EasyVVUQ Scaling and Performance

[orbitfold]

Discuss!

https://ucl-ccs.github.io/EasyVVUQ/dev/bench/

[orbitfold]

See https://github.com/UCL-CCS/EasyVVUQ/blob/dev/tests/test_db_benchmark.py for an example of how to do benchmarks with pytest

[djgroen]

Hi @orbitfold. I do think this is extremely useful, and this will also be beneficial to benchmark the multi-threaded job submission of FabSim3, which is similarly performance critical. :)
I'd like to include Hamid, but he's not in this repo ;).

[orbitfold]

ok I'll add him

[djgroen]

I notified Hamid of this activity now. But before we can use this for FabSim3 we are actually going to port the multithreading implementation to multiprocessing.

[orbitfold]

I recommend you look into python futures https://rednafi.github.io/digressions/python/2020/04/21/python-concurrent-futures.html

[DavidPCoster]

Plot of time taken for a PCE campaign with increasing PCE order.
I have plotted the total time and then the time taken for the three most expensive phases.

[DavidPCoster]

Results collation seems to be scaling with the square of the number of points, and the analysis phase even faster ...

[orbitfold]

https://www.whoishostingthis.com/compare/sqlite/optimize/

Some SQLite optimisation ideas. Some of them we already do. But the caching stuff is interesting to look into.