Parallel MATLAB Instructions for Discover

Index

1. Log Into Discover

Log into Discover using ssh with X11 forwarding with the following command:

  $ ssh -X userid@login.nccs.nasa.gov

| Top of Page |

2. Submit an Interactive PBS Job to Access MATLAB

The following command will submit an interactive job to the correct node that has access to run the MATLAB client with all the correct licenses. Issue the following command:

  $ xsub -I -W group_list=computational_project -l select=trial=true -l walltime=01:00:00 -q visual

OUTPUT

  Establishing X forwarding and submitting batch job...
  The authenticity of host 'discover07 (10.1.50.7)' can't be
  established.
  RSA key fingerprint is
  05:85:dd:d2:f7:b8:b7:38:14:f1:92:7f:24:9d:05:4c.
  Are you sure you want to continue connecting (yes/no)? yes
  qsub: waiting for job 768102.borgmg to start
  qsub: job 768102.borgmg ready

  Thu Jul 24 10:33:32 EDT 2008
  $

Note that if asked to verify the authenticity of the login host, type .yes.. Once a prompt is returned, verify that you are on the visor09 node by the following:

  $ hostname
  visor09
  $

| Top of Page |

3. What do I do if I get the following error? "It doesn't look like your $DISPLAY is set appropriately. Please contact User Support for assistance.."

You did not ssh into Discover with the -X option. Log out or exit, and issue the ssh command with the -X option as described above.

| Top of Page |

4. What do I do if I get the following error? "It doesn't look like your ssh keys are setup properly. Please contact User Support for assistance.."

Users must set up password-less ssh keys in order to use PBS and allow the X11 forwarding to be set up without having to enter a password. If a user receives this error, follow these steps (Click here for more information):

  $ cd .ssh
  $ ssh-keygen -t dsa
  $ cp id_dsa.pub authorized_keys

Once this is done, re-issue the xsub command from above.

| Top of Page |

5. Load MATLAB Module

Once the interactive PBS job starts, users will be logged on to one of the visualization nodes. The first step will be to load the MATLAB module using the following command:

  $ module load tool/matlab-R2008a
  $ which matlab
  /discover/vis/mathworks/matlab_R2008a/bin/matlab
  $

| Top of Page |

6. Launch MATLAB

Run the following command to launch a MATLAB session:

  $ matlab

| Top of Page |

7. Import the PBS MATLAB Configuration File

Once the MATLAB window has appeared on the screen, users must import a PBS configuration file by clicking on the following tabs: Parallel -> Managed Configuration. A new window will appear called the Configurations Manager, and click on File -> Import. Under the Import Configuration window, enter the following path and file name:

  /discover/vis/mathworks/matlab_R2008a/share/PBS.mat

Then exit the Import Configuration window. This configuration will allow for up to 15 workers across 4 nodes to be used in parallel. Users should only have to do this the first time that parallel MATLAB is invoked.

| Top of Page |

8. Verify MATLAB is Using the PBS Configuration File

After the configuration file has been entered, verify that the current instance of MATLAB is using the PBS parallel configuration file by clicking on Parallel -> Select Configuration and choosing the PBS configuration.

| Top of Page |

9. Simple Parallel MATLAB Test

  sched = findResource('scheduler','configuration','PBS');
  pj = createParallelJob(sched);
  createTask(pj, @labindex, 1, {});

  set(pj, 'MaximumNumberOfWorkers', 12);
  set(pj, 'MinimumNumberOfWorkers', 12);
  submit(pj)
  waitForState(pj, 'finished', 60)
  results = getAllOutputArguments(pj)

| Top of Page |

10. Task-Parallel Example with Jobs and Tasks

  sched = findResource('scheduler','configuration','PBS');
  job2 = createJob(sched);
  createTask(job2, @sum, 1, {[1 1]});
  createTask(job2, @sum, 1, {[2 2]});
  createTask(job2, @sum, 1, {[3 3]});
  submit(job2)
  waitForState(job2, 'finished')
  results = getAllOutputArguments(job2)

| Top of Page |

11. Task-Parallel Example with a MATLAB pool

  mpj = createMatlabPoolJob('configuration','PBS');
  createTask(mpj, @sum, 1, {[1 1]});
  set(mpj, 'MaximumNumberOfWorkers', 3);
  set(mpj, 'MinimumNumberOfWorkers', 3);
  submit(mpj)
  waitForState(mpj, 'finished')   results = getAllOutputArguments(mpj)

| Top of Page |