Hide Forgot
++ This bug is a clone of bug 780100 ++ project_key: SOA Attached test report contains 4 scenarios tested against default EDS repository. I tried to run tests against PostgreSQL and they exhibits the same behaviour 1) DepthOnlyPerformanceTest - new nodes are added to increase the depth of tree. The nodes are added in batches of ten nodes. Both write and read operations shows exponential increase of operation time with enlarging tree depth. 2) BreadthOnlyPerformanceTest - new nodes are added to simulate enlarging number of data stored in repository. The read operation time is constant, write operation time is linearly increasing with the number of records in repository. 3) PropertiesOnlyPerformanceTest - new properties are added to a single node. No issue - constant read and write time with increased number of properties. 4) ComplexPerfromaceTest - new nodes are added with depth 5, the leaf node always contains five properties, one of which is bigger. The scenario shows the same behaviour as BreadthOnlyPerformanceTest. I am attaching also test report charts and source code.
Attachment: Added: source.zip Attachment: Added: perfresult.ods
Candidate for SOA 5.1.0. Van will see if Brian can take a look.
Link: Added: This issue depends MODE-923
Changes have been committed to fix a performance issue. Can the tests be rerun to verify improvement?
Link: Added: This issue related JBQA-4030
Link: Added: This issue related BRMS-502
I reran the the tests and the results are almost same as before. (graphs are in attached ods file). There are some small changes, but the tests was ran on the diffenrent databases, so this could be the reason. Note: Original tests were run with local PostgreSQL DB. I ran them with remote Oracle 10g DB.
Attachment: Added: perfresult_modeshape_from_SOAER6.ods
Security: Removed: Public Added: JBoss Internal
Link: Added: This issue is related to BRMS-502
There is retesting targeted for 5.2 based on the BRMS-502
Retesting required
Randall Hauch <rhauch> made a comment on [bug 724554|https://bugzilla.redhat.com/show_bug.cgi?id=724554] I would agree that "createSnapshot" does perform significantly slower than JR, but the other performance numbers are not significantly higher than with JR. Obviously we're still trying to improve performance, though it's not likely we can get much improvement by the release.
Release Notes Docs Status: Added: Not Required Writer: Added: dlesage
Bugzilla References: Added: https://bugzilla.redhat.com/show_bug.cgi?id=724554
Bugzilla References: Removed: https://bugzilla.redhat.com/show_bug.cgi?id=724554 Added: https://bugzilla.redhat.com/show_bug.cgi?id=724554
Tried in 5.2 ER5, behaviour is almost the same, see attached results. Tests were done with PostgreSQL.Did you try to analyze SQL commands to verify that all is optimal and properly indexed?
Attachment: Added: perfresult_5.2_ER5.ods
I run the tests against 510GA and 520ER5 with Oracle10g and the results are almost the same too.
Actually there is a performance regression in case of "Depth only" scenario. See the "ModeShape Charts" sheet of the results document (https://docspace.corp.redhat.com/docs/DOC-24649).
The next major version of ModeShape will target this area.
*** This bug has been marked as a duplicate of bug 780100 ***