Question about the job migration inside of Matlab over Xgrid.

By joseph400 at Mon, Sep 27 2010 2:59pm |

Hi all,

My simulation on Matlab takes about couple of days (or even weeks).
I'm considering setting up Xgrid hoping to fully utilize all the labs resources.
I found how to run Matlab program over Xgrid here: http://hints.macworld.com/article.php?story=20041125065000639
It's basically to submit a script that opens matlab with specific matlab code.
However, it doesn't look that Xgrid can control what's really happening inside of the Matlab.

Can the jobs inside of the Matlab be migrated to another agent?
Or, does the newly assigned agent open Matlab and start from the beginning?
For example, I submitted 10-hours job and it was running on an agent.
Suppose the user of the agent comes back when half of the computation is done (after 5 hours).
After the job is returned to the Xgrid server and assigned to another agent, does the new agent take over what's being done inside of the Matlab, or just relaunch the Matlab and start from scratch?
If the latter is the case, I don't see any reason to use Xgrid.
Is there better way to run large scale Matlab codes efficiently?

Thanks!

‹ How to compile MacPyMOL? Mathematica run remotely with X11 ›

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.
Mon, 2010-09-27 23:15 — scottyb

You really need the

You really need the Distributed/Parallel Computing toolbox, but it is $$$$$$.

Wed, 2010-09-29 09:48 — joseph400

Thanks for the reply. Yes,

Thanks for the reply. Yes, I'm also considering the toolbox as an option. I asked linux cluster administrator to install trial version of it. In addition, I want to take advantage of all the machines that are idle in the lab. That's why I'm considering grid-computing stuff. Coming back to my question, do you mean the Xgrid just reopens the Matlab and starts from the scratch after migration?

Wed, 2010-09-29 23:26 — scottyb

Well, I can only comment

Well, I can only comment using the SGE, and not Xgrid, as cluster-management software, but with Matlab and SGE, one can run batch jobs fine (i.e., one CPU/core per job), but as soon as one wants to run "explicit parallel" jobs (i.e., over multiple compute nodes), one absolutely has to use the Parallel/Distributed Computing toolbox.