Oracle FAQ | Your Portal to the Oracle Knowledge Grid |
Home -> Community -> Usenet -> c.d.o.server -> Re: ORACLE on Linux - IO bottleneck
Mladen Gogala wrote:
> How exactly did you test? Did you use elvtune (RH3) or did you put an
> appropriate scheduler in grub.conf (RH4)? This is a very special case of
> a sequential I/O which doesn't tell you much. Usually, I/O patterns
> resulting from the database use are a bit different.
FWIW, and assuming RHAS3upd.3 and upd.5 and ext3 fs:
I've recently done some extensive IO testing as part of a
data centre upgrade.
Using a home-brew blockread program (O_DIRECT or not
depending on switch, 8K reads with buffers aligned on 8K
boundaries, variable seed random reads), I've found that playing
with elvtune gains nothing. With a dd-based exerciser (sequential
reads) I can see a difference but not enough to warrant too
much fussing about. Only in multi-process situations.
I think in slower hardware IO configurations than hyperchannel SAN, it might be more relevant. Or maybe the 2.6 kernel io scheduler would change this picture somewhat.
If anyone knows of detailed benchmark data that lead to the io elevators being developed in kernel 2.4, I'd love to have a look at them. So far, I've been unable to create any test case where they would be of a definite advantage.
Unfortunately, Oracle's ORION exerciser is useless for this kind of testing: last time I looked it concentrated on raw devices only. Ie, about 5% of the Linux dbserver population... Probably appropriate for those considering ASM, OCFS and such but useless for everyone else. Received on Fri Feb 10 2006 - 23:45:56 CST