mpirun mpirexec 的PBS script 是不同的
A Simple MPI Program Execution
Systems & Access :: Learning to Use :: Linux Computing Resource
Example of an MPI Program mpitest.c
/* program hello */
/* Adapted from mpihello.f by drs */
#include <mpi.h>
#include <stdio.h>
int main(int argc, char **argv)
{
int *buf, i, rank, nints, len;
char hostname[256];
MPI_Init(&argc,&argv);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
gethostname(hostname,255);
printf("Hello world! I am process number: %d on host %s\n", rank, hostname);
MPI_Finalize();
return 0;
}
Example of MPI Program Execution Using mpiexec
The preferred means of starting MPI programs on the Linux cluster is mpiexec. The main advantage of using mpiexec over mpirun is that there is no need for sourcing any setup file during the program execution. However, you still must source the setup file before compiling your program. It is a lot easier to write the PBS script file using mpiexec than mpirun. The example MPI program above could be executed with the following PBS script that uses mpiexec.
Sample mpitest.pbs Using mpiexec
#!/bin/csh #*** The "#PBS" lines must come before any non-blank non-comment lines *** #PBS -l walltime=30:00,nodes=4:myri:ppn=2 set mycommand = "./mpitest" # note "./" is used to not rely on $PATH set myargs = "" # arguments for $mycommand set infiles = "" # list of input files set outfiles = "" # list of output files if ($?PBS_NODEFILE && $?PBS_JOBID) then #count the number of processors assigned by PBS set NP = `wc -l < $PBS_NODEFILE` echo "Running on $NP processors: "`cat $PBS_NODEFILE` else echo "This script must be submitted to PBS with 'qsub -l nodes=X'" exit 1 endif # set up the job and input files, exiting if anything went wrong cd "$PBS_O_WORKDIR" || exit 1 if ($?infiles) then foreach file ($infiles) cp -v $file /scratch/ || exit 1 end endif cd /scratch || exit 1 # run the command, saving the exit value mpiexec $PBS_O_WORKDIR/$mycommand $myargs set ret = $? # get the output files cd "$PBS_O_WORKDIR" if ($?outfiles) then foreach file ($outfiles) cp -vf /scratch/`basename $file` $file end endif echo "Done " `date` exit $ret
Sample Run
To run a job, the first step is to source the setup file:
hpc-master: source /usr/usc/mpich/1.2.5..12/gm-cdk/setup.csh
The second step is to compile the program:
hpc-master: mpicc -o mpitest mpitest.c
The third step is to submit the .pbs file to the queue. When the job submission is successful, the qsub command returns the job's process number:
hpc-master: qsub mpitest.pbs 12240.hpc-pbs.usc.edu
The output files for the above run are created in the same directory where the qsub command was executed. The output files are created using the same filename as your PBS script filename with an extension that includes "e" for "error" or "o" for "output" followed by the process number. In our example above the file mpitest.pbs.e12240 contains the error messages and the file mpitest.pbs.o12240 contains the output of the program. The output of the program can be viewed by listing the output files:
hpc-master: ls -l mpitest.pbs.* -rw------- 1 nirmal rds 0 Jul 16 14:59 mpitest.pbs.e12240 -rw------- 1 nirmal rds 1442 Jul 16 14:59 mpitest.pbs.o12240
hpc-master: more mpitest.pbs.* :::::::::::::: mpitest.pbs.e12240 :::::::::::::: :::::::::::::: mpitest.pbs.o12240 :::::::::::::: ---------------------------------------- Begin PBS Prologue Wed Dec 8 16:49:24 PST 2004 Job ID: 12240.hpc-pbs.usc.edu Username: nirmal Group: rds Nodes: hpc0632 hpc0633 hpc0634 hpc0636 PVFS: mounting at /scratch End PBS Prologue Wed Dec 8 16:49:27 PST 2004 ---------------------------------------- Warning: no access to tty (Bad file descriptor). Thus no job control in this shell. Starting 12240.hpc-pbs.usc.edu Wed Dec 8 16:49:27 PST 2004 Initiated on hpc0636 Running on 8 processors: hpc0636 hpc0636 hpc0634 hpc0634 hpc0633 hpc0633 hpc0632 hpc0632 Hello world! I am process number: 4 on host hpc0632 Hello world! I am process number: 3 on host hpc0634 Hello world! I am process number: 0 on host hpc0635 Hello world! I am process number: 2 on host hpc0634 Hello world! I am process number: 6 on host hpc0631 Hello world! I am process number: 1 on host hpc0635 Hello world! I am process number: 5 on host hpc0632 Hello world! I am process number: 7 on host hpc0631 Done Wed Dec 8 16:49:28 PST 2004 -------------------------------------------------- Begin PBS Epilogue Wed Dec 8 16:49:33 PST 2004 Job ID: 12240.hpc-pbs.usc.edu Username: nirmal Group: rds Job Name: mpitest.pbs Session: 4020 Limits: neednodes=4:myri:ppn=2,nodes=4:myri:ppn=2,walltime=00:02:00 Resources: cput=00:00:00,mem=12648kb,vmem=25752kb,walltime=00:00:04 Queue: quick Account: Nodes: hpc0632 hpc0633 hpc0634 hpc0636 Killing leftovers... End PBS Epilogue Wed Dec 8 16:49:35 PST 2004 --------------------------------------------------