Re: [PATCH] sched, fair: Allow a small load imbalance between low utilisation SD_NUMA domains v3

[Valentin Schneider]

On 07/01/2020 10:16, Mel Gorman wrote:
> I think running tasks at least is the least bad metric. idle CPUs gets
> caught up in corner cases with bindings and util_avg can be skewed by
> outliers. Running tasks is a sensible starting point until there is a
> concrete use case that shows it is unworkable.

I'd tend to agree with you here. Also; this being in the group_has_spare
imbalance type we have some guarantees that the group is not overutilized.

If we keep some threshold of 'sum_nr_running < group_weight / 2', then this
"naturally" puts a hard limit of 50% total group utilization (corner case
where all tasks are 100% util).

> Lets see what you think of
> the other untested patch I posted that takes the group weight and child
> domain weight into account.
>