Re: Linux 2.6.0-test6

[Nick Piggin]

bill davidsen wrote:

> In article <3F77BB2C.7030402@xxxxxxxxxxxxxxx>,
> Nick Piggin  <piggin@xxxxxxxxxxxxxxx> wrote:
>
> | AFAIK, Con's scheduler doesn't change the nice implementation at all.
> | Possibly some of his changes amplify its problems, or, more likely they
> | remove most other scheduler problems leaving this one noticable.
> | 
> | If X is running at -20, and xmms at +19, xmms is supposed to still get
> | 5% of the CPU. Should be enough to run fine. Unfortunately this is
> | achieved by giving X very large timeslices, so xmms's scheduling latency
> | becomes large. The interactivity bonuses don't help, either.
>
> Clearly the "some is good, more is better" approach doesn't provide
> stable balance between sound and cpu hogs. It isn't a question of "how
> much" cpu, just "when"which works or not.
>
> This is sort of like the deadline scheduler in that it trades of
> throughput for avoiding jackpot cases. I think that's desired behaviour
> in a CPU schedular too, at least if used by humans.

I'm not sure what you mean. There is nothing good to say about Ingo's
nice mechanism though (sorry Ingo, its otherwise a very nice
scheduler!).

In my scheduler, nice -20 processes get small timeslices so scheduling
latency stays low or even gets lower, while nice +19 ones get large
timeslices for lower context switches and better cache efficiency. As
you would like.


-
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/