Question

I have an urgent problem because my time is running out: I let my calculations process on a server with 8 cores therefore I’m using openMP in my c++ code and it works fine. Of course I’m not the only one who is using the server, so my capacity is not always 800%CPU.

But it happened now several times that someone who started his python prog on the machine paralyzed mine and his prog completely: Although I was still using around 500%CPU the code was running approx. 100x slower – for me and the other guy. Do you have an idea what the reason could be, how to prevent it?

Answer 1

As Emil Vikström listed, there are several reasons which can cause such a paralysis. I could fix the problem by enabling dynamic adjustment of the number of threads available:

omp_set_dynamic(1);

I assumed that it’s on by default. It is not.