Details
-
Brainstorming
-
Status: Closed
-
Minor
-
Resolution: Fixed
-
None
-
None
Description
This Jira covers experimental, disabled code done as an experiment to explore performance implications of putting in barriers in a UIMA pipeline, so that multiple threads would wait at the barrier until a bunch of them had arrived, and then they would be released together. The intuition motivating this was that in highly multi-threaded machines, running the same annotators together, might improve locality-of-reference for l1/l2/l3 memory caches, potentially speeding up execution.
After one set of experiments which failed to show improvement (for possibly many reasons, not yet investigated), this code is disabled, but kept in the project in case of future interest.