On Mon, Apr 04, 2016 at 03:45:16PM -0400, Rik van Riel wrote: > An optimal solution is NP hard. > > Heuristics that "move tasks with pressure" may be > much more doable, and lead to perfectly satisfactory > results, especially if most migrations happen within > a socket (and the same shared L3 cache).
Right; trick will be finding something that mostly works without making the regular balance paths increase in complexity. As per the argument in kernel/sched/fair.c:5694 the current load-balancing averages out to O(n), and I would very much like to keep it that way.
