lkml.org 
[lkml]   [2018]   [Sep]   [5]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC PATCH 3/4] sched/topology: remove smt_gain
* Vincent Guittot <vincent.guittot@linaro.org> [2018-09-05 11:11:35]:

> On Wed, 5 Sep 2018 at 10:50, Srikar Dronamraju
> <srikar@linux.vnet.ibm.com> wrote:
> >
> > * Vincent Guittot <vincent.guittot@linaro.org> [2018-09-05 09:36:42]:
> >
> > > >
> > > > I dont know of any systems that have come with single threaded and
> > > > multithreaded. However some user can still offline few threads in a core
> > > > while leaving other cores untouched. I dont really know why somebody
> > > > would want to do it. For example, some customer was toying with SMT 3
> > > > mode in a SMT 8 power8 box.
> > >
> > > In this case, it means that we have the same core capacity whatever
> > > the number of CPUs
> > > and a core with SMT 3 will be set with the same compute capacity as
> > > the core with SMT 8.
> > > Does it still make sense ?
> > >
> >
> > To me it make sense atleast from a power 8 perspective, because SMT 1 >
> > SMT 2 > SMT 4 > SMT8. So if one core is configured for SMT 2 and other
> > core is configured for SMT4; all threads being busy, the individual
> > threads running on SMT2 core will complete more work than SMT 4 core
> > threads.
>
> I agree for individual thread capacity but at core group level, the
> core SMT 1 will have the same capacity as core group SMT 8 so load
> balance will try to balance evenly the tasks between the 2 cores
> whereas core SMT 8 > core SMT1 , isn't it ?
>

I believe that Core capacity irrespective of the number of threads
should be similar. We wanted to give a small benefit if the core has
multiple threads and that was smt_gain. Lets say we have 8 equal sw
threads running on 2 cores; one being SMT 2 and other being SMT4.
then 4 threads should be spread to each core. So that we would be fair
to each of the 8 SW threads.

--
Thanks and Regards
Srikar Dronamraju

\
 
 \ /
  Last update: 2018-09-05 13:14    [W:0.086 / U:1.796 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site