Re: [PATCH] sched, fair: Allow a small degree of load imbalance between SD_NUMA domains v2

[Peter Zijlstra]

On Tue, Jan 07, 2020 at 09:56:55AM +0000, Mel Gorman wrote:

> Much more importantly, doing what you suggest allows an imbalance
> of more CPUs than are backed by a single LLC. On high-end AMD EPYC 2
> machines, busiest->group_weight scaled by imbalance_pct spans multiple L3
> caches. That is going to have side-effects. While I also do not account
> for the LLC group_weight, it's unlikely the cut-off I used would be
> smaller than an LLC cache on a large machine as the cache.
> 
> These two points are why I didn't take the group weight into account.
> 
> Now if you want, I can do what you suggest anyway as long as you are happy
> that the child domain weight is also taken into account and to bound the
> largest possible allowed imbalance to deal with the case of a node having
> multiple small LLC caches. That means that some machines will be using the
> size of the node and some machines will use the size of an LLC. It's less
> predictable overall as some machines will be "special" relative to others
> making it harder to reproduce certain problems locally but it would take
> imbalance_pct into account in a way that you're happy with.
> 
> Also bear in mind that whether LLC is accounted for or not, the final
> result should be halved similar to the other imbalance calculations to
> avoid over or under load balancing.

> +		/* Consider allowing a small imbalance between NUMA groups */
> +		if (env->sd->flags & SD_NUMA) {
> +			struct sched_domain *child = env->sd->child;

This assumes sd-child exists, which should be true for NUMA domains I
suppose.

> +			unsigned int imbalance_adj;
> +
> +			/*
> +			 * Calculate an acceptable degree of imbalance based
> +			 * on imbalance_adj. However, do not allow a greater
> +			 * imbalance than the child domains weight to avoid
> +			 * a case where the allowed imbalance spans multiple
> +			 * LLCs.
> +			 */

That comment is a wee misleading, @child is not an LLC per se. This
could be the NUMA distance 2 domain, in which case @child is the NUMA
distance 1 group.

That said, even then it probably makes sense to ensure you don't idle a
whole smaller distance group.

> +			imbalance_adj = busiest->group_weight * (env->sd->imbalance_pct - 100) / 100;
> +			imbalance_adj = min(imbalance_adj, child->span_weight);
> +			imbalance_adj >>= 1;
> +
> +			/*
> +			 * Ignore small imbalances when the busiest group has
> +			 * low utilisation.
> +			 */
> +			if (busiest->sum_nr_running < imbalance_adj)
> +				env->imbalance = 0;
> +		}
> +
>  		return;
>  	}
>