It seems that the time.sleep(0.000001) hack has some unwanted consequences on Windows10 in Python 3. A simple
%timeit time.sleep(0.000001)
reveals that this takes ~15ms on my workstation, which is going to take up the bulk of execution time for fast distributions. Decreasing the fraction of time does not help. Since this is related to the MPI issue, I can imagine that for under Windows the trick is not even necessary.
It seems that the time.sleep(0.000001) hack has some unwanted consequences on Windows10 in Python 3. A simple
%timeit time.sleep(0.000001)
reveals that this takes ~15ms on my workstation, which is going to take up the bulk of execution time for fast distributions. Decreasing the fraction of time does not help. Since this is related to the MPI issue, I can imagine that for under Windows the trick is not even necessary.