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
When running GST using MPI with a very large number of cores we encounter what appears to be an edge case with the processor distribution heuristics that results in a distribution of processors that fails the layout creation stage. Attached is a cleaned up log along with a script and related files needed for reproducing this error.
I was running on feature-globally-germ-aware-fpr, but this should be reproducible on the tip of develop. Other relevant parameters:
20-nodes with 36 cores each for a total of 720 processors.
python 3.9.16
When running GST using MPI with a very large number of cores we encounter what appears to be an edge case with the processor distribution heuristics that results in a distribution of processors that fails the layout creation stage. Attached is a cleaned up log along with a script and related files needed for reproducing this error.
I was running on feature-globally-germ-aware-fpr, but this should be reproducible on the tip of develop. Other relevant parameters:
20-nodes with 36 cores each for a total of 720 processors.
python 3.9.16
Manually specifying a processor grid that is 20x36 looks to alleviate this error.
proc_dist_heuristic_failure.zip
The text was updated successfully, but these errors were encountered: