Changeset 5954

Timestamp:

09/22/10 13:19:26 (14 years ago)

Author:

Eric.Larour

Message:

New framework for launching a job on different clusters, based on the use of classes for each cluster.

Location:

issm/trunk/src/m

Files:

• clusters (added)
• clusters/QueueRequirements.m (added)
• clusters/README (added)
• clusters/astrid.m (added)
• clusters/castor.m (added)
• clusters/cluster.rc (added)
• clusters/cosmos.m (added)
• clusters/eric-mac.m (added)
• clusters/gemini.m (added)
• clusters/larsen.m (added)
• clusters/none.m (added)
• clusters/pfe.m (added)
• clusters/pollux.m (added)
• clusters/wilkes.m (added)
• model/LaunchQueueJob.m (added)
• model/ismodelselfconsistent.m (modified) (2 diffs)
• model/loadresultsfromcluster.m (modified) (3 diffs)
• model/solve.m (modified) (1 diff)
• model/solveparallel.m (modified) (2 diffs)
• model/solvers/solversettoasm.m (modified) (1 diff)
• model/waitonlock.m (modified) (1 diff)

issm/trunk/src/m/model/ismodelselfconsistent.m

@@ -167 +167 @@
 end
 %}}}
-%SCALAR
+%SCALAR {{{
 if ~isscalar(md.control_type),
         error('model not consistent: md.control_type should be a scalar');
@@ -190 +190 @@
 %}}}
 %PARALLEL{{{1
-if ~strcmpi(md.cluster,'none'),
-
-        %NAN VALUES
-        fields={'time','np'};
-        checknan(md,fields);
-
-        %FIELD > 0
-        fields={'time','np'};
-        checkgreaterstrict(md,fields,0);
-
-end
+md.cluster.IsConsistent;
 %}}}
 

issm/trunk/src/m/model/loadresultsfromcluster.m

@@ -5 +5 @@
 %      md=loadresultsfromcluster(md,runtimename);
 
-%Get cluster.rc location
-cluster_rc_location=which('cluster.rc');
-
 %Figure out parameters for this particular cluster
-[codepath,executionpath,login,port]=ClusterParameters(md.cluster,cluster_rc_location);
+codepath=md.cluster.codepath;
+executionpath=md.cluster.executionpath;
+login=md.cluster.login;
+port=md.cluster.port;
+clustername=md.cluster.name;
 
 if nargin==2,
@@ -37 +38 @@
 
 %copy files from cluster to present directory
-issmscpin(md.cluster, login, port, directory, packages);
+issmscpin(clustername, login, port, directory, packages);
 
 %read log files onto  fields
@@ -62 +63 @@
 %erase input file if run was carried out on same platform.
 hostname=oshostname();
-if strcmpi(hostname,md.cluster),
+if strcmpi(hostname,clustername),
         if md.qmu_analysis,
                 delete([['qmu' num2str(GetPId) '/'] md.name '.bin']);

issm/trunk/src/m/model/solve.m

@@ -47 +47 @@
 %If running in parallel, we have a different way of launching the solution
 %sequences. 
-if ~strcmpi(md.cluster,'none'),
+if ~strcmpi(md.cluster.name,'none'),
         md=solveparallel(md,options);
         return;

issm/trunk/src/m/model/solveparallel.m

@@ -10 +10 @@
 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
-cluster_rc_location=which('cluster.rc');
-
-%Figure out parameters for this particular cluster
-[codepath,executionpath,login,port]=ClusterParameters(md.cluster,cluster_rc_location);
-
 %Marshall model data into a binary file.
 marshall(md);
@@ -25 +19 @@
 
 %Now, we need to build the queuing script, used by the cluster to launch the job.
-BuildQueueingScript(md,executionpath,codepath);
+BuildQueueScript(md.cluster,md.name,md.analysis_type,md.mem_debug);
 
 %Now, launch the queueing script
-md=LaunchQueueJob(md,executionpath,login,port,options);
+LaunchQueueJob(md,options);
 
 %Do we return, or just wait for results?
 if (md.waitonlock~=0 &  ~strcmpi(options.batch,'yes')),
         %we wait for the done file
-        islock=waitonlock(md,executionpath,login,port);
+        islock=waitonlock(md);
         if islock==0,
                 %no results to be loaded

issm/trunk/src/m/model/solvers/solversettoasm.m

@@ -7 +7 @@
 %md.petscoptions={{'mat_type','aij'},{'ksp_type','cgs'},{'pc_type','asm'},{sub_mat_type','mumps'},{'sub_pc_type','lu'},{'pc_asm_overlap',4},{'pc_factor_shift_positive_definite','true'}};
 %md.petscoptions={{'mat_type','aij'},{'ksp_type','cgs'},{'pc_type','asm'},{'sub_pc_type','lu'},{'pc_asm_overlap',4}};
-md.petscoptions={{'mat_type','aij'},{'ksp_type','gmres'},{'pc_type','asm'},{'sub_pc_type','lu'},{'pc_asm_overlap',4}};
+md.petscoptions={{'mat_type','aij'},{'ksp_type','gmres'},{'pc_type','asm'},{'sub_pc_type','lu'},{'pc_asm_overlap',3},{'ksp_max_it',100},{'ksp_rtol',1e-30'}};

issm/trunk/src/m/model/waitonlock.m

@@ -9 +9 @@
 
 %Get filename (lock file) and options
+executionpath=md.cluster.executionpath;
+cluster=md.cluster.name;
+login=md.cluster.login;
+port=md.cluster.port;
+timelimit=md.waitonlock;
 filename=[executionpath '/' md.runtimename '/' md.name '.lock'];
-cluster=md.cluster;
-timelimit=md.waitonlock;
 
 %waitonlock will work if the lock is on the same machine only: