multiprocessing

[quantenschaum]

This one is a bit tricky. One would think that if I used 2 processes to compute the alamanc, it would only take half the time. Unfortunately this is not the case, because SkyField internally seems already to use all available cores (the sipce kernel I assume). So, splitting the work does just put more load on the already fully loaded cores, which will not reduce wall clock time. :( There is some reduction of the time due to filling the small gaps when the load drops a bit.

Generating one month on my 4 core machine takes 1m37s singlethreaded and 55s multithreaded, not 24s as expected.

[aendie]

I once looked into the possible processing speed enhancements using Skyfield and my report is here:
skyfielders/python-skyfield#612

[quantenschaum]

Very interesting. What puzzles me is, that SkyField is "pure-Python" but when generating an almanac in a single process, it still uses multiple cores. Any idea why?

SkyField uses numpy and numpy uses native parallel matrix vector operations, that might be the reason.

[quantenschaum]

IMHO the parallelization could be simplified. Concerning the fact that you do not see a gain in speed on Linux, you may want to read my comment skyfielders/python-skyfield#612 (comment).