DragonFly commits List (threaded) for 2005-06
[
Date Prev][
Date Next]
[
Thread Prev][
Thread Next]
[
Date Index][
Thread Index]
Re: cvs commit: src/sys/sys usched.h src/sys/kern usched_bsd4.c kern_clock.c src/bin/ps extern.h keyword.c print.c
Matthew Dillon wrote:
:
:David Rhodus wrote:
:> That has already been the case for over a year now that each cpu has had
:> it own scheduler.
:>
:
: The above statement is only true for the in-kernel LWKT scheduler.
:
: As for the userland scheduler, it's a totally different beast.
:
: -Hiten
: Hiten Pandya
Right. the LWKT scheduler (which does the actual real life scheduling
of the thread) is a per-cpu scheduler. The userland scheduler, whos
job it is to select which of the runnable userland processes should be
passed on to LWKT, currently requires the Big Giant Lock.
Note that all kernel threads are pure LWKT threads. So while I want
to fix the userland scheduler issue I don't think it would actually
make much difference in performance.
-Matt
Matthew Dillon
<dillon@xxxxxxxxxxxxx>
But I suppose once the whole path becomes BGL-free, it will certainly
make a difference.
-Hiten
[
Date Prev][
Date Next]
[
Thread Prev][
Thread Next]
[
Date Index][
Thread Index]