[
https://issues.apache.org/jira/browse/OPENNLP-1816?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18069599#comment-18069599
]
Kristian Rickert commented on OPENNLP-1816:
-------------------------------------------
Real quick - the PR guidelines say that we need to run checkstyle - but almost
all the classes I changed have a ton of existing checkstyle violations in them
to begin with. So my commit isn't going to try to fix those, just the code we
have done.
> Make the ME object design thread-safe and faster
> ------------------------------------------------
>
> Key: OPENNLP-1816
> URL: https://issues.apache.org/jira/browse/OPENNLP-1816
> Project: OpenNLP
> Issue Type: Improvement
> Affects Versions: 3.0.0-M1
> Reporter: Kristian Rickert
> Priority: Major
>
> Heya - so I'm working on a PR to make OpenNLP ME objects thread-safe. so t
> I'll provide the benchmarks in a few moments -
> # Old way — new ME instance per call (what module-chunker does today)
> # New way, instance-per-thread — one ME per thread, reused across calls
> # New way, shared single ME — one ME instance shared across all threads
> So this would mean you don't need to pool ME instances or worry about
> creating new ones every time.
> So far my initial tests show a 1.67x speedup in processing times for the
> tests I'm doing.. I'll have a better benchmark and create the PR soon.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
