[21764] | 1 | %localpfe cluster class definition
|
---|
| 2 | %
|
---|
| 3 | % Usage:
|
---|
| 4 | % cluster=localpfe('name','astrid','np',3);
|
---|
| 5 | % cluster=localpfe('name',oshostname(),'np',3,'login','username');
|
---|
| 6 |
|
---|
| 7 | classdef localpfe
|
---|
| 8 | properties (SetAccess=public)
|
---|
| 9 | % {{{
|
---|
| 10 | name = '';
|
---|
| 11 | login = '';
|
---|
| 12 | np = 1;
|
---|
[21853] | 13 | npocean = 0;
|
---|
[21764] | 14 | port = 0;
|
---|
| 15 | interactive = 1;
|
---|
| 16 | codepath = [IssmConfig('ISSM_PREFIX') '/bin'];
|
---|
| 17 | etcpath = [issmdir() '/etc'];
|
---|
| 18 | executionpath = [issmdir() '/execution'];
|
---|
| 19 | valgrind = [issmdir() '/externalpackages/valgrind/install/bin/valgrind'];
|
---|
| 20 | valgrindlib = [issmdir() '/externalpackages/valgrind/install/lib/libmpidebug.so'];
|
---|
| 21 | valgrindsup = [issmdir() '/externalpackages/valgrind/issm.supp'];
|
---|
| 22 | verbose = 1;
|
---|
| 23 | shell = '/bin/sh';
|
---|
| 24 | %}}}
|
---|
| 25 | end
|
---|
| 26 | methods
|
---|
| 27 | function cluster=localpfe(varargin) % {{{
|
---|
| 28 |
|
---|
| 29 | %Change the defaults if ispc
|
---|
| 30 | if ispc,
|
---|
| 31 | cluster.codepath = [issmdir() '\bin'];
|
---|
| 32 | cluster.etcpath = [issmdir() '\etc'];
|
---|
| 33 | cluster.executionpath = [issmdir() '\execution'];
|
---|
| 34 | end
|
---|
| 35 |
|
---|
| 36 | %use provided options to change fields
|
---|
| 37 | options=pairoptions(varargin{:});
|
---|
| 38 |
|
---|
| 39 | %get name
|
---|
| 40 | cluster.name=getfieldvalue(options,'name',oshostname());
|
---|
| 41 |
|
---|
| 42 | %initialize cluster using user settings if provided
|
---|
| 43 | if (exist([cluster.name '_settings'])==2), eval([cluster.name '_settings']); end
|
---|
| 44 |
|
---|
| 45 | %OK get other fields
|
---|
| 46 | cluster=AssignObjectFields(pairoptions(varargin{:}),cluster);
|
---|
| 47 | end
|
---|
| 48 | %}}}
|
---|
| 49 | function disp(cluster) % {{{
|
---|
| 50 | % display the object
|
---|
| 51 | disp(sprintf('class ''%s'' object ''%s'' = ',class(cluster),inputname(1)));
|
---|
| 52 | disp(sprintf(' name: %s',cluster.name));
|
---|
| 53 | disp(sprintf(' login: %s',cluster.login));
|
---|
| 54 | disp(sprintf(' np: %i',cluster.np));
|
---|
| 55 | disp(sprintf(' port: %i',cluster.port));
|
---|
| 56 | disp(sprintf(' codepath: %s',cluster.codepath));
|
---|
| 57 | disp(sprintf(' executionpath: %s',cluster.executionpath));
|
---|
| 58 | disp(sprintf(' etcpath: %s',cluster.etcpath));
|
---|
| 59 | disp(sprintf(' valgrind: %s',cluster.valgrind));
|
---|
| 60 | disp(sprintf(' valgrindlib: %s',cluster.valgrindlib));
|
---|
| 61 | disp(sprintf(' valgrindsup: %s',cluster.valgrindsup));
|
---|
| 62 | disp(sprintf(' verbose: %s',cluster.verbose));
|
---|
| 63 | disp(sprintf(' shell: %s',cluster.shell));
|
---|
| 64 | end
|
---|
| 65 | %}}}
|
---|
| 66 | function md = checkconsistency(cluster,md,solution,analyses) % {{{
|
---|
| 67 | if cluster.np<1
|
---|
| 68 | md = checkmessage(md,['number of processors should be at least 1']);
|
---|
| 69 | end
|
---|
| 70 | if isnan(cluster.np),
|
---|
| 71 | md = checkmessage(md,'number of processors should not be NaN!');
|
---|
| 72 | end
|
---|
| 73 | end
|
---|
| 74 | %}}}
|
---|
[21765] | 75 | function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{
|
---|
[21764] | 76 |
|
---|
| 77 | %write queuing script
|
---|
| 78 | %what is the executable being called?
|
---|
| 79 | executable='issm.exe';
|
---|
| 80 | if isdakota,
|
---|
| 81 | version=IssmConfig('_DAKOTA_VERSION_'); version=str2num(version(1:3));
|
---|
| 82 | if (version>=6),
|
---|
| 83 | executable='issm_dakota.exe';
|
---|
| 84 | end
|
---|
| 85 | end
|
---|
| 86 |
|
---|
| 87 | fid=fopen([modelname '.queue'],'w');
|
---|
| 88 | fprintf(fid,'#!%s\n',cluster.shell);
|
---|
| 89 | fprintf(fid,'mpiexec -np %i %s/%s %s %s %s \n',cluster.np,cluster.codepath,executable,solution,cluster.executionpath,modelname);
|
---|
| 90 | fclose(fid);
|
---|
| 91 |
|
---|
| 92 |
|
---|
| 93 | %in interactive mode, create a run file, and errlog and outlog file
|
---|
| 94 | if cluster.interactive,
|
---|
| 95 | fid=fopen([modelname '.errlog'],'w'); fclose(fid);
|
---|
| 96 | fid=fopen([modelname '.outlog'],'w'); fclose(fid);
|
---|
| 97 | end
|
---|
| 98 | end
|
---|
| 99 | %}}}
|
---|
| 100 | function BuildQueueScriptMultipleModels(cluster,dirname,modelname,solution,dirnames,modelnames,nps) % {{{
|
---|
| 101 |
|
---|
| 102 | %some checks:
|
---|
| 103 | if isempty(modelname), error('BuildQueueScriptMultipleModels error message: need a non empty model name!');end
|
---|
| 104 |
|
---|
| 105 | %what is the executable being called?
|
---|
| 106 | executable='issm_slr.exe';
|
---|
| 107 |
|
---|
| 108 | if ispc(), error('BuildQueueScriptMultipleModels not support yet on windows machines');end;
|
---|
| 109 |
|
---|
| 110 | %write queuing script
|
---|
| 111 | fid=fopen([modelname '.queue'],'w');
|
---|
| 112 |
|
---|
| 113 | fprintf(fid,'#!%s\n',cluster.shell);
|
---|
| 114 |
|
---|
| 115 | %number of cpus:
|
---|
| 116 | mpistring=sprintf('mpiexec -np %i ',cluster.np);
|
---|
| 117 |
|
---|
| 118 | %executable:
|
---|
| 119 | mpistring=[mpistring sprintf('%s/%s ',cluster.codepath,executable)];
|
---|
| 120 |
|
---|
| 121 | %solution name:
|
---|
| 122 | mpistring=[mpistring sprintf('%s ',solution)];
|
---|
| 123 |
|
---|
| 124 | %execution directory and model name:
|
---|
| 125 | mpistring=[mpistring sprintf('%s/%s %s',cluster.executionpath,dirname,modelname)];
|
---|
| 126 |
|
---|
| 127 | %inform main executable of how many icecaps, glaciers and earth models are being run:
|
---|
| 128 | mpistring=[mpistring sprintf(' %i ',length(dirnames))];
|
---|
| 129 |
|
---|
| 130 | %icecaps, glaciers and earth location, names and number of processors associated:
|
---|
| 131 | for i=1:length(dirnames),
|
---|
| 132 | mpistring=[mpistring sprintf(' %s/%s %s %i ',cluster.executionpath,dirnames{i},modelnames{i},nps{i})];
|
---|
| 133 | end
|
---|
| 134 |
|
---|
| 135 | %log files:
|
---|
| 136 | if ~cluster.interactive,
|
---|
| 137 | mpistring=[mpistring sprintf('2> %s.errlog> %s.outlog',modelname,modelname)];
|
---|
| 138 | end
|
---|
| 139 |
|
---|
| 140 | %write this long string to disk:
|
---|
| 141 | fprintf(fid,mpistring);
|
---|
| 142 | fclose(fid);
|
---|
| 143 |
|
---|
| 144 | %in interactive mode, create a run file, and errlog and outlog file
|
---|
| 145 | if cluster.interactive,
|
---|
| 146 | fid=fopen([modelname '.errlog'],'w'); fclose(fid);
|
---|
| 147 | fid=fopen([modelname '.outlog'],'w'); fclose(fid);
|
---|
| 148 | end
|
---|
| 149 | end
|
---|
| 150 | %}}}
|
---|
[21852] | 151 | function BuildQueueScriptIceOcean(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota) % {{{
|
---|
| 152 |
|
---|
| 153 | %write queuing script
|
---|
| 154 | %what is the executable being called?
|
---|
| 155 | executable='issm_ocean.exe';
|
---|
| 156 |
|
---|
| 157 | fid=fopen([modelname '.queue'],'w');
|
---|
| 158 | fprintf(fid,'#!%s\n',cluster.shell);
|
---|
[22664] | 159 | fprintf(fid,'mpiexec -np %i %s/%s %s %s %s : -np %i ./mitgcmuv\n',cluster.np,cluster.codepath,executable,solution,cluster.executionpath,modelname,cluster.npocean);
|
---|
[21852] | 160 | fclose(fid);
|
---|
| 161 |
|
---|
| 162 | %in interactive mode, create a run file, and errlog and outlog file
|
---|
| 163 | if cluster.interactive,
|
---|
| 164 | fid=fopen([modelname '.errlog'],'w'); fclose(fid);
|
---|
| 165 | fid=fopen([modelname '.outlog'],'w'); fclose(fid);
|
---|
| 166 | end
|
---|
| 167 | end
|
---|
| 168 | %}}}
|
---|
[21764] | 169 | function BuildKrigingQueueScript(cluster,modelname,solution,io_gather,isvalgrind,isgprof) % {{{
|
---|
| 170 |
|
---|
| 171 | %write queuing script
|
---|
| 172 | if ~ispc(),
|
---|
| 173 |
|
---|
| 174 | fid=fopen([modelname '.queue'],'w');
|
---|
| 175 | fprintf(fid,'#!/bin/sh\n');
|
---|
| 176 | if ~isvalgrind,
|
---|
| 177 | if cluster.interactive
|
---|
| 178 | fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s ',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname);
|
---|
| 179 | else
|
---|
| 180 | fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s 2> %s.errlog >%s.outlog ',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname,modelname,modelname);
|
---|
| 181 | end
|
---|
| 182 | elseif isgprof,
|
---|
| 183 | fprintf(fid,'\n gprof %s/kriging.exe gmon.out > %s.performance',cluster.codepath,modelname);
|
---|
| 184 | else
|
---|
| 185 | %Add --gen-suppressions=all to get suppression lines
|
---|
| 186 | fprintf(fid,'LD_PRELOAD=%s \\\n',cluster.valgrindlib);
|
---|
| 187 | fprintf(fid,'mpiexec -np %i %s --leak-check=full --suppressions=%s %s/kriging.exe %s %s 2> %s.errlog >%s.outlog ',...
|
---|
| 188 | cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,[cluster.executionpath '/' modelname],modelname,modelname,modelname);
|
---|
| 189 | end
|
---|
| 190 | if ~io_gather, %concatenate the output files:
|
---|
| 191 | fprintf(fid,'\ncat %s.outbin.* > %s.outbin',modelname,modelname);
|
---|
| 192 | end
|
---|
| 193 | fclose(fid);
|
---|
| 194 |
|
---|
| 195 | else % Windows
|
---|
| 196 |
|
---|
| 197 | fid=fopen([modelname '.bat'],'w');
|
---|
| 198 | fprintf(fid,'@echo off\n');
|
---|
| 199 | if cluster.interactive
|
---|
| 200 | fprintf(fid,'"%s/issm.exe" %s "%s" %s ',cluster.codepath,solution,[cluster.executionpath '/' modelname],modelname);
|
---|
| 201 | else
|
---|
| 202 | fprintf(fid,'"%s/issm.exe" %s "%s" %s 2> %s.errlog >%s.outlog',...
|
---|
| 203 | cluster.codepath,solution,[cluster.executionpath '/' modelname],modelname,modelname,modelname);
|
---|
| 204 | end
|
---|
| 205 | fclose(fid);
|
---|
| 206 | end
|
---|
| 207 |
|
---|
| 208 | %in interactive mode, create a run file, and errlog and outlog file
|
---|
| 209 | if cluster.interactive,
|
---|
| 210 | fid=fopen([modelname '.errlog'],'w'); fclose(fid);
|
---|
| 211 | fid=fopen([modelname '.outlog'],'w'); fclose(fid);
|
---|
| 212 | end
|
---|
| 213 | end
|
---|
| 214 | %}}}
|
---|
| 215 | function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{
|
---|
| 216 | if ~ispc,
|
---|
| 217 |
|
---|
| 218 | %compress the files into one zip.
|
---|
| 219 | compressstring=['tar -zcf ' dirname '.tar.gz '];
|
---|
| 220 | for i=1:numel(filelist),
|
---|
| 221 | compressstring = [compressstring ' ' filelist{i}];
|
---|
| 222 | end
|
---|
| 223 | if cluster.interactive,
|
---|
| 224 | compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog '];
|
---|
| 225 | end
|
---|
| 226 | system(compressstring);
|
---|
| 227 |
|
---|
| 228 | if cluster.verbose, disp('uploading input file and queueing script'); end
|
---|
| 229 | issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']});
|
---|
| 230 | end
|
---|
| 231 | end %}}}
|
---|
| 232 | function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{
|
---|
| 233 |
|
---|
| 234 | %figure out what shell extension we will use:
|
---|
| 235 | if isempty(strfind(cluster.shell,'csh')),
|
---|
| 236 | shellext='sh';
|
---|
| 237 | else
|
---|
| 238 | shellext='csh';
|
---|
| 239 | end
|
---|
| 240 |
|
---|
| 241 | if cluster.verbose, disp('launching solution sequence on remote cluster'); end
|
---|
| 242 |
|
---|
| 243 | launchcommand=['cd ' cluster.executionpath ' && rm -rf *.lock && rm -rf ADOLC* && tar -zxf ' dirname '.tar.gz && rm -rf *.tar.gz'];
|
---|
| 244 | issmssh(cluster.name,cluster.login,cluster.port,launchcommand);
|
---|
| 245 |
|
---|
| 246 | end %}}}
|
---|
[21852] | 247 | function LaunchQueueJobIceOcean(cluster,modelname,dirname,filelist,restart,batch)% {{{
|
---|
| 248 |
|
---|
| 249 | %figure out what shell extension we will use:
|
---|
| 250 | if isempty(strfind(cluster.shell,'csh')),
|
---|
| 251 | shellext='sh';
|
---|
| 252 | else
|
---|
| 253 | shellext='csh';
|
---|
| 254 | end
|
---|
| 255 |
|
---|
| 256 | if cluster.verbose, disp('launching solution sequence on remote cluster'); end
|
---|
| 257 |
|
---|
| 258 | launchcommand=['cd ' cluster.executionpath ' && rm -rf *.lock && tar -zxf ' dirname '.tar.gz && rm -rf *.tar.gz'];
|
---|
| 259 | issmssh(cluster.name,cluster.login,cluster.port,launchcommand);
|
---|
| 260 |
|
---|
| 261 | end %}}}
|
---|
[21764] | 262 | function Download(cluster,dirname,filelist)% {{{
|
---|
| 263 |
|
---|
| 264 | if ispc(),
|
---|
| 265 | %do nothing
|
---|
| 266 | return;
|
---|
| 267 | end
|
---|
| 268 |
|
---|
| 269 | %copy files from cluster to current directory
|
---|
| 270 | issmscpin(cluster.name,cluster.login,cluster.port,cluster.executionpath,filelist);
|
---|
| 271 | end %}}}
|
---|
| 272 | end
|
---|
| 273 | end
|
---|