|
From: Philippe W. <phi...@sk...> - 2011-11-10 21:41:41
|
> I've implemented what Julian suggested on the bug tracker > (https://bugs.kde.org/show_bug.cgi?id=270006): fair scheduling is off > by default but can be enabled via a command-line option. This is very good (a.o. making it very to compare these). > > Personally I'm not very enthusiast about the sched_setaffinity() > approach - I'd rather prefer to document the interaction with > frequency scaling. Documenting this interaction is I think good in any case, as the cpu freq scaling also has some influence on the pipe scheduler. And will probably also badly influence a process which is doing in a loop: burn some cpu sleep for a small time (but enough to let the cpu scaling down) There are a bunch of parameters to tune the way the cpu freq scaling is done. But I suppose that an average Valgrind user might perhaps see the degradation of performance, and either will not know why and/or will not be able (EPERM or similar) to change the cpu governor. > > Another option, if someone has the time, is to report on the LKML that > we (Valgrind developers) have found a scenario in which the Linux CFS > scheduler performs suboptimal. The CFS maintainer, Ingo Molnar, has > asked to report such cases some time ago. Maybe Ingo has a suggestion > about how to address the performance degradation we observed. If I understand well, the idea would be that the CFS should handle specially a process made of multiple threads, all burning CPUs, but only one thread at a time burning CPU. The CFS would have to avoid switching them from cpu, as this interacts badly with cpu frequency scaling. |