Can you provide a vxprint -ht for each of the volumes you mention below ? Would be interesting to see what kind of a stripe width was used especially since you mention thousands of smaller files of ~ 20k each. Also , whenever you are running these tests you will be able to see i/o performance on a per volume basis in the disk group using "vxstat -g <dgname> -i <interval> ".
| "senthil ramanujam" <[EMAIL PROTECTED]>
Sent by: [EMAIL PROTECTED] 11/08/2006 07:42 PM |
To: Veritas-vx@mailman.eng.auburn.edu cc: Subject: [Veritas-vx] performance degradation using less number of volumes |
Hi,
I have been seeing an interesting performance issue. My search at the
archives and Google didn't point me to any helpful hint. Please let me
know if this is an inappropriate forum or if there is a better forum
to discuss these issues.
Allow me to explain the configuration I used. Redhat Advanced Server
is running on a 2-socket system with 16GB memory in it. The system is
attached to 2 SCSI JBOD arrays, each with 12x15Krpm (146GB) drives.
Veritas Storage Foundation (volume manager) is being used to access
these drives and provided the structure(Raid10) we needed.
We use a workload to benchmark the above configuration. The workload
requires 8 directories. There are thousands of smaller-sized (~20k)
files spread across these 8 directories. The workload writes and reads
from these files. It is safe to assume the reads and writes are almost
equally distributed. I think that is enough to say about the workload
used.
The volumes are configured as follows:
Run-A: There are 4 diskgroups, each diskgroup has 6 spindles. There
are 2 volumes built on top of each diskgroup. So, that means, there
are 8 volumes. vxassist is used to build these volumes.
Run-B: There are 4 diskgroups, each diskgroup has 6 spindles. There is
one volume built on top of each diskgroup. This makes totally 4
volumes, each volume has 2 directories. vxassist is used to build
these volumes.
In the above runs, the workload sees 8 directories that sit on top of
ext2 filesystem. This is where the performance issue shows up. Run-A
(8 volumes) performs 10-15% better than Run-B (4 volumes). The *stat
(iostat, mpstat, vmstat) looks almost the same between these runs.
Nothing stands out. I even parsed the iostat data and checked the
reads and writes at the volume and spindles level, which look
more-or-less the same.
I just started working with Veritas, so, it is possible that I may
have overlooked some tuning bits and pieces. Looking at the 8-volume
performance number, I have no reason why we can't get that for
4-volume. One of the most important goals is performance, which is
more important than high-availability. If there is a missing piece of
info, I should be able to get that for you although I can't have both
the configuration at the same time.
I am wondering that someone might be able to provide better insight.
Has anyone seen or heard this before? Any pointers or inputs would be
appreciated.
thanks.
senthil
_______________________________________________
Veritas-vx maillist - Veritas-vx@mailman.eng.auburn.edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-vx
This transmission may contain information that is privileged,
confidential, legally privileged, and/or exempt from disclosure
under applicable law. If you are not the intended recipient, you
are hereby notified that any disclosure, copying, distribution, or
use of the information contained herein (including any reliance
thereon) is STRICTLY PROHIBITED. Although this transmission and
any attachments are believed to be free of any virus or other
defect that might affect any computer system into which it is
received and opened, it is the responsibility of the recipient to
ensure that it is virus free and no responsibility is accepted by
JPMorgan Chase & Co., its subsidiaries and affiliates, as
applicable, for any loss or damage arising in any way from its use.
If you received this transmission in error, please immediately
contact the sender and destroy the material in its entirety,
whether in electronic or hard copy format. Thank you.
_______________________________________________ Veritas-vx maillist - Veritas-vx@mailman.eng.auburn.edu http://mailman.eng.auburn.edu/mailman/listinfo/veritas-vx
