Killing .exe program started from Matlab

I am doing an optimization with ga (with matlabpool parallel computing). From within my function evaluation, I call an external .exe program (XFoil, somebody might know it). At the moment I am simply using "system" for running the .exe program.

system(['xfoil.exe < prof_name.cmd > prof_name.out']);

where prof_name.cmd is an input file for the program and prof_name.out a file that I ask only for avoiding having my screen full of useless messages.

My problem is that sometimes the .exe program gets in a loop and, therefore, the control does not go back to Matlab and my optimization gets stuck. Thus, I would like to kill the .exe after a given time but this is not possible using the system command.

Does anybody have a suggestion?

Thanks

P.S. I considered assembling a job (one per generation) calling several tasks (the evaluation of the individuals) and using the timeout property:

set(j, 'Timeout', 60)

to end the task. Although this would unlock the optimization, it would also leave several instances of the .exe program running (thus quickly reducing the available computational power)

[Jan]

Which OS are you using? Do you start the exe files locally or remotely? Why do they fall into an infinite loop? A straight approach would be to avoid that the exe files are blocked, while killing them externally is more a crude and brute workaround.

[Francesco]

By the way.

The solution reported here: http://www.mathworks.ch/matlabcentral/answers/43823-running-external-program-from-matlab-for-a-given-amount-of-time is not working since I have several instances of the .exe program running and I want to kill just one of them.

Concerning the more general purpose answer, I fear it might be a bit too generic for my understanding...

[Francesco]

Thanks Jan Simon.

I am using Matlab 2012a on a Mindows 7 64bit. I am using a local matlabpoool, so the .exe runs on the same computer running matlab. By the way, the .exe is located in the work folder.

Unfortunately I was not able to avoid the .exe getting stuck (it is happening to others as well). I know it is not nice but at the end I had to accept it..

Answer 1

[Jan]

You can obtain the ProcesID using the OS-command tasklist. Then the task can be kill by taskkill.

The identification of a started task in the tasklist output is not trivial. It would be smarter to start the exe using spawn in a C-mex function, such that you get the ProcessID directly. Perhaps these FEX submissions help: FEX: spawn and FEX: clientserver-framework.

[Francesco]

Thank you Jan Simon

As you said, the tricky part is to get the PID. A colleague already suggested running tasklist before and after system in order to see the PID of the created task. However, due to the parallel nature it is quite probably that I will have several jobs starting at the same time and the trick would probably not work.

I had a look at the FEX submissions you suggested. If I am not wrong, the first one does not give back the PID. The second is unfortunately for Linux only...

[Jan]

Althout it does not reply the PID directly, you could be able to modify the code such that it does. If the C-stuff is not your business, adjust the code as far as you can and post a new question in this forum.

Answer 2

[Francesco]

I finally decided to save time and go for a dirty solution.

I replicated the .exe file (one per lab). Now every lab calls program_labID, so that I can always kill the program_labID task being sure that it will only affect the lab I want.

I know it's super ugly, but unfortunately I was unable to get anything cleaner. Thank you anyhow for the help!

[Jan]

No, this is not super ugly. You could create hard-links to avoid data duplication.