Jesse Keating wrote on Friday 29 April 2005 09:11:
On Fri, 2005-04-29 at 11:18 -0400, Joshua Baker-LePain wrote:
> Err, really? I saw exactly the opposite. Now, this was on centos 4,
> but
> these numbers are for a software RAID0 across 2 9500-12 cards in 11
> disk
> RAID5 mode. I'd also done 'blockdev --setra 16384' on both 3ware
> devices, and this is with a kernel modified to include XFS:
I can't help but wonder if the software RAID 0 is interfering with your
setup. Any chance you can test w/out the software RAID 0 involved?
Just a straight RAID 5 across 11 disks?
Just last night I did essentially that on my system with a 9500S-12 with
11-disk hw RAID5 + hot spare. The command I ran was:
sync ; time dd if=/dev/zero of=/scratch1/8G-2 bs=1M count=8192 ; time sync
Typical rate for the dd was 12-15MB/sec, right in line with what Joshua saw
(he saw roughly double this for two such arrays RAID0'd together). This rate
is terrible, no? :) I chose to write an 8GB file because the system memory
is 4GB. I made several other observations on this system; I made lengthy
comments on
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=121434.
I can run other tests; I'm open to suggestions. Since this box has a second
identical array, I plan to try that one with sw RAID5 instead of hw RAID5.
My test was to an XFS filesystem, using a RHEL4 kernel modified to have XFS
enabled (much like Joshua, I suspect). I don't have numbers for earlier
firmware/drivers.
David