Re: Progress on scaling of FreeBSD on 8 CPU systems

On Mon, Feb 26, 2007 at 07:08:37PM +1100, Peter Jeremy wrote:
On 2007-Feb-24 16:31:11 -0500, Kris Kennaway <kris@xxxxxxxxxxxxxx> wrote:
We have recently made significant progress on optimizing for MySQL
running on an 8-core amd64 system. The graph of results may be found
here:

http://www.freebsd.org/~kris/scaling/scaling.png

Those results look wonderful. Have you tried increasing the number of
threads to see if there's any nasty knee further to the right? Also,
is there any chance of repeating this testing on one of the big Suns
(or a T2000) to see how this scales to lots of cores?

I didnt try at much higher loads yet, that will be interesting to
explore.

Currently on the Sun T2000 (32 hardware threads = virtual CPUs) we
don't have good scaling (it's hard even to saturate all CPUs for
kernel workloads): the current thinking is that this is largely
because of contention on the global scheduler lock. Even when idle
there is a lot of contention on the sched_lock coming from e.g. all 32
CPUs running statclock at once HZ=1000 times a second and fighting for
the sched_lock.

Fixing this is something Jeff and Attilio are working on (Kip Macy
also did a lot of exploratory work last year), so we hope to make
further progress over the coming weeks.

Kris
_______________________________________________
freebsd-hackers@xxxxxxxxxxx mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-hackers
To unsubscribe, send any mail to "freebsd-hackers-unsubscribe@xxxxxxxxxxx"

Relevant Pages

  • Re: Progress on scaling of FreeBSD on 8 CPU systems
    ... The graph of results may be found ... is there any chance of repeating this testing on one of the big Suns ... Currently on the Sun T2000 (32 hardware threads = virtual CPUs) we ... because of contention on the global scheduler lock. ...
    (freebsd-current)
  • Re: Lockfree or not to lockfree thats the question !
    ... 100 CPUs the FSB will become the bottleneck. ... contention while locks minimize contention. ... If the object is handled in a lock free way, one-third of the time, on ...
    (comp.programming.threads)
  • Re: Lockfree or not to lockfree thats the question !
    ... So if you have, say, 8 CPUs. ... contention while locks minimize contention. ... If the object is handled in a lock free way, one-third of the time, on ... level algorithms such as memory management, ...
    (comp.programming.threads)
  • Re: Lockfree or not to lockfree thats the question !
    ... of the CPUs run threads that use the lock free structure and the other ... any way, shape or form, will cast a marked negative effect upon the threads ... However, yes, I am saying that many lock-free algorithms, when used as ... This increases contention if there are other threads ...
    (comp.programming.threads)
  • Re: Lockfree or not to lockfree thats the question !
    ... So if you have, say, 8 CPUs. ... contention while locks minimize contention. ... If the object is handled in a lock free way, one-third of the time, on ... level algorithms such as memory management, ...
    (comp.programming.threads)