Re: [PATCH] O2int 0307041440 for 2.5.74-mm1

From: Wiktor Wodecki (wodecki@gmx.de)
Date: Sun Jul 06 2003 - 07:36:35 EST


On Fri, Jul 04, 2003 at 02:59:08PM +1000, Con Kolivas wrote:
Content-Description: clearsigned data
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> Here is a patch against the current O1int patch in 2.5.74-mm1.
> Since the O1int didn't mean anything I thought I'd call this O2int.
>
> This one wont blow you away but tames those corner cases.
>
> Changes:
> The child penalty is set on 80% which means that tasks that wait on their
> children have children forking just on the edge of the interactive delta so
> they shouldn't starve their own children.
>
> The non linear sleep avg boost is scaled down slightly to prevent this
> particular boost from being capable of making a task highly interactive. This
> makes very new tasks less likely to have a little spurt of too high priority.
>
> Idle tasks now get their static priority over the full time they've been
> running rather than starting again at 1 second. This makes it harder for idle
> tasks to suddenly become highly interactive and _then_ fork an interactive
> bomb. Not sure on this one yet.
>
> The sched_exit penalty to parents of cpu hungry children is scaled accordingly
> (was missed on the original conversion so works better now).
>
> Hysteresis on interactive buffer removed (was unecessary).
>
> Minor cleanup.
>
> Known issue remaining:
> Mozilla acts just like X in that it is mostly interactive but has bursts of
> heavy cpu activity so it gets the same bonus as X. However it makes X jerky
> during it's heavy cpu activity, and might in some circumstances make audio
> skip. Fixing this kills X smoothness as they seem very similar to the
> estimator. Still haven't sorted a workaround for this one but I'm working on
> it. Ingo's original timeslice granularity patch helps a little and may be
> worth resuscitating (and the desktop only people can change the granularity
> down to 10ms to satisfy their needs).
>
> Con

this one behaves worse than the previou one. When loading a big page
(such as www.heise.de or www.ebay.de) and moving the mouse in circles
it gets very jumpy.

-- 
Regards,

Wiktor Wodecki


- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Mon Jul 07 2003 - 22:00:26 EST