Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/12/84; site nbs-amrf.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!gamma!epsilon!zeta!sabre!petrus!bellcore!decvax!decwrl!pyramid!ut-sally!seismo!umcp-cs!nbs-amrf!libes
From: libes@nbs-amrf.UUCP (Don Libes)
Newsgroups: net.unix
Subject: Question on client/server model in 4.2
Message-ID: <101@nbs-amrf.UUCP>
Date: Thu, 2-Jan-86 08:40:50 EST
Article-I.D.: nbs-amrf.101
Posted: Thu Jan  2 08:40:50 1986
Date-Received: Sat, 4-Jan-86 05:34:25 EST
Distribution: net
Organization: National Bureau of Standards
Lines: 16

Using the 4.2 implementation of TCP, each server is restricted to a
very limited number of simultaneous conversations.  One alternative
is to simulate many virtual circuits with sendto and recvfrom,
however I would guess that's very slow.

The IP protocol has its own drawbacks with its reliability and short
packet size.

What am I supposed to use if I want a server with 50 (or even 100)
clients and (reasonably) unlimited-size message lengths?  I am
willing to sacrifice speed.  Is the client/server (using circuits)
model wrong for clients > 32 and if so why?

Also, has anyone actually done any benchmarking of how much more
slowly communications occur when a connection has to be established
for each message?  If not, I will do this.