Comments on: Opteron NUMA-effects http://www.thinkingparallel.com/2006/07/30/opteron-numa-effects/ A Blog on Parallel Programming and Concurrency by Michael Suess Sat, 04 Feb 2012 06:11:44 +0000 http://wordpress.org/?v=2.7 hourly 1 By: Michael Suess http://www.thinkingparallel.com/2006/07/30/opteron-numa-effects/comment-page-1/#comment-20 Michael Suess Wed, 23 Aug 2006 20:33:26 +0000 http://www.thinkingparallel.com/2006/07/30/opteron-numa-effects/#comment-20 Dieter, thanks for your comment. I should have probably mentioned in this article that this is not a comprehensive evaluation of NUMA-effects on the Opteron. If it were, I would not have published it here but in a research paper :-), like e.g. <a href="http://www.nathanr.net/honours/ICCSnuma.pdf" rel="nofollow" rel="nofollow" rel="nofollow">these guys</a> (a good read by the way). I have only investigated the part of the effects related to memory latency and left out the part about memory bandwidth (mainly, because I do not have any memory-bound applications and therefore this case is not so interesting for me). Reading through the article again, I should have made that more clear, as the last sentence is probably not enough of a warning. Anyways, I will not change the article now, as the correction can be seen in the comments section. Dieter,
thanks for your comment. I should have probably mentioned in this article that this is not a comprehensive evaluation of NUMA-effects on the Opteron. If it were, I would not have published it here but in a research paper :-), like e.g. these guys (a good read by the way). I have only investigated the part of the effects related to memory latency and left out the part about memory bandwidth (mainly, because I do not have any memory-bound applications and therefore this case is not so interesting for me). Reading through the article again, I should have made that more clear, as the last sentence is probably not enough of a warning. Anyways, I will not change the article now, as the correction can be seen in the comments section.

]]> By: Dieter http://www.thinkingparallel.com/2006/07/30/opteron-numa-effects/comment-page-1/#comment-18 Dieter Wed, 23 Aug 2006 16:22:23 +0000 http://www.thinkingparallel.com/2006/07/30/opteron-numa-effects/#comment-18 We typically observer ccNUMA effects on the Opteron when using more than 2 threads. It seems that the memory architecture can sufficiently support 2 threads, whereever they are running. But with more than 2 threads things are quite different!! It may get worse with memory benchmarks like stream. THere you can see the difference between good and bad memory placement with 2 threads already See http://www.rz.rwth-aachen.de/computing/events/2005/parco05/drops_slides.pdf We typically observer ccNUMA effects on the Opteron when using more than 2 threads.
It seems that the memory architecture can sufficiently support 2 threads, whereever they are running.
But with more than 2 threads things are quite different!!

It may get worse with memory benchmarks like stream. THere you can see the difference between good and bad memory placement with 2 threads already

See
http://www.rz.rwth-aachen.de/computing/events/2005/parco05/drops_slides.pdf

]]> By: Thinking Parallel » Blog Archive » More information on pthread_setaffinity_np and sched_setaffinity http://www.thinkingparallel.com/2006/07/30/opteron-numa-effects/comment-page-1/#comment-15 Thinking Parallel » Blog Archive » More information on pthread_setaffinity_np and sched_setaffinity Fri, 18 Aug 2006 16:01:56 +0000 http://www.thinkingparallel.com/2006/07/30/opteron-numa-effects/#comment-15 [...] Skimming through the activity logs of this blog, I can see that many people come here looking for information about pthread_setaffinity_np. I mentioned it briefly in my article about Opteron NUMA-effects, but barely touched it because I had found a more satisfying solution for my personal use (taskset). And while I do not have in depth knowledge of the function, maybe the test programs I wrote will be of some help to someone to understand the function better. I will also post my test program for sched_setaffinity here while I am at it, simply because the two offer similar functionality. [...] [...] Skimming through the activity logs of this blog, I can see that many people come here looking for information about pthread_setaffinity_np. I mentioned it briefly in my article about Opteron NUMA-effects, but barely touched it because I had found a more satisfying solution for my personal use (taskset). And while I do not have in depth knowledge of the function, maybe the test programs I wrote will be of some help to someone to understand the function better. I will also post my test program for sched_setaffinity here while I am at it, simply because the two offer similar functionality. [...]

]]>