Performance benchmark of NumPy shared memory.
First, install Python virtual environment and numpy-sharedmem:
make
Run using threads on vanilla NumPy arrays of size 100K, using 8 processes and 20 tasks per process:
time ./python main.py Thread numpy 8 20 100000
Run the same thing, this time using multiprocessing.Process class instead:
time ./python main.py Process numpy 8 20 100000
And now using shared memory for the two cases above:
time ./python main.py Thread sharedmem 8 20 100000
time ./python main.py Process sharedmem 8 20 100000