Re: Scheduler(?) regression from 2.6.22 to 2.6.24 for short-livedthreads

[Mike Galbraith]

On Tue, 2008-02-12 at 10:23 +0100, Mike Galbraith wrote:

> If you plunk a usleep(1) in prior to calling thread_func() does your
> testcase performance change radically?  If so, I wonder if the real
> application has the same kind of dependency.

The answer is yes for 2.6.22, and no for 2.6.24, which surprised me.

2.6.22.17-smp
homer:..local/tmp # ./threadtest2
10000000 loops time 16215 ms
10000000 loops time 16268 ms
10000000 loops time 16246 ms

homer:..local/tmp # ./threadtest3 (with usleep(1))
10000000 loops time 13938 ms
10000000 loops time 13921 ms
10000000 loops time 13898 ms

2.6.24.1-smp
homer:..local/tmp # ./threadtest2
10000000 loops time 14663 ms
10000000 loops time 14523 ms
10000000 loops time 14466 ms

homer:..local/tmp # ./threadtest3
10000000 loops time 14513 ms
10000000 loops time 14500 ms
10000000 loops time 14464 ms

echo 0 >  /proc/sys/kernel/sched_child_runs_first

homer:..local/tmp # ./threadtest2
10000000 loops time 14157 ms
10000000 loops time 14097 ms
10000000 loops time 14153 ms

homer:..local/tmp # ./threadtest3
10000000 loops time 14065 ms
10000000 loops time 14075 ms
10000000 loops time 14018 ms



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/