On Wednesday 08 January 2003 20:32, Con Kolivas wrote:
> > Could you add a time per load metric? (I.E. 86.9/21=4.14 seconds. Yeah,
> > I could do the math myself, but that and total time are usually what I'm
> > trying to compare when I look at these. Maybe it's just me...)
>
> If you look at the information carefully the meaningful number is
>
> (Loads ) / ( process_load_time - no_load_time)
Hmmm... Have to think about this a sec...
So far I've just been looking at the deltas between versions, like I said,
with the implicit assumption that no_load_time remains roughly constant
(after all, kernel build time is what everybody's been optimizing for since
the 2.0 era).
There are really two things it would be nice to isolate: one is the amount of
thrashing the extra processing introduces, slowing down the whole system.
The other is the balancing decisions that are made (the amount of work done
by io_load or mem_load varies and has no impact on the termination of the
test as a whole...) I sort of want to isolate out the balancing decisions a
bit, or at least have a metric to look at them and compare them. (I.E. "yeah
it got slower, but it did more work overall". Now is this what everybody
WANTS, and could we maybe twiddle this with precedence in the scheduler or
something if it isn't?)
I suppose your metric is a more accurate way of measuring that. Cool.
> but keep an eye out for a new version soon.
>
> Con
Of course, :)
Rob
-- penguicon.sf.net - A combination Linux Expo and Science Fiction Convention with GOHs Terry Pratchett, Eric Raymond, Pete Abrams, Illiad & CmdrTaco. - 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 : Wed Jan 15 2003 - 22:00:33 EST