old batch job blocking new ones

Hi,

Since I have code which involves big 'for' loops that seemed to me ideal for parallel computation, I was trying to set up a parallel batch job. With only little experience in parallel computing, I was exercising, hoping, there'll be some "learning-by-doing", following some examples and "first steps" in the MATLAB help browser.

I cannot tell what went wrong but it seems ever since my practice session, there is a batch job running, blocking all new jobs I want to set up. Command output from 'findJob' and 'findRessource' suggests that this blocking job always starts whenever I open my matlab pool. Interestingly, looking at the processor load, this job seems to do nothing at all.

I tried to set up a new configurations, to delete the 'local_scheduler_data' folder and even to change the user. Nothing helped. I couldn't find a solution searching Google and MathWorks so I really would appreciate if anyone could help me. Currently, I have MATLAB R2011a installed and I'm using OpenSuse 11.4.

Thanks in advance, Mathias

Answer 1

[Edric Ellis]

When you open a MATLABPOOL, you are running a job on your scheduler. That job is connected up to your MATLAB client to provide workers to run the body of any PARFOR loops you may execute. I would suggest you follow one or other of the following paths:

1. Change your code to use PARFOR loops rather than FOR loops, and open a MATLABPOOL prior to running stuff
2. Submit batch jobs using the 'batch' function, but do not open a MATLABPOOL.

Answer 2

[Mathias]

Hi Edric,

Thanks for your answer. Changing the appropriate for-loops to parfor-loops was the first thing I did when preparing my code for parallel execution. But using the batch command, apparently, I really don't have to open a matlabpool at all -- even for parallel jobs (command: batch('script','matlabpool',#workers) ). Does MATLAB use the default pool automatically or does it run the code sequentially in this case?

Anyway, I figured out that in my state of experience, spmd may be a more convenient choice.