This miniproject is a test of the abbysal multithreading performance of Python for CPU bound tasks. It contains three implementations of the same simple algorithm:

1. In C++ as baseline.
2. In Python, using multithreading.
3. In Python, using multiprocessing.

The goal is to show how badly Pythons multithreading scales for CPU bound tasks and how the situation can be somehow remedied using multiprocessing instead.

Compile and run with make. The output should be something along these lines:

C++,1,10000000,649
C++,2,10000000,360
C++,4,10000000,314
Python (mt),1,10000000,5934
Python (mt),2,10000000,10429
Python (mt),4,10000000,13316
Python (mp),1,10000000,7386
Python (mp),2,10000000,4109
Python (mp),4,10000000,3934

The first column identifies the implementation, the second the number of threads, the third is the result of the calculation and the fourth and last is the run time in milliseconds.

The machine has two cores. As expected, in C++ and in Python with the multiprocessing library, using two tasks almost halves the runtime. Using more tasks is of no help as the cores are already as busy as they can be. Python with multithreading on the other hand, instead of reducing the runtime takes almost twice as much time to execute the same computation. The GIL is to blame for this.