View ATM in shades of gray

A TM seems to have fallen into the category of technologies that everyone loves to hate.

We are assailed by regular trade press reports of ATM's woes. In a recent Network World article, the inimitable Bob Metcalfe, one of the fathers of Ethernet, repeated his mantra: "Gigabit Ethernet is kicking ATM's butt." Even the ATM Year 98 conference earlier this month indulged in a bit of ATM bashing, with sessions such as "Did the ATM Forum ruin ATM?"

Is ATM dead? No. Will it take over the world? Clearly not. No technology can. But for some reason, we in the technology industry build up unrealistic expectations of a technology and then blame it when it doesn't rise to the occasion. We paint things in black and white: Either a technology is a success or it's a failure. However, with network technology, as in all of life, few things are black and white.

ATM certainly has its faults. It is relatively expensive and complex. It is hard to understand - let alone implement - the nuances of ATM's classes of service, token-bucket and leaky-bucket traffic policing schemes, and the routing intricacies defined in the Private Network-to-Network Interface specification. And LAN Emulation (LANE) is something of a hack, even if it does work.

But ATM also has many strengths. It has proven to be a solid technology for many enterprise and service providers.

At the enterprise level, ATM is the natural successor to FDDI. It offers scalability and fault tolerance, and acts as an integration point for technologies ranging from Ethernet and token ring to frame relay. Many organizations are using ATM's circuit emulation capabilities to combine their voice and data traffic over metropolitan-area networks. Universities and teaching hospitals are deploying ATM for distance learning and related applications.

Over the years, I've spoken with many IT managers who have deployed ATM. While admitting they had to climb a learning curve with the technology, all are pleased with how easy it is to operate their ATM networks. Service providers appear to have similar positive operational experience with ATM. In one case reported by Ron Jeffries, principal of Jeffries Research, an ISP noted that it takes the same number of people to support seven routers as it does to support 34 ATM switches.

The reality is there will always be the next Holy Grail of technology, be it Gigabit Ethernet, IP over Synchronous Optical Network (SONET) or Multi-protocol Label Switching. With each new technology, there's enormous optimism about what it can do. But once we begin deploying these technologies, their warts begin to show. Already, we've seen that Gigabit Ethernet has some growing up to do.

In various press reports, vendors of Gigabit Ethernet adapters have confirmed that their throughput is typically less than 500M bit/sec, with Unix systems performing better than Windows NT ones. Even Alteon Networks, using its proprietary large-packet format in conjunction with its Gigabit Ethernet adapter, can push only a little more than 600M bit/sec of traffic on and off a Sun Microsystems Solaris-based machine. In contrast, The Tolly Group found that 622M bit/sec ATM adapters from FORE Systems, configured with extended packets, achieved throughput in excess of 1G bit/sec between a Unix client and a server, and about half that performance for NT-based machines. (While the Unix machines are clearly superior for network I/O, we all know that Unix is dead and Windows NT is its successor. But I digress.)

Will Gigabit Ethernet survive these growing pains? Of course. But the reality is that if Ethernet endures another 25 years, it will be Ethernet in marketing terms only. Already one of Ethernet's defining characteristics - its carrier-sense multiple access with collision detection access method - has been all but eliminated by the widespread adoption of switching.

Vendors have also begun tinkering with another of Ethernet's associated technologies, the Spanning Tree Algorithm. By the time they're done, Spanning Tree will no longer be Spanning Tree. Likewise, vendors have already begun experimenting with larger packet sizes. (At least in ATM, extended packet sizes have already been defined under LANE.)

ATM, like Ethernet, must go through a natural maturation cycle and evolve to accommodate market realities. And the industry has already learned a lot from defining and implementing ATM in terms of quality of service issues, routing and switching architectures.

ATM will continue to influence the direction of other technology developments. Far from being dead, ATM is growing slowly but steadily, with the market for LAN-based ATM switches projected to break $1 billion this year, according to Dell'Oro Group in Portola Valley, Calif.

Rather than continue to view ATM in black and white, the industry would be well served to appreciate its many shades of gray.

Mary Petrosky is a senior analyst with The Burton Group, an information services firm that provides in-depth technology analysis. She can be reached at (415) 572-0560 or petrosky@tbg.com.