[3858] in linux-net channel archive
Re: 2.0.x TCP problem?
daemon@ATHENA.MIT.EDU (Marek Michalkiewicz)
Thu Jul 25 23:17:35 1996
From: Marek Michalkiewicz <marekm@i17linuxb.ists.pwr.wroc.pl>
To: schenk@cs.toronto.edu (Eric Schenk)
Date: Thu, 25 Jul 1996 09:20:49 +0200 (MET DST)
Cc: linux-net@vger.rutgers.edu
In-Reply-To: <96Jul24.122111edt.15370@dvp.cs.toronto.edu> from "Eric Schenk" at Jul 24, 96 12:21:05 pm
Eric Schenk:
> Hmm. looking at the log files it looks like a problem with fragmentation.
> Possibly the MTU settings are misconfigured somewhere on your network?
MTU is 1500 at both ends. I don't know about routers (Netware 3.11,
three "real" routers, and then Netware 3.11 again). Turning off path
MTU discovery fixes it.
> (It's a bit difficult to say what is going on because I can't see the
> start of the TCP session in this dump.)
I wasn't sure if it's a good idea to post such a big dump to the
mailing list...
> >20:23:05.796113 B.telnet > A.1038: P 1026:2006(980) ack 2 win 14335 (frag 10235:1000@0+)
>
> Next packet sent goes from 1025:2006, what happened to 983:1026?
Now I ran tcpdump on both ends, and compared the two dumps. 1023
bytes are sent and only 980 received. Looks like it gets fragmented
and one fragment is lost.
> Eventually something goes right and we get:
>
> >20:23:56.986761 B.telnet > A.1038: P 3:1026(1023) ack 2 win 14335
Now it's the same on both ends: 1023 bytes sent and 1023 received.
It looks like someone is doing fragmentation incorrectly, and it
may not be Linux's fault, but I still wonder how it is possible that
packets of the same size (1023 bytes of data) usually get fragmented
(and the second fragment is lost) and sometimes they go the same
path without any problems. Is this a known bug with some routers
(or Netware)?
> and things continue. In any case, what ever is going on, it doesn't
> look like a TCP problem to me.
OK, I should have said "networking problem".
Thanks for your help!
Marek
