Setting Up the Batch Configuration: SGE
For MATLAB R2011a
The following instructions are for the Katana Cluster only; there is no need to set up batch configuration for local computers.
- Log on to Katana with X-window software, such as x-win32.
- Run MATLAB. (Screen shots shown below are for R2011a, the currently installed version on Katana)
- Select “Parallel/Manage Configurations” on the MATLAB task bar.
- Select File/New/Generic.
- A MATLAB Message.
- A “Generic Scheduler Configuration Properties” window pops up. There are three tabs (Scheduler, Jobs, and Tasks) in this pane; it is default to the “Scheduler” tab. First, data entered in the “Scheduler” tab by a user are shown below, in red, together with the screen shot. Take care to enter them EXACTLY.
- Configuration name : SGE
- Description : Sun Grid Engine batch configuration
- Root folder … : /usr/local/apps/matlab-2011a/matlab
- Number of workers available to scheduler: 32
- Function called when submitting parallel jobs: sgeParallelSubmitFcn
- Function called when submitting distributed jobs: sgeSubmitFcn
- Cluster nodes’ OS (ClusterOsType): leave it as Unset
- Job data location is accessible … : select True
Once you have finished filling in the data, don’t click “OK” yet! You need also to fill in some more data in the “Jobs” tab.
If you had clicked the OK button by mistake, you would have come out of the configuration manager. You will need to go back to Parallel/Manage configuration menu action above (see Step 3). Right-click on “SGE” and then select “Properties” will bring you right back to this point.
- After verifying that all data entered is accurate, click on the “Jobs” tab.
- Maximum number of workers that can run job: 4
- Minimum number of workers that can run job: 4
The Min/Max worker numbers are for default purposes. You could routinely override the numbers at run time anyway. No action is required for the third tab, “Tasks.” Just click OK to finish.
- You should now be back to the Manage Configuration panel. Whether you should select SGE or local as your default configuration depends on which one you use more often as well as how you prefer to start matlabpool.Once the default configuration is set, just kill the window. You may notice the “Start Validation” button on the right bottom. This button provides test for all the defined configurations in the Manage Configuration panel. However, since the batch queue is typically fairly busy, running this option may take some time. So, just skip it and exit the Manage Configuration window. PCT is now ready for use.
In the above, whether to select local or SGE as your default configuration is entirely up to you. The recommendation is to set the default to local. This setting makes starting matlabpool (this essentially requests processors) simpler, or more convenient, in interactive code development and testing. For production, or batch, runs, spelling out precisely how you intend to start matlabpool go a long way to avoid any unanticipated consequences.