Changeset 1268

Timestamp:

07/08/09 20:06:39 (16 years ago)

Author:

Eric.Larour

Message:

Better handling of queuing on gemini altix clusters using PBS

Location:

issm/trunk/src/m/classes/public

Files:

• BuildQueueingScriptgemini.m (modified) (2 diffs)
• LaunchQueueJob.m (modified) (2 diffs)
• LaunchQueueJobGeneric.m (modified) (1 diff)
• LaunchQueueJobcosmos.m (modified) (1 diff)
• LaunchQueueJobgemini.m (modified) (2 diffs)
• process_solve_options.m (modified) (2 diffs)
• solveparallel.m (modified) (1 diff)

issm/trunk/src/m/classes/public/BuildQueueingScriptgemini.m

@@ -12 +12 @@
 end
 
-fprintf(fid,'#!/bin/csh\n');
-fprintf(fid,'#PBS -l walltime=%i\n',md.time);
+fprintf(fid,'#!/bin/sh\n');
+fprintf(fid,'#PBS -l walltime=%i\n',md.time*60); %walltime is in seconds.
 fprintf(fid,'#PBS -l ncpus=%i\n',md.np);
 if ~isempty(md.queue),
@@ -21 +21 @@
 fprintf(fid,'#PBS -e %s.errlog \n',md.name);
 
-fprintf(fid,'setenv PBS_O_WORKDIR %s\n',executionpath);
+fprintf(fid,'export PBS_O_WORKDIR=%s\n',executionpath);
 fprintf(fid,'cd $PBS_O_WORKDIR\n');
-fprintf(fid,'setenv OMP_NUM_THREADS 1\n');
+fprintf(fid,'export OMP_NUM_THREADS=1\n');
 fprintf(fid,'mpirun -np %i %s/%s.exe %s %s.bin %s.outbin %s.lock',md.np,codepath,md.analysis_type,executionpath,md.name,md.name,md.name);
 

issm/trunk/src/m/classes/public/LaunchQueueJob.m

@@ -1 @@
-function LaunchQueueJob(md,executionpath)
+function md=LaunchQueueJob(md,executionpath,options)
 %LAUNCHQUEUEJOB - ...
 %
@@ -17 +17 @@
 if exist(function_name,'file'),
         %Call this function:
-        eval([function_name '(md,executionpath);']);
+        eval(['md=' function_name '(md,executionpath,options);']);
 else
         %Call the generic LaunchQueueJob:
-        LaunchQueueJobGeneric(md,executionpath);
+        md=LaunchQueueJobGeneric(md,executionpath,options);
 end

issm/trunk/src/m/classes/public/LaunchQueueJobGeneric.m

@@ -1 @@
-function LaunchQueueJobGeneric(md,executionpath)
+function md=LaunchQueueJobGeneric(md,executionpath,options)
 %LAUNCHQUEUEJOBGENERIC - ...
 %

issm/trunk/src/m/classes/public/LaunchQueueJobcosmos.m

@@ -1 @@
-function LaunchQueueJobcosmos(md,executionpath)
+function md=LaunchQueueJobcosmos(md,executionpath,options)
 %LAUNCHQUEUEJOBCOSMOS - ...
 %

issm/trunk/src/m/classes/public/LaunchQueueJobgemini.m

@@ -1 @@
-function LaunchQueueJobgemini(md,executionpath)
+function md=LaunchQueueJobgemini(md,executionpath,options)
 %LAUNCHQUEUEJOBGEMINI - ...
 %
@@ -15 +15 @@
 
 %jpload both files to cluster
-disp('uploading input file,  queueing script and variables script');
-if md.qmu_analysis, 
-        eval(['!scp ' md.name '.bin' ' ' md.name '.queue '  md.name '.qmu.in ' md.cluster ':' executionpath]);
+if ~strcmpi(options.batch,'yes'),
+        disp('uploading input file,  queueing script and variables script');
+        if md.qmu_analysis, 
+                eval(['!scp ' md.name '.bin' ' ' md.name '.queue '  md.name '.qmu.in ' md.cluster ':' executionpath]);
+        else
+                eval(['!scp ' md.name '.bin' ' ' md.name '.queue '  md.cluster ':' executionpath]);
+        end
 else
-        eval(['!scp ' md.name '.bin' ' ' md.name '.queue '  md.cluster ':' executionpath]);
+        disp('batch mode requested: not launching job interactively');
 end
 
 %new gemini cannot launch across cluster using ssh
 disp(['launch solution sequence on remote cluster by logging into it and typing qsub < ' md.name '.queue']);
-return;
-
-disp('launching solution sequence on remote cluster');
-
-%now call the queuing script to launch the job.
-system(['ssh  ' md.cluster ' ''cd ' executionpath ' && qsub <  ' md.name '.queue ''']);
-
+md.waitonlock=0;

issm/trunk/src/m/classes/public/process_solve_options.m

@@ -53 +53 @@
         end
 end
+
+%batch mode for launching jobs.
+found=0;
+for i=1:size(options,1),
+        if strcmpi(options{i,1},'batch'),
+                batch=options{i,2};
+                found=1;
+        end
+end
+if ~found
+        batch='no';
+end
+
 
 %check solution type is supported
@@ -131 +144 @@
 outoptions.iparams=iparams;
 outoptions.runmpi=runmpi;
+outoptions.batch=batch;

issm/trunk/src/m/classes/public/solveparallel.m

@@ -23 +23 @@
 
 %Now, launch the queueing script
-LaunchQueueJob(md,executionpath);
+md=LaunchQueueJob(md,executionpath,options);
 
 %Do we return, or just wait for results?