In a python script of mine, I currently use the lsim function to simulate a system model. An issue I encountered recently is that lsim spawns a lot of subprocesses on multiple cores, and together they cause heavy CPU load. That fact also shows in the profiling log, where I attached the relevant snippet below. I run this script on a processing machine, which I share with multiple people. If I instead use lsim2, it seems that no subprocesses are spawned, however running the script takes unbearably long. Anyone has an idea how I can run lsim fast, while using fewer resources/just one core?
ncalls tottime percall cumtime percall filename:lineno(function)
3740 25.422 0.007 51.062 0.014 /grid/common/pkgs/python/v3.7.2/lib/python3.7/site-packages/scipy/signal/ltisys.py:1870(lsim)
26753891 21.519 0.000 21.519 0.000 {built-in method numpy.dot}
12 0.001 0.000 21.450 1.788 /grid/common/pkgs/python/v3.7.2/lib/python3.7/subprocess.py:431(run)
12 0.000 0.000 21.265 1.772 /grid/common/pkgs/python/v3.7.2/lib/python3.7/subprocess.py:895(communicate)
24 0.000 0.000 21.265 0.886 /grid/common/pkgs/python/v3.7.2/lib/python3.7/subprocess.py:985(wait)
24 0.000 0.000 21.265 0.886 /grid/common/pkgs/python/v3.7.2/lib/python3.7/subprocess.py:1592(_wait)
12 0.000 0.000 21.264 1.772 /grid/common/pkgs/python/v3.7.2/lib/python3.7/subprocess.py:1579(_try_wait)
Setting the environment variable OPENBLAS_NUM_THREADS=1 did the trick for me