Report of strange TCP/IP behaviour

William Burrow (aa126@fan1.fan.nb.ca)
Sat, 1 Feb 1997 12:17:03 -0400 (AST)


Sender: butenuth@lilienthal (Roger Butenuth)
Message-ID: <5cqhi6$c67$3@news.uni-paderborn.de>
References: <5ci093$n1q@eve.speakeasy.org> <87afpuu5da.fsf@localhost.xs4all.nl> <5clglp$1vt@stingray.net>

In article <5clglp$1vt@stingray.net>, cummings@stingray.net (Matthew Cummings) writes:
>> In article <87afpuu5da.fsf@localhost.xs4all.nl>,
>> Peter Mutsaers <plm@xs4all.nl> writes:
>> > I'm sorry, no. Face the truth: FreeBSD has indeed better networking.
>> I'm curious, could you please post your benchmarks since you seem to know
>> what you're talking about and I'm interested.

At least under some conditions Linux tcp/ip is dead slow. If you write
packets of varying size in one direction, answering every packet with
one byte, performance is bad for sizes in the range MTU + 1 to about
2 * MTU. This can be observed over ethernet or even locally. You can
download the benchmark:

http://www.uni-paderborn.de/fachbereich/AG/heiss/butenuth/tcp_bench.c

I suppose some sort of bug and would be glad if it can be fixed.

--
Roger Butenuth
University of Paderborn
FB 17, Mathematik-Informatik
33095 Paderborn
Tel.: 05251/60-6611
Fax.: 05251/60-6619