Extreme filesystem performance decrese going from FreeBSD4 to
tijl at ulyssis.org
Wed May 13 14:05:11 UTC 2009
On Tuesday 12 May 2009 20:16:21 Andy Dills wrote:
> A customer of a customer needed to expand the size of the filesystem
> on a webserver they run, and since it was still running 4.11, they
> decided to go ahead and upgrade to 7.1.
> Unfortunately, they have seen the performance decrease significantly.
> Now, under 7.1, numerous perl processes will queue up (from their web
> applications), the load average will be in the 40s to 90s, and
> according to top they perl scripts are commonly in the "ufs" state,
> which I assume means stuck waiting for read or write responses.
> I have also experienced pretty abysmal read performance from the
> array, using dd...we're talking sub 1MBps. So, my assumption is that
> something is wrong on the filesystem layer.
> When I look at the CPU utilization in top, I see that 80-90% is
> constantly used by "system".
> However, when I look at iostat, I see very low numbers for the raid,
> in fact across the board.
> It's a Dell server with a Perc4/Di RAID controller, which uses the
> amr driver. They upgraded to larger, 15k rpm ultra 320 disks (from
> smaller 15k rpm ultra 320 disks). They made a backup of their web
> root, which is a seperate partition, using dump to a temporary drive,
> then swapped in the new disks, installed FreeBSD 7.1, and restored
> their webroot to a partition on the new array.
> Does anybody have any insight into what could be going on here?
I believe there was a performance regression in 7.1 for SCSI disks. I
don't know if the amr driver was affected but the issues should be
resolved in FreeBSD 7.2 so you might want to give that a try.
More information about the freebsd-questions