From: Danny ter Haar <dth@picard.cistron.nl>
http://marc.theaimsgroup.com/?t=112280339800001&r=1&w=2

A tyan AMD64 opteron machine functioning as a usenet gateway really
pumps some traffic a day (http://newsgate.newsserver.nl)
Incoming traffic comes through a optical gig-E card (acenic)
and local traffic is fed to our spool boxes through cupper gig-E
(tigon3). Machine uses adaptec onboard scsi disks.
At first i thought i had a hardware problem since no kernel would
survive longer than 30 hours. Ofcourse i ran memtest for a couple
of days. Than i compiled 2.6.12-mm1 and this kernel surviced 18 days
without a problem.

reboot   system boot  2.6.12-mm1       Sun Jul 31 09:47          (01:48)
reboot   system boot  2.6.13-rc4-git2  Sat Jul 30 18:29          (17:07)
reboot   system boot  2.6.12-mm1       Sat Jul 30 14:12          (04:14)
reboot   system boot  2.6.13-rc4       Fri Jul 29 14:16        (1+04:10)
reboot   system boot  2.6.13-rc3-mm3   Fri Jul 29 12:17          (01:50)
reboot   system boot  2.6.12-mm1       Thu Jul 28 00:06        (1+12:09)
reboot   system boot  2.6.13-rc3-mm2   Wed Jul 27 22:27          (01:36)
reboot   system boot  2.6.13-rc3-mm1   Wed Jul 27 11:22          (12:41)
reboot   system boot  2.6.12-mm1       Sun Jul 17 15:51        (9+19:29)

Machine does have serial console (and remote powerboot) but no logging
possibility (portmaster1). When it crashes i think most of the times it
has something to do with IRQ.
2.6.13-rc4-git2 stopped working with the following notice:

Jul 31 03:28:18 newsgate kernel: hw tcp v4 csum failed
Jul 31 05:56:59 newsgate kernel: NETDEV WATCHDOG: eth3: transmit timed out
Jul 31 05:56:59 newsgate kernel: tg3: eth3: transmit timed out, resetting

Serial console kept spitting those messages but it gave no prompt
anymore. remote powercycle was needed to get it back.

More info/config can be found at: http://newsgate.newsserver.nl/kernel/