kern/85820: 1.5 times slower performance with SCHED_ULE than SCHED_4BSD

[O. Hartmann]

Ceri Davies wrote:
> On Sun, Oct 08, 2006 at 07:51:41PM +0200, O. Hartmann wrote:
>   
>> Ceri Davies wrote:
>>     
>>> Synopsis: 1.5 times slower performance with SCHED_ULE than SCHED_4BSD
>>>
>>> State-Changed-From-To: open->closed
>>> State-Changed-By: ceri
>>> State-Changed-When: Sun Oct 8 17:19:36 UTC 2006
>>> State-Changed-Why: 
>>> ULE is no longer the default scheduler, and no longer has a maintainer.
>>> This is an interesting test case though.
>>>
>>>
>>> Responsible-Changed-From-To: freebsd-amd64->freebsd-bugs
>>> Responsible-Changed-By: ceri
>>> Responsible-Changed-When: Sun Oct 8 17:19:36 UTC 2006
>>> Responsible-Changed-Why: 
>>> Scheduler problem.
>>>
>>> http://www.freebsd.org/cgi/query-pr.cgi?pr=85820
>>> _______________________________________________
>>> freebsd-amd64 at freebsd.org mailing list
>>> http://lists.freebsd.org/mailman/listinfo/freebsd-amd64
>>> To unsubscribe, send any mail to "freebsd-amd64-unsubscribe at freebsd.org"
>>>   
>>>       
>> Very interesting to read. And this is now the end of the newly
>> introduced allround weapon for more scalability?
>>     
>
> There is some one working on SCHED_ULE, but they are not a committer and
> they suggested that this PR could be closed.  It contains an interesting
> test case but no real problem as such, only that one scheduler is slower
> that the other for certain loads.  That is always going to be true.
>
> Ceri
>   
Sorry for my smug comment.
Today I was searching the web for scaleability benchmarks on the
different flavors of BSD and Linux and I read very much about the great
benefits of ULE, and it is really interesting to read about ULE what has
been written in 2004 and January 2005.

Well, today I exchanged SCHED_ULE in my kernel config back to SCHED_4BSD
on my UP FreeBSD 6.2-PRE/AMD64 box and on normal desktop usage I
experience a significant performance impact (Firefox/Thunderbird react
both much smoother).
Tomorrow I will change the same on my lab's i386 box (also UP) and see,
whether some calculations also will run faster.

Well, I'm not the benchmark crack anyhow and as I knew SCHED_ULE was the
defualt for 6.1 and it was said that even UP systems also benefit from
the improvements. At this point, I see that there is a great need in
work to be done.

Regards,
Oliver