[torqueusers] running matlab distributed computing toolbox jobs
Andrew J Caird
acaird at umich.edu
Tue Sep 26 19:43:16 MDT 2006
For the archives, I've documented what we've done to use Matlab's
Distributed Computing Toolbox (DCT) with Torque at:
If anyone has any feedback, I'm open to suggestions.
On Wed, 20 Sep 2006, Andrew J Caird wrote:
> I think we've got this sorted out. I'll take a look at it to
> make sure I don't embarass myself too much, try to document it,
> and put it somewhere useful sometime soon.
> If you can't wait a few days, I'll email you the two files and
> that might get you started.
> Torque having job arrays would make this a little easier, since
> Mathworks provides a set of scripts for PBSPro that take
> advantage of that functionality. Nonetheless, people here are
> using what we put together.
> On Wed, 20 Sep 2006, Steven A. DuChene wrote:
>> We have a need here to use a product called Matlab DCT or
>> distributed computing toolbox on our cluster. The product from
>> Mathworks includes it's own proprietary scheduler but we need
>> to implement running jobs through our existing Moab/Torque
>> setup rather than with their packaged scheduler. Now Mathworks
>> supplied some VERY minimal instructions on how to get this to
>> work with other scheduler environments (they don't make the
>> distinction between a scheduler and a resource manager) but the
>> supplied examples/instructions are very minimal and quite
>> cryptic. These examples sort of cover getting it to work with
>> LSF, PBSPro (using job arrays), and SGE.
>> Has anyone else here had any experience with this product? I am
>> referring to Matlab DCT rather than the more generic Matlab
>> product suite.
>> Any help or hints on making this function in a moab/torque
>> environment would be most appreciated. The documentation they
>> provide says there has to be a submit function on the client
>> end and a worker decode function on the compute node end. I
>> have those example submit and worker decode functions but they
>> seem to be basic but incomplete templates rather than
>> functional examples. -- Steven A. DuChene
> torqueusers mailing list
> torqueusers at supercluster.org
More information about the torqueusers