Eric Sandeen wrote on Thursday 26 May 2005 21:23:
> David Kewley wrote:
> > On Thursday 26 May 2005 10:34, Eric Sandeen wrote:
> >>If you just run local xfs without stacking other drivers above/below
> >>it in the IO chain, you'll be less likely to hit a problem.
> >
> > Thanks Eric. Unfortunately, my typical application has a stack like
> > this:
> >
> > 3w-9xxx (JBOD since hw RAID5 is extremely slow in my hands)
> > md (RAID 6)
> > lvm2
> > xfs
> > nfs
> >
> > What's your educated guess about the likelihood of hitting problems with
> > this stack, assuming ~10 busy nfs clients?
>
> on ia32 you're doomed. on x86_64 I'd test it. :)
I'd love to test it thoroughly, but my customers want this NOW, and I'm not
willing to take the risk with their data. :/ So I'm going with the RHEL4
default, which is ext3, on the expectation that RH has done thorough testing
for its customers. All the same, I think I'll ask on the RHEL4 mailing list
(nahant-list).
If I manage to free up some testing time with a similar setup in the future, I
certainly will test it & let y'all know what I find. I *do* seem to have a
lot of requests for multi-TB servers these days. :)
Thanks for your feedback!
David
|