On Sun, 30 Jan 2000, Jamie Lokier wrote:
> Rik van Riel wrote:
> > > Are you sure that CPU measurement is real? These sort of changes
> > > tend to break process accounting.
> >
> > I agree that process accounting is kind of random.
> > However, it is exactly this randomness that makes
> > me believe that it really isn't a measurement fault.
>
> With your changes, I wouldn't be surprised to find an entirely
> non-random measurement error. You're messing with scheduling
> quanta, and thats exactly where measurement tends to misrepresent
> what's going on. But...
I'm not at all measuring with scheduling quanta. All my patch
changes is _when_ tasks get scheduled, not how much time they
get.
> > Before the patch, xmms cpu usage varied between 70
> > and 95%. Now it varies between 50 and 75%, this
> > looks like too big a thing to be random.
>
> One way to test this directly is to run a second process at the
> same time that does:
>
> while (1) {
> i = 0;
> time = read_tsc ();
> while (read_tsc () - time < 1 second)
> i++;
> printf( "%d loops per second\n", i );
> }
>
> The inner loop should be written to use as little cache footprint
> as possible of course.
>
> If your patch really works, then xmms will use less CPU and this
> will get more cycles.
Indeed. Maybe I'll try this when I'm fully awake again
(feeling very tired now)...
cheers,
Rik
-- The Internet is not a network of computers. It is a network of people. That is its real strength.- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Mon Jan 31 2000 - 21:00:25 EST