN, and home network. IP is supported over ATM at the endpoints (set-top boxes and servers) to provide connectivity to the Internet and other non-ATM external networks for these endpoints. Residential Access Networks - Since the last mile to the residence is one of the most expensive components of residential broadband service deployment, each network operator chooses the RAN architectures that minimize the cost and risks according to the operator's driving etonomic factors, business models, and existing infrastructure. As a result, the RAN contains the largest variation among different residential broadband network architectures. Due to the cost of dedicating high bandwidth to every individual residence, a significant amount of design involves efficient sharing of the RAN. Deploying ATTH over various RAN architectures is discussed in a later section. Service Distribution Networks - The SDN is responsible for interconnecting service providers

IEEE Network SeptembedOctober 1995

Publicnetwork

Home IlNlr

a Figure 3. Residential broadband network architecture model. I

to the RAN for providing services and applications to the residences. The SDN is a pure (public) ATM network. The SDN can be physically contained in a single headend or central office. or distributed across a metropolitan area connected to Servers in different locations. The regi6nal hub architecture proposed by CableLabs is an example of a distribution network connecting different headends to regional headends for better resource allocation. The interface between the SDN and the RAN should be a network-net[ 141, because both the work interface ("1) SDN and the RAN are part of the public ATM network. NNI allows resource allocation information to be communicated to set up an endto-end ATM connection through the entire public residential network, which includes both SDN and RAN. On the other hand, the interface between the SDN and the Internet, RSTN, and SPN are public user-network interfaces (.UNIs). because they are either non-ATM networks or do not belong to the public network. Service Provider Network - The SPN is a private network that belongs to the service provider. The SPN is responsible for connecting all the servers of the service provider to the SDN (the public network), as well as providing internal communications among the Servers of the service provider. For the ATTH architecture, the SPN is a pure ATM network. The interface between the SPN and the SDN should be a public UN1 [ 141 because the SPN is similar to a traditional private ATM network. which operates independently of the public ATM network. (Since bandwidth allocation decisions in the SPN and SDN are independent, there is no need for an NNI interface between these two networks.)

Home Nehvorks In most current residential broadband network deployment, the set-top boxes are directly connected to the RAN. Unfortunately, this has the disadvantage of requiring different set-top boxes for different RAN architectures. T h e set-top boxes do not have any portability because UN1 interfaces are different for different RAN architectures such as HFC, FITC, FlTH, and ADSL. Due to a lack of standards in RAN design, even for a particular RAN architecture, the network interfaces to the RAN are different from different network vendors.

23

Y t

HFCis a

modular architecture whereby the serving population is

About 500 homes passed

segmented into neighbor-

IIFigure 4. Hybrid fiber-coar network architrctiire.

hood areas,

Ideally, there should be a single UNLr for each type of RAN (x can be HFC, FTTC, FTTH, ADSL, and so forth) between the home network (can be a simple NT) and the RAN. Furthermore, between the home network and all intelligent home devices, there should be a standard interface (UNI), which can be defined for different speeds, such that any set-top box can be plugged into the home network through such a standard interface, independent of the RAN connecting the home network. This vision of a universal broadband home interface is similar to the RJ-11 universal interface today for POTS.

each of which

by in dividuaI is served

fiber trun ks reaching the corresponding fiber node.

Standard Interfaces - To enable a large residential broadband market that can be as big, if not bigger, than today's PC market, the availability of networking equipment from a large number of vendors is critical. This can only be ensured by the availability of standard interfaces or interoperability agreements (such as the specifications from the ATM Forum). Hence, standard interfaces at the boundaries of SPN, SDN, RAN, and the home network are fundamental to supporting a multivendor environment for the residential broadband market. By adopting the ATTH architecture as the universal residential broadband network model, many of these interfaces can be defined immediately using existing standard ATM interfaces such as [14]. This expedites the development of standards in the residential broadband market, and thereby expedites its deployment.

Residential Access Network Architectures

c

.

Lirrently, one of the most active technical debates in the residential broadband deployment is the choice of RAN architectures. Since different RAN architectures have different intrinsic networking characteristics, they each dictate a set of possible applications that can be supported. These networking characteristics include upstream and downstream bandwidth for individual homes, the degree of switching (interactiveness), and dedicated versus shared bandwidth, error characteristics and reliability, power

24

issues, and others. The choice of one RAN architecture over another depends not only on technical merits, but also business models, and the types of applications the network operators intend to support in their residential broadband networks [4].

HfC - In the past few years, the HFC network architecture has emerged to become the cable networkarchitecture for the '90s [ I l l . The HFC architecture consists of fiber trunks from the headend to individual neighborhoods, replacing coaxial trunks of traditional cable network. Each trunk terminates at a fiber node (Fig. 4). The main function of the fiber node is for opto-electronic Qonversion between fiber and coaxial cable. From the fiber node, multiple coaxial drops pass through the homes in the neighborhood (typically 500 homes), similar to traditional tree and branch cable architecture, except with shorter coaxial runs and hence with fewer amplifiers and lower noise. HFC is a modular architecture (similar to cellular) whereby the serving population is segmented into neighborhood areas, each of which is served by individual fiber trunks reaching the corresponding fiber node. The spectrum allocation for the HFC architecture to support switched broadband services typically uses between 450 Bnd 750 MHz for new interactive services in the downstream, while 50 to 450 MHz remains for standard analog broadcast services. However, it is up to network operators to choose the number of channels allocated to both analog and new digital interactive services, based on business reasons. Upgrades from 450 to 550 or even 750 MHz are being carried out in many cable plants across the United States to support these new services. At the same time, the return frequency from 5 to 40 MHz (or 30 MHz) is also being enabled during the upgrade process. HFC is apassband transmission system [15]. The receiver must have a tuner that can tune to one of the fixed frequency bandwidth channels (6 MHz in the United States) between 50 and 750 MHz to receive any downstream signals (analog or digital). To support digital services over HFC, a digital modulation technique must be used to carry digital data across the channels. Two of the

IEEE Network

SeptemberiOctober 1995

-9

*

most popular digital modulation techniques for HFC are QAM [I61 and VSB [ 171. 64 QAM modulation carries about 27 Mbls user data rate (after forward error correction, FEC) over a 6 MHz channel, while the emerging 256 QAM technology can support about 36 Mbls of user data after FEC. The 16 VSB modulation supports about 38 Mb/s of user data rate, which is equivalent to the 256 QAM technology. The HFC network, though broadcast in nature, can be upgraded to support fully switched broadband service by adding switching capability at the headend. To implement the ATTH architecture, an ATM switch interconnecting headend servers to the HFC cable plant can provide interactive services over the cable network. Each port on the ATM switch can be logically connected to a 6 MHz (in the United States) downchannel of t h e cable plant. To receive ATM cells, the set-top box (if connected directly to the RAN) must first tune to the right frequency channel before decoding the information. Each home device can set up its virtual channels (VCs) within its 6 MHz downchannel. In other words, each home device has a virtual link (carrying multiple VCs) between the home and the headend. This is equivalent to a point-topoint switched network; the only difference is that the link is virtual and has variable bandwidth. Each downchannel is shared by multiple home devices to increase efficiency. Each set-top box would receive its information by demultiplexing ATM cells based on VPI/VCI that has been preallocated for its communications. The return channel is used in a similar fashion to multiplex the return data from each home device, except a MAC protocol is required to manage bandwidth sharing in the upstream direction. Hence, standard ATM protocols must be enhanced to support the shared return channels and the passband architecture [4].

FTTC - Fiber-to-the-curb (FTTC) RANs (also known as switched digital video, or SDV) are baseband access networks [15]. F l T C networks run fibers from the central office all the way to the curbside of the houses they service (Fig. 5). Fibers are terminated at the optical network units (ONUs), which typically support between eight and 24 homes. If a passive optical network (PON) architecture is used, a fiber from the central office is split passively into multiple fibers to reach multiple ONUs to reduce laser cost [18]. From t h e O N U to each home, there can be both twisted pair and coaxial cable. Twisted pair is to support POTS, while coaxial cable is to support analog video services. The new digital interactive services can be supported on either twisted pair or coaxial cable. If analog services are not supported, only twisted pair is necessary to support telephony and digital interactive services. Since FTTC is baseband, the multiplexing mechanism for distribution to different homes is by time division multiplexing (TDM). Hence, in addition to opto-electronic conversion, routing (or demultiplexing) is one of the key functions of the ONU; this can be accomplished using a n ATM switch. The home devices demultiplex their information based on ATM cell headers.

IEEE Network

September/October 1995

ir

, Figure 5. Fiber-to-the-curbnetwork architecture. The downstream bandwidth of an FTFC network can be as high as 51 Mbls on coaxial cable or even on twisted pair, especially for short distances such as 500 ft. If no PON architecture is used in the FTTC RAN, FTTC is a point-to-point switched network and implementing ATTH is trivial (using standard ATM protocols). If a PON is used, ATM protocols must be extended to sdpport the shared nature of PON architecture [4].

FJJH - Fiber-to-the-home (FTTH) architecture is a point-to-point network topolocv being implemented by some telcos such ;I' TT [19] and Deutch Telecom [2U]. However, initial deployment of FTTH may not be a truly point-to-point network because, again, a PON may be used to split each fiber from the central office to multiple fibers using a passive optical coupler. This reduces the cost of deployment by reducing the number of laser transceivers required for every home. Bi-directional communication is achieved by a MAC protocol that shares bandwidth on the PON among different homes served by these fiber drops [ 181. The bandwidth on each fiber can be as high as OC-3c. Obviously, this gives the highest bandwidth in both directions of all the different access network types. Hence, the home can easily become the source of information and can provide services to other locations by attaching servers to the access network at the residence. Developing ATTH over F l T H is relatively straightforward because FTTH uses a point-to-point ark h i t ec t u r e. Again , standard ATM protocols must be enhanced to allow sharing of upstream tcansmission in the PON architecture [4]. ADS[ - ADSL is a modem technology deployed over a standard POTS line to dramatically increase the data rate for supporting new'applications. The original ADSL technology supports 1.5 Mb/s in the downstream direction and 64 kb/s in the upstream direction. It can carry an additional POTS simultaneously. However, this is not sufficient to support many interactive broadband applications because consumer quali-

25

/he ATM call model operates at the same layer as the

ATM switch call controller’s connection estublishment

ty MPEG-2 requires approximately 3 to 4 Mb/s for most content. Recently, ADSL has been improved to support 6 Mb/s in the downstream direction and 640 kb/s in the upstream direction (ADSL-2). However, the cost of deployment remains so high as to prevent many network operators from deploying this technology widely. As a result, the ADSL Forum was created last year to standardize this technology in order to quickly drive down the costs [21]. ADSL does, however, allow a telco to deploy a broadband access network in the shortest period of time (especially to remote areas), given the current installed base on twisted pair to the home. This opens up many possibilities for supporting fully interactive broadband applications. Since the ADSL access network technology is a point-topoint switched network architecture, implementing the ATTH architecture is straightforward. Only new ATM physical layer interfaces of the corresponding data rates need to be defined [4].

layer. Hence, the

Signaling Requirements for Residential Broadband Services

performance

r

(call and connection setup time] is comparable to the simple connection setup time.

ere are two classes of signaling requirements for residential broadband services: functional and performance. To support a diverse range of residential broadband services and applications, a number of connection types must be supported by the signaling functions. In addition, to satisfy individual applications requirements and to provide scalability to support millions of subscribers in the residential broadband networks with guaranteed performance (even under heavy load), there are associated performance requirements for each signaling function.

Functional Requirements ATM is a connection-oriented protocol. A virtual connection (VC) must be established before information transfer can occur. Originally, only a simple point-to-point connection was defined for ATM. Also, the concept of call is not distinguished from a connection. When an ATM call is made, it always refers to a single connection being established. To support residential broadband services, more sophisticated connection types and a generalized call model are required. These are discussed below. Multi-Party Connection - The ITU-T has defined the following connection types, which are all important for residential broadband services. Type 1: Bidirectional Point-to-Point - The type 1 connection is a bidirectional point-topoint connection between two parties. Bandwidth is independently specified in each direction,. Asymmetric bandwidth can be supported, and unidirectional connection is the special case with zero bandwidth in one direction. Type 1 also specifies that the physical route taken by the connection in each direction must be identical. Type 1 is the fundamental connection type for ATM and, obviously, must be supported for residential broadband services. Type 2 Unidirectional Point-to-MultipointThe type 2 connection is a point-to-multipoint connection that involves three or more parties,

26

one of which is the root. This is a unidirectional connection for which the root is the only source, and multicasts its information over the type 2 connection to all the other parties of the connection (called leaves). The root is responsible for adding and dropping parties. The leaf parties can also drop from such a connection independently. It is up to the root to specify during the connection setup time whether it allows independent add from leaf parties (known as leaf initiated join), and whether the root should be notified of such actions. The type 2 connection is required to support multicast services, such as broadcast TV. T h e leaf initiated join mode allows viewers to independently join a particular TV channel multicast (i.e., tune to that channel). Type 3: Unidirectional Multipoint-to-PointThe type 3 connection is a unidirectional connection between three or more parties with multiple sources and a single destination. This has the same logical configuration as type 2 connection, except the information flow is reversed, from the leaves back fo the root. Since the cells from different sources (leaves) arrive at the destination (root) carrying the same VPI/VCI, the root cannot reassemble the AAL PDU because the root will receive interleaved cells from different leaves. This cell interleaving problem is the main challenge for designing the signaling prot?cols to support the type 3 connection. There are two special cases, however, for which interleaving is not a problem; the first exception is when AAL 314 is used (which has MID field for demultiplexing), and the second is if only single cell messages are sent. The type 3 connection is important for a number of residehtial broadband applications. One application is advertisement insertion. This requires merging of multiple ads (video clips) from potentially different servers. It is very useful to have a multipoint-to-point connection from all these servers to each viewer. The reason is that at any one instant, there is only one advertisement playing from the viewer’s point-of-view, so the type 3 connection is the most efficient connection type for this application. Otherwise, if a separate point-to-point connection is set up from each server to the viewers, at any one time, only one of the connections is a d v e and the rest are wasted. Type 4 Bidirectional Multipoint-to-Multipoint - Type 4 is the bidirectional multipoint-to-multipoint connection that allows three or more parties to communicate with each other. The information sent by any party will be received by all other parties. This connection type faces the same challenge of cell interleaving as the type 3 connection, and hence can be solved similarly. The type 4 connection is important for supporting group communication, such as multi-user games and multiparty video calls. Multiconnection Call - Originally, the concept of an ATM call was the same as that of an ATM connection [14]. Recently, ITU-T has extended the model of an ATM call such that a call can include multiple connections or even no connection [22]. As such, the concept of connections remains as before (i.e., a virtual connection). But a call between two users becomes an association

IEEE Network

September/October 1995

f t

that describes a set of connections between them. Of course, this can be easily extended to three or more users to become a multiparty/multiconnection call. Also, there can be multiple calls between two users. By extending the call between two users to support multiple connections, multimedia applications can be supported more effectively. Since each media type of a multimedia communication application has its own communications requirements (bandwidth and QoS), it is desirable to carry each media type over a separate VC that satisfies its corresponding communications requirements. In addition, since these different connections belong to the same application, by defining a call that encompasses all these connections, management of this communication is simplified significantly. All these connections can be set up and torn down in parallel. Hence, this also results in performance improvements. Since the call concept can also include multiple parties (three or more), it can support a very rich and flexible application paradigm. This is important for residential broadband service because many applications running on the settop box require communication with multiple servers in the headend or central office. For example, it might be desirable to manage the collection of all connections that are set up as part of a particular application running on a settop box. In this case, we can setup a call that includes all of these connections, and which may connect to more than one server at the headend or central office. In the telco environment, such a call can be billed to the service provider of that application, which in turn bills the subscriber of that set-top box for the application. The concept of a call without a connection is also very useful. When a user makes a call to another user, it may or may not include a connection setup request. A user can set up a call without any connection. This is desirable because it allows prenegotiation of endpoint resources and determination of each other’s capabilities without actually setting up the connection. Otherwise, if a connection setup is issued first, network resources could be held up temporarily, and the user could later discover that such a call cannot be completed because endpoint capabilities do not match, even though network resources are available. Another-advance feature supported by the multiconnection call model is the Common Routing Group (CRG) capability. The CRG capability is very important for routing multimedia calls, because different media types that are carried on different connections of the same call may need’ to be synchronized. For example, the audio and video portion of a movie must be synchronized. By assigning these related connections to the same CRG, they will be routed through the same physical path to minimize the differential cell transfer delay between them.

ATM Call versus DSM-CC Session - Recently, the MPEG standard committee began developing the digital stored media command and control (DSM-CC) protocol that supports a concept called the session. The goal of the DSM-CC session is to allow resource allocation across a het-

IEEE Network SeptembedOctober 1995

erogeneous network architecture, especially one that includes a networking technology that does not support QoS and bandwidth allocation such as the MPEG-Zto-the-home architecture. The session concept attempts to define a higher layer association of resources across multiple networks. Although this solves the general problem of such heterogeneous networks such as MPEG-2-to-the-home, such a session concept is redundant in the case of a pure ATM network such as the ATTH architecture. The reason is that t h e ATM call model discussed above is already a superset of the functions provided by the DSM-CC concept. Furthermore, if the DSMCC session is implemented over t h e A T T H architecture, a performance degradation will result because every connection setup has to go through the centralized DSM-CC controller that operates at a higher layer, external to the ATM network. O n t h e other hand, t h e ATM call model operates at the same layer as the ATM switch call controller’s connection establishment layer. Hence, the performance (call and connection setup time) is comparable to the simple connection setup time.

Performance Requirements To support interactive services, there is a tight delay requirement on the end-to-end (between servers and set-top boxes) connection setup time. For viewers to perceive their actions a s interactive, the response time (from the instant the viewer presses the remote to the appearance of the results on the screen) should be on the order of 100 ms. In a fully distributed operating system environment, many connections can be set up and torn down simultaneously to satis@ a user action on the screen (such as pressing a series of buttons). In general, a 10 ms end-toend call setup time is required for a connection within the residential broadband network (within about 200 miles) across less than five ATM switches. From a call throughput point of view, about 300 calls per second should be achieved on a per OC-3c switch port basis. This assumes a conservative scenario with 30 set-top boxes sharing an OC-3c port through the access network, each requiring about a ten call/second peak calling rate.

For viewers to perceive their actions as interactive, the response time (from the instant the

viewer presses the remote to the appearance of the results on the screen) should be on the order of

700 ms.

Conclus ion t has become clear that the VOD application Iinvestment by itself cannot justify the billions of dollars of in various residential broadband networks being deployed worldwide. To achieve the grand vision of the residential broadband market, thousands of applications must be enabled on these networks. Instead of constantly searching for the single killer application through market studies and speculative guesses, the best insurance policy is to build a platform of thousands of applications such that, collectively, these applications all become killer applications. This article presents a vision of a residential broadband network architecture based on an ATM-to-the-home architecture, which is not only the best technical solution available, but also the best strategic solution for deploying residential broadband services. It proposes a uni-

27

To achieve the grand

versa1 residential model based on ATTH that allows content and applications be developed with the same abstraction. Finally, it discusses the signaling requirements that are critical to enable the vision of thousands of residential broadband services and applications.

vision of the

References [I] Electronic Industries Association.

res iden tial

[2]”Are PCs too powerful?“ Information Week, Apr. 3, 1995. [3]Jupiter 1995 Consumer On-line Services Report. [4]T. Kwok, ATM: private, public ond residential broodbond networks, (Prentice Hall, to be published). [5]T. Kwok, “Communications Requirements of Multimedia

broadband market, thousands of applications must be enabled on

Applications: A preliminary study,” Proc. Inf’l Conf. on Selected Topics in Widess Commun., Vancouver, 1992,pp.

138-142. [6]T. little and A. Ghafoor, “Network Considerations for Distributed Multimedia Object Composition and Communicotion,” IEEE Nefwork, vol. 4, no. 4,Nov. 1990,pp. 32-49. [7]T. Kwok, “Tandem Bonyan Switching Fabric: A New Hi h Performance Fast Packet Switch and its Performance h o t ation,” Ph.D Dissertation, Stanford University, 1 990. [8] G. Wallace, “The JPEG still picture compression standard,” Commun. of the ACM, vol. 34,no. 4,April 1991,pp. 30-

44. [9]T. Kwok, “Broadband ISDN to support Multimedia and Collaborative applications,” Proc. /E€€ Compcon ’92,Feb. 1992,p . 21-24. [ 101 T. Kwot, ” A Vision for Residentiol Broadbond Services:

these networks.

[l

ATM-to-the-home,” Proc. on Noro Int’l Symposium on Emerging Multimedio Communi?, through AJM Networking, Naro, Jopan, May 24-26,1995. 1 1 J. Chiddix, J. Vaughon, and R. Wolfe, “The Use of Fiber Optics in Cable Communication Networks,” / E € € 1. o f lightwave Technologies, vol. 1 1, no. 1 , Jan. 1993,pp.

154-166. [12]T. Kwok, “A Vision for Residential Broadband Services: ATM-to-the-home,” Proc. /E€€ /nf/ Workshop on Community Networking ’95, Princeton, New Jersey, June 20-22,

1995. [13]T. Kwok, “End-to-End Requirements for Supporting MPEG 2 over ATM Networks,“ ATM Forum Contribution 941 109R2,Kyoto, Japan, Nov., 1994. [ 141 The ATM Forum Technical Committee, User-Network Inter-

28

face (UNI) Specification Version 3.1,Se t 1994. [15]J. R. Jones, “Baseband and PossboncfiransportSystems for Interactive Video Services,” IEEE Commun. Mog., vol. 32,no. 5, May 1994,pp. 90-101. [ 161 B. Bauer, “Drop system ond Component Performance: Emergin Requirements in a high bandwidth 64 QAM digital worl$” NCTA Technical Papers, 1993,pp. 244-270. [ 171 R. Cito and R. Lee, ”Practical Implementation of a 43 Mbitlsec (8bit/Hz) digital modem for cable television,” NCTA Technical Papers, 1993,pp. 271 -279. Y. Mochido, “Technologies for Local Access Fibering,” /€E€ Commun. Mog., vol. 32,no. 2, Feb. 1994,pp. 64-

[la]

73. [19]T. Miki, “Toward the Service-Rich Era,” I€€€ Commun. Mag., vol. 32,no. 2,Feb., 1994,pp. 34-39. (201W. Weippert, ‘The Evolution of the Access Network in Germany,” /€E€ Commun. Mog., vol. 32,no. 2,Feb. 1994, pp. 50-55.

[21]P. Kyees, R. McConnell, and K. Sistonizadeh, “ADSL A New Twisted-Pair Access to the Information Highway,” I€€€ Commun. Mag., vol. 33, no. 4,April, 1995. [22]ITU-T Draft Recommendation Q.298~.

Biography TIM KWOK is the chief ATM architect for Microsoh, where he is res onsible for developing Microsoft’s ATM strategy. He arclitected Microsoft‘s ATM-to-the-Homevision to support the residential broadband services. This architecture i s bein deployed in Seattle, Richardson, and Jopan for HFC, FllC, an! FTTH networks, respectively. Before the formation of the ATM Forum in 1991,he was instrumental in organizing the LATM group among Apple, Sun, Bellcore and Xerox PARC to pioneer ATM for LANs (while at Apple), and co-authored the landmark document “Network Compatible local ATM” which became the basis of the first ATM Forum specification (2.0). He has been the principal representative to the ATM Forum since its inception, actively particpating in its technical committee. He pro osed the support of o new closs of service (UBR) for IAN trokc in 1992,in addition to CBR or VBR. In 1995,he wos elected to the Boord of Directors of the ATM Forum and also serves as vice- resident of Business Development Strategies. He wos invited gy Prentice Hall to author ATM: Privote, Public ond Residentiol Broadbond Networks. He received o Ph.D. from Stanford Universiv in 1990,for his invention of o highJperformance ATM switch architecture. His e-moil address is: [email protected].

IEEE Network

SeptembedOctober 1995