On 15/06/16 17:03, Mike Galbraith wrote: > On Wed, 2016-06-15 at 16:32 +0100, Dietmar Eggemann wrote: > >>> In general, the fuzz helps us to not be so spastic. I'm not sure that >>> we really really need to care all that much, because I strongly suspect >>> that it's only gonna make any difference at all in corner cases, but >>> there are real world cases that matter. I know for fact that schbench >>> (facebook) which is at least based on a real world load fails early due >>> to us stacking tasks due to that fuzzy view of reality. In that case, >>> it's because the fuzz consists of a high amplitude aging sawtooth.. >> >> ... only for fork/exec? > > No. Identical workers had longish work/sleep cycle, aging resulted in > weights that ranged from roughly 300-700(ish), depending on when you > peeked at them. > > -Mike >
Isn't there a theoretical problem with the scale_load() on CONFIG_64BIT machines on tip/sched/core? load.weight has a higher resolution than runnable_load_avg (and so the values in the rq->cpu_load[] array). Theoretically because [forkexec|wake]_idx is 0 so [target|source]_load() is nothing else than weighted_cpuload().
