search

GEGlobalResearch / DARPA-ASKE-TA1

ANSWER Project to demonstrate knowledge-driven extraction of scientific models from code and texts
Other
8 stars 5 forks source link

performance investigations #77

Closed kittaakos closed 4 years ago

kittaakos commented 4 years ago

@crapo, I am opening this thread to share my findings regarding the performance. It turned out very quickly, the bottle-neck is the Turbo.sadl, opening and touching (without saving) the file requires validation, the validation of this file takes ~7-8 seconds on my environment:

MacBook Pro (15-inch, 2016)
2.9 GHz Quad-Core Intel Core i7
16 GB 2133 MHz LPDDR3

I started to profile Turbo.sadl. I open the file wait for the validation, start profiler, touch the file by adding a single whitespace stop the profiler.

We spend most of the time at com.ge.research.sadl.reasoner.utils.SadlUtils.classIsSubclassOf(OntClass, OntResource, boolean, List<OntResource>). See the attached HTML (inside the zip).

Call-tree-–-All-threads-merged.html.zip

I am going to play with this and try to figure out where can we speed it up.

kittaakos commented 4 years ago

Upstream PR: https://github.com/crapo/sadlos2/pull/405

crapo commented 4 years ago

Great job Akos!! This is refreshingly better!

kittaakos commented 4 years ago

Great job Akos!! This is refreshingly better!

@crapo, if we want to further increase the performance numbers, we can still disable the "deep-validation" you have suggested last time. Let me know if you need it.

crapo commented 4 years ago

For the moment I think we're good. Let's see if we need further improvement before taking that complicating (for the user) step.