Changeset 26994

Timestamp:

05/06/22 11:51:47 (3 years ago)

Author:

Mathieu Morlighem

Message:

CHG: updating discovery

File:

• issm/trunk-jpl/src/m/classes/clusters/discovery.m (modified) (6 diffs)

issm/trunk-jpl/src/m/classes/clusters/discovery.m

@@ -11 +11 @@
                 name          = 'discovery'
                 login         = '';
-                numnodes      = 20;
-                cpuspernode   = 8;
-                port          = 8000;
+                numnodes      = 1;
+                cpuspernode   = 16;
+                port          = 0;
                 codepath      = '';
                 executionpath = '';
                 interactive   = 0;
-                time          = 10; % hours
-                memory        = 2;
+                time          = 10; %in hours
+                memory        = 2;  %in Gb
+                email         = 'END,FAIL';
 
         end
@@ -39 +40 @@
                         disp(sprintf('    numnodes: %i',cluster.numnodes));
                         disp(sprintf('    cpuspernode: %i',cluster.cpuspernode));
-                        disp(sprintf('    np: %i',cluster.nprocs()));
                         disp(sprintf('    port: %i',cluster.port));
                         disp(sprintf('    codepath: %s',cluster.codepath));
@@ -45 +45 @@
                         disp(sprintf('    interactive: %i',cluster.interactive));
                         disp(sprintf('    time: %i hours',cluster.time));
-                        disp(sprintf('    memory: %i',cluster.memory));
+                        disp(sprintf('    memory: %i Gb',cluster.memory));
                 end
                 %}}}
@@ -58 +58 @@
                         if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end
                         if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end
-
                 end
                 %}}}
@@ -75 +74 @@
                         fprintf(fid,'#SBATCH --ntasks-per-node=%i\n',cluster.cpuspernode);
                         fprintf(fid,'#SBATCH --time=%s\n',datestr(cluster.time/24,'HH:MM:SS')); %walltime is in HH:MM:SS format. cluster.time is in hour
-                        fprintf(fid,'#SBATCH --mem=%iG\n\n',cluster.memory);
+                        fprintf(fid,'#SBATCH --mem=%iG\n',cluster.memory);
+                        if ~isempty(cluster.email)
+                                fprintf(fid,'#SBATCH --mail-type=%s\n',cluster.email);
+                        end
+                        fprintf(fid,'\n');
+
                         fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath);
                         fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n');      
@@ -87 +91 @@
                 %}}}
                 function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{
-
-                        if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end
-                        if(isgprof),    disp('gprof not supported by cluster, ignoring...'); end
-
-                        %write queuing script 
-                        fid=fopen([modelname '.queue'],'w');
-                        fprintf(fid,'#!/bin/bash\n');
-                        fprintf(fid,'#SBATCH --job-name=%s\n',modelname);
-                        fprintf(fid,'#SBATCH -o %s.outlog \n',modelname);
-                        fprintf(fid,'#SBATCH -e %s.errlog \n',modelname);
-                        fprintf(fid,'#SBATCH --nodes=%i\n',cluster.numnodes);
-                        fprintf(fid,'#SBATCH --ntasks-per-node=%i\n',cluster.cpuspernode);
-                        fprintf(fid,'#SBATCH --time=%s\n',datestr(cluster.time/24,'HH:MM:SS')); %walltime is in HH:MM:SS format. cluster.time is in hour
-                        fprintf(fid,'#SBATCH --mem=%iG\n\n',cluster.memory);
-                        fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath);
-                        fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n');       
-                        fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname);
-                        fprintf(fid,'mpirun -n %i %s/issm.exe %s %s %s\n',cluster.nprocs(), cluster.codepath,solution,[cluster.executionpath '/' dirname],modelname);
-                        if ~io_gather, %concatenate the output files:
-                                fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname);
-                        end
-                        fclose(fid);
-
-                        %in interactive mode, create a run file, and errlog and outlog file
-                        if cluster.interactive,
-                                fid=fopen([modelname '.run'],'w');
-                                fprintf(fid,'mpirun -n %i %s/issm.exe %s %s %s\n',cluster.nprocs(), cluster.codepath,solution,[cluster.executionpath '/' dirname],modelname);
-                                if ~io_gather, %concatenate the output files:
-                                        fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname);
-                                end
-                                fclose(fid);
-                                fid=fopen([modelname '.errlog'],'w');
-                                fclose(fid);
-                                fid=fopen([modelname '.outlog'],'w');
-                                fclose(fid);
-                        end
+                        error('not implemented yet');
                 end %}}}
                 function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{