On Wed, 2002-10-16 at 03:12, Seth Mos wrote:
> On 15 Oct 2002, Austin Gonyou wrote:
>
> > I don't know if the issue is XFS, but I don't know it's *not*
> either.
> > The server stats right now, are that I have two Dell 6650's with 8GB
> > ram, RH 7.2 from the xfs 1.1 installer base, with most recent redhat
> > updates, QLogic 2300Fs atached to ~1TB - 3TB of FC2 storage using
> LVM
> > for volume management.
Lets start here, right here. I got a message last night from Michael
Clark on the LKML which described a very similar configuration to the
above.
> Does the box use interrupt sharing? There are also a lot of people on
> linux-poweredge@xxxxxxxx that see interrupt only handled by the first
> processor.
>
> > The symptoms are that after about 6 days of uptime with loadavg
> ~10-12%
> > most of the time, running oracle8i and some other apps we have, the
> box
> > locks.
The symptoms described above seem to be an interaction of the kernel
stack useage with LVM, Qlogic drivers, and a journaling FS. I see the
problem using XFS, he sees the problem using RHAS with EXT3. Both
kernels have LVM. His solution, so far, was to remove LVM from the mix
and now has >17 days uptime, or since his last boot.
This is more noted by the fact that the only volumes which needed
repair, and have their log dumped entirely, are those volumes under LVM
control.
It is yet unknown whether LVM2 or EVMS interactions show similar
symptoms with the described configuration.
--
Austin Gonyou <austin@xxxxxxxxxxxxxxx>
Coremetrics, Inc.
signature.asc
Description: This is a digitally signed message part
|