{"id":3959,"date":"2021-03-08T09:22:15","date_gmt":"2021-03-08T07:22:15","guid":{"rendered":"http:\/\/ucthpc.uct.ac.za\/?page_id=3959"},"modified":"2021-04-09T16:40:23","modified_gmt":"2021-04-09T14:40:23","slug":"matlab-standard-job-submission","status":"publish","type":"page","link":"https:\/\/ucthpc.uct.ac.za\/index.php\/matlab-standard-job-submission\/","title":{"rendered":"MATLAB standard job submission"},"content":{"rendered":"

Getting Started with Parallel Computing using MATLAB on the HPC Cluster<\/strong><\/p>\n

Note:<\/strong> If you are not making use of parallel functions then you can ignore this document and just launch standard MATLAB jobs with the following in your shell script:<\/p>\n

module load software\/matlab-R2020b\r\nmatlab -batch my_matlab_script<\/pre>\n
 <\/p>\n
This document provides the steps to configure MATLAB to submit jobs to a cluster, retrieve results, and debug errors.<\/p>\n
CONFIGURATION<\/strong><\/p>\n
After logging into the cluster use the module command to select the version of MATLAB you require:<\/p>\n
module load software\/matlab-R2019b\r\nor\u00a0\r\nmodule load software\/matlab-R2020b<\/pre>\n
Then configure MATLAB to run parallel jobs on your cluster by calling the shell script configCluster.sh\u00a0\u00a0 This only needs to be called once per version of MATLAB.<\/p>\n
$ module load software\/matlab-R2019b\r\n$ configCluster.sh<\/pre>\n
Jobs will now default to the cluster rather than submit to the local machine.<\/p>\n
CONFIGURING JOBS<\/strong><\/p>\n
Prior to submitting the job, we can specify various parameters to pass to our jobs, such as queue, e-mail, walltime, etc.\u00a0 Only WallTime is required.<\/em><\/p>\n
>> % Get a handle to the cluster<\/span>\r\n>> c = parcluster;\r\n\r\n[REQUIRED<\/strong>]<\/span>\r\n\r\n>> % Specify the walltime (e.g. 5 hours)<\/span>\r\n>> c.AdditionalProperties.WallTime = '05:00:00';\r\n\r\n[OPTIONAL<\/strong>]<\/span>\r\n\r\n>> % Specify an account to use for MATLAB jobs<\/span>\r\n>> c.AdditionalProperties.AccountName = 'account-name';\r\n\r\n>> % Specify e-mail address to receive notifications about your job<\/span>\r\n>> c.AdditionalProperties.EmailAddress = 'user-id@uct.ac.za';\r\n\r\n>> % Request 2 Fermi GPU cards per node<\/span>\r\n>> c.AdditionalProperties.GpuCard = 'fermi';\r\n>> c.AdditionalProperties.GpusPerNode = 2;\r\n\r\n>> % Specify memory to use for MATLAB jobs, per core (MB)<\/span>\r\n>> c.AdditionalProperties.MemUsage = '4000';\r\n\r\n>> % Specify a queue to use for MATLAB jobs\u00a0<\/span>\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\u00a0\r\n>> c.AdditionalProperties.QueueName = 'queue-name';<\/pre>\n
Save changes after modifying AdditionalProperties for the above changes to persist between MATLAB sessions.<\/p>\n
>> c.saveProfile<\/pre>\n
To see the values of the current configuration options, display AdditionalProperties.<\/p>\n
>> % To view current properties\r\n>> c.AdditionalProperties<\/pre>\n
Unset a value when no longer needed.<\/p>\n
>> % Turn off email notifications<\/span>\r\n>> c.AdditionalProperties.EmailAddress = '';\r\n>> c.saveProfile<\/pre>\n
 <\/p>\n
INTERACTIVE JOBS<\/strong><\/p>\n
To run an interactive pool job on the cluster, continue to use parpool as you\u2019ve done before.<\/p>\n
>> % Get a handle to the cluster<\/span>\r\n>> c = parcluster;\r\n\r\n>> % Open a pool of 4 workers on the cluster<\/span>\r\n>> p = c.parpool(4);<\/pre>\n
Rather than running local on the local machine, the pool can now run across multiple nodes on the cluster.<\/p>\n
>> % Run a parfor over 1000 iterations<\/span>\r\n>> parfor idx = 1:1000\r\n      a(idx) = \u2026\r\n   end<\/pre>\n
Once we\u2019re done with the pool, delete it.<\/p>\n
>> % Delete the pool<\/span>\r\n>> p.delete<\/pre>\n
 <\/p>\n
EXAMPLES<\/strong><\/p>\n
Running a job on multiple cores on a single cluster node:<\/p>\n
c = parcluster;\r\nc.AdditionalProperties.ProcsPerNode = 4\r\nc.saveProfile\r\np = c.parpool(4);\r\nspmd, A = rand(1000); b = rand(1000,1); x = A\\b; end<\/pre>\n
 <\/p>\n
Running a job across multiple nodes:<\/p>\n
c = parcluster;\r\nc.AdditionalProperties.ProcsPerNode = 1\r\nc.saveProfile\r\np = c.parpool(4);\r\nspmd, A = distributed.rand(10000); b = distributed.rand(10000,1); x = A\\b; end<\/pre>\n
 <\/p>\n
TO LEARN MORE<\/strong><\/p>\n
To learn more about the MATLAB Parallel Computing Toolbox, check out these resources:<\/p>\n