Parallel execution of external application with different parameters

Im trying to run an optimization were for each iteration an external simulation program is run from within MATLAB. I basically communicate with the external program through ascii files (to send parameters to the application and to later read the results back).

If I try to run it parallel just as it is, I would get some race conditions because of the usage of the same filename. How could I retrieve a unique ID based on the core thats runing the fitting functions?

[Jan]

How do you determine the core that's running the functions?

Answer 1

[Jan]

If all files are created by the same Matlab session, you can create unique IDs based on the current time:

function ID = TimeID
S  = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789_';
D  = T(4) * 360000 + T(5) * 6000 + T(6) * 100;  % 100th of a second
ID = S(1 + rem(fix([D / 250047, D / 3969, D / 63, D]), 63));
pause(0.01);  % Optional

The IDs have 4 characters which are allowed in file names, e.g.: 'Z_Hs'. They are unique inside a single Matlab session and inside an interval of 24 hours.

If you create more than 100 IDs per second, the PAUSE command cares for uniqueness. The IDs can be non-unique, if the function runs more than 24 hours.

Do not use the 1000th of seconds under Matlab. There is a strong correlation between the values in subsequent calls of CLOCK, even if a random pause is inserted between the calls.

This method does not take the core into account, but perhaps you can add this feature.

Answer 2

[Edric Ellis]

I would use the TEMPNAME function, as recent versions of MATLAB generate names that are highly unlikely to clash.