CPU affinity in new ULE scheduler

[Robert Watson]

On Wed, 9 Nov 2005, Ian Lord wrote:

> Are you saying that ULE is slower then 4BSD ?
>
> I'm new to this and when I compiled my kernel, it was "clear" ULE was a 
> better alternative for performance then 4BSD

Schedulers are one of the hardest things to do right in OS design, as they 
rely a great deal on how workloads behave and interact.  I've seen 
significantly varied performance between the two -- there are a lot of 
anecdotal reports that ULE is better for "interactive" workloads on a busy 
desktop machine, but keep in mind that 4BSD has seen a number of 
improvements in the last few years also.  Right now, 4BSD is considered 
the "production" scheduler for FreeBSD, although there's continuing 
interest in improving ULE, as well as integrating some of the techniques 
used in ULE into 4BSD.  For example, ULE used to see a significant 
performance win over 4BSD on SMP as it did a better job of identifying 
idle CPUs and migrating work to those CPUs.  4BSD has improved a lot on 
this front in the last year or two, and so has caught up with some of 
those benefits.

In the end, only by measuring will you be able to tell if ULE is better 
for your workload.  Measurement can mean qualitative experience 
(everything seems snappier) or quantitative (I get 14% more transactions 
per second with scheduler X).

Robert N M Watson

>
> Thanks
>
> At 19:05 2005-11-09, Kris Kennaway wrote:
>> On Wed, Nov 09, 2005 at 04:16:31PM -0600, Jon Brisbin wrote:
>> > I can't find any information on how to set the CPU affinity for processes 
>> in the
>> > FreeBSD 6 ULE scheduler.
>> 
>> That's because you can't.  ULE gives lower performance on the
>> workloads I have tested anyway.  This may be fixed in the future.
>> 
>> Kris
>> 
> _______________________________________________
> freebsd-questions at freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-questions
> To unsubscribe, send any mail to "freebsd-questions-unsubscribe at freebsd.org"
>