Closed psavery closed 5 years ago
@psavery What about this one?
A side note about this PR: because we now have lists of calculations running in a single container, one thing that might be beneficial is that containers now have more flexibility when deciding how to run the list of calculations. Currently, they all just loop through the calculations and run them one by one (see here, for instance).
But there is potential for the containers to split up the list of calculations and run each calculation on a different thread (especially if the individual calculations are single-threaded), or maybe even run each calculation on a different compute node or cluster.
There is potential, but we want to avoid making the containers themselves too complex. Some of that might better live a level above them, where there is more visibility of the overall compute resources available, and let the containers themselves be relatively simple running through a serial list.
This is still a work in progress. It depends on #68.