When I use openmp with ode integrators I don't get a linear increase in performance. i'm running monte-carlo dynamics: random initial conditions and i run trajectories.

one of the people that works with openMP in CS dept suggested that there might be calls that block threads; for example, exceptions.

I spent some time on the issue but it was taking too much time. the only way to debug was to measure the time for each call by hand.

any help or advice?

i forgot to add that in parallel mode, each thread takes more then twice longer then running one thread.