hal@GATEWAY.MITRE.ORG (Hal Feinstein) (12/09/88)
A few weeks ago I put up a query about slow TCP. My thanks to those who replied. A few have asked me to see the replies so I provide a sample of each for the record. The original problem we faced was a non-cooperative link media which simply refused to be tamed. Now every bit becomes hard to move from one side of the network to the other. Its expensive to devote much bandwidth to upper layers but we need end-to-end reliability. What to do? People suggested looking at SLIP. (cblank@mitre.gateway.org, hooper@mitre.gateway.org, steve@mimsy.umd.edu, karn@ka9s.belcore.com) A number sited Van Jacobsons work on head prediction and compression. (craig@nnsc.nsf.net, steve@umiacs.umd.edu) Employ circuit switching for reduced overhead. (lazear@mitre.gateway.org) Avoid TCP altogether, use the ISO transport instead... (anonymous@osinet) > ...I belive TCP is in fact one of the least inefficient > protocols for the problem it tries to solve -- a highly > reliable stream given a complex packet based internet with > widely varying loss characteristics. (jqj@hogg.cc.uoregon.edu)