Le 15/05/2013 23:26, Bastien Chevreux a écrit :
On May 15, 2013, at 10:32 , Laurent MANCHON <lmanchon@xxxxxxxxxxxxxx> wrote:I played with the minimum and maximum number of threads and in the end i found out that for this version of Mira the best throughput was somewhere between 36 and 40 threads. Anything outside those boundaries performed worse.Thanks for the info. Even if this boundary may be specific to your machine (disk speed, RAM speed etc.), it provides a good first estimate for an upper boundary. I have no idea what could be the bottleneck, it could be everything: CPU, disk, RAM, or bad MIRA programming. I'm not likely to find out soon though, the maximum number of cores I have access to is 8 :-) B.
It's easy to check the code with a profiler to find out the bottleneck and then optimize serial and parallel performance, For Intel you have Intel Vtune Amplifier (http://software.intel.com/en-us/intel-vtune-amplifier-xe) is free for non-commercial users, now for GCC you can use Valgrind + Kcachegrind for this type of visualization (both open source). However, Valgrind is more efficient (especially for debugging the code), but slowed considerably code execution.
Laurent -- -- You have received this mail because you are subscribed to the mira_talk mailing list. For information on how to subscribe or unsubscribe, please visit http://www.chevreux.org/mira_mailinglists.html
