Sure thing!

MDS
====

Dell PE2950
2 x Quad Core Xeon 5355
16 GB Ram
Infinihost III DDR HCA
2 x 143GB SAS 13k RPM disks for the metadata volume ( in a hardware perc5 raid 1)

OSS1
====

Dell PE1950
2 x Dual Core Xeon 5150
16 GB Ram
Infinihost III DDR HCA
3 x MD1000s w/1TB Drives
Each MD1000 is connected to a single perc5 and has two 6tb raid5 arrays.

OSS2
=====

Dell PE1950
2 x Dual Core Xeon 5130
8 GB Ram
Infinihost III DDR HCA
2 x MD1000s w/750GB Drives
Each MD1000 is connected to a single perc6 and has two 4.4tb raid5 arrays.


IMPORTANT --
For a significant increase in read performance run this command on every block device exported by the perc5/6. A wiki page on the Dell perc5/md1000 combination could be very useful specifically for this purpose, or perhaps some mention in the user manual. Am I able to create wiki pages?

/sbin/blockdev --setra 16384

Thanks again!

-Aaron

On Feb 5, 2008, at 2:43 PM, Andreas Dilger wrote:

On Feb 05, 2008  09:42 -0500, Aaron Knister wrote:
We have a relatively small lustre setup (50TB) with infiniband as the
interconnect. Well this morning we saw 1.2Gigabytes per second reads
and 500MB/s writes from one client. I literally had to stand up and
leave the room. I was absolutely floored. I wanted to thank the lustre
developers for their outstanding work and continued dedication to
supporting those (like myself) who can't afford a support contract.

Aaron,
thanks for the positive feedback. People always have time to complain,
but far fewer take the time to report when things are going well, or
better than expected.

Can you please elaborate a bit on your system configuration (HCA, SDR/DDR,
switch, CPUs, RAM, #OSS, #OSTs, etc) for reference.

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


Aaron Knister
Associate Systems Analyst
Center for Ocean-Land-Atmosphere Studies

(301) 595-7000
[EMAIL PROTECTED]




_______________________________________________
Lustre-discuss mailing list
[email protected]
http://lists.lustre.org/mailman/listinfo/lustre-discuss

Reply via email to