cd> Benny Amorsen <amorsen@sscnet.com> wrote:
>> As it is, the kernel locks often become a bottleneck on systems
>> with lots of penguins. Look at this scenario:
cd> Just curious, what exactly do you mean by "lots of penguins"? I
cd> assume you mean processes? What do penguins have to do with
cd> anything. ;)
Sorry, I meant lots of processors. Does Ultralinux still display "N
penguins found" these days, or did that go the way of "lpN on fire"?
cd> Actually running? That would be bad. It would mean that we'd have
cd> to preempt another processor, which I gather is a very expensive
cd> thing to do.
I don't know how expensive it is, you are probably right.
cd> If you mean that when scheduling is already
cd> happening, run processes with a kernel lock before processes
cd> without one, I don't know. It may be worth benchmarking to decide.
This is the approach Rik van Riel's patch takes, but only with
SCHED_IDLE (based on my reading). It looks like it could easily be
adapted to do what I propose. -- the big question is whether it would
be any help.
To test the effect, we would need a benchmark for which the kernel
currently scales badly due to lock contention, but which also does a
significant amount of user space processing.
Benny
-
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/