Running MATLAB on HPC

Hello all.

I am using MATLAB R2012b on High Performance Computing facility which I guess is a unix system. I am not able to save my data. Specifically, when I use commands such as 'save', 'xlswrite', I find "No window system found. Java option 'MWT' ignored." written in the "out" file though the code runs perfectly on my local machine with Windows/Linux platform. Due to this I tried writing in a text file (with mixed type of data, namely string, character, float and integers). Still I get the message "No window system found. Java option 'MWT' ignored.". Please help me getting rid of this difficulty.

Answer 1

[Jason Ross]

I'm suspecting the "out" file is from the scheduler, telling you job output. If you are executing a "save" command in MATLAB, it's saving the file to the file system, and that has nothing to do with the job output file -- the output file just saves any output from (most likely) standard out of the console where the process was launched. So what you are seeing is MATLAB starting and not finding a window system -- it's running in "console" mode.

So if you are running on the cluster, it's saving the file somewhere on the filesystem, and that location is determined by whatever code you are running. Since you aren't getting an error, it's likely succeeding -- which means you aren't trying to do something like write to a location you don't have permission to write to.

You should probably also check if there is an .err file and see if there are any errors in there.

I'm also guessing that whatever facility you are using likely has documentation and best practices for you to run MATLAB jobs on the cluster, as well. I've seen a few facilities that have very specific guidelines and howtos that cover things like where output goes, how to submit the job, etc.

[Tahariet Sharon]

A follow-up question: is it possible to track the output of the console when submitting Matlab batch jobs?

[Seung-Goo Kim]

@Tahariet Sharon Oh sure. Everything that you'd see on the Command Window will be stored in the log files of the scheduler (e.g., SLURM), which is key information for debugging.