Matlab Production Server and parfor loops

Hi all,

I have a question regarding parallel programming and the MPS. I know that the with a single MPS license, the MPS can run 24 workers which can be called upon request by a client application. On a multi-core system, multiple request can be handled by the MPS through different workers, and hence by different cores (if available). You can manually break-up a single request in multiple sub-requests and send them to the MPS to gain a performance difference. My question however: would it be possible to implement a parfor loop in the code? In my first tests with the MPS, I see that a single request only uses up to 50% of the CPU power (on a four-core system), leaving the rest of the CPU power idle.

This would simplify the code on the client-side (not having to deal with breaking up the client's request), while gaining a performance boost. Any experiences on this subject are welcome!

Thanks in advance,

Robert

Answer 1

[wangpenghn wang]

hello, do you have any progress on this question? if have, could you please share it?

thank you so much

[Robert Vullings]

In the end I optimized the code such that any parallel computing was not necessary anymore. I haven't tried compiling the CTF with the parfor loop. I would figure that setting up the parallel pool might even make the computation time longer. In my case the request only took about 10 secs, which I wanted to bring down a bit more.