Network Computingwww.networkcomputing.com

As a result of these meetings, and particularly because of my discussion with Molta, I came away feeling even more sure that ATM would be used for carrier-class applications and a few niche campus applications. Here's why.

When we met with the Cisco folks, we saw a technology demonstration where they'd put, on a single Catalyst 5500 board, nine Gigabit Ethernet interfaces with three additional gigabit paths to the backplane. So, something like 12 Gbps could be moved across this board. Because it was a technology demonstration, no pricing was discussed, but I assume it will be in the ballpark with other gigabit products on the market and cost a few thousand dollars per port.

FORE's people had some interesting news for us, too. They talked about their PNNI implementation, which they claim is more complete than anyone else's in the industry. They also talked about the company's LANE (LAN Emulation) implementation with its redundant LANE servers and support for their emerging MPOA (Multiprotocol Over ATM) strategy. They told us how--with the company's latest offer ings--very large, highly fault-tolerant networks could be built. We asked specifically about the threat of Gigabit Ethernet, and they pointed out that with ATM, you could build a totally fault-tolerant network. They also pointed out that this wasn't possible with Ethernet and that with the length restrictions of Gigabit Ethernet, things get really tricky.

Later, as Molta and I strolled the show, we ventured upon the booth of one of the newer vendors, and we saw a sign that read "28 MPPS." We both did a bit of a double take, mostly because it took a second to figure out that MPPS stood for million packets per second. Molta shook his head and told me that at Syracuse University when their backbone router is busy it passes about 50,000 packets per second (pps), and their main backbone switch does about the same. So at Syracuse University, with its 20,000 users and more, they manage to generate some 100,000 pps during peak loads. He shook his head and said, "I just don't know who would need to switch 28 MPPS."

It became evident to me that Gigabit Ethernet, as opposed to ATM, would be the next decade's technology. ATM solves complex, world-class problems. Ethernet is a cheap technology that has evolved to solve the problems of the Syracuse Universities of the world. I've been to Syracuse. The network isn't simple, and the users have extraordinary demands. In fact, the network at Syracuse University looks like that of a typical medium-to-large organization--corporate, university or government.

Great Expectations So, though a few leading-edge concerns might need to move a million packets a second, the majority--probably 90 percent of the market or more--are just like the folks at Syracuse. You'll need to move a million packets a second sometime near the turn of the century. And when you do, Ethernet (and TCP/IP) will have evolved to meet your needs. In fact, it'll be old hat. You won't need MPOA or emulated LANs, and you won't need to concern yourself with redundant LANE servers. You'll just upgrade to the next Ethernet speed, and with it you'll get all the traffic shaping tools you need to keep your network running. Carriers and huge ISPs? They'll buy ATMÉbut they're solving a different problem.

Art Wittmann can be reached at awittmann@nwc.com.