Messages in this thread | | | From | Paul Turner <> | Date | Mon, 22 Apr 2013 10:49:50 -0700 | Subject | Re: [PATCH] sched: wake-affine throttle |
| |
On Mon, Apr 22, 2013 at 3:23 AM, Peter Zijlstra <peterz@infradead.org> wrote: > > OK,.. Ingo said that pipe-test was the original motivation for > wake_affine() and since that's currently broken to pieces due to > select_idle_sibling() is there still a benefit to having it at all? > > Can anybody find any significant regression when simply killing > wake_affine()? >
This matches our experience; we've always tuned toward a more aggressive select_idle_sibling() for in-socket wake-ups.
We did recently turn it back on in the cross-socket case (e.g. SD_NODE) however as we were able to observe a benefit on a few workloads.
| |