ATM's exaggerated demise: interest in ATM appears to be returning as carriers and operators begin to focus on the problems of multipurpose network design

Telecom Asia, Feb, 2004 by Stewart Fist

When confronted by his obituary Mark Twain once said: "The rumours of my demise are greatly exaggerated," and it's much the same for asynchronous transfer mode (ATM).

Cell-switching was promoted in the 1990s as the universal panacea for both carrier and private networks. ATM was believed to be a transmission protocol equally adept at handling voice, video and data in large and small networks.

ATM depends on simplified switching functions because the basic idea is to break streams of data into small standardized micro-packets, called cells, each exactly 53 bytes in length. Cells have no error correction data or sequence controls, and they carry a virtual path or channel number rather than a full address.

Standardized short cells meant that ATM switches could be fabricated as single units, entirely in silicon, and this created a greatly simplified transport system that someone once called "RISC Networking." Error-checking functions are transferred to the receiving terminal rather than intermediate touters.

The micropacket approach, however, carries a higher cost since the 5-byte header is wasted when only 48 bytes of payload data are being carried--a 10% loss of efficiency. The more conventional packet protocols get their efficiencies by storing and forwarding variable length packets up to thousands of bytes in length, but such efficiencies are directly related to erratic delays.

Telephony is where ATM shines, and here we have little tolerance for delay. The small cell payload can quickly be filled, and cells are despatched every few microseconds, overcoming the so-called "queuing problem." As voice and video compression systems improved, problems of packet-filling latency increased because fewer bits were being generated each second. Today a standard mobile phone codec would take a second to fill a long-IP packet, while an ATM cell would be released in one-twentieth of this time.

The ATM approach is further simplified by establishing guaranteed connections (virtual circuits), end-to-end, so that voice and video pass through without intermediate storage. Nor are ATM cells contending for priority with other micropackets.

At the same time, the small-cell approach makes ATM ideal for handling the bursty nature of highly interactive data communications, variable demands for bandwidth (granularity) and the content-dependent variability of compressed audio and video streams. All of these will be needed even more in the future.

Swamped by IP hype

ATM was only beginning to establish a firm foothold when the Internet boom took off, and like most of the other network protocols it was swamped by the flood of IP hype. This was a period of short-termism, and from the viewpoint of the chip makers, IP protocols became much more important than ATM, so cell-switching was relegated to the backburner.

However, judging by recent discussion forums, interest in ATM appears to be returning as the carriers and major private-network operators begin to focus once more on the long-term problems of multipurpose network design.

Since it is silicon-based, the fortunes of ATM are closely tied to chip foundry techniques and questions of design and manufacturing priority. These chips are specialized, so it is only recently that ATM has caught up with the data rates now being used over optical fiber. But in July last year, Marconi released the first commercial 10-Gbps ATM port card for its high-bandwidth (480 Gbps) BXR-48000 switcher-router.

During the Internet boom it became fashionable to believe that VoIP would drive out conventional telephony, and that circuit-switched technologies would be entirely replaced by packet routers. The arguments were based on dubious calculations of bandwidth efficiencies.

When voice is compressed, long packets simply aren't an option. And the need to give voice packets special priority through the routers is contrary to the basic Internet principles of packet-equality and distributed-management. So while VoIP is a viable corporate and Internet option, it has had little impact on mainstream carrier systems. IP has little in the way of QoS management, while ATM is excellent at this.

The idea that ATM was in decline has been regularly reinforced by the success of IP and MPLS in corporate networks, and by promotional PR known in the ATM Forum as "Cisco-spin." MPLS combines features of ATM and IP, and makes IP traffic look like frame relay; it therefore saved money in large-scale network upgrades since frame-relay over an ATM core had become popular in corporate networks.

But while ATM may have been neglected recently at the corporate level, it is still embedded in carrier systems and strong in intergovernment-agency networks, including the US Defense Department and the larger university networks.

Also, at the carrier level, cell-efficiency questions become insignificant when they have DWDM and optical amplification. Today carriers can double, quadruple or "multi-tuple" their fiber bandwidth overnight by adding more lasers or faster terminals.