[8586] | 1 | %GENERIC cluster class definition
|
---|
[8578] | 2 | %
|
---|
| 3 | % Usage:
|
---|
[8586] | 4 | % cluster=generic('name','astrid','np',3);
|
---|
| 5 | % cluster=generic('name',oshostname(),'np',3,'login','username');
|
---|
[8578] | 6 |
|
---|
| 7 | classdef generic
|
---|
[13395] | 8 | properties (SetAccess=public)
|
---|
| 9 | % {{{
|
---|
| 10 | name='';
|
---|
| 11 | login='';
|
---|
| 12 | np=1;
|
---|
| 13 | port=0;
|
---|
| 14 | interactive=1;
|
---|
| 15 | codepath=[issmdir() '/bin'];
|
---|
[13975] | 16 | etcpath=[issmdir() '/etc'];
|
---|
[13395] | 17 | executionpath=[issmdir() '/execution'];
|
---|
| 18 | valgrind=[issmdir() '/externalpackages/valgrind/install/bin/valgrind'];
|
---|
| 19 | valgrindlib=[issmdir() '/externalpackages/valgrind/install/lib/libmpidebug.so'];
|
---|
| 20 | valgrindsup=[issmdir() '/externalpackages/valgrind/issm.supp'];
|
---|
| 21 | %}}}
|
---|
| 22 | end
|
---|
| 23 | methods
|
---|
| 24 | function cluster=generic(varargin) % {{{
|
---|
[8586] | 25 |
|
---|
[13395] | 26 | %use provided options to change fields
|
---|
| 27 | options=pairoptions(varargin{:});
|
---|
[8586] | 28 |
|
---|
[13395] | 29 | %get name
|
---|
| 30 | cluster.name=getfieldvalue(options,'name',oshostname());
|
---|
[8586] | 31 |
|
---|
[13395] | 32 | %initialize cluster using user settings if provided
|
---|
| 33 | if (exist([cluster.name '_settings'])==2), eval([cluster.name '_settings']); end
|
---|
[8578] | 34 |
|
---|
[13395] | 35 | %OK get other fields
|
---|
| 36 | cluster=AssignObjectFields(pairoptions(varargin{:}),cluster);
|
---|
| 37 | end
|
---|
| 38 | %}}}
|
---|
| 39 | function disp(cluster) % {{{
|
---|
| 40 | % display the object
|
---|
| 41 | disp(sprintf('class ''%s'' object ''%s'' = ',class(cluster),inputname(1)));
|
---|
| 42 | disp(sprintf(' name: %s',cluster.name));
|
---|
| 43 | disp(sprintf(' login: %s',cluster.login));
|
---|
| 44 | disp(sprintf(' np: %i',cluster.np));
|
---|
| 45 | disp(sprintf(' port: %i',cluster.port));
|
---|
| 46 | disp(sprintf(' codepath: %s',cluster.codepath));
|
---|
| 47 | disp(sprintf(' executionpath: %s',cluster.executionpath));
|
---|
[13975] | 48 | disp(sprintf(' etcpath: %s',cluster.executionpath));
|
---|
[13395] | 49 | disp(sprintf(' valgrind: %s',cluster.valgrind));
|
---|
| 50 | disp(sprintf(' valgrindlib: %s',cluster.valgrindlib));
|
---|
| 51 | disp(sprintf(' valgrindsup: %s',cluster.valgrindsup));
|
---|
| 52 | end
|
---|
| 53 | %}}}
|
---|
| 54 | function md = checkconsistency(cluster,md,solution,analyses) % {{{
|
---|
| 55 | if cluster.np<1
|
---|
| 56 | md = checkmessage(md,['number of processors should be at least 1']);
|
---|
| 57 | end
|
---|
| 58 | if isnan(cluster.np),
|
---|
| 59 | md = checkmessage(md,'number of processors should not be NaN!');
|
---|
| 60 | end
|
---|
| 61 | end
|
---|
| 62 | %}}}
|
---|
| 63 | function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof) % {{{
|
---|
[8578] | 64 |
|
---|
[13395] | 65 | %write queuing script
|
---|
| 66 | if ~ispc(),
|
---|
[12706] | 67 |
|
---|
[13395] | 68 | fid=fopen([modelname '.queue'],'w');
|
---|
| 69 | fprintf(fid,'#!/bin/sh\n');
|
---|
| 70 | if ~isvalgrind,
|
---|
| 71 | if cluster.interactive
|
---|
[16137] | 72 | if IssmConfig('_HAVE_MPI_'),
|
---|
[13395] | 73 | fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s ',cluster.np,cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname],modelname);
|
---|
| 74 | else
|
---|
| 75 | fprintf(fid,'%s/issm.exe %s %s %s ',cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname],modelname);
|
---|
| 76 | end
|
---|
| 77 | else
|
---|
[16137] | 78 | if IssmConfig('_HAVE_MPI_'),
|
---|
[13395] | 79 | fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',cluster.np,cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname],modelname,modelname,modelname);
|
---|
| 80 | else
|
---|
| 81 | fprintf(fid,'%s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname],modelname,modelname,modelname);
|
---|
| 82 | end
|
---|
| 83 | end
|
---|
| 84 | elseif isgprof,
|
---|
| 85 | fprintf(fid,'\n gprof %s/issm.exe gmon.out > %s.performance',cluster.codepath,modelname);
|
---|
| 86 | else
|
---|
| 87 | %Add --gen-suppressions=all to get suppression lines
|
---|
| 88 | fprintf(fid,'LD_PRELOAD=%s \\\n',cluster.valgrindlib);
|
---|
| 89 | if ismac,
|
---|
[16137] | 90 | if IssmConfig('_HAVE_MPI_'),
|
---|
[13395] | 91 | fprintf(fid,'mpiexec -np %i %s --leak-check=full --dsymutil=yes --suppressions=%s %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',...
|
---|
| 92 | cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname], modelname,modelname,modelname);
|
---|
| 93 | else
|
---|
| 94 | fprintf(fid,'%s --leak-check=full --dsymutil=yes --suppressions=%s %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',...
|
---|
| 95 | cluster.valgrind,cluster.valgrindsup,cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname], modelname,modelname,modelname);
|
---|
| 96 | end
|
---|
| 97 | else
|
---|
[16137] | 98 | if IssmConfig('_HAVE_MPI_'),
|
---|
[13395] | 99 | fprintf(fid,'mpiexec -np %i %s --leak-check=full --suppressions=%s %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',...
|
---|
| 100 | cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname],modelname,modelname,modelname);
|
---|
| 101 | else
|
---|
| 102 | fprintf(fid,'%s --leak-check=full --suppressions=%s %s/issm.exe %s %s %s 2> %s.errlog >%s.outlog ',...
|
---|
| 103 | cluster.valgrind,cluster.valgrindsup,cluster.codepath,EnumToString(solution),[cluster.executionpath '/' dirname],modelname,modelname,modelname);
|
---|
| 104 | end
|
---|
| 105 | end
|
---|
| 106 | end
|
---|
| 107 | if ~io_gather, %concatenate the output files:
|
---|
| 108 | fprintf(fid,'\ncat %s.outbin.* > %s.outbin',modelname,modelname);
|
---|
| 109 | end
|
---|
| 110 | fclose(fid);
|
---|
[12706] | 111 |
|
---|
[13395] | 112 | else % Windows
|
---|
[12706] | 113 |
|
---|
[13395] | 114 | fid=fopen([modelname '.bat'],'w');
|
---|
| 115 | fprintf(fid,'@echo off\n');
|
---|
[15396] | 116 |
|
---|
| 117 | if cluster.np>1,
|
---|
| 118 | fprintf(fid,'"C:\\Program Files\\MPICH2\\bin\\mpiexec.exe" -n %i "%s/issm.exe" %s ./ %s ',cluster.np,cluster.codepath,EnumToString(solution),modelname);
|
---|
[13395] | 119 | else
|
---|
[15396] | 120 | fprintf(fid,'"%s/issm.exe" %s ./ %s ',cluster.codepath,EnumToString(solution),modelname);
|
---|
[13395] | 121 | end
|
---|
| 122 | fclose(fid);
|
---|
| 123 | end
|
---|
[8578] | 124 |
|
---|
[13395] | 125 | %in interactive mode, create a run file, and errlog and outlog file
|
---|
| 126 | if cluster.interactive,
|
---|
| 127 | fid=fopen([modelname '.errlog'],'w'); fclose(fid);
|
---|
| 128 | fid=fopen([modelname '.outlog'],'w'); fclose(fid);
|
---|
| 129 | end
|
---|
| 130 | end
|
---|
| 131 | %}}}
|
---|
| 132 | function BuildKrigingQueueScript(cluster,modelname,solution,io_gather,isvalgrind,isgprof) % {{{
|
---|
[12706] | 133 |
|
---|
[13395] | 134 | %write queuing script
|
---|
| 135 | if ~ispc(),
|
---|
[12706] | 136 |
|
---|
[13395] | 137 | fid=fopen([modelname '.queue'],'w');
|
---|
| 138 | fprintf(fid,'#!/bin/sh\n');
|
---|
| 139 | if ~isvalgrind,
|
---|
| 140 | if cluster.interactive
|
---|
| 141 | fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s ',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname);
|
---|
| 142 | else
|
---|
| 143 | 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);
|
---|
| 144 | end
|
---|
| 145 | elseif isgprof,
|
---|
| 146 | fprintf(fid,'\n gprof %s/kriging.exe gmon.out > %s.performance',cluster.codepath,modelname);
|
---|
| 147 | else
|
---|
| 148 | %Add --gen-suppressions=all to get suppression lines
|
---|
| 149 | fprintf(fid,'LD_PRELOAD=%s \\\n',cluster.valgrindlib);
|
---|
| 150 | fprintf(fid,'mpiexec -np %i %s --leak-check=full --suppressions=%s %s/kriging.exe %s %s 2> %s.errlog >%s.outlog ',...
|
---|
| 151 | cluster.np,cluster.valgrind,cluster.valgrindsup,cluster.codepath,[cluster.executionpath '/' modelname],modelname,modelname,modelname);
|
---|
| 152 | end
|
---|
| 153 | if ~io_gather, %concatenate the output files:
|
---|
| 154 | fprintf(fid,'\ncat %s.outbin.* > %s.outbin',modelname,modelname);
|
---|
| 155 | end
|
---|
| 156 | fclose(fid);
|
---|
[11527] | 157 |
|
---|
[13395] | 158 | else % Windows
|
---|
[8578] | 159 |
|
---|
[13395] | 160 | fid=fopen([modelname '.bat'],'w');
|
---|
| 161 | fprintf(fid,'@echo off\n');
|
---|
| 162 | if cluster.interactive
|
---|
| 163 | fprintf(fid,'"%s/issm.exe" %s "%s" %s ',cluster.codepath,EnumToString(solution),[cluster.executionpath '/' modelname],modelname);
|
---|
| 164 | else
|
---|
| 165 | fprintf(fid,'"%s/issm.exe" %s "%s" %s 2> %s.errlog >%s.outlog',...
|
---|
| 166 | cluster.codepath,EnumToString(solution),[cluster.executionpath '/' modelname],modelname,modelname,modelname);
|
---|
| 167 | end
|
---|
| 168 | fclose(fid);
|
---|
| 169 | end
|
---|
[8578] | 170 |
|
---|
[13395] | 171 | %in interactive mode, create a run file, and errlog and outlog file
|
---|
| 172 | if cluster.interactive,
|
---|
| 173 | fid=fopen([modelname '.errlog'],'w'); fclose(fid);
|
---|
| 174 | fid=fopen([modelname '.outlog'],'w'); fclose(fid);
|
---|
| 175 | end
|
---|
| 176 | end
|
---|
| 177 | %}}}
|
---|
| 178 | function LaunchQueueJob(cluster,modelname,dirname,filelist)% {{{
|
---|
[8578] | 179 |
|
---|
[15396] | 180 | if ~ispc,
|
---|
| 181 | %compress the files into one zip.
|
---|
| 182 | compressstring=['tar -zcf ' dirname '.tar.gz '];
|
---|
| 183 | for i=1:numel(filelist),
|
---|
| 184 | compressstring = [compressstring ' ' filelist{i}];
|
---|
| 185 | end
|
---|
| 186 | if cluster.interactive,
|
---|
| 187 | compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog '];
|
---|
| 188 | end
|
---|
| 189 | system(compressstring);
|
---|
[11527] | 190 |
|
---|
[15396] | 191 | disp('uploading input file and queueing script');
|
---|
| 192 | issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']});
|
---|
[11527] | 193 |
|
---|
[15396] | 194 | disp('launching solution sequence on remote cluster');
|
---|
| 195 | launchcommand=['source ' cluster.etcpath '/environment.sh && cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ...
|
---|
| 196 | ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && source ' modelname '.queue '];
|
---|
| 197 | issmssh(cluster.name,cluster.login,cluster.port,launchcommand);
|
---|
| 198 | else
|
---|
| 199 | system([modelname '.bat']);
|
---|
| 200 | end
|
---|
[13395] | 201 | end %}}}
|
---|
| 202 | function Download(cluster,dirname,filelist)% {{{
|
---|
[11527] | 203 |
|
---|
[13395] | 204 | if ispc(),
|
---|
[12706] | 205 | %do nothing
|
---|
| 206 | return;
|
---|
[8578] | 207 | end
|
---|
| 208 |
|
---|
[12706] | 209 | %copy files from cluster to current directory
|
---|
| 210 | directory=[cluster.executionpath '/' dirname '/'];
|
---|
| 211 | issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist);
|
---|
[11527] | 212 | end %}}}
|
---|
[8578] | 213 | end
|
---|
| 214 | end
|
---|