The following text is copyright 1998 by Network World, permission is hearby given for reproduction, as long as attribution is given and this notice is included.
Colliding with reality
By Scott Bradner
Did someone skip a decade somewhere? I thought that the Ethernet vs. token ring wars were finished a very long time ago and Ethernet won hands down. If that is the case why are we seeing so much of the hype over high-speed token ring presented in ways that seem to be echoes of contests long ago fought and forgotten?
There has been a spate of articles in this and other publications that explain in varying amounts of detail how token ring is superior to Ethernet. The most commonly mentioned differences are the relatively small maximum frame sizes for Ethernet, the built-in support in token ring for link-level traffic prioritization and the ability for token ring networks to have multiple simultaneous paths between switches when switched Ethernet environments can only have one. Less frequently mentioned these days but showing up from time to time are the claims that Ethernet collapses under high load and is less predictable than token ring.
There are problems with most of these arguments. Strings of large sized packets are quite rare, and simple delayed-interrupt tricks in the interface cards can deal with most performance issues. Most corporate network designs do not have multiple paths between switches and even fewer will as level-3 switches continue to get deployed. The disputable claims that Ethernet collapses or is less predictable than token ring do not apply in the increasingly common full duplex switched Ethernet networks. The link-level prioritization is interesting but the Ethernet people are also working on that.
But my problem is not that the claims are false or irrelevant but that they are context free. They are presented in a way that implies that network designers should seriously consider moving their Ethernet networks to token ring. It is very silly indeed to imagine that much of that will happen. The disparity in cost between Ethernet and token ring is just too great. 4/16 Mb token ring PC interfaces are more than 6 times as expensive as 100 Mb full-duplex Ethernet interfaces. Token ring switches are 4 times the per port cost of 100 Mb full-duplex Ethernet switches. Ethernet is vastly outselling token ring. I do not have the actual numbers but in a data communications catalog I got yesterday there were 53 pages of Ethernet products and one third of a page that had some token ring products. If you think that the token-ring-is-better argument is relevant, check out the local video store to see how many Betamax tapes they have.
Don't get me wrong. I think the development of fast and gigabit token ring is a fine thing. Those sites, and there are quite a few, that currently use token ring will benefit greatly from devices that could provide them an upgrade path. I just think that those who are touting the wonderfulness of the prospects should be a bit more realistic in describing just who should be paying attention. To do otherwise, as a co-worker once said, is to do data-free analysis and the result is a collision with reality.
disclaimer: Harvard people collide with reality often and occasionally emerge winners but the above expresses my own frustration.