On optimising the scheduler for large run queues

[Mark Hahn]

Jamie Lokier said:
> They need good process _selection_ criteria for that maximum pleasure,
> deep interactive experience.  (I.e. move the mouse and the pointer moves
> in X even though you've got 3 parallel compiles running).

good point.  I mentioned something like it a while back in a post
about how cacheflush_time is miscalculated.  I was pondering the fact
that under SMP, the scheduler keeps an estimate of the length each
process keeps its timeslice, in cycles.  it uses that only when considering
whether to bounce a process to another CPU, though.  but perhaps another good
heuristic would be to keep the measure for UP as well, and give short-slice
tasks a goodness() benefit.  it sort of does that with counter, but that's
only 10ms granularity, and interacts with niceness...

regards, mark hahn.


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo en vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/