The sun begins to set on ATM

ATM has been dead in the enterprise network for a while, a casualty of extremely cost-effective Ethernet-based LAN switching and a confused standards effort that has delivered too little, too late.

ATM interfaces are still much more expensive than their Ethernet equivalents (millions of which are shipped each month), and software support in operating systems has yet to emerge from its infancy. We will never see end-to-end ATM networks built, thereby severely curtailing ATM's theoretical quality of service (QOS) advantages. Because of this, network managers are turning to Internet protocol-based QOS techniques to deal with multimedia traffic.

However, conventional wisdom says that ATM will still win out in carrier networks, driven by network operators' need to be able to aggregate multiple services on a common transmission infrastructure. With carrier capacity being stretched to the breaking point by explosive growth in data services, the need to efficiently combine Internet services, frame relay services and other data services on a common transmission infrastructure is critical. It is essential to maintaining effective, reliable networking, at least until wavelength division multiplexing (WDM) technology can be deployed.

Still, high-speed ATM switching equipment that supports multiple interfaces at OC-48 speeds has not materialized in the marketplace, and ATM may yet again be too little, too late.

As carriers deploy WDM transmission systems, the desire to aggregate multiple services on a common ATM/Sonet infrastructure will be overtaken by WDM equipment that delivers services over individual "colors" rather than one fat pipe.

The ability to get 16 or more OC-48s on a given optical span dramatically increases transmission capacity, and legacy voice and private line services can easily migrate to their own colors. Frame relay-oriented services can operate on different colors without the risk of passing voice traffic on ATM switching facilities, which have yet to prove their suitability in this role.

Already, established companies such as Cisco and Ascend and start-ups such as Juniper Networks are moving aggressively to develop products with native OC-48 interfaces, which will eliminate the need to use ATM switching between the router or switch and the WDM equipment. Of course, protection switching is still needed, but it likely will be added as part of the core switching equipment or in very simple Sonet equipment optimized for this specific function without add-drop multiplexer functionality.

WDM vendors will probably integrate protection switching into their optical transmission equipment, where it can be implemented cost-effectively and reduce network management complexity.

Either way, the fundamental need to use a common, consolidated switching infrastructure such as ATM will pass, in favor of networks optimized for specific services. Carriers that move in this direction are much more likely to deliver best-of-breed services and integrate new capabilities faster and more reliably than those having a "common" ATM-based infrastructure.

ATM is not going to disappear completely, but it will be just another network service-snot the centerpiece of a unified network architecture. This proves yet again that simpler and less ambitious architectures usually win over complex total solutions driven by conflicting objectives, which rarely ship before the technology changes out.