Ethernet: A Personal History

Few things make me feel older than realizing that technologies I helped popularize are now older than the engineers that keep them running today. The latest example is Ethernet. Its name and concept first burst on the scene 40 years ago in a memo from Bob Metcalfe to management at Xerox PARC. Today's Ethernet is very different from what Metcalfe first envisioned, and I've installed and managed just about all the variations.

Ethernet wasn't my first LAN technology. I suffered through such disasters as Corvus Omninet, the original IBM PC LAN, and XNET, a 500-Kbps technology that used parallel connections over ribbon cable. Imagine the joy of installing a 1-inch-wide rainbow cable to the PC on some attorney's Chippendale desk.

Luckily, Ethernet came along. Even though the main trunk cable was yellow coax that looked like a garden hose, at least we could hide it in the ceiling, along with the vampire tap, and have a much skinnier AUI cable to the desktop. Cheapernet, more formally known as thin Ethernet, made life even easier for the installer thanks to reasonably thin RG-58 cable and tees at each PC. The problem with Cheapernet was that because the main trunk cable was daisy chained through tees at each PC, any user could--and often did--disconnect the wrong thing to clean behind a PC and take down the network segment.

One of Ethernet's archrivals was Token Ring. Token Ring had theoretical advantages, but it was hamstrung by low bandwidth, IBM trying too hard to control the technology, and the incredibly stupid MAU that used mechanical relays to connect and disconnect PCs from the ring when they were powered up.

I used to be a big fan of token passing. It provided well-controlled access to the shared LAN media, unlike Ethernet's unruly CSMA/CD method. A Token Ring network could run at 95% of capacity perfectly well while Ethernets started to bog down with collisions and retransmissions at 60% or 70%. Of course, 60% of 10 Mbps was still more than 100% of 4 Mbps, and IBM never figured out how to mix PCs with 4 and 16-Mbps cards on the same network. So Ethernet won.

For several years I made a good chunk of my living teaching networking and NetWare classes, and I remember saying that because collision detection required the first bit of a frame to traverse the entire net before the last bit was transmitted, Ethernet would never run faster than 100 Mbps. Logic dictated that faster Ethernet would mean smaller networks as the propagation delay limits were reached. Local Ethernet, where each device has its own switch port, made this argument look a bit foolish.

Today's Ethernet shares few attributes with Metcalfe's original invention, or even Cheapernet. Thanks to Synoptics we can now run Ethernet over twisted pair. The folks at Kalpana brought us the switch, so we no longer have to worry about CSMA/CD and can run our connection at full duplex--the latest gear can't even shift down to 10 Mbps. In fact, pretty much the only things today's Ethernet and that old yellow stuff have in common are the frame format and the name Ethernet. Metcalfe himself has been quoted as saying, "I don't know what networks we'll use in the future, but I know we'll call them Ethernet".

Many of today's network engineers have never even seen a shared media network like Metcalfe's Ethernet. They think of Ethernet as a switched, not shared, medium. And, of course, in today's world it is.

What will tomorrow bring for Ethernet? Faster speeds, of course. Forty and 100-Gbps Ethernet devices are available now, if you have deep enough pockets, and engineers are working on terabit versions with low-cost silicon photonics drivers. Of course the wireless guys are right behind them--sending data through the ether may be the only real way to replace Ethernet.