sched: weaken sync hint
authorMike Galbraith <efault@gmx.de>
Fri, 24 Oct 2008 09:06:14 +0000 (11:06 +0200)
committerIngo Molnar <mingo@elte.hu>
Fri, 24 Oct 2008 10:51:01 +0000 (12:51 +0200)
Mysql+oltp and pgsql+oltp peaks are still shifted right. The below puts
the peaks back to 1 client/server pair per core.

Use the avg_overlap information to weaken the sync hint.

Signed-off-by: Mike Galbraith <efault@gmx.de>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
kernel/sched_fair.c

index b27ccc52f6aa626f152f95ca19af7768ada4ff74..b71ee2c62297b46713230a753f6beb7417fc07ec 100644 (file)
@@ -1123,10 +1123,9 @@ wake_affine(struct sched_domain *this_sd, struct rq *this_rq,
        if (!(this_sd->flags & SD_WAKE_AFFINE) || !sched_feat(AFFINE_WAKEUPS))
                return 0;
 
-       if (!sync && sched_feat(SYNC_WAKEUPS) &&
-           curr->se.avg_overlap < sysctl_sched_migration_cost &&
-           p->se.avg_overlap < sysctl_sched_migration_cost)
-               sync = 1;
+       if (sync && (curr->se.avg_overlap > sysctl_sched_migration_cost ||
+                       p->se.avg_overlap > sysctl_sched_migration_cost))
+               sync = 0;
 
        /*
         * If sync wakeup then subtract the (maximum possible)