Re: [patch][rfc] quell interactive feeding frenzy

[Con Kolivas <kernel@kolivas.org>]

On Thursday 13 April 2006 17:41, Mike Galbraith wrote:
> This way also allowed me to eliminate the interactive agony of an array
> switch when at 100% cpu.  Seems to work well.  No more agony, only tiny
> pin pricks.
>
> Anyway, interested readers will find a copy of irman2.c, which is nice
> for testing interactive starvation, attached.   The effect is most
> noticeable with something like bonnie, which otherwise has zero chance
> against irman2.  Just about anything will do though.  Trying to fire up
> Amarok is good for a chuckle.  Whatever.  (if anyone plays with irman2
> on 2.6.16 or below, call it with -S 1)

Comments.

> +repeat:
> +	while ((idx = find_next_bit(bitmap, MAX_PRIO, idx)) < MAX_PRIO) {

...

> +		goto repeat;

...

> +               if (rq->nr_running > 1)
> +                       requeue_starving(rq, now);

An O(n) function in scheduler_tick is probably not the way to tackle this.

-- 
-ck