December 1997

BY BROUGH TURNER

ATM was going to be the ultimate solution for supporting mixed voice and data traffic. But it hasn’t quite worked out that way. ATM (Asynchronous Transfer Mode) is a networking technology that’s been on the verge of happening for several years now. It’s been hailed as the ultimate solution for voice and data integration — but always in the future. Now ATM technology is actually being deployed, by major service providers like MCI, as an integral part of the Internet backbone. Does this mean that ATM has finally arrived? Will ATM in fact become the dominant infrastructure for the Internet and other networks?

SOME BACKGROUND
What we know today as ATM originally began as broadband ISDN, a concept that was being explored in the 1980s by traditional telephone service providers through the International Telecommunications Union (ITU) in Geneva. The technology was intended to be the ultimate solution for integrating voice and data on high-speed networks. In 1991, a group of companies, frustrated at the slow pace of broadband ISDN standards development, formed a trade association with the objective of accelerating the market through a “rapid convergence of interoperability specifications.” They took their name from the technology underlying broadband ISDN and called their association the ATM Forum. Started in the United States, the ATM Forum quickly became a worldwide organization. While the ITU’s efforts were focused on integrating data with voice communications, the ATM Forum’s initial focus was on using the technology to transport data packets at very high speeds in LAN and WAN backbones. The ATM Forum started with the broadband ISDN standards that the members of the ITU had agreed upon, and then determined which portions to use and where extensions were needed to make the technology useful for real business applications.

In the mid-1990s, the ATM Forum gained momentum and many believed that ATM would be the wave of the future. ATM has a lot of strengths. It is highly scalable and its switching functions can be implemented in hardware. This makes it relatively easy to build extremely fast networks. ATM has been defined to run on top of a number of different basic transports from T1 up through the various synchronous fiber optic systems — called SONET (Synchronous Optical Network) in the U.S. and SDH (Synchronous Digital Hierarchy) in the rest of the world. These fiber optic networks support a variety of speeds, including OC3 (Optical Carrier level 3) at 155 Mbps, OC12 at 622 Mbps, OC24 at 1.2 Gbps, and OC192 at 10.8 Gbps, a speed not yet supported by ATM switches. Finally, ATM has built-in mechanisms to provide the appropriate (and different) quality of service (QoS) levels needed when voice and data operate on the same network. Based on these features, many people thought ATM was going to be the ultimate solution for supporting mixed voice and data traffic. But it hasn’t quite worked out that way.

REALITY SETS IN
The first sign that ATM would not be the cornerstone of voice and data infrastructures occurred when ATM vendors introduced ATM to the desktop. It basically came in two flavors — 25 Mbps ATM and 155 Mbps ATM. The former option was proposed as a low-cost solution providing a step up after 10 Mbps Ethernet. However, 100 Mbps Ethernet appeared on the scene at about the same time. It used the identical protocol and administrative interface as 10 Mbps Ethernet. It worked the same way existing Ethernet did. In other words, 100 Mbps Ethernet provided higher speeds without the complexity and administrative changes that using ATM would introduce. So while 25 Mbps ATM worked, it was never deployed.

100 Mbps Ethernet also met the needs of most applications initially targeted for 155 Mbps ATM to the desktop. 155 Mbps ATM costs much more than 100 Mbps Ethernet and has the same management issues as 25 Mbps ATM. In addition, it appears that by the time increased capacity is really needed, Gigabit Ethernet will be available. So ATM to the desktop is essentially dead, a fact that was finally recognized, even in ATM circles, in the first half of 1996. Now the question is — does ATM have a place elsewhere?

There’s been a lot of press about the fact that public Internet backbones are being upgraded with ATM technology. It’s widely reported that MCI has been using, first 155 Mbps ATM, and then 622 Mbps ATM, for their Internet backbone. All the signs are that ATM is finally taking over a mainstream application. Is this strategic for MCI and the other backbone providers? I suggest it’s just tactical — a temporary phenomenon that is based on the fact that ATM equipment is available today at the necessary speeds. There are fundamental issues, however, that will prevent ATM from retaining this foothold.

ATM BASICS
ATM takes packet data or realtime voice data and divides it into small (48 byte) fixed-size cells, each with a 5 byte header for a total of 53 bytes. Because the cells are a fixed size, it is much easier to design dedicated hardware to switch them at high speeds — much faster than traditional routers could route IP packets. ATM also includes definitions for different “adaptation layers,” which are essentially protocols that define how data is taken apart and packed into ATM cells. ATM Adaptation Layers (AAL1 through AAL5) are the feature that allows ATM to be a voice and data solution. By using one type of adaptation layer, you can take apart IP packets and put them into cells. By using a slightly different adaptation layer you can accumulate samples of live voice traffic from a T1 trunk and put that into cells. A single switching fabric can then handle these ATM cells. In addition, ATM was designed, from the start, to support differential quality of service. This allows time-critical traffic such as live voice conversations to get priority over bulk file transfers. Differential QoS is a feature the IP world is currently scrambling to implement.

While ATM is an elegant design, the ATM Forum did not make the wisest choice in one of the adaptation layers it chose to implement first — Adaptation Layer 1 (AAL1). AAL1 provides a way to transport existing 64 Kbps voice circuits over an ATM network. To do this, AAL1 adds a 5 byte header onto every 48 bytes of payload and steals additional bytes from the payload, causing the actual data rate on the ATM network to rise to more than 70 Kbps per voice channel. Meanwhile, modern DSPs allow toll-quality voice transmission at much lower data rates (6 Kbps, for example). So, while the ATM Forum was focusing on 70+ Kbps voice transmission, IP telephony got started using 6.4 Kbps in packets. And with silence suppression, IP packets are only sent during actual speech, making average data rates even lower.

ATM faired better with AAL 5, which is the adaptation layer for the transport of packets (particularly IP packets). AAL 5 is the technology being deployed in the Internet backbone right now. But does it have a longlived future?

MOVE OVER ATM
As stated earlier, I believe that ATM in the Internet backbone is a temporary phenomenon. One reason is that IP — Internet Protocol — is already acting as the universal spanning layer. That is, IP can run over any transport including ATM, Ethernet, T1/E1, FDDI, and so on. Any desired service or application can be built on top of IP. David Clark of MIT coined the term “universal spanning layer.” No previous open horizontal interface was completely successful at isolating the functional layers below. IP does this. The ATM Forum is working to make ATM into a universal spanning layer, but for now ATM only spans a specific set of ATM bearer services that are not widely deployed. The real universal spanning layer is IP. It is ubiquitous and appears to be the layer upon which most, if not all, future applications will be built.

The key requirement for the Internet backbone is the ability to transport and appropriately route IP packets at ever increasing rates. One way to accomplish this is to use ATM, dividing IP packets into AAL5 cells. This provides a very high-speed solution. But that is not the only way. In 1994, several individuals who were previously active in the ATM Forum left to form a company called Ipsilon Networks ( www.ipsilon.com ). Ipsilon developed a family of products to directly route IP packets using silicon developed for ATM switching but bypassing the ATM overhead. Ipsilon’s efforts attracted the attention of major networking companies like Cisco ( www.cisco.com ) who then announced competing technologies. The key in each of these cases is that IP traffic is directly routed over SONET or SDH fiber networks without the ATM overhead. None of these direct IP over SONET initiatives is very widely deployed. In 1997, the Internet backbone is adopting ATM technology. But product introductions happening now and through the first half of 1998 will support extremely high-speed IP switching and routing directly over SONET without the intervening ATM layer. And, with Cisco routers currently handling the majority of Internet backbone traffic, Cisco’s product direction has a major impact on the evolution of the Internet backbone.

THE FUTURE OF ATM
While we have to give the ATM Forum a lot of credit for their pioneering work and for spawning an industry, the reality is that, at this point, the future of ATM is not very bright. ATM to the desktop is essentially dead. ATM as the ideal scalable backbone for the public Internet appears to be a 1997 event that will be superseded in 1998 by IP over SONET. And ATM as the direct transport for packet voice is being blown away by IP telephony. There is a continuing opportunity for ATM in the wireless industry, due to work on a new adaptation layer, AAL2. AAL2 provides a way to put highly compressed speech into very small packets and put multiple packets into an ATM cell. AAL2 is of great interest to the wireless market where packetized speech is widely used, but not yet transported over IP or ATM. AAL2 provides a way to transport wireless voice packets between base stations and base station controllers. But while there may be a market for ATM in part of the wireless infrastructure, ATM will be largely irrelevant to the computer telephony market and irrelevant to the emerging voice over IP market. Over time, voice over IP will dominate in the Internet; wireless Internet (IP) connectivity will expand, and then voice over IP will take over, even in the wireless space.

I’m not suggesting ATM will vanish, but it does appear ATM will be a minority player at every level. ATM remains of interest to a select group of people and its technology is being reused in other venues, but ATM itself is never going to be the dominant networking technology people once thought. Brough Turner is senior vice president of technology at Natural MicroSystems, a leading provider of hardware and software technologies for developers of highvalue telecommunications solutions. For more information, call Natural MicroSystems at 508-620-9300 or visit the company’s Web site at www.nmss.com . E-mail to the author ([email protected]) is also welcome.