On Mon, 23 Jul 2001, Steve Lord wrote:
> No performance is not 'fixed' but it is way better than it was in the
> earlier kernels, we had a scenario where it would just stop dead in
> its tracks and not make any forward progress.
Agreed. I did see that behavior with older kernels.
> It is more complex to setup with the external log, but it looks like a
> pretty viable solution - see Tru Huynh's numbers with a raid1 journal,
> this config still has everything protected and is getting good numbers.
Yes, I'm going to rebuild my RAID setup so that I have mirrored
SCSI partitions for my external log (in the same SCSI disks where the
RAID5 array lives, so the performance won't be optimal either, but I
guess it's the best single disk failure protected setup I can manage with
the hw I have).
I'll post my benchmarks as soon as I get the new setup running.
- Jani
|