Messages in this thread | | | Subject | Re: [patch] Re: scheduler oddity [bug?] | From | Mike Galbraith <> | Date | Mon, 09 Mar 2009 18:28:14 +0100 |
| |
On Mon, 2009-03-09 at 17:12 +0100, Peter Zijlstra wrote: > On Mon, 2009-03-09 at 16:30 +0100, Mike Galbraith wrote: > > +static void put_prev_task(struct rq *rq, struct task_struct *prev) > > +{ > > + if (prev->state == TASK_RUNNING) { > > + u64 runtime = prev->se.sum_exec_runtime; > > + > > + runtime -= prev->se.prev_sum_exec_runtime; > > + runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost); > > + > > + /* > > + * In order to avoid avg_overlap growing stale when we are > > + * indeed overlapping and hence not getting put to sleep, grow > > + * the avg_overlap on preemption. > > + */ > > + update_avg(&prev->se.avg_overlap, runtime); > > + } > > + prev->sched_class->put_prev_task(rq, prev); > > +} > > Right, so we both found it worked quite well, I'm still slightly puzzled > but it. > > If something gets preempted a lot and will therefore have short runtimes > it will be seen as sync even though it might not at all be.
Yes, and the netperf on 2 CPUs with shared cache numbers show that's happening. It just so happens that in the non-shared case, netperf's cache pain far outweighs the benefit of having more CPU available :-/
-Mike
| |