Re: ULE scheduling oddity

On Thu, 17 Jul 2008, Steve Kargl wrote:

> On Thu, Jul 17, 2008 at 09:12:45AM -0700, Barney Cordoba wrote:
>>
>>> Actually, 10 copies of the little app are the only things
>>> running except
>>> top(1) and few sleeping system services (e.g., nfsd and
>>> sshd).  Apparently,
>>> you missed the "41 processes:  11 running, 30
>>> sleeping" line above.
>>>
>>
>> Your apparent argument that somehow every cpu cycle can be
>> sliced equally and automagically is as silly
>
> I do not expect a single cpu cycle to be split evenly
> between the running processes.  I do however expect that
> 8e12 cpu cycles to be split in a better distribution.
>
>> as the expectation that a first generation scheduler will
>> exhibit 100% efficiency across 8 cpus.
>
> ULE in -current is no longer 1st generation.  I tested the
> original ULE when jeffr committed and reported a few panics
> and provided some of the first feedback of interactivity
> problems.
>
> Perhaps, I should have sent my original email directly to
> jeffr instead of the freebsd-current list where others
> might find the observation of interest.  If one expects to
> see future improvements in ULE, then providing feedback
> is crucial.  Apparently, you have a different opinion.

Hey Steve,

Thanks again for providing valuable feedback.  The new cpu topology aware 
load balancing could've disturbed the ever delicate balancing algorithm. 
Do you know if this is better in 7.0 than current?  I know at one time 
I had examined this very workload.

Can you try increasing the balance frequency? (lessening the interval 
value in sysctl)

I haven't been reading current much lately, I must confess.  Please 
contact me directly if I don't respond here in a timely fashion.

Thanks!
Jeff

>
> -- 
> Steve
> _______________________________________________
> freebsd-current_at_freebsd.org mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-current
> To unsubscribe, send any mail to "freebsd-current-unsubscribe_at_freebsd.org"
>