Changeset 5052

Timestamp:

Jul 15, 2024, 3:30:17 PM (8 weeks ago)

Author:

abarral

Message:

Fix CPU binding on Adastra

Location:

BOL/LMDZ_Setup_amaury

Files:

• lmdz_env.sh (modified) (2 diffs)
• script_SIMU (modified) (2 diffs)
• slurm_set_cpu_binding.sh (added)

BOL/LMDZ_Setup_amaury/lmdz_env.sh

@@ -73 +73 @@
       module purge
       module load PrgEnv-gnu  # we need to load the env because lmdz links some shared libraries
-      module load gcc/12.1.0  # required, see https://dci.dci-gitlab.cines.fr/webextranet/user_support/index.html#prgenv-and-compilers
+      module load gcc/13.2.0  # required, see https://dci.dci-gitlab.cines.fr/webextranet/user_support/index.html#prgenv-and-compilers
 
       function cdo {  # cdo is available as a spack cmd which requires a specific, incompatible env
@@ -103 +103 @@
       NB_OMP_MAX=200
       NB_CORE_PER_NODE_MAX=192
-      N_HYPERTHREADING=2
+      N_HYPERTHREADING=1  # Adastra has SMT=2 enabled, but we found no actual performance improvement for the latlon model. Maybe useful for Dynamico ?
       MPICMD="srun -n"
 #      RUNBASHCMD="srun --label --account=$account --constraint=GENOA --ntasks-per-node=1 -n 1 --time=00:15:00 -c"
       RUNBASHCMD="bash"  # On Adastra the docs says we can use login nodes for compilation
-      SUBMITCMD="env $(env | grep -E "SLURM_|SBATCH_|SRUN_" | cut -d= -f1 | awk '{print "-u " $0}' | tr '\n' ' ' ) sbatch --constraint=GENOA --account=$account"  # we need to remove the existing SLURM variables otherwise they lay be unexpectedly inherited by the submitted script
+      SUBMITCMD="env $(env | grep -E "SLURM_|SBATCH_|SRUN_" | cut -d= -f1 | awk '{print "-u " $0}' | tr '\n' ' ' ) sbatch --constraint=GENOA --account=$account"  # we need to remove the existing SLURM variables otherwise they may be unexpectedly inherited by the submitted script
       ;;
     *) echo "WARNING: RUNNING THIS SCRIPT ON A LOCAL COMPUTER IS DISCOURAGED (lackluster cpus and memory)"

BOL/LMDZ_Setup_amaury/script_SIMU

@@ -84 +84 @@
 SCRIPTDIR=$SCRATCHD/$MAINDIR
 
-cp $STORED/$MAINDIR/lmdz_env.sh . ; . lmdz_env.sh
+cp "$STORED/$MAINDIR/lmdz_env.sh" .; . lmdz_env.sh
+cp "$STORED/$MAINDIR/slurm_set_cpu_binding.sh" .
+
 ERADIR=$STORED/$MAINDIR/GUIDE
 if [ "$aerosols" = "spla" ] ; then ERA10mDIR=$STORED/$MAINDIR/ERA10m ; fi
@@ -320 +322 @@
 echo '##################################################################'
 
+#@ADS if 1; then
 time $MPICMD $ntasks ./gcm.e > listing
+#@ADS else
+#@ADS srun --cpu-bind=none --mem-bind=none -- ./slurm_set_cpu_binding.sh ./gcm.e > listing
+#@ADS fi
+
 if [ ! -f restartphy.nc ] ; then
 echo PROBLEME PAS DE FICHIER RESTARTPHY