[mvapich-discuss] MVAPICH2 098 Question

[wei huang]

Hi David,

MVAPICH2-1.0 does not change any startup process as compared to
mpich2-1.0.5. So if mpiexec-0.82 supports mpich2, you should not face any
problem running mvapich2-1.0 with that.

Regarding the mpi_sendrecv problem, we ran up to 16 processes using
IMB-3.0 and up to 32MB messages. We do not see any problem. What is the
smallest job scale that you are able to see the hanging?

Thanks.

Regards,
Wei Huang

774 Dreese Lab, 2015 Neil Ave,
Dept. of Computer Science and Engineering
Ohio State University
OH 43210
Tel: (614)292-8501


On Wed, 17 Oct 2007, David Gunter wrote:

> The MVAPICH2 version is the 0.9.8p3 version.  The mpiexec being used
> is the OSC mpiexec-0.82 that doesn't support the latest MVAPICH2 1.0
> yet.
>
> There are other reasons we have not upgraded to MVAPICH2 1.0, namely
> that we are experiencing hangs with mpi_sendrecv with message sizes
> above 4MB, something I reported earlier.
>
> Thanks,
> david
>
> --
> David Gunter
> HPC-4: HPC Environments: Parallel Tools Team
> Los Alamos National Laboratory
>
>
> On Oct 16, 2007, at 5:24 PM, Dhabaleswar Panda wrote:
>
> > Hi,
> >
> >> We are running Mvapich2 .098. It works great up to ~2400
> >> Processors PEs.
> >> When we try to scale to 4192 PEs it hangs.
> >> Are there some variables or tuning parameters we need to tweak to get
> >> it to scale?
> >
> > Not sure which version of MVAPICH2 0.9.8 you are using. We had a p3
> > version available with several bug fixes for MVAPICH2 long time back.
> > Are you using this version or the original 0.9.8 released version?
> >
> > Since MVAPICH2 1.0 has been released for some time and it incorporates
> > all bug fixes for 0.9.8 and also new features, I will strongly
> > recommend you to use this version. Since MVAPICH2 1.0 release, a few
> > more bug fixes have taken place to the 1.0 branch. I will recommend
> > taking the version from the branch. We plan to make a release of
> > MVAPICH2 1.0.1 in the near future with all these fixes.
> >
> > Let us know if you see any scaling issues for 1.0 version (from the
> > branch) and we will investigate this issue further.
> >
> > Thanks,
> >
> > DK
> >
> >> I see in 1.0 there are some but not sure about 098.
> >>
> >> Please respond to me directly, I haven't subscribed to the mailing
> >> list yet.
> >>
> >> thanks in advance.
> >> parks
> >>
> >>
> >>                     ***** Correspondence *****
> >>
> >> This email contains no programmatic content that requires independent
> >> ADC review
> >>
> >>
> >> _______________________________________________
> >> mvapich-discuss mailing list
> >> mvapich-discuss at cse.ohio-state.edu
> >> http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
> >>
> >
> > _______________________________________________
> > mvapich-discuss mailing list
> > mvapich-discuss at cse.ohio-state.edu
> > http://mail.cse.ohio-state.edu/mailman/listinfo/mvapich-discuss
>
>