[gmx-developers] Running Gromacs on GPUs on multiple machines

[Mark Abraham]

The hash function is there, and is used for communicator splitting, but
does not (yet) implement gmx_hostname_num in release-5-0.

Mark


On Thu, May 29, 2014 at 7:02 PM, Szilárd Páll <pall.szilard at gmail.com>
wrote:

> On Thu, May 29, 2014 at 6:56 PM, Mark Abraham <mark.j.abraham at gmail.com>
> wrote:
> >
> > On May 29, 2014 4:53 PM, "Vedran Miletić" <rivanvx at gmail.com> wrote:
> >>
> >> 2014-05-29 16:36 GMT+02:00 Anders Gabrielsson <andgab at kth.se>:
> >> > You'll probably have to supply -npernode/-ppn too so that your mpirun
> >> > doesn't put all MPI ranks on the same host.
> >> >
> >> > /Anders
> >> >
> >> > On 29 May 2014, at 13:17, Vedran Miletić <rivanvx at gmail.com> wrote:
> >> >
> >> > Fatal error:
> >> > Incorrect launch configuration: mismatching number of PP MPI processes
> >> > and GPUs per node.
> >> > mdrun_mpi was started with 5 PP MPI processes per node, but you
> provided
> >> > 1
> >> > GPU.
> >>
> >> Hi Anders,
> >>
> >> thanks for your response. Weirdly enough, mpirun actually doesn't run
> >> processes all on one node, it distributes them as equally as possible,
> >> going around your hostfile in a round-robin fashion. (I verified this
> >> by running hostname.)
> >
> > What were the hostnames? There is some logic that tries to reconstruct
> the
> > nodes from hostnames as supplied by MPI, but GROMACS assumes the form
> >
> > yourmachinename043
> >
> > with the digits varying across nodes. So far, this has worked, since
> > sysadmins are orderly types... IIRC we use a hash in GROMACS 5, or at
> least
> > we talked about it!
>
> There is hostname hashing implementation (a gmx_hostname_num()) which
> I don't remember anymore what was it implemented for, but neither 4.6
> nor (the current) 5.0 uses this is for splitting the MPI_comm_world
> into groups of ranks per node.
>
> > Mark
> >
> >> However, it seems that for some reason Gromacs assumes mpirun does run
> >> 5 processes on a single node. Regardless, i tried
> >>
> >> mpirun -np 5 -npernode 1 -hostfile ... mdrun_mpi -v -deffnm ... -gpu_id
> 0
> >>
> >> but it produces the same error. Anything else I could try?
> >>
> >> Regards,
> >> Vedran
> >> --
> >> Gromacs Developers mailing list
> >>
> >> * Please search the archive at
> >> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_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-developers
> or
> >> send a mail to gmx-developers-request at gromacs.org.
> >
> >
> > --
> > Gromacs Developers mailing list
> >
> > * Please search the archive at
> > http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_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-developers
> or
> > send a mail to gmx-developers-request at gromacs.org.
> --
> Gromacs Developers mailing list
>
> * Please search the archive at
> http://www.gromacs.org/Support/Mailing_Lists/GMX-developers_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-developers
> or send a mail to gmx-developers-request at gromacs.org.
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20140529/378b8d90/attachment.html>