[gmx-developers] Trying to understand performance issue - due to all threads pinning to first core?
Shirts, Michael (mrs5pt)
mrs5pt at eservices.virginia.edu
Wed Dec 19 20:16:50 CET 2012
-nopin indeed gives full utilization, with a 10% or so performance hit
(sample size = 1). We will investigate reading off the slot from the PBS
environment to set -pinoffset. Might be good to provide some examples of
this all on:
Department of Chemical Engineering
University of Virginia
michael.shirts at virginia.edu
> From: Roland Schulz <roland at utk.edu>
> Date: Wed, 19 Dec 2012 14:01:11 -0500
> To: "michael.shirts at virginia.edu" <michael.shirts at virginia.edu>
> Cc: Discussion list for GROMACS development <gmx-developers at gromacs.org>
> Subject: Re: [gmx-developers] Trying to understand performance issue - due to
> all threads pinning to first core?
> depending on the jobs scheduler you use, you can get get the slot number
> you got assigned for the shared node, so that your jobscript an set the
> correct pinoffset.
> On Wed, Dec 19, 2012 at 1:55 PM, Shirts, Michael (mrs5pt) <
> mrs5pt at eservices.virginia.edu> wrote:
>>> you can use -nopin but you will get slightly lower performance.
>> Ah, great! I'll try this. I didn't see this documented on the page --
>> perhaps it should be added -- if, of course, I didn't just miss it.
>>> You can
>>> also use pinoffset and number the different gromacs instances you are
>>> running any way you wish. Most programs don't pin so it is unlikely that
>>> going to conflict with other programs if you are running both Gromacs and
>>> other programs.
>> But that doesn't solve situations where I have four students plus myself
>> submitting Gromacs runs to the same cluster :)
>> Michael Shirts
>> Assistant Professor
>> Department of Chemical Engineering
>> University of Virginia
>> michael.shirts at virginia.edu
> ORNL/UT Center for Molecular Biophysics cmb.ornl.gov
> 865-241-1537, ORNL PO BOX 2008 MS6309
More information about the gromacs.org_gmx-developers