Running multiple SharedInterpreters, one per thread, does not seem to scale up

[andre...@gmail.com]

Hi,

I'm trying to parallelize a calculation that depends on python libraries.

To do so, I've created a thread pool and associated one SharedInterpreter

for each of the threads. 

The calculation works, but it does not scale up, and indeed it's not any faster than doing it sequentially: while a jstack does not report any visible synchronization at the Java level, I'm suspecting some sort of synchronization at a lower level.

Is this normal? Can I do anything about it?

Cheers

Andrea

[Ben Steffensmeier]

Since jep uses cpython directly, it is limited in concurrency by the cpython GIL. See the wiki page for more information.

Ben