Multithreading on new MultiLCA in Ubuntu

[igt-misc]

Hi team,
I've noticed since the change to MultiLCA, it doesn't appear to be using multiprocessing the way it used to, and is primarily running on a single thread.

System in question is Ubuntu 22.04 LTS, on Python 3.12. I have an AMD CPU and have installed Pypardiso, and have tested both running a standalone script as well as in a Jupyter notebook. All BW25 packages up to date via pip. On Windows it appears to be running as expected, though I haven't tested as thoroughly.

I haven't done anything outside of the normal workflow:

• Make datapackages
• mlca.lci()
• mlca.lcia()
• Process results

Could it be a configuration issue on my side? Or is it just not possible to parallelize the new setup? For what it's worth, the MLCA I'm trying to calculate is between 1 and 7 activities with all 250-ish ReCiPe indicators, so I'm well aware I'm a very, very fringe case.

[cmutel]

pypardiso should use multiple cores, at least while solving Ax=b. The other operations are single-threaded.

As both pypardiso and modern scikits_umfpack are multithreaded, and maintaining the multiprocessing code was a pain, this was removed quite a while ago.

In your case I would recommend a different approach which can be much faster than the MultiLCA method - see here: https://gist.github.com/cmutel/159a99be89139a381a243035eb8d4c05. The main advantages of the MultiLCA approach were:

• Labelled functional unit groups
• Correlated sampling of normalization and weighting values

Pretty sure this code will evolve over time as what we have right now isn't really solving real user problems.