Re: [patch] sched/fair: Add SIS_MIN_LAT feature to mitigate PELT lag induced stacking

[Mike Galbraith]

On Mon, 2018-01-01 at 11:37 +0100, Mike Galbraith wrote:
> 
> 
> Here 0-day bot, here boy..

Forget it 0-day bot, and assorted crickets :)

> +	/*
> +	 * When heavily loaded, stacking may be the best/only latency option,
> +	 * and LAST_BUDDY the only hope of preventing throughput collapse.
> +	 */
> +	if (unlikely(task_cfs_rq(current)->nr_running >= sched_nr_latency) &&
> +		     task_cfs_rq(p)->nr_running >= sched_nr_latency)
> +		return target;

With stock knob settings, that's too late to switch from llc -> l2
affinity for sync wakeups, and completely demolished tbench top end on
huge socket NUMA box with lots of bandwidth.  Lovely for desktop,
somewhere below gawd-awful for big box performance.

	-Mike