David M. Piscitello, Core Competence, Inc.

In the early to mid '90s, ATM earned and steadfastly defended the high ground against SMDS, frame relay, and other switched WAN technology. ATM was the media darling. It was the glorious, ultimate, seamless networking solution. ATM was to be a universal solution, like the foamy white stuff whimsically advertised in a classic Saturday Night Live commercial as both dessert topping and floor wax. ATM was good for the desktop, the campus LAN, and the WAN.

By 1995, ATM had an enormous market edge and a considerable lead over every conceivable and competing technology. With no obvious competition for the LAN or WAN, analysts began to describe every technology other than ATM as legacy. FDDI? Dead. SMDS and Frame Relay? Relevant for a time, until they become services offered over or through ATM.

Ethernet? Absolutely dead. Count on it.

Maybe not.

Through no fault of ATM technology, many of the killer applications that only ATM-based services could deliver to the desktop " integrated voice, video and data " did not materialize as quickly as advertised. What did materialize were enough choices for emulating, bridging and routing NOS protocols and TCP/IP over ATM to make early enablers hesitate rather than deploy.

Those "multimedia" applications that did emerge seemed to run promisingly enough on (surprise!) Ethernet LAN segments. Especially lightly loaded ones.

Like, LAN segments with two communicating stations.

The Ethernet community took note and asked, "What if we take multiport repeaters, make each repeater port behave like a bridge, and switch 'em all across a fast backplane?" Everyone gets uncontended 10 Mbps to the desktop.

For the vast majority of PCs, 10 Mbps dedicated is either more than the applications, operating systems and network drivers can handle; or it's enough to satisfy the needs of the end user of that PC. No change to adapter cards, drivers, or administration required. Swap a hub for a switch. This is too simple.

Networld+Interop Las Vegas '97 attendees were among the first to test-drive applications over switched 10 Mbps Ethernet equipment. The first serious challenge to ubiquitous and seamless ATM deployment had been issued.

ATM to the desktop? Let's think about this some more.

But what do we use for uplinks from Ethernet switch to Ethernet switch? We need at least an order of magnitude more bandwidth on inter-departmental and campus LAN segments, right? FDDI? Maybe. Of course, you can get better than 150 megabits per second using ATM OC-3, so let's put this between Ethernet switches.

But...could we run Ethernet faster if we shortened the cable run or used fiber?

Enter 100 Mbps Ethernet. Twenty-kilometer runs using single-mode fiber satisfy many campus connectivity needs. Even 100-meter runs of 100 Mbps Ethernet over UTP or STP/Coax make a pretty convincing counter-argument to the claim that ATM is the only scalable bandwidth solution. And if you can switch 10 Mbps Ethernet, switching 100 Mbps is a logical next step.

Switched 10/100 and 100 Mbps Ethernet debuted at NetWorld+Interop in 1996. Scalable bandwith had been had been demostrated using Ethernet, and ATM conceded more of the desktop and a share of the campus LAN.

The story becomes as repetitive as a children's book. My 10-year-old son is already accustomed to switched Ethernet running to his PowerMac. I can hear him say, "But what do we use for uplinks between 100 Mbps Ethernet switches?"

And he's right. We need at least an order of magnitude more bandwidth on inter-departmental and campus LAN segments, right? You do get better than six hundred megabits per second using ATM OC-12, so let's put this between Ethernet switches.

But... could we run Ethernet even faster if we shortened the fiber run?

Enter 1000 Base-T, Gigabit Ethernet. Three to ten kilometer runs using single- mode fiber again satisfy a fair number of campus connectivity needs.

On the show floor at NetWorld+Interop 97 Las Vegas, Gigabit Ethernet unequivocally stole the show. ATM concedes more of the campus LAN to Gigabit Ethernet, astonishingly because it offers bandwidth at a greater scale. Many of the same analysts and pundits who pronounced Ethernet a legacy technology now openly question whether ATM is a tractable LAN technology.

Controversy makes for very good reading, and it's entertaining. But it's important to look past the hype and yarn-spinning here and elsewhere and put what has transpired on and off the InteropNet into context:

The pronouncement of the death of Ethernet was absurd and premature. The crowning of ATM as the one true and peerless interconnection technology for all networks was absurd and premature. A pronouncement of the death or even demise of ATM as a LAN solution is also absurd and premature.

There is no reason to argue whether ATM will succeed or fail on the campus and the WAN simply because it must share the market with Ethernet. History suggests that no one technology will satisfy every need for all eternity. Perhaps the most important lesson is that it's not as important to identify a single technology that suits all as it is to identify the one technology that solves a particular need.

A parting sidebar. There is a certain irony in the fact that correct use of the term legacy indeed applies to Ethernet technology. A legacy is a gift handed down by an ancestor or predecessor. Fast and Gigabit Ethernet inherit the same characteristics that made their shared 10 Mbps ancestor the most widely used LAN and internetworking technology: low latency, high bandwidth, datagram delivery. Same framing, same management and administration, and characteristically suited for "plug-and-play" networking. Irony indeed that by attempting to disparage Ethernet, those who labeled it legacy actually paid Ethernet a compliment.