ARP Table Corruption Sles 9
 

Running into an issue which seems to be growing.

Seeing this on at least 4 different SLES 9 servers.

Nics are:
Model: "IBM NetXtreme BCM5704 Gigabit Ethernet"
Vendor: pci 0x14e4 "Broadcom"

Os:
Linux hosta1231 2.6.5-7.283-bigsmp #1 SMP Wed Nov 29 16:55:53 UTC 2006 i686 i686 i386 GNU/Linux

Kern:
SUSE LINUX Enterprise Server 9 (i586)
VERSION = 9
PATCHLEVEL = 3

What we're seeing:

Host linux server is communicating just great, all of a sudden an arp entry will go incomplete for a specific destination server (has been linux, windows, os400). If you try to access that destination host from the source linux server you time out. However if you reverse the traffic.. ie send from the destination back to the host linux server the arp table updates on the source linux server and everything works until the arp entry is cleared or timed out. Placing a static arp entry temporarily fixes the problem.

Any Ideas?

Have you searched the SUSE bugzilla?
http://lists.opensuse.org/cgi-bin/se...&list=opensuse

There are many references and just as many comments..

Checked out that link. Saw several cases of the same issue but nobody has had any ideas of where to look...

After running through the novel (suse) network pdf and checking through everything included I noticed the interupts are pretty high up in the 20- 30's for the nics. But still stumped as to where else we can look.