Entering edit mode
4.8 years ago
daveturner
•
0
When I looked into ABySS a few years ago (version 1.5.2), several post-processing steps used OpenMP but the parallelization of ABYSS-P was done in MPI which I think is still the case in the current code. My performance tests showed less than ideal scaling on single node runs due to the enormous number of very small, latency-bound MPI messages being passed between MPI tasks on the same node. Has there been any thought of doing an OpenMP version of ABYSS-P to provide better scaling efficiency for single node runs?
You might need to edit the post (not to use the code chunk mode) to increase the readability.
No worries, updated by lieven.sterck already.