Changeset 6378

Timestamp:

10/21/10 14:41:06 (15 years ago)

Author:

Mathieu Morlighem

Message:

cluster pfe is by default empty. User dependent settings are to be found in pfe_settings of the user contrib directory that has been added to the matlab path (startup).
Also, waitonlock waits for y/n to load results automatically from cluster. Use waitonlck=0 to disable this functionnality.

Location:

issm/trunk/src/m

Files:

• classes/clusters/pfe.m (modified) (5 diffs)
• contrib/larour (added)
• contrib/larour/pfe_settings.m (added)
• contrib/morlighe (added)
• contrib/morlighe/pfe_settings.m (added)
• contrib/seroussi (added)
• contrib/seroussi/pfe_settings.m (added)
• model/solveparallel.m (modified) (1 diff)
• model/waitonlock.m (modified) (1 diff)

issm/trunk/src/m/classes/clusters/pfe.m

@@ -10 +10 @@
                  % {{{1
                  name='pfe'
-                 login='elarour';
+                 login='';
                  numnodes=20;
                  cpuspernode=8; 
@@ -17 +17 @@
                  time=12*60;
                  processor='neh';
-                 codepath='/staff/elarour/trunk/bin';
-                 executionpath='/nobackupp10/elarour/execution/pleiades';
+                 codepath='';
+                 executionpath='';
                  interactive=0;
                  bbftp=0;
@@ -30 +30 @@
          methods
                  function cluster=pfe(varargin) % {{{1
+
+                         %initialize cluster using default settings if provided
+                         if (exist('pfe_settings')==2), pfe_settings; end
+
+                         %use provided options to change fields
                          options=pairoptions(varargin{:});
                          for i=1:size(options.list,1),
@@ -92 +97 @@
                                  error('IsConsistent error message: unknown processor type, should be ''neh'',''wes'' or ''har''');
                          end
+
+                         %Miscelaneous
+                         if isempty(cluster.login), error('IsConsistent error message: login empty'); end
+                         if isempty(cluster.codepath), error('IsConsistent error message: codepath empty'); end
+                         if isempty(cluster.executionpath), error('IsConsistent error message: executionpath empty'); end
 
                  end
@@ -182 +192 @@
                                 disp('launching solution sequence on remote cluster');
                                 issmssh(cluster.name,cluster.login,cluster.port,launchcommand);
-                                        
-                                if cluster.interactive,
-                                        error('done uploading files to interactive directory on remote cluster');
-                                end
 
                         else

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

@@ -29 +29 @@
 
         %Do we return, or just wait for results?
-        if (md.waitonlock~=0 &  ~strcmpi(options.batch,'yes')),
+        if (md.waitonlock>0 &  ~strcmpi(options.batch,'yes')),
                 %we wait for the done file
                 islock=waitonlock(md);

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

@@ -19 +19 @@
 if ~strcmpi(oshostname(),cluster),
 
-        if port, 
-                %there is a tunnel, so we have a short at looking for the lock file.
-
-                disp(['waiting for ' filename ' hold on... (Ctrl+C to exit)'])
-                time=0; ispresent=0;
-                while (ispresent==0 & time<timelimit)
-                        [status, result]=system(['ssh -q -p ' num2str(port) ' ' login '@localhost "if ( -e ' executionpath '/' md.runtimename '/' md.name '.lock ) echo 1"']);
-                        if ~isempty(result),
-                                if ismember('1',result),
-                                        ispresent=1;
-                                else
-                                        ispresent=0;
-                                end
-                        else
-                                ispresent=0;
-                        end
-                        pause(10); %tunnel can be unstable, let's not use it too much
-                        time=time+1/60;
-                end
-                disp(['waitonlock: lock detected, with value of result:|' result '|']);
-                
-                %build output
-                if (time>timelimit),
-                        disp('Time limit exceeded. Increase md.waitonlock');
-                        disp('The results must be loaded manually with md=loadresultsfromcluster(md).');
-                        error(['waitonlock error message: time limit exceeded']);
-                        flag=0;
-                else
-                        flag=1;
-                end
+        disp('solution launched on remote cluster. log in to detect job completion.');
+        choice=input('Is the job successfully completed? (y/n)','s');
+        if ~strcmp(choice,'y'), 
+                disp('Results not loaded... exiting'); 
+                flag=0;
         else
-                disp('solution launched on remote cluster. log in to detect job completion.');
-                choice=input('Is the job successfully completed? (y/n)','s');
-                if ~strcmp(choice,'y'), 
-                        disp('Results not loaded... exiting'); 
-                        flag=0;
-                else
-                        flag=1;
-                end
+                flag=1;
         end