[gmx-users] multi threads pinning problem

[Zhenyu Meng]

Hi,
Thanks for your help. The tutorial is very helpful for me.

2015-07-24 19:22 GMT+08:00 Mark Abraham <mark.j.abraham at gmail.com>:

> Hi,
>
> Check out
>
> http://manual.gromacs.org/documentation/5.1-rc1/user-guide/mdrun-performance.html
> and
>
> http://manual.gromacs.org/documentation/5.1-rc1/user-guide/mdrun-features.html#running-multi-simulations
> .
> It's hard to be definitive about the correct thing to do, since it depends
> very much on all of your infrastructure. You should observe the performance
> when using all the cores (pinned by default) and when using a single
> (pinned) core and expect performance to be reasonable consistent as you
> change the number of cores.
>
> Mark
>
> On Fri, Jul 24, 2015 at 12:18 PM <terrencesun at gmail.com> wrote:
>
> > Hi,
> >
> > The -pin is used to prevent the OS switching threads across physical
> > cores, which may result in performance loss. If you want to pin your
> > threads to cores, use -pin on. -pinoffset can help you set to which cores
> > your threads are pinned, which is stated in mdrun -h.
> >
> > Regards
> > Terry
> >
> >
> > > On 24 Jul 2015, at 11:53 am, Zhenyu Meng <fdmm1989 at gmail.com> wrote:
> > >
> > > Dear all,
> > > When I'm running gromacs with mdrun, I always receive such message:
> > >
> > > The number of threads is not equal to the number of (logical) cores
> > > and the -pin option is set to auto: will not pin thread to cores.
> > > This can lead to significant performance degradation.
> > > Consider using -pin on (and -pinoffset in case you run multiple jobs).
> > >
> > > The server I use has 128 cores, and usually I use mdrun with flag -nt
> 32
> > or
> > > -nt 64 to specify the threads, I also notice there's large performance
> > lost
> > > in each run.
> > > I wonder what's the difference between '-pin on' and '-pinoffset'. I
> > > usually run multi jobs on the server so will '-pin on' or '-pinoffset'
> > > affect the other jobs? If I use '-pinoffset' what value I should
> > > set(suppose I have already use 32 cores for other jobs)?
> > >
> > > --
> > > Sincerely,
> > > Mr. Meng Zhenyu
> > > Division of Chemistry and Biological Chemistry
> > > School of Physical and Mathematical Sciences
> > > Nanyang Technological University
> > > --
> > > Gromacs Users mailing list
> > >
> > > * Please search the archive at
> > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> > posting!
> > >
> > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> > >
> > > * For (un)subscribe requests visit
> > > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> > send a mail to gmx-users-request at gromacs.org.
> > --
> > Gromacs Users mailing list
> >
> > * Please search the archive at
> > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> > posting!
> >
> > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
> >
> > * For (un)subscribe requests visit
> > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> > send a mail to gmx-users-request at gromacs.org.
> >
> --
> Gromacs Users mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before
> posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or
> send a mail to gmx-users-request at gromacs.org.
>



-- 
Sincerely,
Mr. Meng Zhenyu
Division of Chemistry and Biological Chemistry
School of Physical and Mathematical Sciences
Nanyang Technological University