Re: 2.6.8-rc2-mm2

[Nick Piggin]

Con Kolivas wrote:
> Con Kolivas writes:
>
> > Nick Piggin writes:
> >
> > > Con Kolivas wrote:
> > >
> > > > Andrew Morton wrote:
> > >
> > >
> > > > Anyone with feedback on this please cc me. This was developed 
> > > > separately from the -mm series which has heaps of other scheduler 
> > > > patches which were not trivial to merge with so there may be 
> > > > teething problems. Good reports dont hurt either ;)
> > >
> > > I can't get onto the OSDL site now, but I seem to remember staircase
> > > having some performance problems on a few things. Hackbench and reaim
> > > from memory... are these fixed? was I dreaming?
> >
> >
> > Definitely dreaming I'm afraid :D
> >
> > The performance on both reaim and hackbench has always equalled or 
> > exceeded mainline so thanks for bringing it up.

(OSDL's search thingy still isn't working quite right, but I'll get back
to you about this when it does.)


Otherwise, a couple of problems I noticed:

You removed things like this:
-	/*
-	 * The idle thread is not allowed to schedule!
-	 * Remove this check after it has been exercised a bit.
-	 */
-	if (unlikely(current == rq->idle) && current->state != TASK_RUNNING) {
-		printk(KERN_ERR "bad: scheduling from the idle thread!\n");
-		dump_stack();
-	}
-
And child-runs-first in wake_up_new_task. Please don't.

Also, basic interactivity in X is bad with the interactive sysctl set to 0
(is X supposed to be at nice 0?), however fairness is bad when interactive is 1.
I'm not sure if this is an acceptable tradeoff - are you planning to fix it?

It has interactivity problems with "thud". Also the mouse can freeze for .5 to 1
second when moving between windows while there is disk IO going on in the background
(this is with interactive = 1). The test-starve problem is back.

Increasing priority (negative nice) doesn't have much impact. -20 CPU hog only gets
about double the CPU of a 0 priority CPU hog and only about 120% the CPU time of a
nice -10 hog.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/