On 07/06/11 13:00, Steve Kargl wrote: > On Wed, Jul 06, 2011 at 05:05:41PM +0000, Poul-Henning Kamp wrote: >> In message<20110706170132.GA68775_at_troutmask.apl.washington.edu>, Steve Kargl w >> rites: >> >>> I periodically ran the same type test in the 2008 post over the >>> last three years. Nothing has changed. I even set up an account >>> on one node in my cluster for jeffr to use. He was too busy to >>> investigate at that time. >> >> Isn't this just the lemming-syncer hurling every dirty block over >> the cliff at the same time ? > > I don't know the answer. Of course, having no experience in > processing scheduling, I don't understand the question either ;-) > > AFAICT, it is a cpu affinity issue. If I launch n+1 MPI images > on a system with n cpus/cores, then 2 (and sometimes 3) images > are stuck on a cpu and those 2 (or 3) images ping-pong on that > cpu. I recall trying to use renice(8) to force some load > balancing, but vaguely remember that it did not help. I've seen exactly this problem with multi-threaded math libraries, as well. Using parallel GotoBLAS on FreeBSD gives terrible performance because the threads keep migrating between CPUs, causing frequent cache misses. -NathanReceived on Wed Jul 06 2011 - 16:11:24 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:40:15 UTC