hal@GATEWAY.MITRE.ORG (Hal Feinstein) (11/30/88)
A few years ago someone tried to put tcp onto a slow LAN. The conclusion was that a "cut down" version of tcp would be less "header intensive" and probabily be faster. I'm looking for pointers to others who may have done work dealing with tcp/ip over slow, unreliable channels and if they solved the problem by going to a cut down version. Yes, a LAN is somewhat reliable; the media were looking at is not and the link protocol works hard just to get low data rates across. TCP (even IP) overhead is a heavy burden. Maybe some other protocol like XTP is better?
gutman@MANTA.NOSC.MIL (Lewis M. Gutman) (12/01/88)
I'd appreciate it if you would forward any replies you get about low data rate transport protocols. To the best of my knowledge, XTP was developed mainly for short control messages over *very high* bandwidth channels, so it may not be especially appropriate for low data rate RF links, etc. Lew Gutman Naval Ocean Systems Center
CERF@A.ISI.EDU (12/08/88)
XTP was designed for general end/end transport over very high speed channels. Vint
jbn@glacier.STANFORD.EDU (John B. Nagle) (12/12/88)
At 576 bytes per datagram, header overhead is only 7%. Only if you send very tiny packets is header overhead an issue, and we solved the tinygram problem years ago. Don't worry, be happy. John Nagle