Perform experiment:
1) Run fitting n pipelines in multiprocessing mode
2) Run fitting n pipelines in single process mode
3) Calculate overhead for creating a process
This experiment should be done as something reproducible (let's say, as a file in cases or examples) -- so we could run it on different systems if needed and later for new FEDOT versions. Thus we will be able to monitor performance improvements.
Would be good to separate effects from multiprocessing from effects from cache. So trying to run same thing with & without OperationsCache seems a good idea.
A good experimental question is: what's the minimum size of population to win the overhead and get real benefits?
All this would help to make performance improvements in FEDOT with much better understanding.
Perform experiment: 1) Run fitting n pipelines in multiprocessing mode 2) Run fitting n pipelines in single process mode 3) Calculate overhead for creating a process