Changeset 26744 for issm/trunk/src/m/classes/clusters/pfe.m

Timestamp:

12/22/21 10:39:44 (3 years ago)

Author:

Mathieu Morlighem

Message:

merged trunk-jpl and trunk for revision 26742

Location:

issm/trunk

Files:

• . (modified) (1 prop)
• src (modified) (1 prop)
• src/m/classes/clusters/pfe.m (modified) (17 diffs)

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

@@ -1 @@
-%PFE class definition
+%PFE cluster class definition
 %
 %   Usage:
@@ -7 +7 @@
 
 classdef pfe
-        properties (SetAccess=public)  
+        properties (SetAccess=public)
                 % {{{
                 name           = 'pfe'
                 login          = '';
-                modules        = {'comp-intel/2016.2.181' 'mpi-sgi/mpt'};
+                modules        = {'comp-intel/2016.2.181' 'mpt'};
                 numnodes       = 20;
                 cpuspernode    = 8;
@@ -21 +21 @@
                 codepath       = '';
                 executionpath  = '';
-                grouplist     = 's1690';
+                grouplist      = '';
                 interactive    = 0;
                 bbftp          = 0;
@@ -39 +39 @@
                 %}}}
                 function disp(cluster) % {{{
-                        %  display the object
+                        % display the object
                         disp(sprintf('class ''%s'' object ''%s'' = ',class(cluster),inputname(1)));
                         disp(sprintf('    name: %s',cluster.name));
                         disp(sprintf('    login: %s',cluster.login));
-                        modules=''; for i=1:length(cluster.modules), modules=[modules cluster.modules{i} ',']; end; modules=modules(1:end-1); 
-                        disp(sprintf('    modules: %s',modules));
+                        disp(sprintf('    modules: %s',strjoin(cluster.modules,', ')));
+                        disp(sprintf('    numnodes: %s',cluster.numnodes));
+                        disp(sprintf('    cpuspernode: %s',cluster.cpuspernode));
+                        disp(sprintf('    np: %i',cluster.nprocs()));
                         disp(sprintf('    port: %i',cluster.port));
-                        disp(sprintf('    numnodes: %i',cluster.numnodes));
-                        disp(sprintf('    cpuspernode: %i',cluster.cpuspernode));
-                        disp(sprintf('    np: %i',cluster.cpuspernode*cluster.numnodes));
-                        disp(sprintf('    queue: %s',cluster.queue));
+                        disp(sprintf('    queue: %i',cluster.queue));
                         disp(sprintf('    time: %i',cluster.time));
-                        disp(sprintf('    processor: %s',cluster.processor));
-                        disp(sprintf('    srcpath: %s ($ISSM_DIR on pfe)',cluster.srcpath));
-                        disp(sprintf('    codepath: %s  ($ISSM_DIR/bin on pfe)',cluster.codepath));
-                        disp(sprintf('    executionpath: %s  (directory containing issm.exe on pfe)',cluster.executionpath));
+                        disp(sprintf('    processor: %i',cluster.processor));
+                        disp(sprintf('    srcpath: %s',cluster.srcpath));
+                        disp(sprintf('    codepath: %s',cluster.codepath));
+                        disp(sprintf('    executionpath: %s',cluster.executionpath));
                         disp(sprintf('    grouplist: %s',cluster.grouplist));
-                        disp(sprintf('    interactive: %i',cluster.interactive));
-                        disp(sprintf('    hyperthreading: %i',cluster.hyperthreading));
-                end
-                %}}}
-                function numprocs=np(cluster) % {{{
+                        disp(sprintf('    interactive: %s',cluster.interactive));
+                        disp(sprintf('    bbftp: %s',cluster.bbftp));
+                        disp(sprintf('    numstreams: %s',cluster.numstreams));
+                        disp(sprintf('    hyperthreading: %s',cluster.hyperthreading));
+                end
+                %}}}
+                function numprocs=nprocs(cluster) % {{{
                         %compute number of processors
                         numprocs=cluster.numnodes*cluster.cpuspernode;
@@ -71 +72 @@
                         queue_requirements_np=[2048 2048 150 150 2048];
 
-                        QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,cluster.time)
+                        QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.nprocs(),cluster.time)
 
                         %now, check cluster.cpuspernode according to processor type
@@ -128 +129 @@
                         
                         else
-                                md = checkmessage(md,'unknown processor type, should be ''bro'' or ''has'' or ''ivy'' or ''san'' or ''cas_ait''');
+                                md = checkmessage(md,'unknown processor type, should be ''bro'', ''has'', ''ivy'', ''san'', or ''cas_ait''');
                         end
 
@@ -161 +162 @@
                         fprintf(fid,'#PBS -l select=%i:ncpus=%i:model=%s\n',cluster.numnodes,cluster.cpuspernode,cluster.processor);
                         fprintf(fid,'#PBS -l walltime=%i\n',cluster.time*60); %walltime is in seconds.
-                        fprintf(fid,'#PBS -q %s \n',cluster.queue);
+                        fprintf(fid,'#PBS -q %s\n',cluster.queue);
                         fprintf(fid,'#PBS -W group_list=%s\n',cluster.grouplist);
                         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,'#PBS -o %s/%s/%s.outlog \n',cluster.executionpath,dirname,modelname);
+                        fprintf(fid,'#PBS -e %s/%s/%s.errlog \n\n',cluster.executionpath,dirname,modelname);
                         fprintf(fid,'. /usr/share/modules/init/bash\n\n');
                         for i=1:numel(cluster.modules), fprintf(fid,['module load ' cluster.modules{i} '\n']); end
@@ -176 +177 @@
                         fprintf(fid,'cd %s/%s/\n\n',cluster.executionpath,dirname);
                         if ~isvalgrind,
-                                fprintf(fid,'/u/scicon/tools/bin/several_tries mpiexec -np %i /u/scicon/tools/bin/mbind.x -cs -n%i %s/%s %s %s %s\n',cluster.np,cluster.cpuspernode,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname);
-                        else
-                                fprintf(fid,'mpiexec -np %i valgrind --leak-check=full %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname);
+                                fprintf(fid,'/u/scicon/tools/bin/toss3/several_tries mpiexec -np %i %s/%s %s %s/%s %s\n',cluster.nprocs(),cluster.codepath,executable,solution,cluster.executionpath,dirname,modelname);
+                        else
+                                fprintf(fid,'mpiexec -np %i valgrind --leak-check=full %s/%s %s %s %s\n',cluster.nprocs(),cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname);
                         end
                         if ~io_gather, %concatenate the output files:
@@ -190 +191 @@
                                 if cluster.interactive==10,
                                                 fprintf(fid,'module unload mpi-mvapich2/1.4.1/gcc\n');
-                                                fprintf(fid,'mpiexec -np %i %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,solution,[pwd() '/run'],modelname);
+                                                fprintf(fid,'mpiexec -np %i %s/%s %s %s %s\n',cluster.nprocs(),cluster.codepath,executable,solution,[pwd() '/run'],modelname);
                                 else
                                         if ~isvalgrind,
-                                                fprintf(fid,'mpiexec -np %i %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,solution,[cluster.executionpath '/Interactive' num2str(cluster.interactive)],modelname);
+                                                fprintf(fid,'mpiexec -np %i %s/%s %s %s %s\n',cluster.nprocs(),cluster.codepath,executable,solution,[cluster.executionpath '/Interactive' num2str(cluster.interactive)],modelname);
                                         else
-                                                fprintf(fid,'mpiexec -np %i valgrind --leak-check=full %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,solution,[cluster.executionpath '/Interactive' num2str(cluster.interactive)],modelname);
+                                                fprintf(fid,'mpiexec -np %i valgrind --leak-check=full %s/%s %s %s %s\n',cluster.nprocs(),cluster.codepath,executable,solution,[cluster.executionpath '/Interactive' num2str(cluster.interactive)],modelname);
                                         end
                                 end
@@ -214 +215 @@
 
                         %what is the executable being called? 
-                        executable='issm_slr.exe';
-
-                        if ispc(), error('BuildQueueScriptMultipleModels not support yet on windows machines');end;
+                        executable='issm_slc.exe';
+
+                        if ispc & ~ismingw, error('BuildQueueScriptMultipleModels not support yet on windows machines');end;
 
                         %write queuing script 
@@ -318 +319 @@
                         fprintf(fid,'export MPI_GROUP_MAX=64\n\n');
                         fprintf(fid,'cd %s/%s/\n\n',cluster.executionpath,modelname);
-                        fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname); %FIXME
+                        fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.nprocs(),cluster.codepath,[cluster.executionpath '/' modelname],modelname); %FIXME
                         if ~io_gather, %concatenate the output files:
                                 fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname);
@@ -328 +329 @@
                                 fid=fopen([modelname '.run'],'w');
                                 if ~isvalgrind,
-                                        fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname);
-                                else
-                                        fprintf(fid,'mpiexec -np %i valgrind --leak-check=full %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname);
+                                        fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.nprocs(),cluster.codepath,[cluster.executionpath '/' modelname],modelname);
+                                else
+                                        fprintf(fid,'mpiexec -np %i valgrind --leak-check=full %s/kriging.exe %s %s\n',cluster.nprocs(),cluster.codepath,[cluster.executionpath '/' modelname],modelname);
                                 end
                                 if ~io_gather, %concatenate the output files:
@@ -382 +383 @@
 
                 end %}}}
-                function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{
+                function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{
 
                         %compress the files into one zip.
-                        compressstring=['tar -zcf ' dirname '.tar.gz '];
+                        compressstring=['tar -zcf ' dirname '.tar.gz'];
                         for i=1:numel(filelist),
                                 compressstring = [compressstring ' ' filelist{i}];
@@ -403 +404 @@
                         end
 
-                        if ~cluster.bbftp,
+                        if cluster.bbftp,
+                                issmbbftpout(cluster.name,directory,cluster.login,cluster.port,cluster.numstreams,{[dirname '.tar.gz']});
+                        else
                                 issmscpout(cluster.name,directory,cluster.login,cluster.port,{[dirname '.tar.gz']});
-                        else
-                                issmbbftpout(cluster.name,directory,cluster.login,cluster.port,cluster.numstreams,{[dirname '.tar.gz']});
-                        end
-
-                end
-                %}}}
-                function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{
-
-                        %lauch command, to be executed via ssh
-                        if ~cluster.interactive, 
-                                if ~isempty(restart)
-                                        launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && qsub ' modelname '.queue '];
-                                else
-                                        launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
-                                                ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz  && qsub ' modelname '.queue '];
-                                end
-                        else
+                        end
+
+                end
+                %}}}
+                function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{
+
+                        %launch command, to be executed via ssh
+                        if cluster.interactive,
                                 if ~isempty(restart)
                                         launchcommand=['cd ' cluster.executionpath '/Interactive' num2str(cluster.interactive)];
@@ -431 +425 @@
                                         end
                                 end
+                        else
+                                if ~isempty(restart)
+                                        launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && qsub ' modelname '.queue '];
+                                else
+                                        launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
+                                                ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && qsub ' modelname '.queue '];
+                                end
                         end
 
@@ -437 +438 @@
                 end
                 %}}}
-                function Download(cluster,dirname,filelist)% {{{
+                function Download(cluster,dirname,filelist) % {{{
 
                         %copy files from cluster to current directory
@@ -448 +449 @@
                         end
 
-                        if ~cluster.bbftp,
+                        if cluster.bbftp,
+                                issmbbftpin(cluster.name, cluster.login, cluster.port, cluster.numstreams, directory, filelist);
+                        else
                                 issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist);
-                        else
-                                issmbbftpin(cluster.name, cluster.login, cluster.port, cluster.numstreams, directory, filelist);
                         end