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

[Rik van Riel <riel@surriel.com>]

[-- Attachment #1: Type: text/plain, Size: 1482 bytes --]

On Mon, 2020-01-06 at 14:42 +0000, Mel Gorman wrote:
> 
> +
> +		/* Consider allowing a small imbalance between NUMA
> groups */
> +		if (env->sd->flags & SD_NUMA) {
> +			long imbalance_adj, imbalance_max;
> +
> +			/*
> +			 * imbalance_adj is the allowable degree of
> imbalance
> +			 * to exist between two NUMA domains.
> imbalance_pct
> +			 * is used to estimate the number of active
> tasks
> +			 * needed before memory bandwidth may be as
> important
> +			 * as memory locality.
> +			 */
> +			imbalance_adj = (100 / (env->sd->imbalance_pct
> - 100)) - 1;
> +
> +			/*
> +			 * Allow small imbalances when the busiest
> group has
> +			 * low utilisation.
> +			 */
> +			imbalance_max = imbalance_adj << 1;
> +			if (busiest->sum_nr_running < imbalance_max)
> +				env->imbalance -= min(env->imbalance,
> imbalance_adj);
> +		}
> +

Wait, so imbalance_max is a function only of
env->sd->imbalance_pct, and it gets compared
against busiest->sum_nr_running, which is related
to the number of CPUs in the node?

Lets see how this works out for different numbers
of CPUs in each node.

With imbalance_pct == 115, we end up with 
imbalance_adj = (100 / (115 - 100)) - 1 = 0.18,
which gets rounded down to 0.

Wait, never mind the different numbers of CPUs.

Am I overlooking something, or did you mean to have
a factor of number of CPUs in the imbalance_adj
calculation?

kind regards,

Rik
-- 
All Rights Reversed.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 488 bytes --]