DokuWiki

User Tools

Site Tools

Trace:
cluster:96

Warning: Undefined array key -1 in /usr/share/dokuwiki/inc/html.php on line 1458

Differences

This shows you the differences between two versions of the page.

Link to this comparison view

Both sides previous revision Previous revision
Next revision 		Previous revision
cluster:96 [2011/02/09 14:55]
hmeij 		cluster:96 [2015/03/13 13:52] (current)
hmeij [Note]
Line 1: Line 1:
 \\ \\
 **[[cluster:0|Home]]** **[[cluster:0|Home]]**
 +
 +==== Note ====
 +
 +You can also run matlab jobs via scripts like other software.
 +
 +Create a file with the matlab commands and then create a shell script to submit the job.
 +
 +It would look like
 +
 +<code>
 +
 +#!/bin/bash
 +# submit via 'bsub < run.serial'
 +
 +#BSUB -q matlab
 +
 +#BSUB -J test
 +#BSUB -o test.stdout
 +#BSUB -e test.stderr
 +
 +
 +
 +matlab -no display < my_code.m > /dev/null
 +
 +</code>
 +
 +==== Matlab 2013a ====
 +
 +Is now the default matlab version, just change the paths in scripts below
 + --- //[[hmeij@wesleyan.edu|Meij, Henk]] 2013/03/14 14:57//
  
 ==== Matlab 2010b ==== ==== Matlab 2010b ====
Line 11: Line 41:
   * [[http://www.mathworks.com/help/toolbox/distcomp/bqxceii-1.html|Part2]]   * [[http://www.mathworks.com/help/toolbox/distcomp/bqxceii-1.html|Part2]]
  
-In the examples below the code waits for the results.  Don't do that when submitting jobs, it is just for debug purposes. Change the last 3 lines with this 'get(sched)' and then you exit Matlab interactive session.+=== Serial 1 === 
 + 
 +In the example below, the code waits for the results.  Don't do that when submitting jobs, it is just for debugging purposes.
  
 This old way still works for distributed jobs: This old way still works for distributed jobs:
 +
 <code> <code>
  
 % distributed matlab jobs % distributed matlab jobs
-% start 'matlab -nodisplay', issue the command 'newSimple'+% start 'matlab -nodisplay', issue the command 'Simple'
  
 % set up the scheduler and matlab worker environment % set up the scheduler and matlab worker environment
Line 23: Line 56:
 set(sched, 'HasSharedFilesystem', true); set(sched, 'HasSharedFilesystem', true);
 set(sched, 'ClusterMatlabRoot', '/share/apps/matlab/2010b') set(sched, 'ClusterMatlabRoot', '/share/apps/matlab/2010b')
-set(sched, 'SubmitFcn', @distributedSubmitFcn)   +set(sched, 'SubmitFcn', @lsfSimpleSubmitFcn)
  
 % specify location for worker output % specify location for worker output
Line 34: Line 67:
 % submit job, wait for jobs to finish, fetch output % submit job, wait for jobs to finish, fetch output
 submit(j) submit(j)
 +
 % WARNING: this may hang if the workers are busy! % WARNING: this may hang if the workers are busy!
 waitForState(j) waitForState(j)
 results = getAllOutputArguments(j); results = getAllOutputArguments(j);
 results{1:5} results{1:5}
 +% better to do it like Simple2.m
 +
  
 </code> </code>
  
-This old way for parallel jobs does not, not sure why:+=== Serial 2 === 
 + 
 +After you submit the job, you can exit matlab. 
 <code> <code>
-% start 'matlab -nodisplay', issue the command 'myPJob' 
  
-% set up the scheduler and matlab worker environment +% distributed matlab jobs 
-%2007a sched = findResource('scheduler', 'configuration', 'generic') +% start 'matlab -nodisplay', issue the command 'Simple2'
-%2007a set(sched, 'Configuration', 'generic') +
-sched = findResource('scheduler','type','LSF') +
-set(sched, 'SubmitArguments', '-R type==LINUX64') +
-set(sched, 'SubmitArguments', '-q matlab')+
  
 +% set up the scheduler and matlab worker environment
 +sched = findResource('scheduler', 'type', 'generic');
 set(sched, 'HasSharedFilesystem', true); set(sched, 'HasSharedFilesystem', true);
 set(sched, 'ClusterMatlabRoot', '/share/apps/matlab/2010b') set(sched, 'ClusterMatlabRoot', '/share/apps/matlab/2010b')
-set(sched, 'SubmitFcn', @lsfParallelSubmitFcn)   +set(sched, 'SubmitFcn', @lsfSimpleSubmitFcn)
  
 % specify location for worker output % specify location for worker output
Line 60: Line 96:
  
 % create job and assign tasks to be done % create job and assign tasks to be done
 +j = createJob(sched);
 +T = createTask(j, @rand, 1, {{3,3} {3,3} {3,3} {3,3} {3,3}});
  
-set(pj, 'MaximumNumberOfWorkers', 4) +% submit job, then exit matlab interactive session 
-set(pj, 'MinimumNumberOfWorkers', 4)+submit(j) 
 +get(sched)
  
-% create job and assign tasks to be done +</code>
-pj = createParallelJob(sched); +
-T = createTask(pj, @rand, 1, {{3,3} {3,3} {3,3} {3,3} {3,3}});+
  
 +=== Parallel ===
 +
 +Slightly different syntax.  Adjust your number of workers values based on some benchmarking, not always do the most workers finish a job in the shortest amount of time.
 +
 +<code>
 +
 +% parallel matlab jobs
 +% start 'matlab -nodisplay', issue the command 'Simplep'
 +
 +% set up the scheduler and matlab worker environment
 +sched = findResource('scheduler', 'type', 'generic')
 +set(sched, 'HasSharedFilesystem', true);
 +set(sched, 'ClusterMatlabRoot', '/share/apps/matlab/2010b')
 +set(sched, 'ParallelSubmitFcn', @lsfParallelSubmitFcn)
 +
 +% specify location for worker output
 +set(sched, 'DataLocation', '/home/hmeij/hp/matlab')
 +pj = createParallelJob(sched);
 +set(pj, 'MaximumNumberOfWorkers', 8)
 +set(pj, 'MinimumNumberOfWorkers', 8)
 +T = createTask(pj, @rand, 1, {{300,300}});
  
 % submit job, wait to finish, get status jobs, gather results % submit job, wait to finish, get status jobs, gather results
 submit(pj) submit(pj)
 get(sched) get(sched)
- 
 % you can now exit matlab % you can now exit matlab
-% at system prompt type 'bjobs' +% at system prompt type 'bjobs' 
  
 </code> </code>
 +
 +
  
 \\ \\
 **[[cluster:0|Home]]** **[[cluster:0|Home]]**
cluster/96.1297281358.txt.gz · Last modified: 2011/02/09 14:55 by hmeij

Page Tools

Except where otherwise noted, content on this wiki is licensed under the following license: CC Attribution-Share Alike 4.0 International
CC Attribution-Share Alike 4.0 International Donate Powered by PHP Valid HTML5 Valid CSS Driven by DokuWiki