[mpich-discuss] Hydra process manager on Condor

Pavan Balaji balaji at mcs.anl.gov
Tue Dec 27 00:01:27 CST 2011


Hydra does not understand Condor's parameters.  But you can emulate the 
behavior you want by setting these options for mpiexec:

  -outfile-pattern ho.out -errfile-pattern he.err

You can also do more fancy things like:

  -outfile-pattern ho.%r.out -errfile-pattern he.%r.err

which uses different files for each rank.

See mpiexec -outfile-pattern -help for more information on other patterns.

  -- Pavan

On 12/19/2011 12:27 PM, Shrivastava, Savita wrote:
> Hi,
>
> We have installed MPICH2 with Hydra process manager on our Condor
> cluster. When I submit an mpi job to condor, the job is transferred to
> the execute node and executed properly (as I looked in the condor log on
> execute node where the job was executing) but it does not write the
> standard output from script to condor output file mentioned in job
> description file. Please guide me here how to run the mpi job
> successfully on condor using hydra process manager.
>
> My job description as below. The perl script has standard output "Hello".
>
> universe = parallel
>
> executable = /usr/lib64/mpich2/bin/mpiexec
>
> arguments = -n 2 -machinefile machinefile test2.pl
>
> getenv=true
>
> machine_count = 1
>
> should_transfer_files = yes
>
> when_to_transfer_output = on_exit
>
> transfer_input_files = test2.pl
>
> output = ho.out
>
> error = he.err
>
> log = hl.log
>
> Requirements = Memory >= 1024 && Cpus >=2
>
> request_cpus = 2
>
> request_memory = 1024
>
> queue
>
> Thanks
>
> Savita
>
>
>
> _______________________________________________
> mpich-discuss mailing list     mpich-discuss at mcs.anl.gov
> To manage subscription options or unsubscribe:
> https://lists.mcs.anl.gov/mailman/listinfo/mpich-discuss

-- 
Pavan Balaji
http://www.mcs.anl.gov/~balaji


More information about the mpich-discuss mailing list