[mpich-discuss] Problems with mpiexec.hydra in mpich2-1.2

Bryan Putnam bfp at purdue.edu
Mon Oct 19 10:20:40 CDT 2009


On Mon, 19 Oct 2009, Bryan Putnam wrote:

> On Mon, 19 Oct 2009, Pavan Balaji wrote:
> 
> > Bryan,
> > 
> > > I've recently installed mpich2-1.2 on three of our Linux clusters, and 
> > > have found that mpiexec.hydra is no longer working on any of them. One is 
> > > a Chelsio cluster, the other two are Infiniband, and all three are running 
> > > RHEL5.
> > > 
> > > The first few lines of the errors I'm seeing are appended below.
> > > 
> > > Note that if I use the same executable built with mpich2-1.2, and then use 
> > > the previous mpich2-1.1.1p1 version of mpiexec.hydra. it still works fine.
> > 
> > I believe you are running into this issue:
> > http://trac.mcs.anl.gov/projects/mpich2/ticket/893
> > 
> > This has been fixed, but the fix didn't make it into mpich2-1.2; it'll
> > be in 1.2.1. In the meanwhile, can you try using PBS as:
> > 
> > % mpiexec.hydra -rmk pbs ./hellof
> > 
> > .. and let us know if that works OK?
> > 
> > Thanks,
> > 
> >  -- Pavan
> 
> Pavan,
> 
> Using
> 
> mpiexec.hydra -rmk pbs ./hellof
> mpiexec.hydra -f $PBS_NODEFILE -rmk pbs -np 4 ./hellof
> 
> did not have any effect. If the fix is in mpich2-1.2.1 that sounds good. 
> We can still use the OSC mpiexec, or the mpdboot method, or the 
> mpiexec.hydra from mpich2-1.1.1p1.

mpiexec.hydra -rmk ./hellof

does appear to work correctly if I do something like

qsub -I -l select=2:ncpus=1:mpiprocs=1,place=scatter
mpiexec.hydra -rmk ./hellof

In that case, since there is only 1 processor per node, each node name 
only occurs once in the PBS_NODEFILE.

Thanks,
Bryan

> 
> Thanks,
> Bryan
> 
> > 
> > -- 
> > Pavan Balaji
> > http://www.mcs.anl.gov/~balaji
> > _______________________________________________
> > mpich-discuss mailing list
> > mpich-discuss at mcs.anl.gov
> > https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
> > 
> 
> 
> _______________________________________________
> mpich-discuss mailing list
> mpich-discuss at mcs.anl.gov
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss
> 




More information about the mpich-discuss mailing list