[3755] in linux-net channel archive
Re: More data on wizard MTU/Fragmentation problem
daemon@ATHENA.MIT.EDU (Garry Adkins)
Fri Jul 19 15:44:57 1996
From: Garry Adkins <gpa@piranha.ianet.net>
To: linux-net@vger.rutgers.edu
Date: Wed, 17 Jul 1996 08:39:51 -0400 (EDT)
Forwarded message:
From gpa Wed Jul 17 08:38:51 1996
Subject: Re: More data on wizard MTU/Fragmentation problem (fwd)
To: richb@pioneer.ci.net (Rich Braun)
Date: Wed, 17 Jul 1996 08:38:51 -0400 (EDT)
In-Reply-To: <199607171222.IAA03076@pioneer.ci.net> from "Rich Braun" at Jul 17, 96 08:22:31 am
X-Mailer: ELM [version 2.4 PL23]
Content-Type: text
Content-Length: 2074
> You wrote to linux-net@vger.rutgers.edu:
> > I have a fragmentation problem that seems to have surfaced
> > around 1.3.66. It seems to persist even through 2.0.3.
> >
> > When using an mtu on a network smaller than the default 1500,
> > the packets get corrupted. Two internet sites that are unreachable
> > because of this are www.mu.edu, and www.webcrawler.com.
>
> Brian Bartholomew forwarded me your message. Have you gotten confirmation
> from any of the kernel IP gurus? Here's another interesting tidbit on
> the problem:
I just posted it last night, so I haven't gotten any responses yet.
> Dialup users of the Linux server have reported no incidents. Users of
> the FreeBSD server have been griping loudly about unreachable POP3
> mailboxes (on our 2.0 server) and/or web pages, ftp, etc.
This is similar to our original manefestion here. Our News server
(linux 1.2.8) suddenly became unreachable. I was able to kludge a
fix for that by changing the eth0 interface to use the smallest
MTU that any customer had (576). Now packets don't need to be
fragmented. I originally planned to simply change the MTU on
my cisco router to the same, but it doesn't support changing
the mtu on the ethernet port. So I'm still kind of stuck.
> As Brian probably noted, we upgraded to 2.0 on the Monday after it was
> released, having run 1.2.0 on the main server without any changes for about
> 15 months prior. We've been running Linux here since 0.98pl5 (Dec '92).
> We needed 2.0 for its performance and for its module support. It crashed
> about three times in the first two weeks, but has been stable for 12 days
> now on our main server (which has a diverse enough range of heavy
> applications use that it finds most O/S bugs immediately).
I switched to 2.0 because I had converted to elf, and had trouble
getting the 1.2.13 kernel to properly compile with gcc 2.7.2.
The built in samba support helped as well.
I'm glad to see that others are actually seeing the same thing.
I'm very interested in any other people with similar problems.
Garry
