[Lustre-discuss] MDS/MDT Performance

[Canon, Richard Shane]

Rei,

We did some testing with a TMS SSD array (over SRP) about a year ago.
We did not see a substantial performance difference versus a write
cache-on DDN system (I think it was a 9550 attached via IB SRP).  With
write cache turned off we did see a difference (I forget how large).
Most of the tests were done with mdtest from LLNL.  Sarp Oral did the
evaluation.  He can provide more details if you are interested.

We are hoping to repeat these tests soon with some loaner hardware.  We
can share results as we get them, although our intent is to draft a
white paper when we are done.

--Shane

-----Original Message-----
From: lustre-discuss-bounces at clusterfs.com
[mailto:lustre-discuss-bounces at clusterfs.com] On Behalf Of Andreas
Dilger
Sent: Thursday, January 17, 2008 2:31 AM
To: Rei Lee
Cc: lustre-discuss at clusterfs.com
Subject: Re: [Lustre-discuss] MDS/MDT Performance

On Jan 15, 2008  11:08 -0800, Rei Lee wrote:
> We are in the process of building a file system and
> we are wondering whether we should get disk drivers
> of 10K RPM or 15 RPM for the MDT?  How is
> the MDS performance affected by the performance
> of the underlying disk drivers used by MDT?
> Will solid-state disks provide a much better MDS
> performance?

The MDT is mostly IOPS and CPU bound, so faster disks will definitely
make
a difference, as will having a lot of RAM for filesystem metadata cache.

Cheers, Andreas
--
Andreas Dilger
Sr. Staff Engineer, Lustre Group
Sun Microsystems of Canada, Inc.

_______________________________________________
Lustre-discuss mailing list
Lustre-discuss at clusterfs.com
https://mail.clusterfs.com/mailman/listinfo/lustre-discuss