[24307] | 1 | Index: ../trunk-jpl/src/m/classes/clusters/sherlock.m
|
---|
| 2 | ===================================================================
|
---|
| 3 | --- ../trunk-jpl/src/m/classes/clusters/sherlock.m (nonexistent)
|
---|
| 4 | +++ ../trunk-jpl/src/m/classes/clusters/sherlock.m (revision 24160)
|
---|
| 5 | @@ -0,0 +1,160 @@
|
---|
| 6 | +%PFE class definition
|
---|
| 7 | +%
|
---|
| 8 | +% Usage:
|
---|
| 9 | +% cluster=sherlock();
|
---|
| 10 | +% cluster=sherlock('np',3);
|
---|
| 11 | +% cluster=sherlock('np',3,'login','username');
|
---|
| 12 | +
|
---|
| 13 | +classdef sherlock
|
---|
| 14 | + properties (SetAccess=public)
|
---|
| 15 | + % {{{
|
---|
| 16 | + name = 'sherlock'
|
---|
| 17 | + login = '';
|
---|
| 18 | + numnodes = 1;
|
---|
| 19 | + cpuspernode = 24;
|
---|
| 20 | + port = 0;
|
---|
| 21 | + codepath = '';
|
---|
| 22 | + executionpath = '';
|
---|
| 23 | + interactive = 0;
|
---|
| 24 | + time = 30;
|
---|
| 25 | + memory = 2;
|
---|
| 26 | + end
|
---|
| 27 | + %}}}
|
---|
| 28 | + methods
|
---|
| 29 | + function cluster=sherlock(varargin) % {{{
|
---|
| 30 | +
|
---|
| 31 | + %initialize cluster using default settings if provided
|
---|
| 32 | + if (exist('sherlock_settings')==2), sherlock_settings; end
|
---|
| 33 | +
|
---|
| 34 | + %use provided options to change fields
|
---|
| 35 | + cluster=AssignObjectFields(pairoptions(varargin{:}),cluster);
|
---|
| 36 | + end
|
---|
| 37 | + %}}}
|
---|
| 38 | + function disp(cluster) % {{{
|
---|
| 39 | + % display the object
|
---|
| 40 | + disp(sprintf('class ''%s'' object ''%s'' = ',class(cluster),inputname(1)));
|
---|
| 41 | + disp(sprintf(' name: %s',cluster.name));
|
---|
| 42 | + disp(sprintf(' login: %s',cluster.login));
|
---|
| 43 | + disp(sprintf(' port: %i',cluster.port));
|
---|
| 44 | + disp(sprintf(' numnodes: %i',cluster.numnodes));
|
---|
| 45 | + disp(sprintf(' cpuspernode: %i',cluster.cpuspernode));
|
---|
| 46 | + disp(sprintf(' np: %i',cluster.cpuspernode*cluster.numnodes));
|
---|
| 47 | + disp(sprintf(' codepath: %s',cluster.codepath));
|
---|
| 48 | + disp(sprintf(' executionpath: %s',cluster.executionpath));
|
---|
| 49 | + disp(sprintf(' interactive: %i',cluster.interactive));
|
---|
| 50 | + disp(sprintf(' time: %i',cluster.time));
|
---|
| 51 | + disp(sprintf(' memory: %i',cluster.memory));
|
---|
| 52 | + end
|
---|
| 53 | + %}}}
|
---|
| 54 | + function numprocs=np(cluster) % {{{
|
---|
| 55 | + %compute number of processors
|
---|
| 56 | + numprocs=cluster.numnodes*cluster.cpuspernode;
|
---|
| 57 | + end
|
---|
| 58 | + %}}}
|
---|
| 59 | + function md = checkconsistency(cluster,md,solution,analyses) % {{{
|
---|
| 60 | +
|
---|
| 61 | + %Miscelaneous
|
---|
| 62 | + if isempty(cluster.login), md = checkmessage(md,'login empty'); end
|
---|
| 63 | + if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end
|
---|
| 64 | + if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end
|
---|
| 65 | +
|
---|
| 66 | + end
|
---|
| 67 | + %}}}
|
---|
| 68 | + function BuildKrigingQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{
|
---|
| 69 | +
|
---|
| 70 | + if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end
|
---|
| 71 | + if(isgprof), disp('gprof not supported by cluster, ignoring...'); end
|
---|
| 72 | +
|
---|
| 73 | + %write queuing script
|
---|
| 74 | + fid=fopen([modelname '.queue'],'w');
|
---|
| 75 | + fprintf(fid,'#!/bin/bash\n');
|
---|
| 76 | + fprintf(fid,'#SBATCH --job-name=%s\n',mdelname);
|
---|
| 77 | + fprintf(fid,'#SBATCH -p %s \n',cluster.queue);
|
---|
| 78 | + fprintf(fid,'#SBATCH -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode);
|
---|
| 79 | + fprintf(fid,'#SBATCH --time=%i\n',cluster.time*60); %walltime is in seconds.
|
---|
| 80 | + fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory);
|
---|
| 81 | + fprintf(fid,'#SBATCH -o %s.outlog \n',modelname);
|
---|
| 82 | + fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname);
|
---|
| 83 | + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME
|
---|
| 84 | + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME
|
---|
| 85 | + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname);
|
---|
| 86 | + fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname);
|
---|
| 87 | + if ~io_gather, %concatenate the output files:
|
---|
| 88 | + fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname);
|
---|
| 89 | + end
|
---|
| 90 | + fclose(fid);
|
---|
| 91 | + end
|
---|
| 92 | + %}}}
|
---|
| 93 | + function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{
|
---|
| 94 | +
|
---|
| 95 | + if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end
|
---|
| 96 | + if(isgprof), disp('gprof not supported by cluster, ignoring...'); end
|
---|
| 97 | +
|
---|
| 98 | + %write queuing script
|
---|
| 99 | + fid=fopen([modelname '.queue'],'w');
|
---|
| 100 | + fprintf(fid,'#!/bin/bash\n');
|
---|
| 101 | + fprintf(fid,'#SBATCH --job-name=%s\n',modelname);
|
---|
| 102 | + fprintf(fid,'#SBATCH -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode);
|
---|
| 103 | + fprintf(fid,'#SBATCH --time=%i\n',cluster.time*60); %walltime is in seconds.
|
---|
| 104 | + fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory);
|
---|
| 105 | + fprintf(fid,'#SBATCH -o %s.outlog \n',modelname);
|
---|
| 106 | + fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname);
|
---|
| 107 | + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME
|
---|
| 108 | + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME
|
---|
| 109 | + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname);
|
---|
| 110 | + fprintf(fid,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,cluster.codepath,solution,[cluster.executionpath '/' dirname],modelname);
|
---|
| 111 | + if ~io_gather, %concatenate the output files:
|
---|
| 112 | + fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname);
|
---|
| 113 | + end
|
---|
| 114 | + fclose(fid);
|
---|
| 115 | +
|
---|
| 116 | + %in interactive mode, create a run file, and errlog and outlog file
|
---|
| 117 | + if cluster.interactive,
|
---|
| 118 | + fid=fopen([modelname '.run'],'w');
|
---|
| 119 | + fprintf(fid,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,cluster.codepath,solution,[cluster.executionpath '/' dirname],modelname);
|
---|
| 120 | + if ~io_gather, %concatenate the output files:
|
---|
| 121 | + fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname);
|
---|
| 122 | + end
|
---|
| 123 | + fclose(fid);
|
---|
| 124 | + fid=fopen([modelname '.errlog'],'w');
|
---|
| 125 | + fclose(fid);
|
---|
| 126 | + fid=fopen([modelname '.outlog'],'w');
|
---|
| 127 | + fclose(fid);
|
---|
| 128 | + end
|
---|
| 129 | + end %}}}
|
---|
| 130 | + function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{
|
---|
| 131 | +
|
---|
| 132 | + %compress the files into one zip.
|
---|
| 133 | + compressstring=['tar -zcf ' dirname '.tar.gz '];
|
---|
| 134 | + for i=1:numel(filelist),
|
---|
| 135 | + compressstring = [compressstring ' ' filelist{i}];
|
---|
| 136 | + end
|
---|
| 137 | + if cluster.interactive,
|
---|
| 138 | + compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog '];
|
---|
| 139 | + end
|
---|
| 140 | + system(compressstring);
|
---|
| 141 | +
|
---|
| 142 | + disp('uploading input file and queueing script');
|
---|
| 143 | + issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']});
|
---|
| 144 | +
|
---|
| 145 | + end %}}}
|
---|
| 146 | + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{
|
---|
| 147 | +
|
---|
| 148 | + disp('launching solution sequence on remote cluster');
|
---|
| 149 | + if ~isempty(restart)
|
---|
| 150 | + launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && sbatch ' modelname '.queue '];
|
---|
| 151 | + else
|
---|
| 152 | + launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
|
---|
| 153 | + ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && sbatch ' modelname '.queue '];
|
---|
| 154 | + end
|
---|
| 155 | + issmssh(cluster.name,cluster.login,cluster.port,launchcommand);
|
---|
| 156 | + end %}}}
|
---|
| 157 | + function Download(cluster,dirname,filelist)% {{{
|
---|
| 158 | +
|
---|
| 159 | + %copy files from cluster to current directory
|
---|
| 160 | + directory=[cluster.executionpath '/' dirname '/'];
|
---|
| 161 | + issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist);
|
---|
| 162 | +
|
---|
| 163 | + end %}}}
|
---|
| 164 | + end
|
---|
| 165 | +end
|
---|