Changeset 20320

Timestamp:

03/16/16 19:09:34 (9 years ago)

Author:

schlegel

Message:

CHG: update discover class for SLURM

File:

• issm/trunk-jpl/src/m/classes/clusters/discover.m (modified) (8 diffs)

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

@@ -11 +11 @@
                  name=oshostname();
                  login='';
+                 modules        = {};
                  numnodes=20;
                  cpuspernode=8; 
                  port=0;
                  queue='general';
-                 time=12*60;
+                 time=12*60*60;
                  processor='west';
                  codepath='';
@@ -23 +24 @@
                  numstreams=8;
                  hyperthreading=0;
+                 email='';
          end
          %}}}
@@ -40 +42 @@
                          disp(sprintf('    name: %s',cluster.name));
                          disp(sprintf('    login: %s',cluster.login));
+                         disp(sprintf('    modules: %s',strjoin(cluster.modules,', ')));
                          disp(sprintf('    port: %i',cluster.port));
                          disp(sprintf('    numnodes: %i',cluster.numnodes));
@@ -51 +54 @@
                          disp(sprintf('    interactive: %i',cluster.interactive));
                          disp(sprintf('    hyperthreading: %i',cluster.hyperthreading));
+                         disp(sprintf('    email: %s',cluster.email));
                  end
                  %}}}
@@ -60 +64 @@
                  function md = checkconsistency(cluster,md,solution,analyses) % {{{
 
-                         available_queues={'general_long','general','debug'};
-                         queue_requirements_time=[24*60 12*60 60];
-                         queue_requirements_np=[516 1024 32];
+                         available_queues={'long','general','debug'};
+                         queue_requirements_time=[24*60*60 12*60*60 60];
+                         queue_requirements_np=[4116 532 532];
 
                          QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,cluster.time)
@@ -71 +75 @@
                                          md = checkmessage(md,'cpuspernode should be between 1 and 16 for ''sand'' processors');
                                  end
-                         elseif strcmpi(cluster.processor,'west'),
-                                 if ((cluster.cpuspernode>12 ) | (cluster.cpuspernode<1)),
-                                         md = checkmessage(md,'cpuspernode should be between 1 and 12 for ''west'' processors');
-                                 end
-                         else
-                                 md = checkmessage(md,'unknown processor type, should be ''sand'' or ''west'' ');
+                         elseif strcmpi(cluster.processor,'hasw'),
+                                 if ((cluster.cpuspernode>28 ) | (cluster.cpuspernode<1)),
+                                         md = checkmessage(md,'cpuspernode should be between 1 and 28 for ''hasw'' processors');
+                                 end
+                         else
+                                 md = checkmessage(md,'unknown processor type, should be ''sand'' or ''hasw'' ');
                          end
 
@@ -90 +94 @@
                          if(isgprof),    disp('gprof not supported by cluster, ignoring...'); end
 
-                         %write queuing script 
+                         %write queuing script
                          fid=fopen([modelname '.queue'],'w');
+
                          fprintf(fid,'#!/bin/bash\n');
-%                        fprintf(fid,'#PBS -N %s\n',modelname);
-                         fprintf(fid,'#PBS -l select=%i:mpiprocs=%i:proc=%s\n',cluster.numnodes,cluster.cpuspernode,cluster.processor);
-                         fprintf(fid,'#PBS -l walltime=%i:00\n',cluster.time); %SLURM walltime is in minutes:
-                         if ~strcmp(cluster.queue,'general'),
-                                 fprintf(fid,'#PBS -q %s \n',cluster.queue);
-                         end
-                         fprintf(fid,'#PBS -W group_list=s1010\n');
-                         fprintf(fid,'#PBS -m e\n');
-                         fprintf(fid,'#PBS -o %s.outlog \n',[cluster.executionpath '/' dirname '/' modelname]);
-                         fprintf(fid,'#PBS -e %s.errlog \n\n',[cluster.executionpath '/' dirname '/' modelname]);
+                         fprintf(fid,'#SBATCH -J %s \n',modelname);
+                         fprintf(fid,'#SBATCH --qos=%s \n',cluster.queue);
+                         fprintf(fid,'#SBATCH -o %s.outlog \n',modelname);
+                         fprintf(fid,'#SBATCH -e %s.errlog \n',modelname);
+                         fprintf(fid,'#SBATCH -n %i \n',cluster.numnodes*cluster.cpuspernode);
+                         fprintf(fid,'#SBATCH -N %i \n',cluster.numnodes);
+                         fprintf(fid,'#SBATCH -t %02i:%02i:00 \n',floor(cluster.time/3600),floor(mod(cluster.time,3600)/60));
+                         fprintf(fid,'#SBATCH -A s1010 \n\n');
+                         for i=1:numel(cluster.modules),
+                                 fprintf(fid,['module load ' cluster.modules{i} '\n']);
+                         end
+                         if length(find(cluster.email=='@'))>0
+                                 fprintf(fid,'#SBATCH --mail-user=%s \n',cluster.email);
+                                 fprintf(fid,'#SBATCH --mail-type=end \n\n');
+                         end
                          fprintf(fid,'. /usr/share/modules/init/bash\n\n');
-                         fprintf(fid,'module purge\n');
-                         fprintf(fid,'module load comp/intel-11.1.038\n');
+                         fprintf(fid,'module load comp/intel-15.0.0.090\n');
                          fprintf(fid,'module load mpi/impi-4.0.3.008\n');
-                         fprintf(fid,'module load lib/mkl-10.1.2.024\n\n');
                          fprintf(fid,'export PATH="$PATH:."\n\n');
-                         fprintf(fid,'export MPI_GROUP_MAX=64\n\n');
-                         fprintf(fid,'cd %s/%s/\n\n',cluster.executionpath,dirname);
+                         fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME
+                         fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n');       %FIXME
+                         fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname);
+
                          fprintf(fid,'mpirun -np %i %s/issm.exe %s %s %s\n',cluster.np,cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname],modelname);
                          if ~io_gather, %concatenate the output files:
@@ -162 +172 @@
                  end
                  %}}}
-                 function LaunchQueueJob(cluster,modelname,dirname,filelist,restart)% {{{
+                 function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{
 
                          if ~cluster.interactive,