[patch] 2.6.21-rc4 nicksched v32

[Al Boldi]

Nick Piggin wrote:
> Forward port of nicksched to 2.6.21-rc4.

Great!

> Can't find my old
> design/description document for it, but it is yet another scheduler. Focus
> is on simplicity and fairness.

Simplicity is really key.

> This one tends to require X to be reniced to -10 or so for best results
> (renice -10 `pidof X`).

No problem.

> Timeslices get scaled by /proc/sys/kernel/base_timeslice. If you have bad
> interactivity you could try lowering this and see if it helps.

As I am on 2.6.20, I can't really test this patch, but I tried your sched 
from PlugSched and its not bad.

I'm getting strange numbers with chew.c, though.  Try this:
Boot into /bin/sh
Run ./chew on one console.
Run ./chew on another console.
Watch latencies.

Console 1:
pid 671, prio   0, out for  799 ms, ran for  800 ms, load  50%
pid 671, prio   0, out for  799 ms, ran for  801 ms, load  50%
pid 671, prio   0, out for  799 ms, ran for  799 ms, load  49%
pid 671, prio   0, out for  800 ms, ran for  800 ms, load  49%

Console 2:
pid 672, prio   0, out for  800 ms, ran for  799 ms, load  49%
pid 672, prio   0, out for  799 ms, ran for  800 ms, load  50%
pid 672, prio   0, out for  799 ms, ran for  800 ms, load  50%
pid 672, prio   0, out for  799 ms, ran for  799 ms, load  49%
pid 672, prio   0, out for  799 ms, ran for  799 ms, load  49%

Looks good, but now add a cpu-hog, ie. while :; do :; done

Console 1:

pid 671, prio   0, out for  799 ms, ran for  399 ms, load  33%
pid 671, prio   0, out for  799 ms, ran for  399 ms, load  33%
pid 671, prio   0, out for  799 ms, ran for  399 ms, load  33%
pid 671, prio   0, out for  799 ms, ran for  399 ms, load  33%

Console 2:
pid 672, prio   0, out for 1599 ms, ran for  799 ms, load  33%
pid 672, prio   0, out for 1599 ms, ran for  799 ms, load  33%
pid 672, prio   0, out for 1599 ms, ran for  800 ms, load  33%
pid 672, prio   0, out for 1599 ms, ran for  799 ms, load  33%

It's smooth, but latencies are double on console2.  Easy enough to fix, 
though, just press scrollock to induce a sleep and then release.

Also, latencies are huge.  Is there a way to fix latencies per nice?


Thanks!

--
Al


/*
 * original idea by Chris Friesen.  Thanks.
 */

#include <stdio.h>
#include <sys/time.h>
#include <sys/resource.h>

#define THRESHOLD_USEC 2000

unsigned long long stamp()
{
        struct timeval tv;
        gettimeofday(&tv, 0);
        return (unsigned long long) tv.tv_usec + ((unsigned long long) tv.tv_sec)*1000000;
}


int main()
{
        unsigned long long thresh_ticks = THRESHOLD_USEC;
        unsigned long long cur, last, start, act;
        struct timespec ts;

        sched_rr_get_interval(0, &ts);
        printf("pid %d, prio %3d, interval of %d nsec\n", getpid(), getpriority(PRIO_PROCESS, 0), ts.tv_nsec);

        start = last = stamp();
        while(1) {
                cur = stamp();
                unsigned long long delta = cur-last;
                if (delta > thresh_ticks) {
			act = last - start;
                        printf("pid %d, prio %3d, out for %4llu ms, ran for %4llu ms, load %3llu%\n"
			, getpid(), getpriority(PRIO_PROCESS, 0), delta/1000, act/1000,(act*100)/(cur-start));
                        start = cur = stamp();
                }
                last = cur;
        }

        return 0;
}