[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Condor-users] openMPI wrapper



I'm trying to get condor to work with openmpi as well. So far not
succesfull. It seems you don't need a special wrapper. The following
condor submission file will start and run an openmpi job:

---------

## without using a shared file system
######################################
universe = parallel
executable = hello
log = log/logfile
output = output/outfile.$(NODE)
error = output/errfile.$(NODE)
machine_count = 4
environment =
LD_LIBRARY_PATH=/usr/lib/openmpi/1.2.5-gcc/lib:/opt/intel_fc_80/li
b;PATH=/usr/lib/openmpi/1.2.5-gcc/bin
queue

---------

where the output and log dirs must exist. The only special thing I need
to do is to set the path for openmpi, because it's not the default
mpi-installation. The executable is simply the executable created with
mpif90 (make sure again that it's the openmpi one)
The problem I'm still facing now is that all nodes get rank 0 (zero). So
really it's impossible to really communicate. Another very annoying
feature is that the job is killed as soon as the first node finishes.

Maybe these things can be solved by some clever wrapper, but I wouldn't
know what to put in there (actually the problems I'm having with openmpi
_also_ happen with mpich2) I mentioned these problems on this mail list
before, but didn't get any response. It seems it is impossible to use
condor with MPI.

Yours sincerely,
Jakob


Majid Arabgol wrote:
> Hi All,
> 
> Anyone has experienced with opnempi and condor ?  I am looking for
> openmpi wrapper
> I appreciate any helps.
> 
> 
> Regards
> Majid
> _______________________________________________
> Condor-users mailing list
> To unsubscribe, send a message to condor-users-request@xxxxxxxxxxx with a
> subject: Unsubscribe
> You can also unsubscribe by visiting
> https://lists.cs.wisc.edu/mailman/listinfo/condor-users
> 
> The archives can be found at: 
> https://lists.cs.wisc.edu/archive/condor-users/