Received:  by oss.sgi.com id <S553782AbQKOUda>;
	Wed, 15 Nov 2000 12:33:30 -0800
Received: from ns1.engrs.org ([216.54.29.10]:28683 "EHLO ns1.engrs.org") convert rfc822-to-8bit17243i
	by oss.sgi.com with ESMTP id <S553770AbQKOUdS>;
	Wed, 15 Nov 2000 12:33:18 -0800
Received: (from uucp@localhost)
	by ns1.engrs.org (8.9.3/8.9.3) id PAA22762;
	Wed, 15 Nov 2000 15:35:14 -0500
Received: from UNKNOWN(216.54.52.141), claiming to be "mal.sourcesoft.org"
 via SMTP by ns1.engrs.org, id smtpdvXT81m; Wed Nov 15 20:35:11 2000
From:   Frank Byrum <greatbridge.com@ns1.engrs.org>
Date:   Wed, 15 Nov 2000 21:33:59 GMT
Message-ID: <20001115.21335900@mal.sourcesoft.org>
Subject: Re: Neighbour Table Overflow
To:     Mark Spencer <markster@linux-support.net>
CC:     netdev@oss.sgi.com
In-Reply-To: 
        <Pine.LNX.4.21.0011151319110.6256-100000@hoochie.linux-support.net>
References: <Pine.LNX.4.21.0011151319110.6256-100000@hoochie.linux-support.net>
X-Mailer: Mozilla/3.0 (compatible; StarOffice/5.2;Linux)
X-Priority: 3 (Normal)
MIME-Version: 1.0
Content-Type: text/plain; charset=ISO-8859-1
Content-Transfer-Encoding: 8BIT
Sender: owner-netdev@oss.sgi.com
Precedence: bulk
Return-Path: <owner-netdev@oss.sgi.com>
X-Orcpt: rfc822;netdev-outgoing

I have noticed something like this which seems to be related to the ARP 
cache entry expiring and so the packet can not be delivered.  This 
usually happens somewhere between 12 and 15 seconds.  If you hardcode the 
ARP entries the problems goes away.  

BTW, how long does an ARP entry live in the cache?

Frank

>>>>>>>>>>>>>>>>>> Original Message <<<<<<<<<<<<<<<<<<

On 11/15/00, 1:33:20 PM, Mark Spencer <markster@linux-support.net> wrote 
regarding Neighbour Table Overflow:


> I've been running some tests on an x86 Linux based router with basically
> an AMD 486 DX4-100 processor and two PCNet32 ethernet controllers.

> The kernel on the router is 2.2.17 with "optimize as router".

> For these tests, I'm using netperf (tcp stream test) running on two PC's,
> and asking the router in the middle to forward the packets between the 
two
> hosts.

> After delivering packets successfully for about 15 seconds, the unit will
> cease to respond to packets for several seconds.  The only clue that I
> have to the source of the problem is that I receive the message 
"Neighbour
> table overflow".  My understanding of the neighbour table is that it is
> used primarily for ARP'ing.  If so, this is a surprising message since 
the
> router should only be aware of one machine on each subnet, and indeed if 
I
> cat /proc/net/arp I see only the two machines.

> Does anyone have any suggestions at what to look for?

> Mark