Differences

This shows you the differences between two versions of the page.

--- strutture:cnaf:cnaf_rd:infrastruttura:utilizzo_cluster:job_mpi [2013/10/31 08:07] – created caberletti@infn.it
+++ strutture:cnaf:cnaf_rd:infrastruttura:utilizzo_cluster:job_mpi [2013/11/22 14:44] (current) – caberletti@infn.it
@@ Line 1: / Line 1: @@
+====== Job MPI ======
+===== Intel MPI =====
+Programma di test:
+<code C>
+#include "mpi.h"
+#include <stdio.h>
+#include <string.h>
+int
+main (int argc, char *argv[])
+{
+    int i, rank, size, namelen;
+    char name[MPI_MAX_PROCESSOR_NAME];
+    MPI_Status stat;
+    MPI_Init (&argc, &argv);
+    MPI_Comm_size (MPI_COMM_WORLD, &size);
+    MPI_Comm_rank (MPI_COMM_WORLD, &rank);
+    MPI_Get_processor_name (name, &namelen);
+    if (rank == 0) {
+	printf ("Hello world: rank %d of %d running on %s\n", rank, size, name);
+	for (i = 1; i < size; i++) {
+	    MPI_Recv (&rank, 1, MPI_INT, i, 1, MPI_COMM_WORLD, &stat);
+	    MPI_Recv (&size, 1, MPI_INT, i, 1, MPI_COMM_WORLD, &stat);
+	    MPI_Recv (&namelen, 1, MPI_INT, i, 1, MPI_COMM_WORLD, &stat);
+	    MPI_Recv (name, namelen + 1, MPI_CHAR, i, 1, MPI_COMM_WORLD, &stat);
+	    printf ("Hello world: rank %d of %d running on %s\n", rank, size, name);
+	}
+    } else {
+	MPI_Send (&rank, 1, MPI_INT, 0, 1, MPI_COMM_WORLD);
+	MPI_Send (&size, 1, MPI_INT, 0, 1, MPI_COMM_WORLD);
+	MPI_Send (&namelen, 1, MPI_INT, 0, 1, MPI_COMM_WORLD);
+	MPI_Send (name, namelen + 1, MPI_CHAR, 0, 1, MPI_COMM_WORLD);
+    }
+    MPI_Finalize ();
+    return (0);
+}
+</code>
+Compilazione:
+<code>
+$ module load ips-xe-2013 intel-mpi
+$ mpiicc -o test test.c
+</code>
+==== LSF ====
+Esempio di job LSF che fa utilizzo di Intel MPI:
+<code bash>
+#BSUB -J impi		        # Job name
+#BSUB -o %J.out			# Job standard output
+#BSUB -e %J.err			# Job standard error
+#BSUB -N			# Job report
+#BSUB -B			# Send mail
+#BSUB -q normal			# Job queue
+#BSUB -a intelmpi               # Application
+#BSUB -m rd-coka-02             # Host selection
+#BSUB -n 10                     # Number of process
+#
+# commands
+module load ips-xe-2013 intel-mpi
+mpirun.lsf /home/GUEST/mcaberletti/test_intel_mpi/test
+</code>
+Salvare in un file, ad esempio ''impi.job'' e sottometterlo con:
+<code bash>
+$ bsub < impi.job
+</code>
+==== Slurm ====
+Con Slurm occorre tenere presente che il comando ''sbatch'' non fa lo spawn di processi,
+ma riserva solamente le risorse.
+Per questo occorre indicare il numero di processi
+Esempio di job:
+<code bash>
+#!/bin/bash
+#SBATCH -J test_impi                                    # Job name
+#SBATCH -p coka-test                                    # Queue
+#SBATCH -e %J.err                                       # Stderr
+#SBATCH -o %J.out                                       # Stdout
+#SBATCH -n 10                                           # Task number
+#SBATCH -w rd-coka-[01-02]                              # Host selection
+#SBATCH -D /home/GUEST/mcaberletti/test_intel_mpi       # Working dir
+#SBATCH --mail-type=ALL                                 # Send all notification via mail
+#SBATCH --mail-user=mcaberletti@rd-ui
+# commands
+module load ips-xe-2013 intel-mpi
+srun ./test
+</code>
+Salvare in un file, ad esempio ''impi.job'' e sottometterlo con:
+<code bash>
+$ sbatch impi.job
+</code>