Changeset 27919

Timestamp:

09/20/23 23:55:12 (18 months ago)

Author:

vverjans

Message:

CHG: updating pace cluster script

File:

• issm/trunk-jpl/src/m/classes/clusters/pace.m (modified) (4 diffs)

issm/trunk-jpl/src/m/classes/clusters/pace.m

@@ -9 +9 @@
         properties (SetAccess=public)
         % {{{
-                name            = 'login-phoenix-4.pace.gatech.edu' %Phoenix cluster name
+                name            = 'login-phoenix-slurm.pace.gatech.edu' %Phoenix cluster name
                 login           = ''; %personal login
                 numnodes        = 1; %number of nodes requested
@@ -17 +17 @@
                 queue           = 'inferno'; %queue
                 time            = 60; %time requested per run [minutes]
-                accountname     = 'GT-arobel3-atlas'; %group account name
+                accountname     = 'gts-arobel3-atlas'; %group account name
                 codepath        = ''; %path to issm binaries
                 executionpath   = ''; %path for execution folder
@@ -62 +62 @@
                         fid=fopen([modelname '.queue'],'w');
                         fprintf(fid,'#!/bin/sh\n');
-                        fprintf(fid,'#PBS -l walltime=%i\n',cluster.time*60); %conversion of walltime from minutes to seconds.
-                        fprintf(fid,'#PBS -N %s\n',modelname);
-                        fprintf(fid,'#PBS -l nodes=1:ppn=%i\n',cluster.np);
-                        fprintf(fid,'#PBS -l pmem=%igb\n',cluster.mem);
-                        fprintf(fid,'#PBS -q %s\n',cluster.queue);
-                        fprintf(fid,'#PBS -A %s\n',cluster.accountname);
-                
-                        fprintf(fid,'#PBS -o %s/%s/%s.outlog \n',cluster.executionpath,dirname,modelname);
-         fprintf(fid,'#PBS -e %s/%s/%s.errlog \n\n',cluster.executionpath,dirname,modelname);
 
-                        fprintf(fid,'export PBS_O_WORKDIR=%s\n',[cluster.executionpath '/' dirname]);
-                        fprintf(fid,'cd $PBS_O_WORKDIR\n');
-                        fprintf(fid,'mpiexec -np %i %s/%s %s %s %s \n',cluster.np,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname);  
+                        fprintf(fid,'#SBATCH -t%i\n',cluster.time);
+         fprintf(fid,'#SBATCH -J%s\n',modelname);
+         fprintf(fid,'#SBATCH -N 1 --ntasks-per-node=%i\n',cluster.np);
+         %fprintf(fid,'#SBATCH -N %i\n',cluster.numnodes);
+         %fprintf(fid,'#SBATCH --ntasks=1\n');
+         %fprintf(fid,'#SBATCH --cpus-per-task=%i\n',cluster.np);
+         fprintf(fid,'#SBATCH --mem-per-cpu=%iG\n',cluster.mem);
+         fprintf(fid,'#SBATCH -p%s\n',cluster.queue);
+         fprintf(fid,'#SBATCH -A %s\n',cluster.accountname);
+         fprintf(fid,'#SBATCH -o%s/%s/%s.outlog \n',cluster.executionpath,dirname,modelname);
+         fprintf(fid,'#SBATCH -e%s/%s/%s.errlog \n\n',cluster.executionpath,dirname,modelname);
+         fprintf(fid,'export SLURM_SUBMIT_DIR=%s\n',[cluster.executionpath '/' dirname]);
+         fprintf(fid,'cd $SLURM_SUBMIT_DIR\n');
+         fprintf(fid,'export LD_LIBRARY_PATH=/opt/slurm/current/lib:/opt/pmix/current/lib:$LD_LIBRARY_PATH \n');
+         fprintf(fid,'srun --mpi=pmi2 -n %i %s/%s %s %s %s \n',cluster.np,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname);
 
                         fclose(fid);
@@ -97 +100 @@
                         disp('launching solution sequence on remote cluster');
                         if ~isempty(restart)
-                                launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && qsub ' modelname '.queue '];
+                                launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && sbatch ' modelname '.queue '];
                         else
                                 launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
-                                        ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz  && qsub ' modelname '.queue '];
+                                        ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz  && sbatch ' modelname '.queue '];
                         end
                         issmssh(cluster.name,cluster.login,cluster.port,launchcommand);