The following text is copyright 1997 by Network World, permission is hearby given for reproduction, as long as attribution is given and this notice is included.

False idol?

There sure has been a lot of talk of "fast token ring" in this paper over the last month or two. First it looked like nirvana was going to pass us by and now it looks as if it will happen after all. All this talk has been puzzling and frustrating in a number of ways.

Much of the commentary has repeated the old 'Ethernet is bad, token ring is good' falsehoods. It is claimed that Ethernet collapses under high loads, has unpredictable performance and is wasteful of bandwidth with all those collisions. These assertions developed in the early days of token ring and were used by IBM salesmen to try and convince people to use token ring rather than Ethernet. It is amazing to see that people still accept what are essentially old marketing ploys. As early as 1988 DEC published a research report that demonstrated that these assertions were false. ( ftp://gatekeeper.dec.com/pub/DEC/WRL/research-reports/WRL-TR-88.4.ps) and there is a good analysis in Practical Planning for Network Growth by John Blommers.

Even if some of the claims were true, they only effect Ethernet at quite high loads. Do you really want to be running your LANs at sustained high loads? Since traffic on data networks is by nature bursty (see ftp://thumper.bellcore.com/pub/dvw/sigcomm93.ps.Z) the only way to get sustained high utilization is to be spreading out the burstyness by buffering in the interconnect devices. Of course, such buffering will result in increased packet latencies -- hardly a design goal in most networks. If your LANs are that busy, divide them up with routers or switches.

There are many token ring networks in the world and token ring is not about to go away soon. But it is my experience that in most existing large token ring networks parallel Ethernet networks are being constructed-- mostly for cost and availability reasons. Token ring still costs quite a bit more than Ethernet does and while many new PCs come with Ethernet built in, token ring is almost always an extra cost add in.

Fast token ring is being touted as a backbone for token ring LANs. There is a clear reason that it is not a good idea to use Ethernet (fast or gigabit) as a backbone for token ring LANs. Token rings can support much larger packets than Ethernet can and these larger packets can not be transported across an Ethernet backbone without extra work and in the case of non-IP protocols, not at all. In a way we have had fast token ring for quite a while - it is called FDDI. It is a token-based network technology, it runs at 100 Mbps and supports large packet sizes. Many current token ring networks already use FDDI backbones.

Far be it for me to discourage the development of any useful technology but I do wonder just how much fast token ring will add to the networking tableau. I will say that it is not something I would advise a venture capitalist to put much money into.

disclaimer: Harvard has a couple of token ring LANs (gifts of IBM) but is mostly Ethernet so the above are my own frustrations.