[Beowulf] Re: Contents of Beowulf digest Vol 17, Issue 12, hybrid (openmp+mpi) job submit (solution)
Many of your questions may have already been answered in earlier discussions or in the FAQ. The search results page will indicate current discussions as well as past list serves, articles, and papers.
Ufuk Utku Turuncoglu turuncu at be.itu.edu.trThu Jul 14 01:29:01 PDT 2005
- Previous message: [Beowulf] Open position in Clustercomputing/MPI R&D (NEC Europe)
- Next message: [Beowulf] [gorelsky@stanford.edu: CCL:dual-core Opteron275performance]
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
hi,
i found the solition of my problem at the end. The problem is related
with default shell (/bin/ksh, in NIS).
I change my default shell to tcsh and i edit the .tcshrc file to define
the OMP_NUM_THREADS as 2.
that is all,
thanks for your help,
ufuk
my previous mail in list,
i try to run a job that is parallelized using openmp and mpi programming
interfaces (hybrid). I need to run mpi jobs in each node as an openmp job.
for this reason, i have to define OMP_NUM_THREADS environment variable for
each one of the node. first i try to put it into .profile file but it is
not sucessful. also i try to write an LSF job script and i fail too. The
LSF script as fallows,
#!/bin/ksh
#BSUB -J MM5_RUN # job name
#BSUB -n 2 # sum of number of tasks
#BSUB -R "span[ptile=1]" # number of processes per node
#BSUB -m "cn07 cn08" # run host
#BSUB -o mm5lsf.out # output file name
#BSUB -q cigq # queue name
#BSUB -L /bin/bash #
#BSUB -E "export OMP_NUM_THREADS=2"
. ${PWD}/mm5.deck.par
time mpirun -np 2 -machinefile ../machfile ./mm5.mpp
in this case. job run in each of the specified node as a single processor
mode (except execution host, because it is same machine which is login in
and OMP_NUM_THREADS environment variable comes from .profile file).
how can i run a command (or script) in each node just before runing mpi
executable?
thanks,
Ufuk Utku Turuncoglu
Istanbul Technical University
Informatics Institute
- Previous message: [Beowulf] Open position in Clustercomputing/MPI R&D (NEC Europe)
- Next message: [Beowulf] [gorelsky@stanford.edu: CCL:dual-core Opteron275performance]
- Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]
More information about the Beowulf mailing list
