ATM Basics

Asynchronous Transfer Modewhat a mouthful of vague abstractions. If there weren't so much money and job security riding on ATM, it would be awfully tempting for many of us to ignore it. A transfer mode is a method of transmitting, multiplexing, and switching data in a communications network. (Multiplexing is combining multiple streams of data on a single circuit; if it weren't for multiplexing, the view of the sky in our cities would be blocked by telephone wires.)

It's only possible to understand "asynchronous" by getting a grasp on "synchronous." The terms refer to digital signals; more specifically , they identify two ways that units of data are framed or blocked within a stream of bits. Synchronous signals are closely tied to some sort of clock, so each unit of data begins, for example, precisely at 0.0ms, then 7.5ms, then 15.0ms, then 22.5ms, and so forth. Asynchronous signals are not bound tightly to a clockperhaps their data units have a start and a stop bit, or some kind of unique bit pattern to identify the beginning and end of a character or a packet.

Most serial communications and practically all LAN communications are asynchronous, but most data transfers in and out of your microprocessor, the traffic on your parallel port, and the traffic on your computer's bus, are synchronous. Given a steady stream of data, synchronous transmission tends to be more efficient than asynchronous, while asynchronous transmission tends to be more flexible and resilient.

The telephone companies, who incidentally built the infrastructure for wide-area computer data communications when they built their voice networks, can be forgiven for concentrating on a data typehuman speechthat is highly intolerant of timing variations. As they built equipment to handle thousands and millions of simultaneous conversations, they developed techniques for multiplexing numerous digital voice circuits on single linesfirst copper wires, then fiber optic cable.

Time division multiplexing turned out to be the best way to combine many telephone circuits on a single physical cable. At each level of concentration, a 64Kbits/sec telephone circuit is tied to a specific time slot. If the timing of a telephone call is disrupted, perhaps through being routed via a satellite or as a result of some malfunction, we might hear a slightly delayed echo that makes it nearly impossible to keep talking, or the whole conversation may be unintelligible. Synchronous communication is well matched to the voice data type.

Video and much multimedia material also matched well with the characteristics of synchronous communication. Not only is there often human speech or other sound involved, the sound can't be allowed to wander away from the image; if it does, the video or multimedia session starts to feel like a badly dubbed Japanese sci-fi movie.

Unfortunately, most computer data communications do not fit well with synchronous methods . For one thing, they tend to be bursty , meaning the ratio of the peak data rate to the average data rate is high. (Synchronous data links have no time gapsthe peak data rate is the same as the average data rate.) Thus, data communicators face the quandary of either paying huge amounts for data pipes that are mostly idle except at peak usage times or suffering through long delays at peak times with a less-expensive small data pipe. The telephone network is not well suited to supply bandwidth-on-demand.

ATM is the result of a compromise among all of the data-type constituencies to find a single common denominator for all types of data. One alternative to time division multiplexing is to use packet or cell multiplexing. A stream of bits is broken up into discrete packets or cells , each of which has a header indicating its path and other worthwhile information. If the cell size is made small, and the overall throughput of the circuit is high, delay-sensitive traffic can be carried along with bursty types of data successfully, and everyone gets what they need from the data link. Voice and video work without glitches, and data customers ( potentially ) get bandwidth-on-demand. As a universal transport, ATM can plausibly be installed on the desktop, on departmental and campus backbones, on high-capacity wide area services, and even on a global information superhighway system.

During the development of the fundamental ATM definition, the voice interestsparticularly the European telephone providerswanted a 32-byte cell with a 4-byte header, while many North American interests preferred a more efficient 64-byte cell with a 5-byte header. The compromise of a 48-byte cell with a 5-byte header was reached, so, an ATM cell is a 53-byte entity.

ATM And Networks

Like frame relay and X.25, ATM protocols are connection oriented. ATM sessions take place over virtual circuits (virtual because they need not use particular physical paths, although once the virtual circuit is established, it stays in place for the duration of a session). (For a graphical representation, see Figure 1.) Most, if not all, of today's ATM services offer only permanent virtual circuits (PVCs); setting up and tearing down PVCs is a job for the telephone company unless the ATM network is completely private. The real promise of bandwidth-on-demand will be fulfilled when switched virtual circuits (SVCs) become available. PVCs are comparable to leased lines, while SVCs are comparable to dial-up voice service. An ATM SVC will typically take only a fraction of a second to be established, however.

Figure 1: Computer Data: Packets of computer data are chopped up into ATM cells for transit over the ATM virtual circuit, then reassembled at the receiving end. Source: ATM Forum.

With its connection orientation, ATM does not readily compare with shared medium protocols, such as Ethernet and Token Ring, or with connectionless protocols that perform routing, such as IP and IPX. With the development of LAN emulation standards, ATM services can be made available to Ethernet and Token Ring networks. Products for translating frame relay data to ATM have been announced. IP and Address Resolution Protocol over ATM are described in the Internet RFC1577. In general, ATM fits into the data link and physical layers , but because connection-oriented protocols don't require routing, it is possible for ATM to provide services to the upper layer protocols directly. This is the sense in which ATM is supposed to sound the death knell for all routers.

ATM Protocols

The top layer of the ATM protocol stack is the ATM Adaptation Layer (AAL). Different AALs correspond to the different data types ATM supports. Thus AAL1 permits the ATM device to closely resemble a constant bit-rate voice circuit; AAL3/4 and AAL5 are used for variable bit-rate data types, which are those typically found on computer networks. The AAL is also responsible for integrating the inherently connection-oriented ATM with connectionless data sources, enabling ATM clients to emulate broadcasting and multicasting.

The ATM layer is the common core of all ATM technology. There are multiple AALs and multiple physical layer options, but the protocol that describes the cell header layout and governs the actions of switches on the cells is a constant. The ATM layer is responsible for cell routing, multiplexing, and demultiplexing .

Before any user data can flow over an ATM virtual circuit, each intermediate switch must create a local routing table entry that maps the inbound virtual channel identifier to an outbound port. In order to simplify the routing overhead for intermediate links, ATM defines virtual paths (VP), which are basically virtual channels defined over two or more physical links that are treated as a unit. VPs are semi-permanent connections, and their routing tables may be set up in advance. A packet that travels along a VP will not have to be rerouted at each of the component VCs.

Based on the type of data carried in cells, the ATM layer interleaves multiple streams together based on the priority of each type. It is also responsible for identifying congestion, managing faults, and managing traffic.

At the physical layer, now customarily referred to as PHY, ATM supports (or will support) multimode optical fiber, single mode fiber, STP, coaxial cable, and UTP, at throughputs as high as 10Gbits/sec, although the speed of ATM can be extended as far as the market's ability to pay for it. ATM traffic can readily fit into SONET or SDH (Synchronous Digital Hierarchy, the international superset of SONET standards) data streamsthe 155Mbits/sec single- and multimode fiber physical layer standards are based on SONET frames . A 45Mbits/sec standard for the DS3 interface, which uses coaxial cable, has also been defined. DS3 facilities are much more widely installed in North America than SONET facilities. At 100Mbits/sec, ATM can use the physical standards defined for FDDI.

The ATM Forum has chosen to adopt existing physical layer standards wherever it can. Potential future standards may include: 52Mbits/sec over Category 3 (or higher) UTP; 155Mbits/sec over Category 5 UTP; and 1.544Mbits/sec for T1 (or DS1) lines. IBM and several allied companies have also proposed a 25Mbit/sec standard for desktop connections using Category 5 UTP cabling.

This tutorial, number 81, by Steve Steinke, was originally published in the May 1995 issue of LAN Magazine/Network Magazine.