Pretty simple, really:

#!/bin/sh
mpiexec -np 64 binary_executable < input_filename > output_filename 

is all you need. We'll pickup the allocation automatically and do all that is required


On Sep 13, 2012, at 12:37 AM, bharati_singh@jncasr.ac.in wrote:

Hello Openmpi,

I have Torque integrated with Moab cluster.We use msub command with intel mpi . I submit parallel job by using below two scripts.

$ cat submit1.sh
msub -V -N  name_of_job   -q  normal64c  -l nodes=8:ppn=8  -d  directory_path   -e  directory_path/err   ./submit2

$ cat submit2.sh
#!/bin/sh
cat $PBS_NODEFILE |uniq > temp.1
mpdboot -n `cat temp.1|wc -l` -f temp.1 -r ssh -v
mpiexec -genv  I_MPI_DEVICE  rdma -envall  -np  64    binary_executable  < input_filename > output_filename
mpdallexit

We have infiniband interconnection. I have compiled openmpi-1.6 with below options

$ ./configure --prefix=/home/bharati/software/openmpi CC=icc CXX=icpc F77=mpif90 --with-tm=/usr/src/torque-2.4.8/src/ --with-openib=/usr

May I know how can I do the same with openmpi?

Thanks & Regards
Bharati Singh


_______________________________________________
users mailing list
users@open-mpi.org
http://www.open-mpi.org/mailman/listinfo.cgi/users