ATM (Asynchronous Transfer Mode)

Asynchronous Transfer Mode promises to deliver vast amounts of bandwidth to network users. While ATM was envisioned as technology for public network carriers , its application has been recast, and you can expect to see ATM deployed in private as well as public networks over the next decade .

ATM is the purported solution to the LAN/WAN integration quandary . Companies are looking for an efficient and cost-effective method of integrating their dispersed multiprotocol LANs, and frame relay, SMDS, and T3 are vying as contenders. So far, none has been wholly successful. LAN technologies, with their ability to carry large amounts of data over limited distances, are inherently unsuitable in a geographically large network. WAN services, although able to efficiently carry voice and to a lesser extent data over long distances, offer limited bandwidth. ATM, however, can effectively integrate the benefits of LAN and WAN technologies while minimizing the side effects of both.

ATM offers a high bandwidth service that is capable of carrying data, voice, and video over great distances. ATM can provide interfaces to transmission speeds ranging from 1Mbit/sec to 10Gbits/sec. It offers low latency, making it suitable for time-sensitive or isochronous services such as video and voice. Plus, it is protocol- and distance-independent.

What's Wrong With LANs

There's nothing wrong with LANs, as long as their users are local to their server and their applications don't require vast amounts of bandwidth. But beyond that, LANs are shared media that don't scale gracefully.

Most LANs are relatively low speed. Ethernet and Token Ring were designed when LANs were primarily occupied with file transfer. LANs stand to be the delivery mechanism of a whole host of distributed, client-server applications, but these applications must be viable over a wide area network. While 10Mbits/sec may have once seemed extravagant, MIPS and RAM are inexpensive, and workstations and PCs are pumping vast amounts of data onto the network. Five users on an Ethernet or Fast Ethernet aren't uncommon; some Fast Ethernet segments have just one user .

FDDI, designed as a 100Mbit/sec backbone technology and rescaled into a high-speed workgroup technology, can accommodate the high bandwidth needs of workstations. Adapter cards and hubs are costly to purchase, and fiber is expensive to install, although twisted-pair FDDI was once expected to bring down the pricing to be competitive with the more expensive Token Ring products. But at 100Mbits/sec, FDDI is inherently a LAN technology and limited to transmitting information a few kilometers, not cross-country.

LANs don't scale well. When traffic becomes overwhelming, network managers segment the network with bridges and routers, thereby reducing overall traffic. Segmentation can increase delayas anyone sitting on the other side of a slow router from the desired server can attest.

LANs don't scale well when they are based on a shared medium. All users on a shared medium LAN must share the available bandwidth, whether it is a 10Mbit/sec or 100Mbit/sec pipe. When a station transmits, it occupies the entire bandwidth, and all other stations that want to transmit must wait until the sender has finished. Even on a high-speed LAN such as Fast Ethernet or FDDI, only one station may use the bandwidth at any given time. Two stations that want to transmit 50Mbits/sec each cannot transmit in parallel.

The Killer Application

The next "killer" application may do more than spur product sales, as killer applications are supposed to do; it may kill the network. The multimedia applications looming on the horizon will consume every spare Kbit/sec on the LAN but will not be sated. Apple, Microsoft, IBM, and many other developers are writing applications that will integrate voice, video, and data. Image-enabled software, such as Lotus Notes, is only the beginning. Novell is integrating image capability into NetWare. E-mail will come equipped so users can make voice annotations. Imagine having workers improve their job skills or students expand their knowledge by downloading video clips from video servers that reside in different cities. Consider the possibilities of interactive video. Now ruminate on these applications' effects on the LAN and WAN.

FDDI was once supposed to be the medium for multimedia applications, but because of its hard-to-expand bandwidth and insensitivity to time delays, FDDI is suitable in only limited applications. Whereas data can tolerate some delays in transmissions, video and voice cannot. FDDI, like all legacy forms of LAN technologies, cannot guarantee that quality of service.

The proposed FDDI-II is sensitive to the needs of voice and video; however, the specification is still under development and is completely incompatible with FDDI-I (or what we think of as FDDI).

The Technical Details

ATM can be used to integrate disparate LANs across the WAN as well as on the LAN. ATM is a switch technology, not a shared media technology, which diverges from traditional LAN architecture, but is quite common in telecommunications.

Specifically, ATM is a CCITT and ANSI standard for cell switching that operates at speeds from 1.544Mbits/sec to 10Gbits/sec, with several specified interim speeds. Cells are short fixed-length packets, and in ATM, the cells consist of 48 bytes of user information plus 5 bytes for the header. Because the cell size is fixed, network delays and latencies can be predicted , making ATM suitable for carrying real-time information. LANs use variable length packets, which makes delays unpredictable and unsuitable for carrying voice and video. With the fixed cell size, large packet switches can be built rather inexpensively.

With an ATM network, any one user can directly connect to any other user by establishing a virtual circuit, which can transfer data over the link with no added overhead. Contrast this with a LAN internetwork, where each packet must find its way through every intermediate router. In other words, ATM is connection-oriented, while most LAN technologies, including IP and IPX, create connectionless networks.

ATM is scalable. With ATM, additional switches can be added to increase the network capacity. In a switch-based architecture, the aggregate capacity of the network goes up as more ports or lines are added. With a shared medium such as LANs, the aggregate capacity remains the same.

Logical connections between users are made via virtual circuits and virtual paths. Virtual circuits can be permanently established, thereby guaranteeing a level of access, or set up dynamically, allowing for the network service to adjust itself to the demand. For each call, the user application specifies the average and peak traffic rates, peak traffic duration, and burstiness of the traffic. By setting these parameters, network designers can ensure that the voice, video, and data traffic get the required quality of service. For example, the network can respond to a traffic burst by automatically allocating additional bandwidth to a particular virtual circuit. Or, certain types of traffic or calls can be prioritized according to their importance or sensitivity to time delay.

Cell relay separates the relaying of data cells from the management of logical connections. Hardware will process the cells, while software will establish the virtual circuits, manage the resources, route calls, and handle billing. This separation into layers enables you to upgrade the hardware and software separately, thereby allowing a longer life for the network infrastructure.

ATM is independent of the upper-layer network protocols, supporting IP and other Layer 3 protocols. ATM can use a variety of transmission speeds and protocols at the physical layer.

Where Atm Fits

ATM may make its appearance in several places on your network: hubs, routers, to the desktop, and eventually, as a publicly offered service.

ATM has made a marginal appearance on the desktop. Some workstation vendors have equipped their machines with ATM interfaces. Multimedia and other bandwidth intensive applications suitable for ATM to the desktop include engineering, financial analysis, medical imaging, and multimedia. Initially, prices of ATM adapter cards were highin the $1,500 to $2,000 range, but prices decreased, though never to the point where a substantial market imperative developed for desktop ATM. Router vendors are incorporating ATM into their products, primarily to create campus backbones and to access wide area links.

Not Just Another Acronym

It seems like every time you turn around, there's another LAN or WAN technology claiming to be the ultimate solution. A barrage of standards and a slew of acronyms assail network planners and administrators. ISDN didn't work, then there was frame relay and SMDS. Why should ATM be any different?

ATM is designed to handle the needs of both voice and data, thereby reintegrating the communication that was disjointed by computers and telephones. ATM has strong support from manufacturers, telephone companies, and users, both domestically and internationally. Whereas the U.S. and Europe use different speed T1, ISDN, and other WAN services, ATMat 155Mbits/sec and higher ratesprovides a common ground for a single global infrastructure. Also, ATM is at the beginning of its lifecycle. ATM supports speeds up to 10Gbits/sec but also accommodates slower speeds. You can install it today, and with the increasing adoption of SONET, the infrastructure and transmission system should be viable for the next decade.

This tutorial, number 50, by Patricia Schnaidt, was originally published in the October 1992 issue of LAN Magazine/Network Magazine.