Performance degradation due to PrimitiveLongMap
-----------------------------------------------
Key: DROOLS-446
URL: http://jira.codehaus.org/browse/DROOLS-446
Project: drools
Type: Bug
Components: reteoo
Versions: 2.0-final, 2.1
Environment: JDK 1.4.2_08
Reporter: Jerry Cattell
Assigned to: bob mcwhirter
Priority: Critical
Attachments: DroolsPerformanceExample.java, WorkingMemoryImpl.java,
odds.java.drl
I am experiencing performance problems using drools with a very simple use
case. I have 50-100 rules, 5-10 facts, and assert facts and fire the rules
hundreds to thousands of times per second. I have attached a basic example of
the problem I am encountering. You'll see that as it loops, creating new
WorkingMemory objects, asserting facts, and firing rules, the time to do this
takes longer and longer until it gets to an unusable state.
After running this through JProfiler, I found the hot spot to be
PrimitiveLongMap. When the RuleBase is cached and new WorkingMemoryImpls are
created (as the preferred approach seems to be), the generated FactHandle ids
continue to increase. The problem is that as the number gets large, say over 1
million, the PrimitiveLongMap seems to spend all of its time creating Pages, or
alternatively, iterating through all of the pages to return the collection of
values when WorkingMemoryImpl.getObjects is called.
I was able to fix this for my use case by simply replacing PrimitiveLongMap
with HashMap. Of course, I'm assuming PrimitiveLongMap was used to replace a
HashMap for a reason (very large number of facts? Long object creation?), so
this is probably not a long term solution for the drools project. So I see
three other options:
1. Fix PrimitiveLongMap. I haven't spent the time to try and understand what
type of work this would require.
2. Replace PrimitiveLongMap with a different implementation (perhaps one of the
ones listed on
http://blogs.codehaus.org/people/mproctor/archives/000890_primitive_map_implementation_and_benchmark_comparisons.html
).
3. Generate fact handle ids on a per working memory basis (as shown in the
WorkingMemoryImpl.java included). I don't know if the RuleBase ever uses the
FactHandle id in a global way or always just passes the FactHandle object
through to a WorkingMemoryImpl, so that could lead to potential problems.
Let me know if any additional details are required.
--
This message is automatically generated by JIRA.
-
If you think it was sent incorrectly contact one of the administrators:
http://jira.codehaus.org/secure/Administrators.jspa
-
For more information on JIRA, see:
http://www.atlassian.com/software/jira
