Hi all, Attached are my latest sgpdd-surveys of a DDN S2A 9550. I believe we''ve finally solved the performance problems seen with the previous surveys. The solution is to use 2 tiers/LUN to ensure adequate disk bandwidth. At least 2 MB IOs are also needed, which currently requires special patches to Lustre. My next step is to run obdfilter-survey on the same hardware. Thanks to Oak Ridge National Laboratory and Indiana University for the hardware used in these sgpdd-surveys. Cheers, Jody -------------- next part -------------- A non-text attachment was scrubbed... Name: 9500-sgp_dd.xls Type: application/vnd.ms-excel Size: 450560 bytes Desc: not available Url : http://mail.clusterfs.com/pipermail/lustre-discuss/attachments/20070118/8d6a4e79/9500-sgp_dd-0001.xls
Nicholas Henke
2007-Jan-18 13:00 UTC
[Lustre-discuss] Latest sgpdd-surveys of DDN S2A 9550
Jody McIntyre wrote:> Hi all, > > Attached are my latest sgpdd-surveys of a DDN S2A 9550. I believe we''ve > finally solved the performance problems seen with the previous surveys. > The solution is to use 2 tiers/LUN to ensure adequate disk bandwidth.Jody -- is this 2 full tiers per LUN? What size LUN did you end up with ? Nic
Hi Nic, On Thu, Jan 18, 2007 at 11:59:50AM -0600, Nicholas Henke wrote:> Jody -- is this 2 full tiers per LUN? What size LUN did you end up with ?Close to it - there was 32 MB of each tier set aside already in separate LUNs, so I left it and used the rest of the space. LUN capacity is 7630876 MB (this is in the spreadsheet.) Cheers, Jody
I found these ClustreFS performance results in the archives. Can you provide details on the host configuration (processor, OS, storage connection) that was used in these tests? Thx, Tim -----Original Message----- From: lustre-discuss-bounces@clusterfs.com [mailto:lustre-discuss-bounces@clusterfs.com] On Behalf Of Jody McIntyre Sent: Thursday, January 18, 2007 1:42 PM To: lustre-discuss@clusterfs.com Subject: [Lustre-discuss] Latest sgpdd-surveys of DDN S2A 9550 Hi all, Attached are my latest sgpdd-surveys of a DDN S2A 9550. I believe we''ve finally solved the performance problems seen with the previous surveys. The solution is to use 2 tiers/LUN to ensure adequate disk bandwidth. At least 2 MB IOs are also needed, which currently requires special patches to Lustre. My next step is to run obdfilter-survey on the same hardware. Thanks to Oak Ridge National Laboratory and Indiana University for the hardware used in these sgpdd-surveys. Cheers, Jody
Hi Tim, On Thu, Jul 05, 2007 at 07:05:41AM -0600, Snider, Tim wrote:> I found these ClustreFS performance results in the archives. > Can you provide details on the host configuration (processor, OS, > storage connection) that was used in these tests?All results in the latest version of the spreadsheet were measured on Indiana University Data Capacitor OSS nodes, so if Bret or Steve from IU are following this thread, they can likely provide more details. I don''t recall the exact CPU used in these nodes. They are Intel Woodcrest-based servers from Dell. The OS is RHEL 4 and storage is connected via 4 GBit FC. Cheers, Jody> Thx, > Tim > > -----Original Message----- > From: lustre-discuss-bounces@clusterfs.com > [mailto:lustre-discuss-bounces@clusterfs.com] On Behalf Of Jody McIntyre > Sent: Thursday, January 18, 2007 1:42 PM > To: lustre-discuss@clusterfs.com > Subject: [Lustre-discuss] Latest sgpdd-surveys of DDN S2A 9550 > > Hi all, > > Attached are my latest sgpdd-surveys of a DDN S2A 9550. I believe we''ve > finally solved the performance problems seen with the previous surveys. > The solution is to use 2 tiers/LUN to ensure adequate disk bandwidth. > At least 2 MB IOs are also needed, which currently requires special > patches to Lustre. > > My next step is to run obdfilter-survey on the same hardware. > > Thanks to Oak Ridge National Laboratory and Indiana University for the > hardware used in these sgpdd-surveys. > > Cheers, > Jody--