On Wed, 21 Oct 2020, Mel Gorman wrote: > On Wed, Oct 21, 2020 at 05:19:53PM +0200, Vincent Guittot wrote: > > On Wed, 21 Oct 2020 at 17:08, Mel Gorman <mgorman@xxxxxxx> wrote: > > > > > > On Wed, Oct 21, 2020 at 03:24:48PM +0200, Julia Lawall wrote: > > > > > I worry it's overkill because prev is always used if it is idle even > > > > > if it is on a node remote to the waker. It cuts off the option of a > > > > > wakee moving to a CPU local to the waker which is not equivalent to the > > > > > original behaviour. > > > > > > > > But it is equal to the original behavior in the idle prev case if you go > > > > back to the runnable load average days... > > > > > > > > > > It is similar but it misses the sync treatment and sd->imbalance_pct part of > > > wake_affine_weight which has unpredictable consequences. The data > > > available is only on the fully utilised case. > > > > In fact It's the same because runnable_load_avg was null when cpu is idle, so > > if prev_cpu was idle, we were selecting prev_idle > > > > Sync wakeups may only consider this_cpu and the load of the waker but > in that case, it was probably selected already by the sync check in > wake_affine_idle which will pass except when the domain is overloaded. > Fair enough, I'll withdraw any concerns. It could have done with a > comment :/ Sure, I'll resend the patch and extend the log message with this issue. Otherwise, I was wondering are there any particular kinds of applications where gathering the threads back with the waker is a good idea? I've been looking more at applications with N threads on N cores, where it would be best for the threads to remain where they are. thanks, julia
