can parfor in compiled code access a local worker pool?

I run on a dedicated multi-code computer very often mcc-compiled matlab programs, in which I execute parfor. Each program has a relatively slow startup time, I think because it is create the parallel worker pool. For my application, it would be more efficient if the pool could remain open all the time in the background, for example from a headless matlab session. Is something like that possible, or are there other solutions?

Answer 1

[Dasharath Gulvady]

It is definitely possible for a compiled MATLAB code to utilize profiles created using Parallel Computing Toolbox. The below documentation page explains the same:

http://www.mathworks.com/help/compiler/compile-and-deploy-a-standalone-application-with-the-parallel-computing-toolbox.html

[emmenlau]

This does not do what I want. I have followed the instructions and first set up a dedicated local profile, then I started an interactive Matlab with a parpool for this profile, and finally I started my compiled application with the profile. However, the application is not using the parallel workers from the (already running) interactive Matlab, it starts new workers, which slows down startup time significantly.

What I want is that all Matlab share the same workers, to avoid startup time and process creation. From the interactive session, this works perfectly!

[Dasharath Gulvady]

This is not possible, since no two instance of MATLAB can share the same MATLAB workers.

[emmenlau]

Thats strange, because when I open Matlab twice interactively, and create a parpool in both (using the same profile), Matlab will tell me that there is a second pool already open, and that I can fetch and delete jobs on the existing pool! Did you try that? It seems they could in principle share the pool, just I don't understand how it would work.

[Dasharath Gulvady]

Which version of MATLAB are you using? The expected behavior is not to share any workers between the MATLAB sessions. In the second MATLAB session, if you execute

 >>p = gcp('nocreate');

What is the value of p? Note that, you should execute the above command without creating a pool beforehand.