Script: ATM audio primer

For years, managers of corporate communications networks thought of data networks and voice networks as completely separate - one network dedicated to data and one dedicated to voice.

But many now think it would be more efficient if all traffic-data, voice and even video-traveled over the same network.

That kind of streamlining is the ultimate goal of a technology known as Asynchronous Transfer Mode or ATM.

Advertisement:

ATM is a packet technology, which means that all the information sent over an ATM network is broken down into discrete packets. Unlike other packet technologies, ATM employs uniformly sized packets, called cells, each of which is 53 bytes long.

Because the cells are all the same size, cell delay at ATM switches is more predictable and manageable.

ATM was designed specifically to handle broadband applications efficiently and at the same time let users give certain types of traffic priority treatment on the network.

For example, voice traffic, which cannot tolerate much delay, can be marked "high priority" with a guaranteed bandwidth and minimal delay. Less sensitive traffic, such as electronic mail, can be marked for lower priority.

The first five bytes of an ATM cell, known as the header, contain an address that tells where the cell wants to go in the network. The header also specifies the priority of the "payload" information carried in the remaining 48 bytes.

ATM networks are linked together by a series of ATM switches that take in cells from various sources and switch them out again. There are two ways to do establish connections between the switches.

In a permanent virtual circuit, network managers pre-set paths from switch to switch ahead of time. PVCs were the initial way to establish virtual circuits between sites before the technology supporting switched virtual circuits was developed.

In a switched virtual circuit, the cells travel along paths that are set up on the fly and torn down again after a designated period. The advantage of this approach is that switched virtual circuits occupy switching capacity only while they are set up, increasing switch efficiency and so payback on investment.

In either case, when a network device, such as a router, wants to send cells into an ATM network, it requests a virtual circuit. That request is passed from the initial ATM switch to other ATM switches in the network, with each switch determining whether it can handle the request. If all switches along the path can accommodate the request, the virtual circuit is established. If not, the request must be repeated.

That virtual circuit will support a certain quality of service that has been preset in the ATM network switches based on user requirements. The most common ATM qualities of service are known as constant bit rate, variable bit rate, available bit rate and unspecified bit rate.

Constant bit rate guarantees low delay and the ability to handle a specified cell rate all the time. As long as the traffic falls within the cell rate, the network will handle it. It is intended for voice and high-quality video traffic.

Variable bit rate is intended for bursty traffic. It allows traffic to burst up to specified rates for specified periods of time before discarding cells. Variable bit rate comes in two flavors: real-time and non-real-time. Real-time VBR limits delay that would hamper certain applications such as some types of enterprise resource planning systems that require low network delay, plus some forms of video. Non-real-time VBR is for bursty traffic that is unaffected by delay, such as data make it: such as more traditional LAN traffic or e-mail.

Unspecified bit rate is a best-effort quality of service. When there is available bandwidth, the switch forwards UBR traffic. UBR is meant to support non-critical applications like overnight file transfers.

Initially, ATM was regarded as a way to send data faster than conventional means within a local area network. But newer technologies such as Gigabit Ethernet offer the speed advantage without the drawbacks of ATM on the LAN.

One of these potential drawbacks is the amount of ATM traffic used for overhead. In addition to the five bytes of each 53-byte cell devoted to the header, in some cases two more bytes are used for error checking - and some cells are used entirely for sending administrative data. All this can consume up to 20 percent of the throughput on a given circuit. This "cell tax" makes ATM an inefficient transport method for data-only networks where quality of service is not an issue.

But ATM has found a home on the backbone of large wide-area and metropolitan area networks. For example, carriers use ATM networks for transporting multiple types of traffic across long distances. Voice, frame relay and other data traffic are converted into cells and sorted out at the other end of long-haul links.

In fact, large carriers are so heavily invested in ATM that they often mask its use under other names, due to ATM's failure to achieve marketing buzz. Many more carrier services are ATM-based than most users realize.

For example, AT&T's IP-Enabled Frame Relay is a service that attaches special labels representing groups of user IP addresses to create the illusion of an IP virtual private network. But it's really ATM between AT&T switches. SBC's Project Pronto is sold as a DSL connection to a nearby neighborhood gateway, but it's all ATM from there. Sprint's Integrated On-Demand Network, or ION, is pure ATM from the customer premise device out.

And standards groups are working on ways to make ATM more attractive in LANs. Multiprotocol over ATM, or MPoA, standards let non-ATM networks send traffic across an ATM network and tap into ATM quality of service guarantees. ATM has also gained more flexibility in the WAN via recent standards. Some ATM carriers now allow users who can't afford T-3 connections to choose multiples of T-1 bit rates via Inverse Multiplexing over ATM. And all the large carriers now allow users to keep frame relay equipment at some sites and ATM equipment at other sites in a single corporate network. That's via a standard called Frame Relay-to-ATM Interworking.

Few users will ever realize ATM's early promise of sending all types of traffic over a single network. That's probably because ATM took too long to overcome early problems with expensive, hard-to-use equipment, and meanwhile most efforts to create a single global network shifted to IP. But ATM remains a key consideration for users needing to reduce disparate networks today, while still enjoying reliable classes of service for different kinds of traffic.

Back to the main ATM audio primer page

ATM research page

RELATED LINKS