[gmx-users] using_gromacs_with_openmpi

[Mark Abraham]

Hi,

Asking "are you done yet?" and assembling the message while it is coming
off the network is work for a CPU... CPU utilization (e.g. as seen by htop
utility) is necessary for maximal productivity, but it isn't sensitive
about whether the work being done is useful.

Mark

On Wed, Apr 22, 2015 at 11:02 PM, Hossein H <haji309 at gmail.com> wrote:

> But I wonder if the the performance degrade due to network latency why
> mdrun process uses 100% CPU on each node!!!
>
> On Thu, Apr 23, 2015 at 1:05 AM, Justin Lemkul <jalemkul at vt.edu> wrote:
>
> >
> >
> > On 4/22/15 4:24 PM, Hossein H wrote:
> >
> >> the performance degrades even when I only use two nodes!!! and the mdrun
> >> process uses almost 100% of CPU on each node
> >>
> >>
> > Like I said, any benefit you might theoretically get from multiple CPU is
> > being undermined by latency in the gigabit ethernet connection.  It's
> > generally not adequate for high-performance MD simulations.
> >
> >
> > -Justin
> >
> >  On Thu, Apr 23, 2015 at 12:47 AM, Justin Lemkul <jalemkul at vt.edu>
> wrote:
> >>
> >>
> >>>
> >>> On 4/22/15 4:14 PM, Hossein H wrote:
> >>>
> >>>  isn't gigabyte Ethernet adequate even for 2 nodes?
> >>>>
> >>>>
> >>>>  Well, what does your benchmarking show you?  Likely the performance
> >>> degrades because the interconnect is simply too slow to benefit from
> >>> additional CPU horsepower.
> >>>
> >>> -Justin
> >>>
> >>>
> >>>   On Thu, Apr 23, 2015 at 12:36 AM, Justin Lemkul <jalemkul at vt.edu>
> >>> wrote:
> >>>
> >>>>
> >>>>
> >>>>
> >>>>> On 4/22/15 3:59 PM, Hossein H wrote:
> >>>>>
> >>>>>   Dear GROMACS users and developers
> >>>>>
> >>>>>>
> >>>>>> I've compiled GROMACS 4.5.5 with gcc and *enable-mpi* option for a
> >>>>>> small
> >>>>>> cluster e.g 4 nodes on 1GbE network
> >>>>>> The code was compiled without problem and i can use it in sequential
> >>>>>> and
> >>>>>> parallel mode (using mpirun command), but the performance over the
> >>>>>> network
> >>>>>> is very poor. indeed i can run the jobs on one nodes faster than 4
> >>>>>> nodes.
> >>>>>>
> >>>>>> because each nodes has 4 cores, i run jobs over network using
> >>>>>> following
> >>>>>> command:
> >>>>>>
> >>>>>>    mpirun -np 16 --host node01,node02,node03,node04 mdrun -deffnm
> >>>>>> input
> >>>>>> -v
> >>>>>>
> >>>>>>
> >>>>>>>
> >>>>>>>
> >>>>>> Someone has any idea why the performance is so poor?
> >>>>>>
> >>>>>>
> >>>>>>   Because you're using gigabit ethernet as your connection.  That's
> >>>>>> not
> >>>>>>
> >>>>> adequate for parallelization across machines.
> >>>>>
> >>>>> -Justin
> >>>>>
> >>>>> --
> >>>>> ==================================================
> >>>>>
> >>>>> Justin A. Lemkul, Ph.D.
> >>>>> Ruth L. Kirschstein NRSA Postdoctoral Fellow
> >>>>>
> >>>>> Department of Pharmaceutical Sciences
> >>>>> School of Pharmacy
> >>>>> Health Sciences Facility II, Room 629
> >>>>> University of Maryland, Baltimore
> >>>>> 20 Penn St.
> >>>>> Baltimore, MD 21201
> >>>>>
> >>>>> jalemkul at outerbanks.umaryland.edu | (410) 706-7441
> >>>>> http://mackerell.umaryland.edu/~jalemkul
> >>>>>
> >>>>> ==================================================
> >>>>> --
> >>>>> 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.
> >>>>>
> >>>>>
> >>>>>  --
> >>> ==================================================
> >>>
> >>> Justin A. Lemkul, Ph.D.
> >>> Ruth L. Kirschstein NRSA Postdoctoral Fellow
> >>>
> >>> Department of Pharmaceutical Sciences
> >>> School of Pharmacy
> >>> Health Sciences Facility II, Room 629
> >>> University of Maryland, Baltimore
> >>> 20 Penn St.
> >>> Baltimore, MD 21201
> >>>
> >>> jalemkul at outerbanks.umaryland.edu | (410) 706-7441
> >>> http://mackerell.umaryland.edu/~jalemkul
> >>>
> >>> ==================================================
> >>> --
> >>> 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.
> >>>
> >>>
> > --
> > ==================================================
> >
> > Justin A. Lemkul, Ph.D.
> > Ruth L. Kirschstein NRSA Postdoctoral Fellow
> >
> > Department of Pharmaceutical Sciences
> > School of Pharmacy
> > Health Sciences Facility II, Room 629
> > University of Maryland, Baltimore
> > 20 Penn St.
> > Baltimore, MD 21201
> >
> > jalemkul at outerbanks.umaryland.edu | (410) 706-7441
> > http://mackerell.umaryland.edu/~jalemkul
> >
> > ==================================================
> > --
> > 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.
>