Intel C2D COREs not used equally in FreeBSD 7.0-CURRENT i386

Attilio Rao attilio at FreeBSD.org
Mon Jun 4 21:49:57 UTC 2007


Abdullah Ibn Hamad Al-Marri wrote:
> On 6/5/07, youshi10 at u.washington.edu <youshi10 at u.washington.edu> wrote:
>> On Mon, 4 Jun 2007, Abdullah Ibn Hamad Al-Marri wrote:
>>
>> > On 6/4/07, Alexey Tarasov <master at preved.cn> wrote:
>> >> Hi.
>> >>
>> >>  > options         SCHED_4BSD              # 4BSD scheduler
>> >>
>> >> Try to use SCHED_ULE or SCHED_CORE.
>> >>
>> >> ::[ | | | | ]::
>> >> Alexey Tarasov
>> >> master at preved.cn
>> >
>> > I thought Davidxu removed SCHED_CORE for SCHED_ULE, am I missing 
>> something?
>> > --
>> > Regards,
>> >
>> > -Abdullah Ibn Hamad Al-Marri
>> > Arab Portal
>> > http://www.WeArab.Net/
>>
>> I think so. Try SCHED_ULE.
>>
>> -Garrett
> 
> I'm using ULE now, and I see mysqld could use 104%in WCPU.
> 
> But do you know why services still share the cpu0 instead of using CPU1
> 
> PID USERNAME  THR PRI NICE   SIZE    RES STATE  C   TIME   WCPU COMMAND
>  598 mysql      16  44    0   639M   633M ucond  0  50:53  4.83% mysqld
>  740 services    3  44    0 54292K 51352K select 0   3:08  0.00% services
> 
> I wish kernel can balance using both CORES with ULE 2.0

ULE currently uses some tricks in order to avoid too many sched_lock 
acquisitions that pessimize in load balancing and in pick_pri case.
In the night, I think Jeff is going to commit the infrastructure to let 
sched_lock back home and soon we will have ULE sched_lock free. It means 
that we could really have a good balancing and an effective usage of 
pick_pri for it.

As saying: it is a known bug, and partially intentional...

Attilio



More information about the freebsd-current mailing list