> Maybe it's an idea to add this change to kernel/sched.c::goodness()
>
> if (p == prev)
> weight += 1;
> }
> -->
> else if (p->processor != this_cpu)
> weight = -1;
> <--
> return weight;
> }
the problem you are trying to solve is IMO elsewhere. The above change
makes a difference only if all runnable processes have a zero-counter, ie.
counter recalculation is due. The simple solution is to simply repeat the
goodness() run after recalculation (the relevant data structures are
cache-hot already and recalculation is a rare event nevertheless), because
otherwise the first reschedule after recalculation isnt just random. I did
this some time ago and it showed no measureable effect. (i've attached a
patch that does something like that)
-- mingo
--- linux/kernel/sched.c.orig Tue Mar 30 11:01:33 1999
+++ linux/kernel/sched.c Wed Apr 14 12:56:58 1999
@@ -691,6 +691,7 @@
sched_data->prevstate = prev->state;
+repeat:
/* this is the scheduler proper: */
{
struct task_struct * p = init_task.next_run;
@@ -735,6 +736,8 @@
for_each_task(p)
p->counter = (p->counter >> 1) + p->priority;
read_unlock(&tasklist_lock);
+ spin_lock_irq(&runqueue_lock);
+ goto repeat;
}
}
-
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/