[xansys][other] comparing RAID0 runs--need advice

RD
Robert Dillworth
Tue, Nov 28, 2017 8:54 PM

Got a Dell T7910 machine earlier this year, with a LSI Megaraid SAS 9341-8i raid controller.  Ran a job that crawled due to slow i/o-48 hours elapsed time, which was many times the total CPU time.  Ran the same job on older T7600 in the office that has a Dell PERC H710P raid controller-about 9 hours elapsed time, and very similar to the CPU time.

It wasn't a sparse-but-out-of-core type of problem at play-both runs ran incore.  But both were multicore runs using DMP parallel, so each process (six cores so six processes) is reading and writing all of the usual files: multiple *.rst, *.emat, etc.  Lots of i/o going on.  The 9341 handles this poorly; the PERC handles it nicely.

A little research of my own and consultation with ANSYS confirms my fear: my megaraid controller is inferior.  A little more research shows that the 9341 controller has NO cache memory of its own, while the PERC has 1 Gig cache.  Now considering buying the LSI Megaraid SAS 9361-8i controller from the Dell site, which has 1 Gig cache.  Before I ask the boss for the money, though, anyone out there have a similar experience or any owners of the LSI Megaraid SAS 9361-8i controller?

Any related advice would be welcome.  All RAID0/hardware gurus, I need you now more than ever.

P.S. - Having trouble with our system showing my messages on xansys, so I'll try to work with xansys archives to assist communication.  Thanks all.

LPI, Inc.
Robert Dillworth, P.E.
Principal Engineer
304 Hudson Street, 4th Floor
New York, NY 10013

Ph.: (212) 233-2737
Fax: (212) 406-1417

Got a Dell T7910 machine earlier this year, with a LSI Megaraid SAS 9341-8i raid controller. Ran a job that crawled due to slow i/o-48 hours elapsed time, which was many times the total CPU time. Ran the same job on older T7600 in the office that has a Dell PERC H710P raid controller-about 9 hours elapsed time, and very similar to the CPU time. It wasn't a sparse-but-out-of-core type of problem at play-both runs ran incore. But both were multicore runs using DMP parallel, so each process (six cores so six processes) is reading and writing all of the usual files: multiple *.rst, *.emat, etc. Lots of i/o going on. The 9341 handles this poorly; the PERC handles it nicely. A little research of my own and consultation with ANSYS confirms my fear: my megaraid controller is inferior. A little more research shows that the 9341 controller has NO cache memory of its own, while the PERC has 1 Gig cache. Now considering buying the LSI Megaraid SAS 9361-8i controller from the Dell site, which has 1 Gig cache. Before I ask the boss for the money, though, anyone out there have a similar experience or any owners of the LSI Megaraid SAS 9361-8i controller? Any related advice would be welcome. All RAID0/hardware gurus, I need you now more than ever. P.S. - Having trouble with our system showing my messages on xansys, so I'll try to work with xansys archives to assist communication. Thanks all. LPI, Inc. Robert Dillworth, P.E. Principal Engineer 304 Hudson Street, 4th Floor New York, NY 10013 Ph.: (212) 233-2737 Fax: (212) 406-1417