[Scilab-users] scilab and "parallelization" objectives

[Carrico, Paul]

Dear All,

My internship had an interesting request and I'm not able to answer to him   :):)

The context is the following one:

-          We are using a parallel finite element solver on 4 processors (under Linux),

-          We are performing optimization with an external optimizer,

-          Scilab has been interfaced with both the optimizer and the solver,

-          For the moment, the optimization loops are "basically" performed using the parallelization capabilities of the solvers.

In practice, Scilab "runs" the solver and "waits" the end of the simulation (always using 4 processors) before calculating the cost function value, giving the later value to the optimizer and then closing.

Imagine that now we would like to run 4 different simulations each on 1 processor in order to determine the gain (CPU time) ... how to proceed ?

In practice Scilab must be able to :

-          Launch 1rst calculation and to handback ...

-          ... to launch the 2nd one ... and so on

We can imagine creating a function per calculation (and I prefer in order to mix stuffs), then running independently but at the same time the different functions.

Honestly I've never done this before and I'm wondering how to proceed ... or if it is possible : any feedback on it ?

Thanks and regards

Paul


EXPORT CONTROL :
Cet email ne contient pas de données techniques
This email does not contain technical data

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.scilab.org/pipermail/users/attachments/20161020/9842fac6/attachment.htm>