Index: /issm/trunk-jpl/src/m/classes/clusters/aurora.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/aurora.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/aurora.m (revision 26330) @@ -1,3 +1,3 @@ -%AURORA class definition +%AURORA cluster class definition % % Usage: @@ -7,128 +7,128 @@ classdef aurora - properties (SetAccess=public) - % {{{ - name='aurora.jpl.nasa.gov' - login='username'; - numnodes = 1; - cpuspernode = 24; - modules = {'intel/cluster-toolkit-2013.5.192'}; - port=1070; - queue='shortq'; - time=3*60; - codepath='~/issm/trunk/'; - executionpath='~/issm/trunk/execution/'; - mpipath='/opt/intel/impi/4.1.3/intel64/bin/'; - %}}} - end - methods - function cluster=aurora(varargin) % {{{ + properties (SetAccess=public) + % {{{ + name='aurora.jpl.nasa.gov' + login='username'; + numnodes = 1; + cpuspernode = 24; + modules = {'intel/cluster-toolkit-2013.5.192'}; + port=1070; + queue='shortq'; + time=3*60; + codepath='~/issm/trunk/'; + executionpath='~/issm/trunk/execution/'; + mpipath='/opt/intel/impi/4.1.3/intel64/bin/'; + %}}} + end + methods + function cluster=aurora(varargin) % {{{ - %initialize cluster using default settings if provided - if (exist('aurora_settings')==2), aurora_settings; end + %initialize cluster using default settings if provided + if (exist('aurora_settings')==2), aurora_settings; end - cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); - end - %}}} - function disp(cluster) % {{{ - % 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)); - disp(sprintf(' numnodes: %i',cluster.numnodes)); - disp(sprintf(' cpuspernode: %i',cluster.cpuspernode)); - disp(sprintf(' port: %i',cluster.port)); - disp(sprintf(' queue: %s',cluster.queue)); - disp(sprintf(' time: %i',cluster.time)); - disp(sprintf(' codepath: %s',cluster.codepath)); - disp(sprintf(' executionpath: %s',cluster.executionpath)); - end - %}}} - function numprocs=np(cluster) % {{{ - %compute number of processors - numprocs=cluster.numnodes*cluster.cpuspernode; - end - %}}} - function md = checkconsistency(cluster,md,solution,analyses) % {{{ + cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); + end + %}}} + function disp(cluster) % {{{ + % 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)); + disp(sprintf(' numnodes: %i',cluster.numnodes)); + disp(sprintf(' cpuspernode: %i',cluster.cpuspernode)); + disp(sprintf(' port: %i',cluster.port)); + disp(sprintf(' queue: %s',cluster.queue)); + disp(sprintf(' time: %i',cluster.time)); + disp(sprintf(' codepath: %s',cluster.codepath)); + disp(sprintf(' executionpath: %s',cluster.executionpath)); + end + %}}} + function numprocs=nprocs(cluster) % {{{ + %compute number of processors + numprocs=cluster.numnodes*cluster.cpuspernode; + end + %}}} + function md = checkconsistency(cluster,md,solution,analyses) % {{{ - available_queues={'debugq','shortq','mediumq','longq','verylongq'}; - queue_requirements_time=[60*1 60*3 60*12 60*48 60*192]; - queue_requirements_np=[16 256 256 128 128]; + available_queues={'debugq','shortq','mediumq','longq','verylongq'}; + queue_requirements_time=[60*1 60*3 60*12 60*48 60*192]; + queue_requirements_np=[16 256 256 128 128]; - QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.numnodes.*cluster.cpuspernode,cluster.time) - end - %}}} - function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.numnodes.*cluster.cpuspernode,cluster.time) + end + %}}} + 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 - executable='issm.exe'; - if isdakota, - version=IssmConfig('_DAKOTA_VERSION_'); version=str2num(version(1:3)); - if (version>=6), - executable='issm_dakota.exe'; - end - end - if isoceancoupling, - executable='issm_ocean.exe'; - end + if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end + if(isgprof), disp('gprof not supported by cluster, ignoring...'); end + executable='issm.exe'; + if isdakota, + version=IssmConfig('_DAKOTA_VERSION_'); version=str2num(version(1:3)); + if (version>=6), + executable='issm_dakota.exe'; + end + end + if isoceancoupling, + executable='issm_ocean.exe'; + end - %write queuing script - fid=fopen([modelname '.queue'],'w'); - fprintf(fid,'#!/bin/bash\n'); - fprintf(fid,'#PBS -l select=%i:ncpus=%i\n',cluster.numnodes,cluster.cpuspernode); - fprintf(fid,'#PBS -N %s\n',modelname); - 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 -o %s.outlog \n',modelname); - fprintf(fid,'#PBS -e %s.errlog \n',modelname); - fprintf(fid,'source /usr/share/Modules/init/bash\n'); - for i=1:numel(cluster.modules), fprintf(fid,['module load ' cluster.modules{i} '\n\n']); end - fprintf(fid,'export PATH="$PATH:."\n\n'); - fprintf(fid,'export MPI_GROUP_MAX=64\n\n'); - fprintf(fid,'export MPI_UNBUFFERED_STDIO=true\n\n'); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); - fprintf(fid,'export PATH="$PATH:.:%s"\n',cluster.mpipath); - fprintf(fid,'export PBS_O_WORKDIR=%s\n',[cluster.executionpath '/' dirname]); - fprintf(fid,'cd $PBS_O_WORKDIR\n'); - fprintf(fid,'mpirun -n %i %s/%s %s %s %s',cluster.np(),cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname); - fclose(fid); + %write queuing script + fid=fopen([modelname '.queue'],'w'); + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#PBS -l select=%i:ncpus=%i\n',cluster.numnodes,cluster.cpuspernode); + fprintf(fid,'#PBS -N %s\n',modelname); + fprintf(fid,'#PBS -l walltime=%i\n',cluster.time*60); %walltime is in seconds. + fprintf(fid,'#PBS -q %s\n',cluster.queue); + fprintf(fid,'#PBS -o %s.outlog \n',modelname); + fprintf(fid,'#PBS -e %s.errlog \n',modelname); + fprintf(fid,'source /usr/share/Modules/init/bash\n'); + for i=1:numel(cluster.modules), fprintf(fid,['module load ' cluster.modules{i} '\n\n']); end + fprintf(fid,'export PATH="$PATH:."\n\n'); + fprintf(fid,'export MPI_GROUP_MAX=64\n\n'); + fprintf(fid,'export MPI_UNBUFFERED_STDIO=true\n\n'); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); + fprintf(fid,'export PATH="$PATH:.:%s"\n',cluster.mpipath); + fprintf(fid,'export PBS_O_WORKDIR=%s\n',[cluster.executionpath '/' dirname]); + fprintf(fid,'cd $PBS_O_WORKDIR\n'); + fprintf(fid,'mpirun -n %i %s/%s %s %s %s',cluster.nprocs(),cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname); + fclose(fid); - end - %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ + end + %}}} + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ - %compress the files into one zip. - compressstring=['tar -zcf ' dirname '.tar.gz ']; - for i=1:numel(filelist), - compressstring = [compressstring ' ' filelist{i}]; - end - system(compressstring); + %compress the files into one zip. + compressstring=['tar -zcf ' dirname '.tar.gz ']; + for i=1:numel(filelist), + compressstring = [compressstring ' ' filelist{i}]; + end + system(compressstring); - disp('uploading input file and queueing script'); - issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); + disp('uploading input file and queueing script'); + issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); - end %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ + end %}}} + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ - 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 + 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 - disp('launching solution sequence on remote cluster'); - issmssh(cluster.name,cluster.login,cluster.port,launchcommand); + disp('launching solution sequence on remote cluster'); + issmssh(cluster.name,cluster.login,cluster.port,launchcommand); - end %}}} - function Download(cluster,dirname,filelist)% {{{ + end %}}} + function Download(cluster,dirname,filelist) % {{{ - %copy files from cluster to current directory - directory=[cluster.executionpath '/' dirname '/']; - issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); + %copy files from cluster to current directory + directory=[cluster.executionpath '/' dirname '/']; + issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); - end %}}} + end %}}} end end Index: /issm/trunk-jpl/src/m/classes/clusters/camhpc.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/camhpc.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/camhpc.m (revision 26330) @@ -1,3 +1,3 @@ -%PFE class definition +%CAMHPC cluster class definition % % Usage: @@ -7,145 +7,145 @@ classdef camhpc - properties (SetAccess=public) - % {{{ - name = 'camhpc' - login = ''; - numnodes = 20; - cpuspernode = 8; - port = 8000; - project = ''; - partition = ''; - codepath = ''; - executionpath = ''; - interactive = 0; - time = 24*60; - memory = 2; - end - %}}} - methods - function cluster=camhpc(varargin) % {{{ + properties (SetAccess=public) + % {{{ + name = 'camhpc' + login = ''; + numnodes = 20; + cpuspernode = 8; + port = 8000; + project = ''; + partition = ''; + codepath = ''; + executionpath = ''; + interactive = 0; + time = 24*60; + memory = 2; + end + %}}} + methods + function cluster=camhpc(varargin) % {{{ - %initialize cluster using default settings if provided - if (exist('camhpc_settings')==2), camhpc_settings; end + %initialize cluster using default settings if provided + if (exist('camhpc_settings')==2), camhpc_settings; end - %use provided options to change fields - cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); - end - %}}} - function disp(cluster) % {{{ - % 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)); - 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(' project: %s',cluster.project)); - disp(sprintf(' partition: %s',cluster.partition)); - disp(sprintf(' codepath: %s',cluster.codepath)); - disp(sprintf(' executionpath: %s',cluster.executionpath)); - disp(sprintf(' interactive: %i',cluster.interactive)); - disp(sprintf(' time: %i',cluster.time)); - disp(sprintf(' memory: %i',cluster.memory)); - end - %}}} - function numprocs=np(cluster) % {{{ - %compute number of processors - numprocs=cluster.numnodes*cluster.cpuspernode; - end - %}}} - function md = checkconsistency(cluster,md,solution,analyses) % {{{ + %use provided options to change fields + cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); + end + %}}} + function disp(cluster) % {{{ + % 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)); + 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(' project: %s',cluster.project)); + disp(sprintf(' partition: %s',cluster.partition)); + disp(sprintf(' codepath: %s',cluster.codepath)); + disp(sprintf(' executionpath: %s',cluster.executionpath)); + disp(sprintf(' interactive: %i',cluster.interactive)); + disp(sprintf(' time: %i',cluster.time)); + disp(sprintf(' memory: %i',cluster.memory)); + end + %}}} + function numprocs=nprocs(cluster) % {{{ + %compute number of processors + numprocs=cluster.numnodes*cluster.cpuspernode; + end + %}}} + function md = checkconsistency(cluster,md,solution,analyses) % {{{ - available_queues={'ARNOLD-SL3-CPU'}; % Updated for csd3 NSA 28/3/18 - queue_requirements_time=[Inf Inf]; - queue_requirements_np=[80 80]; + available_queues={'ARNOLD-SL3-CPU'}; % Updated for csd3 NSA 28/3/18 + queue_requirements_time=[Inf Inf]; + queue_requirements_np=[80 80]; - QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.project,cluster.np,1) + QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.project,cluster.np,1) - %Miscelaneous - if isempty(cluster.login), md = checkmessage(md,'login empty'); end - if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end - if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end + %Miscelaneous + if isempty(cluster.login), md = checkmessage(md,'login empty'); end + if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end + if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end - end - %}}} - function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + end + %}}} + 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 + if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end + if(isgprof), disp('gprof not supported by cluster, ignoring...'); end - %write queuing script - disp(modelname) - fid=fopen([modelname '.queue'],'w'); - fprintf(fid,'#!/bin/bash\n'); - fprintf(fid,'#SBATCH --job-name=%s\n',modelname); - fprintf(fid,'#SBATCH -p %s \n',cluster.partition); - fprintf(fid,'#SBATCH -A %s \n',cluster.project); - fprintf(fid,'#SBATCH --mail-type=ALL\n'); - fprintf(fid,'#SBATCH -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode); - fprintf(fid,'#SBATCH --time=%i\n',cluster.time*60) %walltime is in seconds. - fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); - fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); - fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); - fprintf(fid,'mpirun -np %i %s/issm.exe %s %s %s\n',cluster.np,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); + %write queuing script + disp(modelname) + fid=fopen([modelname '.queue'],'w'); + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#SBATCH --job-name=%s\n',modelname); + fprintf(fid,'#SBATCH -p %s \n',cluster.partition); + fprintf(fid,'#SBATCH -A %s \n',cluster.project); + fprintf(fid,'#SBATCH --mail-type=ALL\n'); + fprintf(fid,'#SBATCH -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode); + fprintf(fid,'#SBATCH --time=%i\n',cluster.time*60) %walltime is in seconds. + fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); + fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); + fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); + fprintf(fid,'mpirun -np %i %s/issm.exe %s %s %s\n',cluster.np,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 -np %i %s/issm.exe %s %s %s\n',cluster.np,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 - end %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ + %in interactive mode, create a run file, and errlog and outlog file + if cluster.interactive, + fid=fopen([modelname '.run'],'w'); + fprintf(fid,'mpirun -np %i %s/issm.exe %s %s %s\n',cluster.np,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 + end %}}} + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ - %compress the files into one zip. - compressstring=['tar -zcf ' dirname '.tar.gz ']; - for i=1:numel(filelist), - compressstring = [compressstring ' ' filelist{i}]; - end - if cluster.interactive, - compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; - end - system(compressstring); + %compress the files into one zip. + compressstring=['tar -zcf ' dirname '.tar.gz ']; + for i=1:numel(filelist), + compressstring = [compressstring ' ' filelist{i}]; + end + if cluster.interactive, + compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; + end + system(compressstring); - disp('uploading input file and queueing script'); - issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); + disp('uploading input file and queueing script'); + issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); - end %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ + end %}}} + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ - disp('launching solution sequence on remote cluster'); + disp('launching solution sequence on remote cluster'); % % qsub replaced by sbatch for csd3 system NSA 28/3/18 - if ~isempty(restart) - launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && qsub ' modelname '.queue ']; - else - launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... - ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && sbatch ' modelname '.queue ']; - end - issmssh(cluster.name,cluster.login,cluster.port,launchcommand); - end %}}} - function Download(cluster,dirname,filelist)% {{{ + if ~isempty(restart) + launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && qsub ' modelname '.queue ']; + else + launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... + ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && sbatch ' modelname '.queue ']; + end + issmssh(cluster.name,cluster.login,cluster.port,launchcommand); + end %}}} + function Download(cluster,dirname,filelist) % {{{ - %copy files from cluster to current directory - directory=[cluster.executionpath '/' dirname '/']; - issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); + %copy files from cluster to current directory + directory=[cluster.executionpath '/' dirname '/']; + issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); - end %}}} + end %}}} end end Index: /issm/trunk-jpl/src/m/classes/clusters/discover.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/discover.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/discover.m (revision 26330) @@ -1,3 +1,3 @@ -%PFE class definition +%DISCOVER cluster class definition % % Usage: @@ -6,221 +6,221 @@ % cluster=discover('np',3,'login','username'); -classdef discover - properties (SetAccess=public) - % {{{ - name=oshostname(); - login=''; - modules = {}; - numnodes=20; - cpuspernode=8; - port=0; - queue='general'; - time=12*60*60; - processor='west'; - codepath=''; - executionpath=''; - interactive=0; - bbftp=0; - numstreams=8; - hyperthreading=0; - email=''; - end - %}}} - methods - function cluster=discover(varargin) % {{{ - - %initialize cluster using default settings if provided - if (exist('discover_settings')==2), discover_settings; end - - %use provided options to change fields - cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); - end - %}}} - function disp(cluster) % {{{ - % 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)); - disp(sprintf(' modules: %s',strjoin(cluster.modules,', '))); - 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(' time: %i',cluster.time)); - disp(sprintf(' processor: %s',cluster.processor)); - disp(sprintf(' codepath: %s',cluster.codepath)); - disp(sprintf(' executionpath: %s',cluster.executionpath)); - disp(sprintf(' interactive: %i',cluster.interactive)); - disp(sprintf(' hyperthreading: %i',cluster.hyperthreading)); - disp(sprintf(' email: %s',cluster.email)); - end - %}}} - function numprocs=np(cluster) % {{{ - %compute number of processors - numprocs=cluster.numnodes*cluster.cpuspernode; - end - %}}} - function md = checkconsistency(cluster,md,solution,analyses) % {{{ - - available_queues={'long','allnccs','debug'}; - queue_requirements_time=[24*60*60 12*60*60 1*60*60]; - queue_requirements_np=[560 6000 532]; - - QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,cluster.time) - - %now, check cluster.cpuspernode according to processor type - if ( strcmpi(cluster.processor,'sand')), - if ((cluster.cpuspernode>16 ) | (cluster.cpuspernode<1)), - md = checkmessage(md,'cpuspernode should be between 1 and 16 for ''sand'' processors'); - end - elseif strcmpi(cluster.processor,'hasw'), - if ((cluster.cpuspernode>28 ) | (cluster.cpuspernode<1)), - md = checkmessage(md,'cpuspernode should be between 1 and 28 for ''hasw'' processors'); - end - else - md = checkmessage(md,'unknown processor type, should be ''sand'' or ''hasw'' '); - end - - %Miscelaneous - if isempty(cluster.login), md = checkmessage(md,'login empty'); end - if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end - if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end - - end - %}}} - function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ - - if(isgprof), disp('gprof not supported by cluster, ignoring...'); end - - executable='issm.exe'; - if isdakota, - version=IssmConfig('_DAKOTA_VERSION_'); version=str2num(version(1:3)); - if (version>=6), - executable='issm_dakota.exe'; - end - end - if isoceancoupling, - executable='issm_ocean.exe'; - end - - %write queuing script - fid=fopen([modelname '.queue'],'w'); - - fprintf(fid,'#!/bin/bash\n'); - fprintf(fid,'#SBATCH -J %s \n',modelname); - fprintf(fid,'#SBATCH --qos=%s \n',cluster.queue); - fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); - fprintf(fid,'#SBATCH -e %s.errlog \n',modelname); - fprintf(fid,'#SBATCH -n %i \n',cluster.numnodes*cluster.cpuspernode); - fprintf(fid,'#SBATCH -N %i \n',cluster.numnodes); - fprintf(fid,'#SBATCH -t %02i:%02i:00 \n',floor(cluster.time/3600),floor(mod(cluster.time,3600)/60)); - fprintf(fid,'#SBATCH -A s1690 \n\n'); - for i=1:numel(cluster.modules), - fprintf(fid,['module load ' cluster.modules{i} '\n']); - end - if length(find(cluster.email=='@'))>0 - fprintf(fid,'#SBATCH --mail-user=%s \n',cluster.email); - fprintf(fid,'#SBATCH --mail-type=end \n\n'); - end - fprintf(fid,'. /usr/share/modules/init/bash\n\n'); - fprintf(fid,'module load comp/intel/20.0.0.166\n'); - fprintf(fid,'module load mpi/sgi-mpt/2.17\n'); - fprintf(fid,'module load cmake/3.17.0\n'); - fprintf(fid,'export MPI_GROUP_MAX=64\n\n'); - fprintf(fid,'export MPI_UNBUFFERED_STDIO=true\n\n'); - fprintf(fid,'export PATH="$PATH:."\n\n'); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); - - fprintf(fid,'mpiexec -np %i %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,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'); - if ~isvalgrind, - fprintf(fid,'mpiexec -np %i %s/%s %s %s %s\n',cluster.np,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); - end - 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 - end %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ - - %compress the files into one zip. - compressstring=['tar -zcf ' dirname '.tar.gz ']; - for i=1:numel(filelist), - compressstring = [compressstring ' ' filelist{i}]; - end - if cluster.interactive, - compressstring = [compressstring ' ' modelname '.run ' modelname '.errlog ' modelname '.outlog ']; - end - system(compressstring); - - disp('uploading input file and queueing script'); - if cluster.interactive, - directory=[cluster.executionpath '/Interactive' num2str(cluster.interactive)]; - else - directory=cluster.executionpath; - end - - if ~cluster.bbftp, - 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)% {{{ - - if ~cluster.interactive, - if ~isempty(restart) - launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && sbatch ' modelname '.queue ']; - else - launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... - ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && sbatch ' modelname '.queue ']; - end - else - if ~isempty(restart) - launchcommand=['cd ' cluster.executionpath '/Interactive' num2str(cluster.interactive)]; - else - launchcommand=['cd ' cluster.executionpath '/Interactive' num2str(cluster.interactive) ' && tar -zxf ' dirname '.tar.gz']; - end - end - - disp('launching solution sequence on remote cluster'); - issmssh(cluster.name,cluster.login,cluster.port,launchcommand); - end - %}}} - function Download(cluster,dirname,filelist)% {{{ - - %copy files from cluster to current directory - if ~cluster.interactive, - directory=[cluster.executionpath '/' dirname '/']; - else - directory=[cluster.executionpath '/Interactive' num2str(cluster.interactive) '/']; - end - - if ~cluster.bbftp, - issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); - else - issmbbftpin(cluster.name, cluster.login, cluster.port, cluster.numstreams, directory, filelist); - end - - end %}}} +classdef discover + properties (SetAccess=public) + % {{{ + name=oshostname(); + login=''; + modules = {}; + numnodes=20; + cpuspernode=8; + port=0; + queue='general'; + time=12*60*60; + processor='west'; + codepath=''; + executionpath=''; + interactive=0; + bbftp=0; + numstreams=8; + hyperthreading=0; + email=''; + end + %}}} + methods + function cluster=discover(varargin) % {{{ + + %initialize cluster using default settings if provided + if (exist('discover_settings')==2), discover_settings; end + + %use provided options to change fields + cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); + end + %}}} + function disp(cluster) % {{{ + % 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)); + disp(sprintf(' modules: %s',strjoin(cluster.modules,', '))); + 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(' time: %i',cluster.time)); + disp(sprintf(' processor: %s',cluster.processor)); + disp(sprintf(' codepath: %s',cluster.codepath)); + disp(sprintf(' executionpath: %s',cluster.executionpath)); + disp(sprintf(' interactive: %i',cluster.interactive)); + disp(sprintf(' hyperthreading: %i',cluster.hyperthreading)); + disp(sprintf(' email: %s',cluster.email)); + end + %}}} + function numprocs=nprocs(cluster) % {{{ + %compute number of processors + numprocs=cluster.numnodes*cluster.cpuspernode; + end + %}}} + function md = checkconsistency(cluster,md,solution,analyses) % {{{ + + available_queues={'long','allnccs','debug'}; + queue_requirements_time=[24*60*60 12*60*60 1*60*60]; + queue_requirements_np=[560 6000 532]; + + QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,cluster.time) + + %now, check cluster.cpuspernode according to processor type + if ( strcmpi(cluster.processor,'sand')), + if ((cluster.cpuspernode>16 ) | (cluster.cpuspernode<1)), + md = checkmessage(md,'cpuspernode should be between 1 and 16 for ''sand'' processors'); + end + elseif strcmpi(cluster.processor,'hasw'), + if ((cluster.cpuspernode>28 ) | (cluster.cpuspernode<1)), + md = checkmessage(md,'cpuspernode should be between 1 and 28 for ''hasw'' processors'); + end + else + md = checkmessage(md,'unknown processor type, should be ''sand'' or ''hasw'' '); + end + + %Miscelaneous + if isempty(cluster.login), md = checkmessage(md,'login empty'); end + if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end + if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end + + end + %}}} + function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + + if(isgprof), disp('gprof not supported by cluster, ignoring...'); end + + executable='issm.exe'; + if isdakota, + version=IssmConfig('_DAKOTA_VERSION_'); version=str2num(version(1:3)); + if (version>=6), + executable='issm_dakota.exe'; + end + end + if isoceancoupling, + executable='issm_ocean.exe'; + end + + %write queuing script + fid=fopen([modelname '.queue'],'w'); + + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#SBATCH -J %s \n',modelname); + fprintf(fid,'#SBATCH --qos=%s \n',cluster.queue); + fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); + fprintf(fid,'#SBATCH -e %s.errlog \n',modelname); + fprintf(fid,'#SBATCH -n %i \n',cluster.numnodes*cluster.cpuspernode); + fprintf(fid,'#SBATCH -N %i \n',cluster.numnodes); + fprintf(fid,'#SBATCH -t %02i:%02i:00 \n',floor(cluster.time/3600),floor(mod(cluster.time,3600)/60)); + fprintf(fid,'#SBATCH -A s1690 \n\n'); + for i=1:numel(cluster.modules), + fprintf(fid,['module load ' cluster.modules{i} '\n']); + end + if length(find(cluster.email=='@'))>0 + fprintf(fid,'#SBATCH --mail-user=%s \n',cluster.email); + fprintf(fid,'#SBATCH --mail-type=end \n\n'); + end + fprintf(fid,'. /usr/share/modules/init/bash\n\n'); + fprintf(fid,'module load comp/intel/20.0.0.166\n'); + fprintf(fid,'module load mpi/sgi-mpt/2.17\n'); + fprintf(fid,'module load cmake/3.17.0\n'); + fprintf(fid,'export MPI_GROUP_MAX=64\n\n'); + fprintf(fid,'export MPI_UNBUFFERED_STDIO=true\n\n'); + fprintf(fid,'export PATH="$PATH:."\n\n'); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); + + fprintf(fid,'mpiexec -np %i %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,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'); + if ~isvalgrind, + fprintf(fid,'mpiexec -np %i %s/%s %s %s %s\n',cluster.np,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); + end + 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 + end %}}} + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ + + %compress the files into one zip. + compressstring=['tar -zcf ' dirname '.tar.gz ']; + for i=1:numel(filelist), + compressstring = [compressstring ' ' filelist{i}]; + end + if cluster.interactive, + compressstring = [compressstring ' ' modelname '.run ' modelname '.errlog ' modelname '.outlog ']; + end + system(compressstring); + + disp('uploading input file and queueing script'); + if cluster.interactive, + directory=[cluster.executionpath '/Interactive' num2str(cluster.interactive)]; + else + directory=cluster.executionpath; + end + + if ~cluster.bbftp, + 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) % {{{ + + if ~cluster.interactive, + if ~isempty(restart) + launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && sbatch ' modelname '.queue ']; + else + launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... + ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && sbatch ' modelname '.queue ']; + end + else + if ~isempty(restart) + launchcommand=['cd ' cluster.executionpath '/Interactive' num2str(cluster.interactive)]; + else + launchcommand=['cd ' cluster.executionpath '/Interactive' num2str(cluster.interactive) ' && tar -zxf ' dirname '.tar.gz']; + end + end + + disp('launching solution sequence on remote cluster'); + issmssh(cluster.name,cluster.login,cluster.port,launchcommand); + end + %}}} + function Download(cluster,dirname,filelist) % {{{ + + %copy files from cluster to current directory + if ~cluster.interactive, + directory=[cluster.executionpath '/' dirname '/']; + else + directory=[cluster.executionpath '/Interactive' num2str(cluster.interactive) '/']; + end + + if ~cluster.bbftp, + issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); + else + issmbbftpin(cluster.name, cluster.login, cluster.port, cluster.numstreams, directory, filelist); + end + + end %}}} end end Index: /issm/trunk-jpl/src/m/classes/clusters/generic.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/generic.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/generic.m (revision 26330) @@ -286,5 +286,5 @@ end %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ if ~ispc, @@ -307,5 +307,5 @@ end end %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ if ~ispc, @@ -336,5 +336,5 @@ end %}}} - function LaunchQueueJobIceOcean(cluster,modelname,dirname,filelist,restart,batch)% {{{ + function LaunchQueueJobIceOcean(cluster,modelname,dirname,filelist,restart,batch) % {{{ if ~ispc, @@ -365,5 +365,5 @@ end %}}} - function Download(cluster,dirname,filelist)% {{{ + function Download(cluster,dirname,filelist) % {{{ if ispc(), Index: /issm/trunk-jpl/src/m/classes/clusters/generic.py =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/generic.py (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/generic.py (revision 26330) @@ -4,4 +4,5 @@ import numpy as np +from fielddisplay import fielddisplay try: from generic_settings import generic_settings @@ -39,18 +40,17 @@ self.shell = '/bin/sh' - #use provided options to change fields + # Use provided options to change fields options = pairoptions(*args) - #get name + # Get name self.name = gethostname() - #initialize cluster using user settings if provided - + # Initialize cluster using user settings if provided try: self = generic_settings(self) except NameError: - print("generic_settings.py not found, using default settings") - - #OK get other fields + print('generic_settings.py not found, using default settings') + + # OK get other fields self = options.AssignObjectFields(self) # }}} @@ -59,15 +59,15 @@ # display the object s = "class '%s' object '%s' = \n" % (type(self), 'self') - s += " name: %s\n" % self.name - s += " login: %s\n" % self.login - s += " np: %i\n" % self.np - s += " port: %i\n" % self.port - s += " codepath: %s\n" % self.codepath - s += " executionpath: %s\n" % self.executionpath - s += " valgrind: %s\n" % self.valgrind - s += " valgrindlib: %s\n" % self.valgrindlib - s += " valgrindsup: %s\n" % self.valgrindsup - s += " verbose: %s\n" % self.verbose - s += " shell: %s\n" % self.shell + s += '{}\n'.format(fielddisplay(self, 'name', '')) + s += '{}\n'.format(fielddisplay(self, 'login', '')) + s += '{}\n'.format(fielddisplay(self, 'np', '')) + s += '{}\n'.format(fielddisplay(self, 'port', '')) + s += '{}\n'.format(fielddisplay(self, 'codepath', '')) + s += '{}\n'.format(fielddisplay(self, 'executionpath', '')) + s += '{}\n'.format(fielddisplay(self, 'valgrind', '')) + s += '{}\n'.format(fielddisplay(self, 'valgrindlib', '')) + s += '{}\n'.format(fielddisplay(self, 'valgrindsup', '')) + s += '{}\n'.format(fielddisplay(self, 'verbose', '')) + s += '{}\n'.format(fielddisplay(self, 'shell', '')) return s # }}} Index: /issm/trunk-jpl/src/m/classes/clusters/greenplanet.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/greenplanet.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/greenplanet.m (revision 26330) @@ -1,3 +1,3 @@ -%PFE class definition +%GREENPLANET cluster class definition % % Usage: @@ -7,171 +7,171 @@ classdef greenplanet - properties (SetAccess=public) - % {{{ - name = 'greenplanet' - login = ''; - numnodes = 20; - cpuspernode = 8; - port = 8000; - queue = 'c6145'; - codepath = ''; - executionpath = ''; - interactive = 0; - time = 24*60; - memory = 2; - end - %}}} - methods - function cluster=greenplanet(varargin) % {{{ + properties (SetAccess=public) + % {{{ + name = 'greenplanet' + login = ''; + numnodes = 20; + cpuspernode = 8; + port = 8000; + queue = 'c6145'; + codepath = ''; + executionpath = ''; + interactive = 0; + time = 24*60; + memory = 2; + end + %}}} + methods + function cluster=greenplanet(varargin) % {{{ - %initialize cluster using default settings if provided - if (exist('greenplanet_settings')==2), greenplanet_settings; end + %initialize cluster using default settings if provided + if (exist('greenplanet_settings')==2), greenplanet_settings; end - %use provided options to change fields - cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); - end - %}}} - function disp(cluster) % {{{ - % 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)); - 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(' codepath: %s',cluster.codepath)); - disp(sprintf(' executionpath: %s',cluster.executionpath)); - disp(sprintf(' interactive: %i',cluster.interactive)); - disp(sprintf(' time: %i',cluster.time)); - disp(sprintf(' memory: %i',cluster.memory)); - end - %}}} - function numprocs=np(cluster) % {{{ - %compute number of processors - numprocs=cluster.numnodes*cluster.cpuspernode; - end - %}}} - function md = checkconsistency(cluster,md,solution,analyses) % {{{ + %use provided options to change fields + cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); + end + %}}} + function disp(cluster) % {{{ + % 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)); + 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(' codepath: %s',cluster.codepath)); + disp(sprintf(' executionpath: %s',cluster.executionpath)); + disp(sprintf(' interactive: %i',cluster.interactive)); + disp(sprintf(' time: %i',cluster.time)); + disp(sprintf(' memory: %i',cluster.memory)); + end + %}}} + function numprocs=nprocs(cluster) % {{{ + %compute number of processors + numprocs=cluster.numnodes*cluster.cpuspernode; + end + %}}} + function md = checkconsistency(cluster,md,solution,analyses) % {{{ - available_queues={'brd2.4','has2.5','ilg2.3','m-c1.9','m-c2.2','nes2.8','sib2.9','m2090','default'}; - queue_requirements_time=[Inf Inf Inf Inf Inf Inf Inf Inf Inf]; - queue_requirements_np=[80 80 80 80 80 80 80 80 80]; + available_queues={'brd2.4','has2.5','ilg2.3','m-c1.9','m-c2.2','nes2.8','sib2.9','m2090','default'}; + queue_requirements_time=[Inf Inf Inf Inf Inf Inf Inf Inf Inf]; + queue_requirements_np=[80 80 80 80 80 80 80 80 80]; - QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,1) + QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,1) - %Miscelaneous - if isempty(cluster.login), md = checkmessage(md,'login empty'); end - if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end - if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end + %Miscelaneous + if isempty(cluster.login), md = checkmessage(md,'login empty'); end + if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end + if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end - end - %}}} - function BuildKrigingQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + end + %}}} + function BuildKrigingQueueScript(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 + 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 -p %s \n',cluster.queue); - fprintf(fid,'#SBATCH -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode); - fprintf(fid,'#SBATCH --time=%i\n',cluster.time*60); %walltime is in seconds. - fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); - fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); - fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); - fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname); - if ~io_gather, %concatenate the output files: - fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); - end - fclose(fid); - end - %}}} - function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + %write queuing script + fid=fopen([modelname '.queue'],'w'); + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#SBATCH --job-name=%s\n',modelname); + fprintf(fid,'#SBATCH -p %s \n',cluster.queue); + fprintf(fid,'#SBATCH -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode); + fprintf(fid,'#SBATCH --time=%i\n',cluster.time*60); %walltime is in seconds. + fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); + fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); + fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); + fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname); + if ~io_gather, %concatenate the output files: + fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); + end + fclose(fid); + end + %}}} + 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 + 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 --partition=%s',cluster.queue{1}); - for i=2:length(cluster.queue) + %write queuing script + fid=fopen([modelname '.queue'],'w'); + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#SBATCH --job-name=%s\n',modelname); + fprintf(fid,'#SBATCH --partition=%s',cluster.queue{1}); + for i=2:length(cluster.queue) fprintf(fid,',%s',cluster.queue{i}); - end - fprintf(fid,'\n'); - %fprintf(fid,'#SBATCH -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode); - %fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); - fprintf(fid,'#SBATCH --nodes=%i\n',cluster.numnodes); % in general, just 1 - fprintf(fid,'#SBATCH --ntasks=%i\n',cluster.cpuspernode); % in general, just 1 - fprintf(fid,'#SBATCH --cpus-per-task=%i\n',1); - fprintf(fid,'#SBATCH --mem=%igb\n',cluster.memory); % minimum total node memory required - 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 -o %s.outlog \n',modelname); - fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); - fprintf(fid,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,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); + end + fprintf(fid,'\n'); + %fprintf(fid,'#SBATCH -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode); + %fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); + fprintf(fid,'#SBATCH --nodes=%i\n',cluster.numnodes); % in general, just 1 + fprintf(fid,'#SBATCH --ntasks=%i\n',cluster.cpuspernode); % in general, just 1 + fprintf(fid,'#SBATCH --cpus-per-task=%i\n',1); + fprintf(fid,'#SBATCH --mem=%igb\n',cluster.memory); % minimum total node memory required + 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 -o %s.outlog \n',modelname); + fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); + fprintf(fid,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,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,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,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 - end %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ + %in interactive mode, create a run file, and errlog and outlog file + if cluster.interactive, + fid=fopen([modelname '.run'],'w'); + fprintf(fid,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,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 + end %}}} + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ - %compress the files into one zip. - compressstring=['tar -zcf ' dirname '.tar.gz ']; - for i=1:numel(filelist), - compressstring = [compressstring ' ' filelist{i}]; - end - if cluster.interactive, - compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; - end - system(compressstring); + %compress the files into one zip. + compressstring=['tar -zcf ' dirname '.tar.gz ']; + for i=1:numel(filelist), + compressstring = [compressstring ' ' filelist{i}]; + end + if cluster.interactive, + compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; + end + system(compressstring); - disp('uploading input file and queueing script'); - issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); + disp('uploading input file and queueing script'); + issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); - end %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ + end %}}} + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ - disp('launching solution sequence on remote cluster'); - if ~isempty(restart) - launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && sbatch ' modelname '.queue ']; - else - launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... - ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && sbatch ' modelname '.queue ']; - end - issmssh(cluster.name,cluster.login,cluster.port,launchcommand); - end %}}} - function Download(cluster,dirname,filelist)% {{{ + disp('launching solution sequence on remote cluster'); + if ~isempty(restart) + launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && sbatch ' modelname '.queue ']; + else + launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... + ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && sbatch ' modelname '.queue ']; + end + issmssh(cluster.name,cluster.login,cluster.port,launchcommand); + end %}}} + function Download(cluster,dirname,filelist) % {{{ - %copy files from cluster to current directory - directory=[cluster.executionpath '/' dirname '/']; - issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); + %copy files from cluster to current directory + directory=[cluster.executionpath '/' dirname '/']; + issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); - end %}}} + end %}}} end end Index: /issm/trunk-jpl/src/m/classes/clusters/hpc.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/hpc.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/hpc.m (revision 26330) @@ -1,3 +1,3 @@ -%HPC class definition +%HPC cluster class definition % % Usage: @@ -7,159 +7,159 @@ classdef hpc - properties (SetAccess=public) - % {{{ - name='hpc.oit.uci.edu' - login=''; - numnodes=1; - cpuspernode=4; - port=0; - queue='pub64'; - codepath=''; - executionpath=''; - interactive=0; - end - %}}} - methods - function cluster=hpc(varargin) % {{{ + properties (SetAccess=public) + % {{{ + name='hpc.oit.uci.edu' + login=''; + numnodes=1; + cpuspernode=4; + port=0; + queue='pub64'; + codepath=''; + executionpath=''; + interactive=0; + end + %}}} + methods + function cluster=hpc(varargin) % {{{ - %initialize cluster using default settings if provided - if (exist('hpc_settings')==2), hpc_settings; end + %initialize cluster using default settings if provided + if (exist('hpc_settings')==2), hpc_settings; end - %use provided options to change fields - cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); - end - %}}} - function disp(cluster) % {{{ - % 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)); - 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(' codepath: %s',cluster.codepath)); - disp(sprintf(' executionpath: %s',cluster.executionpath)); - disp(sprintf(' interactive: %i',cluster.interactive)); - end - %}}} - function numprocs=np(cluster) % {{{ - %compute number of processors - numprocs=cluster.numnodes*cluster.cpuspernode; - end - %}}} - function md = checkconsistency(cluster,md,solution,analyses) % {{{ + %use provided options to change fields + cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); + end + %}}} + function disp(cluster) % {{{ + % 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)); + 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(' codepath: %s',cluster.codepath)); + disp(sprintf(' executionpath: %s',cluster.executionpath)); + disp(sprintf(' interactive: %i',cluster.interactive)); + end + %}}} + function numprocs=nprocs(cluster) % {{{ + %compute number of processors + numprocs=cluster.numnodes*cluster.cpuspernode; + end + %}}} + function md = checkconsistency(cluster,md,solution,analyses) % {{{ - available_queues={'pub64','free64','free48','free*,pub64','free*'}; - queue_requirements_time=[Inf Inf Inf Inf Inf]; - queue_requirements_np=[64 64 48 48 48]; + available_queues={'pub64','free64','free48','free*,pub64','free*'}; + queue_requirements_time=[Inf Inf Inf Inf Inf]; + queue_requirements_np=[64 64 48 48 48]; - QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,1) + QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,1) - %Miscelaneous - if isempty(cluster.login), md = checkmessage(md,'login empty'); end - if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end - if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end + %Miscelaneous + if isempty(cluster.login), md = checkmessage(md,'login empty'); end + if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end + if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end - end - %}}} - function BuildKrigingQueueScript(cluster,modelname,solution,io_gather,isvalgrind,isgprof) % {{{ + end + %}}} + function BuildKrigingQueueScript(cluster,modelname,solution,io_gather,isvalgrind,isgprof) % {{{ - if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end - if(isgprof), disp('gprof not supported by cluster, ignoring...'); end + 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,'#$ -N %s\n',modelname); - fprintf(fid,'#$ -q %s \n',cluster.queue); - fprintf(fid,'#$ -pe one-node-mpi 2-64\n'); - fprintf(fid,'#$ -R y\n'); - fprintf(fid,'#$ -m beas\n'); - fprintf(fid,'#$ -o %s.outlog \n',modelname); - fprintf(fid,'#$ -e %s.errlog \n\n',modelname); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,modelname); - fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname); - if ~io_gather, %concatenate the output files: - fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); - end - fclose(fid); - end - %}}} - function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + %write queuing script + fid=fopen([modelname '.queue'],'w'); + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#$ -N %s\n',modelname); + fprintf(fid,'#$ -q %s \n',cluster.queue); + fprintf(fid,'#$ -pe one-node-mpi 2-64\n'); + fprintf(fid,'#$ -R y\n'); + fprintf(fid,'#$ -m beas\n'); + fprintf(fid,'#$ -o %s.outlog \n',modelname); + fprintf(fid,'#$ -e %s.errlog \n\n',modelname); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,modelname); + fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname); + if ~io_gather, %concatenate the output files: + fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); + end + fclose(fid); + end + %}}} + function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ - if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end - if(isgprof), disp('gprof not supported by cluster, ignoring...'); end + if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end + if(isgprof), disp('gprof not supported by cluster, ignoring...'); end - %write queuing script - fid=fopen([modelname '.queue'],'w'); - fprintf(fid,'#!/bin/bash\n'); - fprintf(fid,'#$ -N %s\n',modelname); - fprintf(fid,'#$ -q %s \n',cluster.queue); - fprintf(fid,'#$ -pe one-node-mpi 2-64\n'); - fprintf(fid,'#$ -R y\n'); - %fprintf(fid,'#$ -m beas\n'); - fprintf(fid,'#$ -o %s.outlog \n',modelname); - fprintf(fid,'#$ -e %s.errlog \n\n',modelname); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); - fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s\n',cluster.np,cluster.codepath,solution,[cluster.executionpath '/' dirname],modelname); - if ~io_gather, %concatenate the output files: - fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); - end - fclose(fid); + %write queuing script + fid=fopen([modelname '.queue'],'w'); + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#$ -N %s\n',modelname); + fprintf(fid,'#$ -q %s \n',cluster.queue); + fprintf(fid,'#$ -pe one-node-mpi 2-64\n'); + fprintf(fid,'#$ -R y\n'); + %fprintf(fid,'#$ -m beas\n'); + fprintf(fid,'#$ -o %s.outlog \n',modelname); + fprintf(fid,'#$ -e %s.errlog \n\n',modelname); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); + fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s\n',cluster.np,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,'mpiexec -np %i %s/issm.exe %s %s %s\n',cluster.np,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 - end %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ + %in interactive mode, create a run file, and errlog and outlog file + if cluster.interactive, + fid=fopen([modelname '.run'],'w'); + fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s\n',cluster.np,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 + end %}}} + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ - %compress the files into one zip. - compressstring=['tar -zcf ' dirname '.tar.gz ']; - for i=1:numel(filelist), - compressstring = [compressstring ' ' filelist{i}]; - end - if cluster.interactive, - compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; - end - system(compressstring); + %compress the files into one zip. + compressstring=['tar -zcf ' dirname '.tar.gz ']; + for i=1:numel(filelist), + compressstring = [compressstring ' ' filelist{i}]; + end + if cluster.interactive, + compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; + end + system(compressstring); - disp('uploading input file and queueing script'); - issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); + disp('uploading input file and queueing script'); + issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); - end %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ + end %}}} + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ - disp('launching solution sequence on remote cluster'); - if ~isempty(restart) - launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && qsub ' modelname '.queue ']; - else - launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... - ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && qsub ' modelname '.queue ']; - end - issmssh(cluster.name,cluster.login,cluster.port,launchcommand); - end %}}} - function Download(cluster,dirname,filelist)% {{{ + disp('launching solution sequence on remote cluster'); + if ~isempty(restart) + launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && qsub ' modelname '.queue ']; + else + launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... + ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && qsub ' modelname '.queue ']; + end + issmssh(cluster.name,cluster.login,cluster.port,launchcommand); + end %}}} + function Download(cluster,dirname,filelist) % {{{ - %copy files from cluster to current directory - directory=[cluster.executionpath '/' dirname '/']; - issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); + %copy files from cluster to current directory + directory=[cluster.executionpath '/' dirname '/']; + issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); - end %}}} + end %}}} end end Index: /issm/trunk-jpl/src/m/classes/clusters/hpc_simba.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/hpc_simba.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/hpc_simba.m (revision 26330) @@ -1,3 +1,3 @@ -%HPC class definition +%HPC_SIMBA cluster class definition % % Usage: @@ -7,176 +7,176 @@ classdef hpc_simba - properties (SetAccess=public) - % {{{ - name='simba00' - login='inwoo'; - numnodes=1; - cpuspernode=36; - np=4; - port=7777; - queue='pub64'; - codepath='/home/DATA/externalpackagesissm/issm/trunk-jpl/bin'; - executionpath='/home/DATA/externalpackages/issm/trunk-jpl/execution/'; - interactive=0; - end - %}}} - methods - function cluster=hpc_simba(varargin) % {{{ + properties (SetAccess=public) + % {{{ + name='simba00' + login='inwoo'; + numnodes=1; + cpuspernode=36; + np=4; + port=7777; + queue='pub64'; + codepath='/home/DATA/externalpackagesissm/issm/trunk-jpl/bin'; + executionpath='/home/DATA/externalpackages/issm/trunk-jpl/execution/'; + interactive=0; + end + %}}} + methods + function cluster=hpc_simba(varargin) % {{{ - %initialize cluster using default settings if provided - if (exist('hpc_settings')==2), hpc_settings; end + %initialize cluster using default settings if provided + if (exist('hpc_settings')==2), hpc_settings; end - %use provided options to change fields - cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); - end - %}}} - function disp(cluster) % {{{ - % 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)); - 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(' codepath: %s',cluster.codepath)); - disp(sprintf(' executionpath: %s',cluster.executionpath)); - disp(sprintf(' interactive: %i',cluster.interactive)); - end - %}}} - %function numprocs=np(cluster) % {{{ - % %compute number of processors - % numprocs=cluster.numnodes*cluster.cpuspernode; + %use provided options to change fields + cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); + end + %}}} + function disp(cluster) % {{{ + % 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)); + 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(' codepath: %s',cluster.codepath)); + disp(sprintf(' executionpath: %s',cluster.executionpath)); + disp(sprintf(' interactive: %i',cluster.interactive)); + end + %}}} + %function numprocs=nprocs(cluster) % {{{ + % %compute number of processors + % numprocs=cluster.numnodes*cluster.cpuspernode; % end - %}}} - function md = checkconsistency(cluster,md,solution,analyses) % {{{ + %}}} + function md = checkconsistency(cluster,md,solution,analyses) % {{{ - available_queues={'pub64','free64','free48','free*,pub64','free*'}; - queue_requirements_time=[Inf Inf Inf Inf Inf]; - queue_requirements_np=[64 64 48 48 48]; + available_queues={'pub64','free64','free48','free*,pub64','free*'}; + queue_requirements_time=[Inf Inf Inf Inf Inf]; + queue_requirements_np=[64 64 48 48 48]; - QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,1) + QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,1) - %Miscelaneous - if isempty(cluster.login), md = checkmessage(md,'login empty'); end - if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end - if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end + %Miscelaneous + if isempty(cluster.login), md = checkmessage(md,'login empty'); end + if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end + if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end - end - %}}} - function BuildKrigingQueueScript(cluster,modelname,solution,io_gather,isvalgrind,isgprof) % {{{ + end + %}}} + function BuildKrigingQueueScript(cluster,modelname,solution,io_gather,isvalgrind,isgprof) % {{{ - if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end - if(isgprof), disp('gprof not supported by cluster, ignoring...'); end + 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,'#$ -N %s\n',modelname); - fprintf(fid,'#$ -q %s \n',cluster.queue); - fprintf(fid,'#$ -pe one-node-mpi 2-64\n'); - fprintf(fid,'#$ -R y\n'); - fprintf(fid,'#$ -m beas\n'); - fprintf(fid,'#$ -o %s.outlog \n',modelname); - fprintf(fid,'#$ -e %s.errlog \n\n',modelname); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,modelname); - fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname); - if ~io_gather, %concatenate the output files: - fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); - end - fclose(fid); - end - %}}} - function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + %write queuing script + fid=fopen([modelname '.queue'],'w'); + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#$ -N %s\n',modelname); + fprintf(fid,'#$ -q %s \n',cluster.queue); + fprintf(fid,'#$ -pe one-node-mpi 2-64\n'); + fprintf(fid,'#$ -R y\n'); + fprintf(fid,'#$ -m beas\n'); + fprintf(fid,'#$ -o %s.outlog \n',modelname); + fprintf(fid,'#$ -e %s.errlog \n\n',modelname); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,modelname); + fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname); + if ~io_gather, %concatenate the output files: + fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); + end + fclose(fid); + end + %}}} + function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ - if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end - if(isgprof), disp('gprof not supported by cluster, ignoring...'); end + if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end + if(isgprof), disp('gprof not supported by cluster, ignoring...'); end - %write queuing script - fid=fopen([modelname '.queue'],'w'); - if 0 - fprintf(fid,'#!/bin/bash\n'); - fprintf(fid,'#$ -N %s\n',modelname); - fprintf(fid,'#$ -q %s \n',cluster.queue); - fprintf(fid,'#$ -pe one-node-mpi 2-64\n'); - fprintf(fid,'#$ -R y\n'); - %fprintf(fid,'#$ -m beas\n'); - fprintf(fid,'#$ -o %s.outlog \n',modelname); - fprintf(fid,'#$ -e %s.errlog \n\n',modelname); - else % FIXME : This script for SIMBA-Kopri - fprintf(fid,'#!/bin/bash\n'); - %fprintf(fid,'#PBS -q workq\n'); - fprintf(fid,'#PBS -N %s\n',modelname); - %fprintf(fid,'#PBS -l nodes=simba01:ppn=%d\n',... - % ceil(cluster.np/cluster.cpuspernode), cluster.np); - fprintf(fid,'#PBS -l nodes=simba01:ppn=%d\n',... + %write queuing script + fid=fopen([modelname '.queue'],'w'); + if 0 + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#$ -N %s\n',modelname); + fprintf(fid,'#$ -q %s \n',cluster.queue); + fprintf(fid,'#$ -pe one-node-mpi 2-64\n'); + fprintf(fid,'#$ -R y\n'); + %fprintf(fid,'#$ -m beas\n'); + fprintf(fid,'#$ -o %s.outlog \n',modelname); + fprintf(fid,'#$ -e %s.errlog \n\n',modelname); + else % FIXME : This script for SIMBA-Kopri + fprintf(fid,'#!/bin/bash\n'); + %fprintf(fid,'#PBS -q workq\n'); + fprintf(fid,'#PBS -N %s\n',modelname); + %fprintf(fid,'#PBS -l nodes=simba01:ppn=%d\n',... + % ceil(cluster.np/cluster.cpuspernode), cluster.np); + fprintf(fid,'#PBS -l nodes=simba01:ppn=%d\n',... cluster.np); - fprintf(fid,'#PBS -o %s.outlog \n',modelname); - fprintf(fid,'#PBS -e %s.errlog \n\n',modelname); - fprintf(fid,'\n'); - end - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'\n'); - fprintf(fid,'module load intel18/impi-18\n'); - fprintf(fid,'\n'); - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); - fprintf(fid,'mpirun -np %i %s/issm.exe %s %s %s\n',cluster.np,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); + fprintf(fid,'#PBS -o %s.outlog \n',modelname); + fprintf(fid,'#PBS -e %s.errlog \n\n',modelname); + fprintf(fid,'\n'); + end + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'\n'); + fprintf(fid,'module load intel18/impi-18\n'); + fprintf(fid,'\n'); + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); + fprintf(fid,'mpirun -np %i %s/issm.exe %s %s %s\n',cluster.np,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,'mpiexec -np %i %s/issm.exe %s %s %s\n',cluster.np,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 - end %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ + %in interactive mode, create a run file, and errlog and outlog file + if cluster.interactive, + fid=fopen([modelname '.run'],'w'); + fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s\n',cluster.np,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 + end %}}} + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ - %compress the files into one zip. - compressstring=['tar -zcf ' dirname '.tar.gz ']; - for i=1:numel(filelist), - compressstring = [compressstring ' ' filelist{i}]; - end - if cluster.interactive, - compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; - end - system(compressstring); + %compress the files into one zip. + compressstring=['tar -zcf ' dirname '.tar.gz ']; + for i=1:numel(filelist), + compressstring = [compressstring ' ' filelist{i}]; + end + if cluster.interactive, + compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; + end + system(compressstring); - disp('uploading input file and queueing script'); - issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); + disp('uploading input file and queueing script'); + issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); - end %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ + end %}}} + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ - disp('launching solution sequence on remote cluster'); - if ~isempty(restart) - launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && qsub ' modelname '.queue ']; - else - launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... - ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && qsub ' modelname '.queue ']; - end - issmssh(cluster.name,cluster.login,cluster.port,launchcommand); - end %}}} - function Download(cluster,dirname,filelist)% {{{ + disp('launching solution sequence on remote cluster'); + if ~isempty(restart) + launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && qsub ' modelname '.queue ']; + else + launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... + ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && qsub ' modelname '.queue ']; + end + issmssh(cluster.name,cluster.login,cluster.port,launchcommand); + end %}}} + function Download(cluster,dirname,filelist) % {{{ - %copy files from cluster to current directory - directory=[cluster.executionpath '/' dirname '/']; - issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); + %copy files from cluster to current directory + directory=[cluster.executionpath '/' dirname '/']; + issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); - end %}}} + end %}}} end end Index: /issm/trunk-jpl/src/m/classes/clusters/lonestar.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/lonestar.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/lonestar.m (revision 26330) @@ -1,3 +1,3 @@ -%LONESTAR class definition +%LONESTAR cluster class definition % % Usage: @@ -7,194 +7,194 @@ classdef lonestar - properties (SetAccess=public) - % {{{ - name = 'ls5.tacc.utexas.edu' - login = ''; - modules = {'intel/18.0.2' 'gsl'}; - numnodes = 1; - cpuspernode = 24; - port = 1099; - queue = 'normal'; - codepath = ''; - executionpath = ''; - interactive = 0; - time = 48*60*60; - email = ''; - end - %}}} - methods - function cluster=lonestar(varargin) % {{{ - - %initialize cluster using default settings if provided - if (exist('lonestar_settings')==2), lonestar_settings; end - - %use provided options to change fields - cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); - - end - %}}} - function disp(cluster) % {{{ - % 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)); - disp(sprintf(' modules: %s',strjoin(cluster.modules,', '))); - disp(sprintf(' port: %i',cluster.port)); - disp(sprintf(' numnodes: %i',cluster.numnodes)); - disp(sprintf(' cpuspernode: %i',cluster.cpuspernode)); - disp(sprintf(' np: %i',cluster.np)); - disp(sprintf(' queue: %s',cluster.queue)); - disp(sprintf(' codepath: %s',cluster.codepath)); - disp(sprintf(' executionpath: %s',cluster.executionpath)); - disp(sprintf(' interactive: %i',cluster.interactive)); - disp(sprintf(' time: %i',cluster.time)); - disp(sprintf(' email: %s',cluster.email)); - end - %}}} - function numprocs=np(cluster) % {{{ - %compute number of processors - numprocs=cluster.numnodes*cluster.cpuspernode; - end - %}}} - function md = checkconsistency(cluster,md,solution,analyses) % {{{ - - available_queues={'normal','development'}; - queue_requirements_time=[48*60*60 2*60*60]; - queue_requirements_np=[4104 264]; - - QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,cluster.time) - - %Miscelaneous - if isempty(cluster.login), md = checkmessage(md,'login empty'); end - if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end - if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end - end - %}}} - function BuildKrigingQueueScript(cluster,modelname,solution,io_gather,isvalgrind,isgprof) % {{{ - - 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,'#$ -N %s\n',modelname); - fprintf(fid,'#$ -q %s \n',cluster.queue); - fprintf(fid,'#$ -pe one-node-mpi 2-64\n'); - fprintf(fid,'#$ -R y\n'); - fprintf(fid,'#$ -m beas\n'); - fprintf(fid,'#$ -o %s.outlog \n',modelname); - fprintf(fid,'#$ -e %s.errlog \n\n',modelname); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,modelname); - fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname); - if ~io_gather, %concatenate the output files: - fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); - end - fclose(fid); - end - %}}} - function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ - - if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end - if(isgprof), disp('gprof not supported by cluster, ignoring...'); end - - executable='issm.exe'; - if isdakota, - version=IssmConfig('_DAKOTA_VERSION_'); version=str2num(version(1:3)); - if (version>=6), - executable='issm_dakota.exe'; - end - end - if isoceancoupling, - executable='issm_ocean.exe'; - end - - %write queuing script - fid=fopen([modelname '.queue'],'w'); - - fprintf(fid,'#!/bin/bash\n'); - fprintf(fid,'#SBATCH -J %s \n',modelname); - fprintf(fid,'#SBATCH -p %s \n',cluster.queue); - fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); - fprintf(fid,'#SBATCH -e %s.errlog \n',modelname); - fprintf(fid,'#SBATCH -n %i \n',cluster.numnodes*max(cluster.np/cluster.numnodes,24)); - fprintf(fid,'#SBATCH -N %i \n',cluster.numnodes); - fprintf(fid,'#SBATCH -t %02i:%02i:00 \n\n',floor(cluster.time/3600),floor(mod(cluster.time,3600)/60)); - for i=1:numel(cluster.modules), - fprintf(fid,['module load ' cluster.modules{i} '\n']); - end - - if isdakota, - fprintf(fid,'export KMP_AFFINITY="granularity=fine,compact,verbose" \n\n'); - end - - if length(find(cluster.email=='@'))>0 - fprintf(fid,'#SBATCH --mail-user=%s \n',cluster.email); - fprintf(fid,'#SBATCH --mail-type=end \n\n'); - - %fprintf(fid,'ssh login1 "mail -s ''SLURM Jobid=${SLURM_JOBID} Name=${SLURM_JOB_NAME} Began on Lonestar 5.'' %s <<< ''Job Started'' " \n\n',cluster.email); - end - - fprintf(fid,'export PATH="$PATH:."\n\n'); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); - fprintf(fid,'ibrun -np %i %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,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,'ibrun -np %i %s/%s %s %s %s\n',cluster.np,executable,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 - end %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ - - %compress the files into one zip. - compressstring=['tar -zcf ' dirname '.tar.gz ']; - for i=1:numel(filelist), - compressstring = [compressstring ' ' filelist{i}]; - end - if cluster.interactive, - compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; - end - system(compressstring); - - disp('uploading input file and queueing script'); - issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); - - end %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ - - disp('launching solution sequence on remote cluster'); - if ~isempty(restart) - launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && sbatch ' modelname '.queue ']; - else - launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... - ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && sbatch ' modelname '.queue ']; - end - issmssh(cluster.name,cluster.login,cluster.port,launchcommand); - end %}}} - function Download(cluster,dirname,filelist)% {{{ - - %copy files from cluster to current directory - directory=[cluster.executionpath '/' dirname '/']; - issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); - - end %}}} + properties (SetAccess=public) + % {{{ + name = 'ls5.tacc.utexas.edu' + login = ''; + modules = {'intel/18.0.2' 'gsl'}; + numnodes = 1; + cpuspernode = 24; + port = 1099; + queue = 'normal'; + codepath = ''; + executionpath = ''; + interactive = 0; + time = 48*60*60; + email = ''; + end + %}}} + methods + function cluster=lonestar(varargin) % {{{ + + %initialize cluster using default settings if provided + if (exist('lonestar_settings')==2), lonestar_settings; end + + %use provided options to change fields + cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); + + end + %}}} + function disp(cluster) % {{{ + % 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)); + disp(sprintf(' modules: %s',strjoin(cluster.modules,', '))); + disp(sprintf(' port: %i',cluster.port)); + disp(sprintf(' numnodes: %i',cluster.numnodes)); + disp(sprintf(' cpuspernode: %i',cluster.cpuspernode)); + disp(sprintf(' np: %i',cluster.np)); + disp(sprintf(' queue: %s',cluster.queue)); + disp(sprintf(' codepath: %s',cluster.codepath)); + disp(sprintf(' executionpath: %s',cluster.executionpath)); + disp(sprintf(' interactive: %i',cluster.interactive)); + disp(sprintf(' time: %i',cluster.time)); + disp(sprintf(' email: %s',cluster.email)); + end + %}}} + function numprocs=nprocs(cluster) % {{{ + %compute number of processors + numprocs=cluster.numnodes*cluster.cpuspernode; + end + %}}} + function md = checkconsistency(cluster,md,solution,analyses) % {{{ + + available_queues={'normal','development'}; + queue_requirements_time=[48*60*60 2*60*60]; + queue_requirements_np=[4104 264]; + + QueueRequirements(available_queues,queue_requirements_time,queue_requirements_np,cluster.queue,cluster.np,cluster.time) + + %Miscelaneous + if isempty(cluster.login), md = checkmessage(md,'login empty'); end + if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end + if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end + end + %}}} + function BuildKrigingQueueScript(cluster,modelname,solution,io_gather,isvalgrind,isgprof) % {{{ + + 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,'#$ -N %s\n',modelname); + fprintf(fid,'#$ -q %s \n',cluster.queue); + fprintf(fid,'#$ -pe one-node-mpi 2-64\n'); + fprintf(fid,'#$ -R y\n'); + fprintf(fid,'#$ -m beas\n'); + fprintf(fid,'#$ -o %s.outlog \n',modelname); + fprintf(fid,'#$ -e %s.errlog \n\n',modelname); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,modelname); + fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname); + if ~io_gather, %concatenate the output files: + fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); + end + fclose(fid); + end + %}}} + function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + + if(isvalgrind), disp('valgrind not supported by cluster, ignoring...'); end + if(isgprof), disp('gprof not supported by cluster, ignoring...'); end + + executable='issm.exe'; + if isdakota, + version=IssmConfig('_DAKOTA_VERSION_'); version=str2num(version(1:3)); + if (version>=6), + executable='issm_dakota.exe'; + end + end + if isoceancoupling, + executable='issm_ocean.exe'; + end + + %write queuing script + fid=fopen([modelname '.queue'],'w'); + + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#SBATCH -J %s \n',modelname); + fprintf(fid,'#SBATCH -p %s \n',cluster.queue); + fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); + fprintf(fid,'#SBATCH -e %s.errlog \n',modelname); + fprintf(fid,'#SBATCH -n %i \n',cluster.numnodes*max(cluster.np/cluster.numnodes,24)); + fprintf(fid,'#SBATCH -N %i \n',cluster.numnodes); + fprintf(fid,'#SBATCH -t %02i:%02i:00 \n\n',floor(cluster.time/3600),floor(mod(cluster.time,3600)/60)); + for i=1:numel(cluster.modules), + fprintf(fid,['module load ' cluster.modules{i} '\n']); + end + + if isdakota, + fprintf(fid,'export KMP_AFFINITY="granularity=fine,compact,verbose" \n\n'); + end + + if length(find(cluster.email=='@'))>0 + fprintf(fid,'#SBATCH --mail-user=%s \n',cluster.email); + fprintf(fid,'#SBATCH --mail-type=end \n\n'); + + %fprintf(fid,'ssh login1 "mail -s ''SLURM Jobid=${SLURM_JOBID} Name=${SLURM_JOB_NAME} Began on Lonestar 5.'' %s <<< ''Job Started'' " \n\n',cluster.email); + end + + fprintf(fid,'export PATH="$PATH:."\n\n'); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); + fprintf(fid,'ibrun -np %i %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,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,'ibrun -np %i %s/%s %s %s %s\n',cluster.np,executable,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 + end %}}} + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ + + %compress the files into one zip. + compressstring=['tar -zcf ' dirname '.tar.gz ']; + for i=1:numel(filelist), + compressstring = [compressstring ' ' filelist{i}]; + end + if cluster.interactive, + compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; + end + system(compressstring); + + disp('uploading input file and queueing script'); + issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); + + end %}}} + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ + + disp('launching solution sequence on remote cluster'); + if ~isempty(restart) + launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && sbatch ' modelname '.queue ']; + else + launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... + ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && sbatch ' modelname '.queue ']; + end + issmssh(cluster.name,cluster.login,cluster.port,launchcommand); + end %}}} + function Download(cluster,dirname,filelist) % {{{ + + %copy files from cluster to current directory + directory=[cluster.executionpath '/' dirname '/']; + issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); + + end %}}} end end Index: /issm/trunk-jpl/src/m/classes/clusters/maui.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/maui.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/maui.m (revision 26330) @@ -1,3 +1,3 @@ -%MAUI class definition +%MAUI cluster class definition % % Usage: @@ -7,146 +7,146 @@ classdef maui - properties (SetAccess=public) - % {{{ - name = 'maui' - login = ''; - numnodes = 1; - cpuspernode = 8; - port = 0; - projectaccount = ''; - partition = 'nesi_research'; - codepath = ''; - executionpath = ''; - interactive = 0; - time = 24*60; - memory = 2; - end - %}}} - methods - function cluster=maui(varargin) % {{{ + properties (SetAccess=public) + % {{{ + name = 'maui' + login = ''; + numnodes = 1; + cpuspernode = 8; + port = 0; + projectaccount = ''; + partition = 'nesi_research'; + codepath = ''; + executionpath = ''; + interactive = 0; + time = 24*60; + memory = 2; + end + %}}} + methods + function cluster=maui(varargin) % {{{ - %initialize cluster using default settings if provided - if (exist('maui_settings')==2), maui_settings; end + %initialize cluster using default settings if provided + if (exist('maui_settings')==2), maui_settings; end - %use provided options to change fields - cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); - end - %}}} - function disp(cluster) % {{{ - % 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)); - 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(' projectaccount: %s',cluster.projectaccount)); - disp(sprintf(' partition: %s',cluster.partition)); - disp(sprintf(' codepath: %s',cluster.codepath)); - disp(sprintf(' executionpath: %s',cluster.executionpath)); - disp(sprintf(' interactive: %i',cluster.interactive)); - disp(sprintf(' time: %i',cluster.time)); - disp(sprintf(' memory: %i',cluster.memory)); - end - %}}} - function numprocs=np(cluster) % {{{ - %compute number of processors - numprocs=cluster.numnodes*cluster.cpuspernode; - end - %}}} - function md = checkconsistency(cluster,md,solution,analyses) % {{{ + %use provided options to change fields + cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); + end + %}}} + function disp(cluster) % {{{ + % 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)); + 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(' projectaccount: %s',cluster.projectaccount)); + disp(sprintf(' partition: %s',cluster.partition)); + disp(sprintf(' codepath: %s',cluster.codepath)); + disp(sprintf(' executionpath: %s',cluster.executionpath)); + disp(sprintf(' interactive: %i',cluster.interactive)); + disp(sprintf(' time: %i',cluster.time)); + disp(sprintf(' memory: %i',cluster.memory)); + end + %}}} + function numprocs=nprocs(cluster) % {{{ + %compute number of processors + numprocs=cluster.numnodes*cluster.cpuspernode; + end + %}}} + function md = checkconsistency(cluster,md,solution,analyses) % {{{ - available_partitions={'nesi_research'} - partition_requirements_time=[24*60]; - partition_requirements_np=[80]; + available_partitions={'nesi_research'} + partition_requirements_time=[24*60]; + partition_requirements_np=[80]; - QueueRequirements(available_partitions,partition_requirements_time,partition_requirements_np,cluster.partition,cluster.np,1) + QueueRequirements(available_partitions,partition_requirements_time,partition_requirements_np,cluster.partition,cluster.np,1) - %Miscelaneous - if isempty(cluster.login), md = checkmessage(md,'login empty'); end - if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end - if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end - end - %}}} - function BuildKrigingQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ - error('not implemented yet'); - end - %}}} - function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + %Miscelaneous + if isempty(cluster.login), md = checkmessage(md,'login empty'); end + if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end + if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end + end + %}}} + function BuildKrigingQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + error('not implemented yet'); + end + %}}} + 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 + 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 --account=%s \n',cluster.projectaccount); - fprintf(fid,'#SBATCH --partition=%s \n',cluster.partition); - fprintf(fid,'#SBATCH --ntasks=%i \n',cluster.numnodes); - fprintf(fid,'#SBATCH --cpus-per-task=%i\n',cluster.cpuspernode); - fprintf(fid,'#SBATCH --time=%i\n',cluster.time); %walltime is in minutes - fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); - fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); - fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'module swap PrgEnv-cray PrgEnv-intel\n'); - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); - fprintf(fid,'srun -n %i %s/issm.exe %s %s %s\n',cluster.np,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); + %write queuing script + fid=fopen([modelname '.queue'],'w'); + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#SBATCH --job-name=%s\n',modelname); + fprintf(fid,'#SBATCH --account=%s \n',cluster.projectaccount); + fprintf(fid,'#SBATCH --partition=%s \n',cluster.partition); + fprintf(fid,'#SBATCH --ntasks=%i \n',cluster.numnodes); + fprintf(fid,'#SBATCH --cpus-per-task=%i\n',cluster.cpuspernode); + fprintf(fid,'#SBATCH --time=%i\n',cluster.time); %walltime is in minutes + fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); + fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); + fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'module swap PrgEnv-cray PrgEnv-intel\n'); + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); + fprintf(fid,'srun -n %i %s/issm.exe %s %s %s\n',cluster.np,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,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,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 - end %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ + %in interactive mode, create a run file, and errlog and outlog file + if cluster.interactive, + fid=fopen([modelname '.run'],'w'); + fprintf(fid,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,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 + end %}}} + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ - %compress the files into one zip. - compressstring=['tar -zcf ' dirname '.tar.gz ']; - for i=1:numel(filelist), - compressstring = [compressstring ' ' filelist{i}]; - end - if cluster.interactive, - compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; - end - system(compressstring); + %compress the files into one zip. + compressstring=['tar -zcf ' dirname '.tar.gz ']; + for i=1:numel(filelist), + compressstring = [compressstring ' ' filelist{i}]; + end + if cluster.interactive, + compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; + end + system(compressstring); - disp('uploading input file and queueing script'); - issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); + disp('uploading input file and queueing script'); + issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); - end %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ + end %}}} + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ - disp('launching solution sequence on remote cluster'); - if ~isempty(restart) - launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && sbatch ' modelname '.queue ']; - else - launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... - ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && sbatch ' modelname '.queue ']; - end - issmssh(cluster.name,cluster.login,cluster.port,launchcommand); - end %}}} - function Download(cluster,dirname,filelist)% {{{ + disp('launching solution sequence on remote cluster'); + if ~isempty(restart) + launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && sbatch ' modelname '.queue ']; + else + launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... + ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && sbatch ' modelname '.queue ']; + end + issmssh(cluster.name,cluster.login,cluster.port,launchcommand); + end %}}} + function Download(cluster,dirname,filelist) % {{{ - %copy files from cluster to current directory - directory=[cluster.executionpath '/' dirname '/']; - issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); + %copy files from cluster to current directory + directory=[cluster.executionpath '/' dirname '/']; + issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); - end %}}} + end %}}} end end Index: /issm/trunk-jpl/src/m/classes/clusters/pfe.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/pfe.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/pfe.m (revision 26330) @@ -1,3 +1,3 @@ -%PFE class definition +%PFE cluster class definition % % Usage: @@ -7,5 +7,5 @@ classdef pfe - properties (SetAccess=public) + properties (SetAccess=public) % {{{ name = 'pfe' @@ -39,23 +39,29 @@ %}}} function disp(cluster) % {{{ - % display the object + % TODO: + % - Correct formatting of modules and cluster.np + % + + % 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(' 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(' time: %i',cluster.time)); - disp(sprintf(' processor: %s',cluster.processor)); - disp(sprintf(' srcpath: %s ($ISSM_DIR on pfe)',cluster.srcpath)); - disp(sprintf(' codepath: %s ($ISSM_DIR/bin on pfe)',cluster.codepath)); - disp(sprintf(' executionpath: %s (directory containing issm.exe on pfe)',cluster.executionpath)); - disp(sprintf(' grouplist: %s',cluster.grouplist)); - disp(sprintf(' interactive: %i',cluster.interactive)); - disp(sprintf(' hyperthreading: %i',cluster.hyperthreading)); + fielddisplay(cluster,'name','name of the cluster'); + fielddisplay(cluster,'login','login'); + modules=''; for i=1:length(cluster.modules), modules=[modules cluster.modules{i} ',']; end; modules=modules(1:end-1); + disp(sprintf(' modules : %s',modules)); + fielddisplay(cluster,'port','machine access port'); + fielddisplay(cluster,'numnodes','number of nodes'); + fielddisplay(cluster,'cpuspernode','number of CPUs per node'); + disp(sprintf(' np: : %i -- %s',cluster.np(),'number of CPUs')); + fielddisplay(cluster,'queue','name of the queue'); + fielddisplay(cluster,'time','walltime requested'); + fielddisplay(cluster,'processor','type of processor'); + fielddisplay(cluster,'srcpath','$ISSM_DIR on pfe'); + fielddisplay(cluster,'codepath','$ISSM_DIR/bin on pfe'); + fielddisplay(cluster,'executionpath','directory containing issm.exe on pfe'); + fielddisplay(cluster,'grouplist','grouplist'); + fielddisplay(cluster,'interactive',''); + fielddisplay(cluster,'bbftp',''); + fielddisplay(cluster,'numstreams',''); + fielddisplay(cluster,'hyperthreading',''); end %}}} @@ -128,5 +134,5 @@ 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 @@ -137,4 +143,5 @@ if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end if isempty(cluster.grouplist), md = checkmessage(md,'grouplist empty'); end + if ~isempty(cluster.interactive), md = checkmessage(md,'interactive mode not implemented'); end end @@ -164,6 +171,6 @@ 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 @@ -176,5 +183,5 @@ 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); + 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 %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); @@ -382,8 +389,8 @@ 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}]; @@ -411,5 +418,5 @@ end %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ %lauch command, to be executed via ssh @@ -437,5 +444,5 @@ end %}}} - function Download(cluster,dirname,filelist)% {{{ + function Download(cluster,dirname,filelist) % {{{ %copy files from cluster to current directory Index: /issm/trunk-jpl/src/m/classes/clusters/pfe.py =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/pfe.py (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/pfe.py (revision 26330) @@ -1,29 +1,29 @@ import subprocess + from fielddisplay import fielddisplay -from pairoptions import pairoptions -from issmssh import issmssh +from IssmConfig import IssmConfig from issmscpin import issmscpin from issmscpout import issmscpout -from QueueRequirements import QueueRequirements -from IssmConfig import IssmConfig +from issmssh import issmssh +from pairoptions import pairoptions try: from pfe_settings import pfe_settings except ImportError: print('You need pfe_settings.py to proceed, check presence and sys.path') - +from QueueRequirements import QueueRequirements class pfe(object): + """PFE cluster class definition + + Usage: + cluster = pfe() + cluster = pfe('np', 3) + cluster = pfe('np', 3, 'login', 'username') """ - PFE cluster class definition - - Usage: - cluster = pfe() - cluster = pfe('np', 3) - cluster = pfe('np', 3, 'login', 'username') - """ - - def __init__(self, *args): # {{{ + + def __init__(self, *args): # {{{ self.name = 'pfe' self.login = '' + self.modules = ['comp-intel/2016.2.181', 'mpi-sgi/mpt'] self.numnodes = 20 self.cpuspernode = 8 @@ -31,9 +31,9 @@ self.queue = 'long' self.time = 12 * 60 - self.processor = 'bro' + self.processor = 'ivy' self.srcpath = '' self.codepath = '' self.executionpath = '' - self.grouplist = 's1010' + self.grouplist = 's1690' self.interactive = 0 self.bbftp = 0 @@ -41,44 +41,54 @@ self.hyperthreading = 0 - #use provided options to change fields + # Use provided options to change fields options = pairoptions(*args) - #initialize cluster using user settings if provided - self = pfe_settings(self) - self.np = self.nprocs() + # Initialize cluster using user settings if provided + try: + self = pfe_settings(self) + except NameError: + print('pfe_settings.py not found, using default settings') #OK get other fields self = options.AssignObjectFields(self) - - # }}} - - def __repr__(self): # {{{ - # display the object - s = "class pfe object:" - s = "%s\n%s" % (s, fielddisplay(self, 'name', 'name of the cluster')) - s = "%s\n%s" % (s, fielddisplay(self, 'login', 'login')) - s = "%s\n%s" % (s, fielddisplay(self, 'numnodes', 'number of nodes')) - s = "%s\n%s" % (s, fielddisplay(self, 'cpuspernode', 'number of nodes per CPUs')) - s = "%s\n%s" % (s, fielddisplay(self, 'np', 'number of CPUs')) - s = "%s\n%s" % (s, fielddisplay(self, 'port', 'machine access port')) - s = "%s\n%s" % (s, fielddisplay(self, 'queue', 'name of the queue')) - s = "%s\n%s" % (s, fielddisplay(self, 'time', 'walltime requested')) - s = "%s\n%s" % (s, fielddisplay(self, 'processor', 'type of processor')) - s = "%s\n%s" % (s, fielddisplay(self, 'codepath', '$ISSM_DIR on pfe')) - s = "%s\n%s" % (s, fielddisplay(self, 'executionpath', 'directory containing issm.exe on pfe')) - s = "%s\n%s" % (s, fielddisplay(self, 'grouplist', 'name of the group')) - s = "%s\n%s" % (s, fielddisplay(self, 'interactive', '')) - s = "%s\n%s" % (s, fielddisplay(self, 'bbftp', '')) - s = "%s\n%s" % (s, fielddisplay(self, 'numstreams', '')) - s = "%s\n%s" % (s, fielddisplay(self, 'hyperthreading', '')) + # }}} + + def __repr__(self): # {{{ + # TODO: + # - Correct formatting of modules and cluster.np + # + + # Display the object + s = 'class pfe object\n' + s += '{}\n'.format(fielddisplay(self, 'name', 'name of the cluster')) + s += '{}\n'.format(fielddisplay(self, 'login', 'login')) + modules = '' + for i in range(len(self.modules)): + modules += self.modules[i] + ',' + modules = modules[:-1] + s += ' modules : {}\n'.format(modules) + s += '{}\n'.format(fielddisplay(self, 'port', 'machine access port')) + s += '{}\n'.format(fielddisplay(self, 'numnodes', 'number of nodes')) + s += '{}\n'.format(fielddisplay(self, 'cpuspernode', 'number of CPUs per node')) + s += ' np: : {} -- {}\n'.format(self.nprocs(), 'number of CPUs') + s += '{}\n'.format(fielddisplay(self, 'queue', 'name of the queue')) + s += '{}\n'.format(fielddisplay(self, 'time', 'walltime requested')) + s += '{}\n'.format(fielddisplay(self, 'processor', 'type of processor')) + s += '{}\n'.format(fielddisplay(self, 'srcpath', '$ISSM_DIR on pfe')) + s += '{}\n'.format(fielddisplay(self, 'codepath', '$ISSM_DIR/bin on pfe')) + s += '{}\n'.format(fielddisplay(self, 'executionpath', 'directory containing issm.exe on pfe')) + s += '{}\n'.format(fielddisplay(self, 'grouplist', 'grouplist')) + s += '{}\n'.format(fielddisplay(self, 'interactive', '')) + s += '{}\n'.format(fielddisplay(self, 'bbftp', '')) + s += '{}\n'.format(fielddisplay(self, 'numstreams', '')) + s += '{}\n'.format(fielddisplay(self, 'hyperthreading', '')) return s # }}} - def nprocs(self): # {{{ - self.np = self.numnodes * self.cpuspernode - return self.np - # }}} - - def checkconsistency(self, md, solution, analyses): # {{{ + def nprocs(self): # {{{ + return self.numnodes * self.cpuspernode + # }}} + + def checkconsistency(self, md, solution, analyses): # {{{ queuedict = {'long': [5 * 24 * 60, 2048], 'normal': [8 * 60, 2048], @@ -87,32 +97,30 @@ QueueRequirements(queuedict, self.queue, self.nprocs(), self.time) - #now, check cluster.cpuspernode according to processor type - if self.processor == 'har' or self.processor == 'neh': + # Now, check cluster.cpuspernode according to processor type + if self.processor == 'ivy': if self.hyperthreading: - if not 0 < self.cpuspernode < 17: + if self.cpuspernode > 40 or self.cpuspernode < 1: + md = md.checkmessage('cpuspernode should be between 1 and 40 for ''ivy'' processors in hyperthreading mode') + else: + if self.cpuspernode > 20 or self.cpuspernode < 1: + md = md.checkmessage('cpuspernode should be between 1 and 20 for ''ivy'' processors') + elif self.processor == 'bro': + if self.hyperthreading: + if self.cpuspernode > 56 or self.cpuspernode < 1: + md = md.checkmessage('cpuspernode should be between 1 and 56 for ''bro'' processors in hyperthreading mode') + else: + if self.cpuspernode > 28 or self.cpuspernode < 1: + md = md.checkmessage('cpuspernode should be between 1 and 28 for ''bro'' processors') + elif self.processor == 'har' or self.processor == 'neh': + if self.hyperthreading: + if self.cpuspernode > 16 or self.cpuspernode < 1: md = md.checkmessage('cpuspernode should be between 1 and 16 for ''neh'' and ''har'' processors in hyperthreading mode') else: - if not 0 < self.cpuspernode < 9: + if self.cpuspernode > 8 or self.cpuspernode < 1: md = md.checkmessage('cpuspernode should be between 1 and 8 for ''neh'' and ''har'' processors') - - elif self.processor == 'bro': - if self.hyperthreading: - if not 0 < self.cpuspernode < 57: - md = md.checkmessage('cpuspernode should be between 1 and 56 for ''bro'' processors in hyperthreading mode') - else: - if not 0 < self.cpuspernode < 29: - md = md.checkmessage('cpuspernode should be between 1 and 28 for ''bro'' processors') - - elif self.processor == 'ivy': - if self.hyperthreading: - if not 0 < self.cpuspernode < 41: - md = md.checkmessage('cpuspernode should be between 1 and 40 for ''ivy'' processors in hyperthreading mode') - else: - if not 0 < self.cpuspernode < 21: - md = md.checkmessage('cpuspernode should be between 1 and 20 for ''ivy'' processors') else: - md = md.checkmessage('unknown processor type, should be ''neh'', ''bro'' or ''har'' or ''ivy''') - - #Miscellaneous + md = md.checkmessage('unknown processor type, should be \'ivy\', \'bro\', \'har\', or \'neh\'') + + # Miscellaneous if not self.login: md = md.checkmessage('login empty') @@ -131,5 +139,8 @@ # }}} - def BuildQueueScript(self, dirname, modelname, solution, io_gather, isvalgrind, isgprof, isdakota, isoceancoupling): # {{{ + def BuildQueueScript(self, dirname, modelname, solution, io_gather, isvalgrind, isgprof, isdakota, isoceancoupling): # {{{ + if isgprof: + print('gprof not supported by cluster, ignoring...') + executable = 'issm.exe' if isdakota: @@ -141,24 +152,25 @@ executable = 'issm_ocean.exe' - #write queuing script + # Write queuing script fid = open(modelname + '.queue', 'w') - fid.write('#PBS -S / bin / bash\n') - fid.write('#PBS -l select=%i:ncpus=%i:model=%s\n' % (self.numnodes, self.cpuspernode, self.processor)) - fid.write('#PBS -l walltime=%i\n' % (self.time * 60)) - fid.write('#PBS -q %s \n' % self.queue) - fid.write('#PBS -W group_list=%s\n' % self.grouplist) + fid.write('#PBS -S /bin/bash\n') + fid.write('#PBS -l select={}:ncpus={}:model={}\n'.format(self.numnodes, self.cpuspernode, self.processor)) + fid.write('#PBS -l walltime={}\n'.format(self.time * 60)) # walltime is in seconds + fid.write('#PBS -q {} \n'.format(self.queue)) + fid.write('#PBS -W group_list={}\n'.format(self.grouplist)) fid.write('#PBS -m e\n') - fid.write('#PBS -o %s/%s/%s.outlog \n' % (self.executionpath, dirname, modelname)) - fid.write('#PBS -e %s/%s/%s.errlog \n\n' % (self.executionpath, dirname, modelname)) - fid.write('./usr/share/modules/init/bash\n\n') - fid.write('module load comp-intel/2015.0.090\n') - fid.write('module load mpi-sgi/mpt.2.11r13\n') + fid.write('#PBS -o {}/{}/{}.outlog \n'.format(self.executionpath, dirname, modelname)) + fid.write('#PBS -e {}/{}/{}.errlog \n\n'.format(self.executionpath, dirname, modelname)) + fid.write('. /usr/share/modules/init/bash\n\n') + for i in range(len(self.modules)): + fid.write('module load {}\n'.format(self.modules[i])) fid.write('export PATH="$PATH:."\n\n') + fid.write('export MPI_LAUNCH_TIMEOUT=520\n') fid.write('export MPI_GROUP_MAX=64\n\n') - fid.write('export ISSM_DIR="%s"\n' % self.srcpath) - fid.write('source $ISSM_DIR/etc/environment.sh\n') + fid.write('export ISSM_DIR="{}"\n'.format(self.srcpath)) # FIXME + fid.write('source $ISSM_DIR/etc/environment.sh\n') # FIXME fid.write('export LD_LIBRARY_PATH="$LD_LIBRARY_PATH:$ISSM_DIR/externalpackages/petsc/install/lib"\n') - fid.write('cd %s/%s/ \n\n' % (self.executionpath, dirname)) - fid.write('mpiexec - np %i %s/%s %s %s/%s %s\n' % (self.nprocs(), self.codepath, executable, str(solution), self.executionpath, dirname, modelname)) + fid.write('cd %s/%s/\n\n'.format(self.executionpath, dirname)) + fid.write('mpiexec -np {} {}/{} {} {}/{} {}\n'.format(self.nprocs(), self.cpuspernode, self.codepath, executable, solution, self.executionpath, dirname, modelname)) fid.close() @@ -166,27 +178,29 @@ # }}} - def UploadQueueJob(self, modelname, dirname, filelist): # {{{ - #compress the files into one zip. - compressstring = 'tar -zcf %s.tar.gz ' % dirname + def UploadQueueJob(self, modelname, dirname, filelist): # {{{ + # Compress the files into one zip + compressstring = 'tar -zcf {}.tar.gz'.format(dirname) for file in filelist: - compressstring += ' %s' % file + compressstring += file subprocess.call(compressstring, shell=True) print('uploading input file and queueing script') - issmscpout(self.name, self.executionpath, self.login, self.port, [dirname + '.tar.gz']) - - # }}} - def LaunchQueueJob(self, modelname, dirname, filelist, restart, batch): # {{{ + directory = self.executionpath + issmscpout(self.name, directory, self.login, self.port, [dirname + '.tar.gz']) + + # }}} + def LaunchQueueJob(self, modelname, dirname, filelist, restart, batch): # {{{ + if restart: + launchcommand = 'cd {} && cd {} && qsub {}.queue'.format(self.executionpath, dirname, modelname) + else: + launchcommand = 'cd {} && rm -rf ./{} && mkdir {} && cd {} && mv ../{}.tar.gz ./ && tar -zxf {}.tar.gz && qsub {}.queue'.format(self.executionpath, dirname, dirname, dirname, dirname, dirname, modelname) + print('launching solution sequence on remote cluster') - if restart: - launchcommand = 'cd %s && cd %s && qsub %s.queue' % (self.executionpath, dirname, modelname) - else: - launchcommand = 'cd %s && rm -rf ./%s && mkdir %s && cd %s && mv ../%s.tar.gz ./ && tar -zxf %s.tar.gz && qsub %s.queue' % (self.executionpath, dirname, dirname, dirname, dirname, dirname, modelname) issmssh(self.name, self.login, self.port, launchcommand) # }}} - def Download(self, dirname, filelist): # {{{ - #copy files from cluster to current directory - directory = '%s/%s/' % (self.executionpath, dirname) + def Download(self, dirname, filelist): # {{{ + # Copy files from cluster to current directory + directory = '{}/{}/'.format(self.executionpath, dirname) issmscpin(self.name, self.login, self.port, directory, filelist) # }}} Index: /issm/trunk-jpl/src/m/classes/clusters/sherlock.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/sherlock.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/sherlock.m (revision 26330) @@ -1,3 +1,3 @@ -%PFE class definition +%SHERLOCK cluster class definition % % Usage: @@ -7,154 +7,154 @@ classdef sherlock - properties (SetAccess=public) - % {{{ - name = 'sherlock' - login = ''; - numnodes = 1; - cpuspernode = 24; - port = 0; - codepath = ''; - executionpath = ''; - interactive = 0; - time = 30; - memory = 2; - end - %}}} - methods - function cluster=sherlock(varargin) % {{{ + properties (SetAccess=public) + % {{{ + name = 'sherlock' + login = ''; + numnodes = 1; + cpuspernode = 24; + port = 0; + codepath = ''; + executionpath = ''; + interactive = 0; + time = 30; + memory = 2; + end + %}}} + methods + function cluster=sherlock(varargin) % {{{ - %initialize cluster using default settings if provided - if (exist('sherlock_settings')==2), sherlock_settings; end + %initialize cluster using default settings if provided + if (exist('sherlock_settings')==2), sherlock_settings; end - %use provided options to change fields - cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); - end - %}}} - function disp(cluster) % {{{ - % 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)); - 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(' codepath: %s',cluster.codepath)); - disp(sprintf(' executionpath: %s',cluster.executionpath)); - disp(sprintf(' interactive: %i',cluster.interactive)); - disp(sprintf(' time: %i',cluster.time)); - disp(sprintf(' memory: %i',cluster.memory)); - end - %}}} - function numprocs=np(cluster) % {{{ - %compute number of processors - numprocs=cluster.numnodes*cluster.cpuspernode; - end - %}}} - function md = checkconsistency(cluster,md,solution,analyses) % {{{ + %use provided options to change fields + cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); + end + %}}} + function disp(cluster) % {{{ + % 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)); + 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(' codepath: %s',cluster.codepath)); + disp(sprintf(' executionpath: %s',cluster.executionpath)); + disp(sprintf(' interactive: %i',cluster.interactive)); + disp(sprintf(' time: %i',cluster.time)); + disp(sprintf(' memory: %i',cluster.memory)); + end + %}}} + function numprocs=nprocs(cluster) % {{{ + %compute number of processors + numprocs=cluster.numnodes*cluster.cpuspernode; + end + %}}} + function md = checkconsistency(cluster,md,solution,analyses) % {{{ - %Miscelaneous - if isempty(cluster.login), md = checkmessage(md,'login empty'); end - if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end - if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end + %Miscelaneous + if isempty(cluster.login), md = checkmessage(md,'login empty'); end + if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end + if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end - end - %}}} - function BuildKrigingQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + end + %}}} + function BuildKrigingQueueScript(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 + 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',mdelname); - fprintf(fid,'#SBATCH -p %s \n',cluster.queue); - fprintf(fid,'#SBATCH -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode); - fprintf(fid,'#SBATCH --time=%i\n',cluster.time*60); %walltime is in seconds. - fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); - fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); - fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); - fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname); - if ~io_gather, %concatenate the output files: - fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); - end - fclose(fid); - end - %}}} - function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + %write queuing script + fid=fopen([modelname '.queue'],'w'); + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#SBATCH --job-name=%s\n',mdelname); + fprintf(fid,'#SBATCH -p %s \n',cluster.queue); + fprintf(fid,'#SBATCH -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode); + fprintf(fid,'#SBATCH --time=%i\n',cluster.time*60); %walltime is in seconds. + fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); + fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); + fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); + fprintf(fid,'mpiexec -np %i %s/kriging.exe %s %s\n',cluster.np,cluster.codepath,[cluster.executionpath '/' modelname],modelname); + if ~io_gather, %concatenate the output files: + fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); + end + fclose(fid); + end + %}}} + 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 + 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 -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode); - fprintf(fid,'#SBATCH --time=%i\n',cluster.time*60); %walltime is in seconds. - fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); - fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); - fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); - fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME - fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME - fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); - fprintf(fid,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,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); + %write queuing script + fid=fopen([modelname '.queue'],'w'); + fprintf(fid,'#!/bin/bash\n'); + fprintf(fid,'#SBATCH --job-name=%s\n',modelname); + fprintf(fid,'#SBATCH -N %i -n %i\n',cluster.numnodes,cluster.cpuspernode); + fprintf(fid,'#SBATCH --time=%i\n',cluster.time*60); %walltime is in seconds. + fprintf(fid,'#SBATCH --mem-per-cpu=%igb\n',cluster.memory); + fprintf(fid,'#SBATCH -o %s.outlog \n',modelname); + fprintf(fid,'#SBATCH -e %s.errlog \n\n',modelname); + fprintf(fid,'export ISSM_DIR="%s/../"\n',cluster.codepath); %FIXME + fprintf(fid,'source $ISSM_DIR/etc/environment.sh\n'); %FIXME + fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); + fprintf(fid,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,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,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,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 - end %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ + %in interactive mode, create a run file, and errlog and outlog file + if cluster.interactive, + fid=fopen([modelname '.run'],'w'); + fprintf(fid,'mpiexec -n %i %s/issm.exe %s %s %s\n',cluster.np,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 + end %}}} + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ - %compress the files into one zip. - compressstring=['tar -zcf ' dirname '.tar.gz ']; - for i=1:numel(filelist), - compressstring = [compressstring ' ' filelist{i}]; - end - if cluster.interactive, - compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; - end - system(compressstring); + %compress the files into one zip. + compressstring=['tar -zcf ' dirname '.tar.gz ']; + for i=1:numel(filelist), + compressstring = [compressstring ' ' filelist{i}]; + end + if cluster.interactive, + compressstring = [compressstring ' ' modelname '.errlog ' modelname '.outlog ']; + end + system(compressstring); - disp('uploading input file and queueing script'); - issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); + disp('uploading input file and queueing script'); + issmscpout(cluster.name,cluster.executionpath,cluster.login,cluster.port,{[dirname '.tar.gz']}); - end %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ + end %}}} + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ - disp('launching solution sequence on remote cluster'); - if ~isempty(restart) - launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && sbatch ' modelname '.queue ']; - else - launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... - ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && sbatch ' modelname '.queue ']; - end - issmssh(cluster.name,cluster.login,cluster.port,launchcommand); - end %}}} - function Download(cluster,dirname,filelist)% {{{ + disp('launching solution sequence on remote cluster'); + if ~isempty(restart) + launchcommand=['cd ' cluster.executionpath ' && cd ' dirname ' && hostname && sbatch ' modelname '.queue ']; + else + launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... + ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && hostname && sbatch ' modelname '.queue ']; + end + issmssh(cluster.name,cluster.login,cluster.port,launchcommand); + end %}}} + function Download(cluster,dirname,filelist) % {{{ - %copy files from cluster to current directory - directory=[cluster.executionpath '/' dirname '/']; - issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); + %copy files from cluster to current directory + directory=[cluster.executionpath '/' dirname '/']; + issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); - end %}}} + end %}}} end end Index: /issm/trunk-jpl/src/m/classes/clusters/tetralith.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/tetralith.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/tetralith.m (revision 26330) @@ -1,3 +1,3 @@ -%teralith class definition +%TERALITH cluster class definition % % Usage: @@ -7,5 +7,5 @@ classdef tetralith - properties (SetAccess=public) + properties (SetAccess=public) % {{{ name = 'tetralith'; @@ -66,5 +66,5 @@ end %}}} - function numprocs=np(self) % {{{ + function numprocs=nprocs(self) % {{{ %compute number of processors numprocs=self.numnodes*self.cpuspernode; @@ -78,5 +78,5 @@ %compute number of processors % cluster.np=cluster.numnodes*cluster.cpuspernode; - np(cluster);%=cluster.numnodes*cluster.cpuspernode; + nprocs(cluster);%=cluster.numnodes*cluster.cpuspernode; %write queuing script @@ -106,10 +106,10 @@ % fprintf(fid,'mpiexec -np %i %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname); % fprintf(fid,'mpiexec_mpt -np %i %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname); - fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s\n',cluster.np,cluster.codepath,solution,[cluster.executionpath '/' dirname],modelname); + fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s\n',cluster.np,cluster.codepath,solution,[cluster.executionpath '/' dirname],modelname); % fprintf(fid,'mpirun -np %i %s/issm.exe %s %s %s\n',cluster.np,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); + if ~io_gather, %concatenate the output files: + fprintf(fid,'cat %s.outbin.* > %s.outbin',modelname,modelname); + end + fclose(fid); end %}}} @@ -132,5 +132,5 @@ %compute number of processors % cluster.np=cluster.numnodes*cluster.cpuspernode; - np(cluster);%=cluster.numnodes*cluster.cpuspernode; + nprocs(cluster);%=cluster.numnodes*cluster.cpuspernode; % shortname = substring(modelname,1,min(12,length(modelname))); @@ -160,5 +160,5 @@ fprintf(fid,'cd %s/%s\n\n',cluster.executionpath,dirname); % fprintf(fid,'mpiexec -np %i %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname); - fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s\n',cluster.np,cluster.codepath,solution,[cluster.executionpath '/' dirname],modelname); + fprintf(fid,'mpiexec -np %i %s/issm.exe %s %s %s\n',cluster.np,cluster.codepath,solution,[cluster.executionpath '/' dirname],modelname); % fprintf(fid,'mpirun -np %i %s/issm.exe %s %s %s\n',cluster.np,cluster.codepath,solution,[cluster.executionpath '/' dirname],modelname); % fprintf(fid,'mpiexec_mpt -np %i %s/%s %s %s %s\n',cluster.np,cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname); @@ -183,5 +183,5 @@ end end %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ %compress the files into one zip. @@ -198,5 +198,5 @@ end %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ disp('launching solution sequence on remote cluster'); @@ -206,5 +206,5 @@ issmssh(cluster.name,cluster.login,cluster.port,launchcommand); end %}}} - function Download(cluster,dirname,filelist)% {{{ + function Download(cluster,dirname,filelist) % {{{ %copy files from cluster to current directory Index: /issm/trunk-jpl/src/m/classes/clusters/yellowstone.m =================================================================== --- /issm/trunk-jpl/src/m/classes/clusters/yellowstone.m (revision 26329) +++ /issm/trunk-jpl/src/m/classes/clusters/yellowstone.m (revision 26330) @@ -1,3 +1,3 @@ -%YELLOWSTONE class definition +%YELLOWSTONE cluster class definition % % Usage: @@ -7,150 +7,150 @@ classdef yellowstone - properties (SetAccess=public) - % {{{ - name = 'yellowstone.ucar.edu' - login = ''; - modules = {'ncarenv/1.0' 'ncarbinlibs/1.1' 'perlmods/5.0' 'gmake/4.1' 'python/2.7.7' 'all-python-libs' 'git/2.3.0' 'intel/15.0.3' 'mkl/11.1.2' 'esmf' 'esmf-6.3.0rp1-defio-mpi-O' 'netcdf-mpi/4.3.3.1' 'pnetcdf/1.6.1' 'ncarcompilers/1.0' 'cmake/3.0.2' 'matlab/R2015b' 'issm'}; - numnodes = 1; - cpuspernode = 8; - port = 0; - queue = 'premium'; - time = 12*60; - processor = 'sandy'; - codepath = ''; - executionpath = ''; - grouplist = 'P93300301'; - end - %}}} - methods - function cluster=yellowstone(varargin) % {{{ + properties (SetAccess=public) + % {{{ + name = 'yellowstone.ucar.edu' + login = ''; + modules = {'ncarenv/1.0' 'ncarbinlibs/1.1' 'perlmods/5.0' 'gmake/4.1' 'python/2.7.7' 'all-python-libs' 'git/2.3.0' 'intel/15.0.3' 'mkl/11.1.2' 'esmf' 'esmf-6.3.0rp1-defio-mpi-O' 'netcdf-mpi/4.3.3.1' 'pnetcdf/1.6.1' 'ncarcompilers/1.0' 'cmake/3.0.2' 'matlab/R2015b' 'issm'}; + numnodes = 1; + cpuspernode = 8; + port = 0; + queue = 'premium'; + time = 12*60; + processor = 'sandy'; + codepath = ''; + executionpath = ''; + grouplist = 'P93300301'; + end + %}}} + methods + function cluster=yellowstone(varargin) % {{{ - %initialize cluster using default settings if provided - if (exist('yellowstone_settings')==2), yellowstone_settings; end + %initialize cluster using default settings if provided + if (exist('yellowstone_settings')==2), yellowstone_settings; end - %use provided options to change fields - cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); - end - %}}} - function disp(cluster) % {{{ - % 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(' 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(' time: %i',cluster.time)); - disp(sprintf(' processor: %s',cluster.processor)); - disp(sprintf(' codepath: %s',cluster.codepath)); - disp(sprintf(' executionpath: %s',cluster.executionpath)); - disp(sprintf(' grouplist: %s',cluster.grouplist)); - end - %}}} - function numprocs=np(cluster) % {{{ - %compute number of processors - numprocs=cluster.numnodes*cluster.cpuspernode; - end - %}}} - function md = checkconsistency(cluster,md,solution,analyses) % {{{ + %use provided options to change fields + cluster=AssignObjectFields(pairoptions(varargin{:}),cluster); + end + %}}} + function disp(cluster) % {{{ + % 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(' 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(' time: %i',cluster.time)); + disp(sprintf(' processor: %s',cluster.processor)); + disp(sprintf(' codepath: %s',cluster.codepath)); + disp(sprintf(' executionpath: %s',cluster.executionpath)); + disp(sprintf(' grouplist: %s',cluster.grouplist)); + end + %}}} + function numprocs=nprocs(cluster) % {{{ + %compute number of processors + numprocs=cluster.numnodes*cluster.cpuspernode; + end + %}}} + function md = checkconsistency(cluster,md,solution,analyses) % {{{ - available_queues={'premium','regular'}; - queue_requirements_time=[12*60 12*650]; - queue_requirements_np=[16384 16384]; + available_queues={'premium','regular'}; + queue_requirements_time=[12*60 12*650]; + queue_requirements_np=[16384 16384]; - 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.np,cluster.time) - %Miscelaneous - if isempty(cluster.login), md = checkmessage(md,'login empty'); end - if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end - if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end - if isempty(cluster.grouplist), md = checkmessage(md,'grouplist empty'); end + %Miscelaneous + if isempty(cluster.login), md = checkmessage(md,'login empty'); end + if isempty(cluster.codepath), md = checkmessage(md,'codepath empty'); end + if isempty(cluster.executionpath), md = checkmessage(md,'executionpath empty'); end + if isempty(cluster.grouplist), md = checkmessage(md,'grouplist empty'); end - end - %}}} - function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ + end + %}}} + function BuildQueueScript(cluster,dirname,modelname,solution,io_gather,isvalgrind,isgprof,isdakota,isoceancoupling) % {{{ - executable='issm.exe'; - if isdakota, - version=IssmConfig('_DAKOTA_VERSION_'); version=str2num(version(1:3)); - if (version>=6), - executable='issm_dakota.exe'; - end - end - if isoceancoupling, - executable='issm_ocean.exe'; - end + executable='issm.exe'; + if isdakota, + version=IssmConfig('_DAKOTA_VERSION_'); version=str2num(version(1:3)); + if (version>=6), + executable='issm_dakota.exe'; + end + end + if isoceancoupling, + executable='issm_ocean.exe'; + end - %write queuing script - fid=fopen([modelname '.queue'],'w'); - fprintf(fid,'#!/bin/tcsh\n'); - fprintf(fid,'#BSUB -P %s\n',cluster.grouplist); - fprintf(fid,'#BSUB -W %i:%i\n',floor(cluster.time/60),cluster.time-floor(cluster.time/60)*60); - fprintf(fid,'#BSUB -n %i\n',cluster.np); - fprintf(fid,'#BSUB -J %s\n',modelname); - fprintf(fid,'#BSUB -o %s.outlog \n',[cluster.executionpath '/' dirname '/' modelname]); - fprintf(fid,'#BSUB -e %s.errlog \n',[cluster.executionpath '/' dirname '/' modelname]); - fprintf(fid,'#BSUB -q %s\n',cluster.queue); + %write queuing script + fid=fopen([modelname '.queue'],'w'); + fprintf(fid,'#!/bin/tcsh\n'); + fprintf(fid,'#BSUB -P %s\n',cluster.grouplist); + fprintf(fid,'#BSUB -W %i:%i\n',floor(cluster.time/60),cluster.time-floor(cluster.time/60)*60); + fprintf(fid,'#BSUB -n %i\n',cluster.np); + fprintf(fid,'#BSUB -J %s\n',modelname); + fprintf(fid,'#BSUB -o %s.outlog \n',[cluster.executionpath '/' dirname '/' modelname]); + fprintf(fid,'#BSUB -e %s.errlog \n',[cluster.executionpath '/' dirname '/' modelname]); + fprintf(fid,'#BSUB -q %s\n',cluster.queue); - fprintf(fid,'module purge\n'); - for i=1:length(cluster.modules), - fprintf(fid,'module load %s\n',cluster.modules{i}); - end + fprintf(fid,'module purge\n'); + for i=1:length(cluster.modules), + fprintf(fid,'module load %s\n',cluster.modules{i}); + end - fprintf(fid,'setenv OMP_STACKSIZE 256M\n'); - fprintf(fid,'setenv MP_LABELIO yes\n'); - fprintf(fid,'setenv MP_INFOLEVEL 2\n'); - fprintf(fid,'setenv MP_SHARED_MEMORY yes\n'); - fprintf(fid,'setenv MP_EUILIB us\n'); - fprintf(fid,'setenv MP_STDOUTMODE unordered\n'); - fprintf(fid,'setenv MP_RC_USE_LMC yes\n'); - fprintf(fid,'setenv MP_MPILIB mpich2\n'); - fprintf(fid,'setenv LD_LIBRARY_PATH ${LD_LIBRARY_PATH}:/opt/ibmhpc/pecurrent/mpich2/intel/lib64/\n'); - fprintf(fid,'setenv LD_LIBRARY_PATH ${LD_LIBRARY_PATH}:/ncar/opt/intel/psxe-2015_update3/impi/5.0.3.048/lib64/\n'); + fprintf(fid,'setenv OMP_STACKSIZE 256M\n'); + fprintf(fid,'setenv MP_LABELIO yes\n'); + fprintf(fid,'setenv MP_INFOLEVEL 2\n'); + fprintf(fid,'setenv MP_SHARED_MEMORY yes\n'); + fprintf(fid,'setenv MP_EUILIB us\n'); + fprintf(fid,'setenv MP_STDOUTMODE unordered\n'); + fprintf(fid,'setenv MP_RC_USE_LMC yes\n'); + fprintf(fid,'setenv MP_MPILIB mpich2\n'); + fprintf(fid,'setenv LD_LIBRARY_PATH ${LD_LIBRARY_PATH}:/opt/ibmhpc/pecurrent/mpich2/intel/lib64/\n'); + fprintf(fid,'setenv LD_LIBRARY_PATH ${LD_LIBRARY_PATH}:/ncar/opt/intel/psxe-2015_update3/impi/5.0.3.048/lib64/\n'); - fprintf(fid,'cd %s/%s/\n\n',cluster.executionpath,dirname); + fprintf(fid,'cd %s/%s/\n\n',cluster.executionpath,dirname); - fprintf(fid,'mpirun.lsf %s/%s %s %s %s\n',cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname); - fclose(fid); + fprintf(fid,'mpirun.lsf %s/%s %s %s %s\n',cluster.codepath,executable,solution,[cluster.executionpath '/' dirname],modelname); + fclose(fid); - end %}}} - function UploadQueueJob(cluster,modelname,dirname,filelist)% {{{ + end %}}} + function UploadQueueJob(cluster,modelname,dirname,filelist) % {{{ - %compress the files into one zip. - compressstring=['tar -zcf ' dirname '.tar.gz ']; - for i=1:numel(filelist), - compressstring = [compressstring ' ' filelist{i}]; - end - system(compressstring); + %compress the files into one zip. + compressstring=['tar -zcf ' dirname '.tar.gz ']; + for i=1:numel(filelist), + compressstring = [compressstring ' ' filelist{i}]; + end + system(compressstring); - disp('uploading input file and queueing script'); - directory=cluster.executionpath; + disp('uploading input file and queueing script'); + directory=cluster.executionpath; - issmscpout(cluster.name,directory,cluster.login,cluster.port,{[dirname '.tar.gz']}); + issmscpout(cluster.name,directory,cluster.login,cluster.port,{[dirname '.tar.gz']}); - end - %}}} - function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch)% {{{ + end + %}}} + function LaunchQueueJob(cluster,modelname,dirname,filelist,restart,batch) % {{{ - %lauch command, to be executed via ssh - launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... - ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && bsub < ' modelname '.queue ']; + %lauch command, to be executed via ssh + launchcommand=['cd ' cluster.executionpath ' && rm -rf ./' dirname ' && mkdir ' dirname ... + ' && cd ' dirname ' && mv ../' dirname '.tar.gz ./ && tar -zxf ' dirname '.tar.gz && bsub < ' modelname '.queue ']; - disp('launching solution sequence on remote cluster'); - issmssh(cluster.name,cluster.login,cluster.port,launchcommand); - end - %}}} - function Download(cluster,dirname,filelist)% {{{ + disp('launching solution sequence on remote cluster'); + issmssh(cluster.name,cluster.login,cluster.port,launchcommand); + end + %}}} + function Download(cluster,dirname,filelist) % {{{ - %copy files from cluster to current directory - directory=[cluster.executionpath '/' dirname '/']; - issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); + %copy files from cluster to current directory + directory=[cluster.executionpath '/' dirname '/']; + issmscpin(cluster.name,cluster.login,cluster.port,directory,filelist); - end %}}} + end %}}} end end