Re: [tlug] File Servers from Hell

[Matt Doughty <mdoughty@example.com>]

On Tue, May 07, 2002 at 05:43:46PM +0900, A.Sajjad Zaidi wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Im having some really weird problems with a file server and was
> wondering if anyone knows what they mean.
> 
> The machine is a Duron with 4 IBM IDE drives (all masters), RH6.2 with a
> 2.4.17 kernel. The drives are running RAID-5. I know IDE dosnt sound
> good on a server, but that part has been quite stable and the problem
> doesnt seem to be related.
> 
> It has been in perfect shape for about 1.5 years and had an uptime of
> about 100days before the problem started.
> 
> All I did was add a SCSI card and upon rebooting, it became unstable as
> hell. It would run fsck (took really long on the 40GB+ data and ext2)
> and fail during the RAID sync. Nothing in the logs. No response from
> even 'Alt+SysRq'.
> 
> The problem persisted upon rebooting and removing the SCSI card. After
> that, I shut off Samba and NFSd and everything went well, but died again
> when I turned them on.
> 
> It has been running fine with Samba after I switched off NFSd altogether,
> but died again the same way when I tried to run backups by piping tar
> through ssh.
> 
> My guess is either a bad mobo(not very likely) or a RAID bug.
> 

It sounds like a shared interupt problem. Its possible that when you 
added the scsi card to the mix it caused the mother board to set two
of your other components to share the same IRQ. Even after removing the
SCSI card the IRQ sharing might have persisted.  I would look in the bios
setup for the option to reset the IRQ settings.

--Matt

-- 
"Take away them collisions and the common channel and it's like Christianity 
 without Christ." -Jim Breen (speaking about "full-duplex" Ethernet)