You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
@AlexReedy has narrowed this down to something about the dask chunksize - works with a chunksize of 3000 but not 50, at least in the set up he was running. @kemccusker maybe this is enough of a clue to figure this out?
iterating a bit, on this 3000 was just some ridiculous value to see if it worked, tried at 200 and that was a little slow and 500 seems to be working pretty quickly.
default chunksize of 500 works on amarel and on docker (for default runs) I have pushed the changes into the development branch and left a further elaborating.
We can either close this or leave it open as the two systems have drastically different memory capacities so 500 may work for a default on docker but probably won't for a full run (66,000 locations) though I don't think anyone is doing that on Docker right now so... It's up to you guys.
May be better to leave open so @kemccusker can confirm the changes work. Note for @kemccusker, last week we just swapped out your totaling script for the old old one. You can
Clone the new dev branch
CP the total_workflow.py script from the current dev branch into yours
If you are actually still using the version of total_workflow.py that uses dask all you have to do is change the chunksize (but I don't think you are)
The use of dask for chunking improves speed greatly on Amarel but does not currently work on docker.
https://github.com/radical-collaboration/facts/blob/development/modules/facts/total/total_workflow.py
The text was updated successfully, but these errors were encountered: