Joe Cooper wrote:
> Is noatime still appropriate/necessary for a COSS object store? I'm
> guessing it would be (and I can't test it on my current test machine
> since my COSS dir is on a normal partition with other normal system stuff).
noatime should not matter much for COSS. It is a problem for the "ufs" family
of stores because of the amount of inodes being touched by atime, but for
COSS there only is one inode.
> Is there anything else that ought to be altered on the system to
> optimize for this kind of workload? Perhaps increasing the disk flush
> buffers, such that writes/reads are more aggregated?
The defaults of Linux should work pretty fine I think.
> Interrupts are still going extraordinarily high on write flushes (like
> 12000-16000 showing up on a 'vmstat 1'). Context switches aren't bad
> though...I would have thought not being kernel level aio would cause
> extra context switches...but then I'm ignorant on this subject so can't
> really judge how things ought to be behaving.
Are you using IDE PIO? If so, try DMA.
Received on Mon Aug 20 2001 - 22:51:23 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:14:13 MST