Changeset 5546

Timestamp:

08/24/10 14:08:32 (15 years ago)

Author:

Eric.Larour

Message:

New runtime name, which is model name + PID + date. Allows multiple runs on same model

Location:

issm/trunk/src/m/classes

Files:

• @model/model.m (modified) (1 diff)
• public/loadresultsfromcluster.m (modified) (2 diffs)
• public/queue/ClusterLaunchCommand.m (modified) (2 diffs)
• public/queue/ClusterScript.m (modified) (3 diffs)
• public/queue/LaunchQueueJob.m (modified) (1 diff)
• public/solveparallel.m (modified) (2 diffs)
• public/waitonlock.m (modified) (2 diffs)

issm/trunk/src/m/classes/@model/model.m

@@ -12 +12 @@
         %Model general information
         md.notes='';
-        md.name='';
+        md.name=''; 
+        md.runtimename=''; %name used when running a parallel job
         md.counter=0; %1 mesh, 2 geography, 3 parameterize
         md.domainoutline=NaN;

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

@@ -10 +10 @@
 %Figure out parameters for this particular cluster
 [codepath,executionpath,login,port]=ClusterParameters(md.cluster,cluster_rc_location);
+
+%Figure out the  directory where all the files are in: 
+directory=[executionpath '/' md.runtimename '/'];
 
 %What packages are we picking up from remote cluster
@@ -26 +29 @@
 
 %copy files from cluster to present directory
-issmscpin(md.cluster, login, port, executionpath, packages);
+issmscpin(md.cluster, login, port, directory, packages);
 
 %read log files onto  fields

issm/trunk/src/m/classes/public/queue/ClusterLaunchCommand.m

@@ -1 @@
-function command=ClusterLaunchCommand(cluster,name,executionpath)
+function command=ClusterLaunchCommand(cluster,runtimename,name,executionpath)
 %CLUSTERLAUNCHCOMMAND: build command to launch job on cluster
 %
@@ -7 +7 @@
 %if cluster is current hostname, just source queue file
 if strcmpi(oshostname(),cluster),
-        command=['cd ' executionpath ' && source ' name '.queue '];
+        command=['cd ' executionpath ' && rm -rf ./' runtimename ' && mkdir ' runtimename ' && cd ' runtimename ' && mv ../' runtimename '.tar.gz ./ && tar -zxvf ' runtimename '.tar.gz  && source  ' name '.queue '];
 else
         if strcmpi(cluster,'cosmos'),
-                command=['cd ' executionpath ' && rm -rf ' name '.lock ' name '.outlog ' name '.errlog ' name '.outbin && qsub -S/bin/sh ' name '.queue '];
+                command=['cd ' executionpath ' && rm -rf ./' runtimename ' && mkdir ' runtimename ' && cd ' runtimename ' && mv ../' runtimename '.tar.gz ./ && tar -zxvf ' runtimename '.tar.gz  && qsub -S/bin/sh ' name '.queue '];
         elseif strcmpi(cluster,'gemini'),
-                command=['cd ' executionpath ' && rm -rf ' name '.lock ' name '.outlog ' name '.errlog ' name '.outbin && qsub ' name '.queue '];
+                command=['cd ' executionpath ' && rm -rf ./' runtimename ' && mkdir ' runtimename ' && cd ' runtimename ' && mv ../' runtimename '.tar.gz ./ && tar -zxvf ' runtimename '.tar.gz  && qsub ' name '.queue '];
         elseif strcmpi(cluster,'pollux'),
-                command=['cd ' executionpath ' && rm -rf ' name '.lock ' name '.outlog ' name '.errlog ' name '.outbin && qsub ' name '.queue '];
+                command=['cd ' executionpath ' && rm -rf ./' runtimename ' && mkdir ' runtimename ' && cd ' runtimename ' && mv ../' runtimename '.tar.gz ./ && tar -zxvf ' runtimename '.tar.gz  && qsub ' name '.queue '];
         elseif strcmpi(cluster,'castor'),
-                command=['cd ' executionpath ' && rm -rf ' name '.lock ' name '.outlog ' name '.errlog ' name '.outbin && qsub ' name '.queue '];
+                command=['cd ' executionpath ' && rm -rf ./' runtimename ' && mkdir ' runtimename ' && cd ' runtimename ' && mv ../' runtimename '.tar.gz ./ && tar -zxvf ' runtimename '.tar.gz  && qsub ' name '.queue '];
         elseif strcmpi(cluster,'pfe'),
-                command=['cd ' executionpath ' && rm -rf ' name '.lock ' name '.outlog ' name '.errlog ' name '.outbin && qsub ' name '.queue '];
+                command=['cd ' executionpath ' && rm -rf ./' runtimename ' && mkdir ' runtimename ' && cd ' runtimename ' && mv ../' runtimename '.tar.gz ./ && tar -zxvf ' runtimename '.tar.gz  && qsub ' name '.queue '];
         else
                 error('ClusterLaunchCommand error message: unknown cluster command');

issm/trunk/src/m/classes/public/queue/ClusterScript.m

@@ -11 +11 @@
         fprintf(fid,'#!/bin/bash\n');
         fprintf(fid,'#PBS -l select=%i:ncpus=1\n',np);
+        fprintf(fid,'#PBS -N %s\n',name);
         fprintf(fid,'#PBS -l walltime=%i\n',time*60); %walltime is in seconds.
         fprintf(fid,'#PBS -q %s\n',queue);
@@ -26 +27 @@
 
         fprintf(fid,'#!/bin/sh\n');
-        fprintf(fid,'rm -rf %s/%s.lock\n',executionpath,name);
         if mem_debug==0,
                 fprintf(fid,'mpirun -np %i %s/issm.exe %s %s %s.bin %s.outbin %s.lock  2> %s.errlog >%s.outlog & ',np,codepath,EnumToString(analysis_type),executionpath,name,name,name,name,name);
@@ -39 +39 @@
         fprintf(fid,'#!/bin/sh\n');
         fprintf(fid,'#PBS -l walltime=%i\n',time*60); %walltime is in seconds.
+        fprintf(fid,'#PBS -N %s\n',name);
         fprintf(fid,'#PBS -l ncpus=%i\n',np);
         if ~isempty(queue),

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

@@ -21 +21 @@
 if ~strcmpi(options.batch,'yes'),
         
-        %what files are we sending?
-        packages={[md.name '.bin'],[md.name '.queue']};
+        %compress the files into one zip.
+        compressstring=['tar -zcvf ' md.runtimename '.tar.gz ' md.name '.bin ' md.name '.queue '];
         if md.qmu_analysis,
-                packages{end+1}=[md.name '.qmu.in'];
+                compressstring=[compressstring md.name '.qmu.in'];
         end
-
+        system(compressstring);
+        
         disp('uploading input file and queueing script');
-        issmscpout(md.cluster,executionpath,login,port,packages);
+        issmscpout(md.cluster,executionpath,login,port,{[md.runtimename '.tar.gz']});
 
         disp('launching solution sequence on remote cluster');
-        issmssh(md.cluster,login,port,ClusterLaunchCommand(md.cluster,md.name,executionpath));
+        issmssh(md.cluster,login,port,ClusterLaunchCommand(md.cluster,md.runtimename,md.name,executionpath));
 
 else

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

@@ -5 +5 @@
 %      md=solveparallel(md);
 
+%First, build a runtime name that is unique, that we will use to create 
+%directories, name jobs, etc ...
+c=clock;
+md.runtimename=sprintf('%s-%i-%i-%i-%i-%i-%i',md.name,GetPId,c(2),c(3),c(1),c(4),c(5));
 
 %Get cluster.rc location
@@ -11 +15 @@
 %Figure out parameters for this particular cluster
 [codepath,executionpath,login,port]=ClusterParameters(md.cluster,cluster_rc_location);
-
-%Append name of directory if specified in options: 
-executionpath=[executionpath '/' options.directory];
 
 %Marshall model data into a binary file.

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

@@ -9 +9 @@
 
 %Get filename (lock file) and options
-filename=[executionpath '/' md.name '.lock'];
+filename=[executionpath '/' md.runtimename '/' md.name '.lock'];
 cluster=md.cluster;
 timelimit=md.waitonlock;
@@ -22 +22 @@
                 time=0; ispresent=0;
                 while (ispresent==0 & time<timelimit)
-                        [status, result]=system(['ssh -q -p ' num2str(port) ' ' login '@localhost "if ( -e ' executionpath '/' md.name '.lock ) echo 1"']);
+                        [status, result]=system(['ssh -q -p ' num2str(port) ' ' login '@localhost "if ( -e ' executionpath '/' md.runtimename '/' md.name '.lock ) echo 1"']);
+                        result
                         if ~isempty(result),
                                 if ismember('1',result),