Re: [PATCH v2] sched/fair: fix 1 task per CPU

[Peter Zijlstra]

On Mon, Sep 10, 2018 at 04:43:09PM +0200, Vincent Guittot wrote:
> When CPUs have different capacity because of RT/DL tasks or
> micro-architecture or max frequency differences, there are situation where
> the imbalance is not correctly set to migrate waiting task on the idle CPU.
> 
> The UC uses the force_balance case:
>
> 	if (env->idle != CPU_NOT_IDLE && group_has_capacity(env, local) &&
> 	    busiest->group_no_capacity)
> 		goto force_balance;
> 
> But calculate_imbalance fails to set the right amount of load to migrate
> a task because of the special condition:
>
>   busiest->avg_load <= sds->avg_load || local->avg_load >= sds->avg_load)
> 
> Add in fix_small_imbalance, this special case that triggered the force
> balance in order to make sure that the amount of load to migrate will be
> enough.

So I think this patch is going in the wrong direction for a number of
reasons:

 - we'd like to get rid of fix_small_imbalance(), and this adds to it;

 - the whole load_per_task stuff is terminally broken, it _cannot_ work
   right.


What I've suggested in the past is parameterizing the load balancer and
picking different criteria to balance on:

 - nr_running ; if there are idle CPUs around
 - utilization ; if there's idle time (u<1)
 - weight

And I suppose you're hitting one of the nr_running cases here.

> Signed-off-by: Vincent Guittot <vincent.guittot@xxxxxxxxxx>
> ---
>  kernel/sched/fair.c | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 309c93f..72bc5e8 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -8048,6 +8048,20 @@ void fix_small_imbalance(struct lb_env *env, struct sd_lb_stats *sds)
>  	local = &sds->local_stat;
>  	busiest = &sds->busiest_stat;
>  
> +	/*
> +	 * There is available capacity in local group and busiest group is
> +	 * overloaded but calculate_imbalance can't compute the amount of load
> +	 * to migrate because load_avg became meaningless due to asymetric
> +	 * capacity between groups. In such case, we only want to migrate at
> +	 * least one tasks of the busiest group and rely of the average load
> +	 * per task to ensure the migration.
> +	 */
> +	if (env->idle != CPU_NOT_IDLE && group_has_capacity(env, local) &&
> +	    busiest->group_no_capacity) {
> +		env->imbalance = busiest->load_per_task;
> +		return;
> +	}
> +
>  	if (!local->sum_nr_running)
>  		local->load_per_task = cpu_avg_load_per_task(env->dst_cpu);
>  	else if (busiest->load_per_task > local->load_per_task)
> -- 
> 2.7.4
>