Re: [patch] "HT scheduler", sched-2.5.63-B3

[Ingo Molnar]

On Thu, 6 Mar 2003, Linus Torvalds wrote:

> > okay, here's a patch that merges Linus' "priority boost backmerging" and
> > my CPU-hog-detection-improvement patches.
> 
> I was actually going to suggest making the CPU-hog detector _worse_, to
> see what the extreme behaviour of the "boost balancing" is. One of the
> things I would hope for is that the interactivity balancing would act as
> a damper on the CPU hug detector, and make it less important to get that
> one right in the first place.

this does not appear to happen, at least with the current incarnation,
make -j5 jobs of the kernel tree (on UP), with your patch alone are
hoovering at a considerably higher dynamic priority than they are with my
patch. But interactivity itself is good as far as i can tell, in both
cases - but i think Andrew would be a better tester in this regard.

plus my patch doesnt really change the core CPU-hog detector, what it does
is that it changes the way children/parents will inherit priorities, plus
changes the order of how children get their first timeslice. The
end-effect is that the 'CPU-hog' make job is better recognized as that -
but X itself will not be recognized in any different way.

	Ingo