> Well, I tried the patch, and it make a considerable difference on io
> based processes.
And on multimedia processes that only use x% of the CPU.
X11amp, mpg123 and mikmod are running quite a lot better
on my system...
(and my editor flies too, it takes some getting used to
when writing this letter :-)
> When I checked to see how many cpu burst that were
> actually less than the quantum I got .8 %.
>
> Which seems really low. Correct me if I am wrong.
> But shouldn't it be more like about 80% of cpu burst < quantum?
Quantum = 200ms (for a 0-niced process)
And on a personal system (with rc5des), there isn't much
going on that uses considerable CPU.
It's a bit like the "20% of the files use 80% of the disk"
story, it looks weird, but when you think about it, it's
quite OK.
> I was running a total count on goodness calls where there was still
> counter, then totalled the amount that relinquished control before the
> end of the quantum.
> inside of goodness function on if (weight)... I did the total count,
> and in if (p == prev) i did the other.
>
> Again if I am in error here please set me strait on where to go. :)
I think your measurements are quite correct, since the number
of context switches also didn't go up.
My patch just gives non-CPU eating processes a larger edge over
niced processes.
Rik.
+-------------------------------------------+--------------------------+
| Linux: - LinuxHQ MM-patches page | Scouting webmaster |
| - kswapd ask-him & complain-to guy | Vries cubscout leader |
| http://www.fys.ruu.nl/~riel/ | <H.H.vanRiel@fys.ruu.nl> |
+-------------------------------------------+--------------------------+
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu