On Wed, May 25, 2016 at 10:51:20AM -0400, Chris Mason wrote:
On Mon, May 09, 2016 at 12:48:07PM +0200, Peter Zijlstra wrote:
> here be a semi coherent patch series for the recent select_idle_siblings()
> tinkering. Happy benchmarking..
This took a while, mostly because my original schbench showed your
patches were just as fast as our internal patch, but our production
benchmark showed a 5-10% regression. My theory was just that schbench
wasn't fine grained enough, and the load would stop before the
regression kicked in.
Just to be sure; regression as per your production kernel, not vs.