Messages in this thread | | | Subject | Re: [git pull] scheduler fixes | From | Mike Galbraith <> | Date | Sat, 17 Jan 2009 17:28:47 +0100 |
| |
On Sat, 2009-01-17 at 17:12 +0100, Ingo Molnar wrote: > * Mike Galbraith <efault@gmx.de> wrote: > > > Dunno about the IO bits, but.. > > > > The problem with the C++ testcases seems to be wake_up_all() plunking a > > genuine thundering herd onto runqueues. The sleeper fairness logic > > places the entire herd left of min_vruntime, meaning N*sched_latency > > pain for the poor sods who are setting the runqueue pace. > > 100 wakeup pairs that all run and ping-pong between each other? > > That creates 200 tasks with an average system load of 100.0, on a > dual-core system. Is that a fair representation of some real workload, or > just an unrealistic "gee, look, given enough tasks running I can overload > the system _this bad_" example?
Looks contrived to me, but it is a hole. Dang sleepers, can't live with 'em can't live without 'em.
-Mike
| |