mo@LBL-UNIX@sri-unix (09/18/82)
From: mo at LBL-UNIX (Mike O'Dell [system]) Date: 10 Sep 1982 23:58:32-PDT In my short note about our pronet performance numbers, I committed a grievous mispeak which I wish to publicly retract and clarify. The offending sentence went something like "IP/TCP is a pig..." This is patently NOT true and not at all what I intended to say. (The main reasons it came out like that were stupidity, damn foolishness, and lack of proofreading, but the excuses don't matter.) There is a great deal of folklore about the performance of IP/TCP and that remark didn't help any. I want to make some positive statements to help set the record straight. The numbers we are seeing with the 4.1a TCP implementation running on the Pronet ring are the same order-of-magnitude as the numbers cited by Bill Croft in his measurements of the Purdue network. That speaks VERY highly of the protocols and implementation. To understand the significance of this, it must be noted that Bill Croft's protocols were oriented towards local area communications, using links which provide hardware support for reliable transmission. (This is no reflection on Bill's fine work, but merely a difference in orientation.) IP/TCP, on the other hand, is designed for a much larger universe of applicability, and takes a much less optomistic view of how well the world works. IP/TCP is running quite well on many different machines and across more different kinds of links than any other protocol I know of: from phone lines to Packet Radio to Hyperchannel to Wide-Band Satellite. Having one powerful protocol suite (not tied to any particular vendor) which can operate sucessfully in such a wide range of environments is incredibly valuable to the network planner. He is truly independent of the choice of wire. I am genuinely sold on the value of IP/TCP . But more importantly, I do not want my previous stupid remark, however unintended, to contribute to the erroneous folklore. (Flaming at his own incredible ineptitude) -Mike O'Dell