[jira] [Commented] (JXPATH-178) Contention on functions can be reduced
[
https://issues.apache.org/jira/browse/JXPATH-178?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14651537#comment-14651537
]
Rico Neubauer commented on JXPATH-178:
--
Only can provide some high-level tests from our application, don't have a
self-contained test - see below.
Right, java.util.concurrent is not available on JXPath's current target,
recognized this after the patch. backport-util-concurrent could be an
alternative.
{panel}
- original (no patch)
1 processes - 7:09 min.
2 processes - 15:36 min.
5 processes - 45:01 min.
- patched version
1 processes - 7:04 min
2 processes - 14:46 min
5 processes - 42:01 min
{panel}
Contention on functions can be reduced
--
Key: JXPATH-178
URL: https://issues.apache.org/jira/browse/JXPATH-178
Project: Commons JXPath
Issue Type: Improvement
Affects Versions: 1.3
Reporter: Rico Neubauer
Priority: Minor
Labels: multithreaded
Several series of thread dumps showed us that some avoidable synchronization
in jxpath causes plenty of blocked threads waiting on the sync block under
heavy load.
Using concurrent map makes the things better.
See this for a proposed fix:
https://github.com/seeburger-ag/commons-jxpath/commit/eeb7f30f761a6a89e03c2c1421c09968412659af.diff
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
[jira] [Commented] (JXPATH-178) Contention on functions can be reduced
[ https://issues.apache.org/jira/browse/JXPATH-178?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanelfocusedCommentId=14650369#comment-14650369 ] Michele Vivoda commented on JXPATH-178: --- Hi, do you have any test that shows the improvement ? I would be curious to see if things get better with 2 maps: one not synchronized for statically initialized classes and interfaces (eg: registerAtomicClass), that does not change, and one synchronized for those registered later. The ideal would be to register statically all the needed classes in addition to those initialized at begin in jxpath code, may be with some property file.. so that is possible to use a map that does not change and does not need to be synchronized. For you changes, I believe JXPath is JDK4 where java.util.concurrent is not available. Contention on functions can be reduced -- Key: JXPATH-178 URL: https://issues.apache.org/jira/browse/JXPATH-178 Project: Commons JXPath Issue Type: Improvement Affects Versions: 1.3 Reporter: Rico Neubauer Priority: Minor Labels: multithreaded Several series of thread dumps showed us that some avoidable synchronization in jxpath causes plenty of blocked threads waiting on the sync block under heavy load. Using concurrent map makes the things better. See this for a proposed fix: https://github.com/seeburger-ag/commons-jxpath/commit/eeb7f30f761a6a89e03c2c1421c09968412659af.diff -- This message was sent by Atlassian JIRA (v6.3.4#6332)
