jackrabbit-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Maxim Zinal (JIRA)" <j...@apache.org>
Subject [jira] [Commented] (JCR-3721) Slow and actively called NodeId.toString()
Date Sun, 09 Feb 2014 06:34:20 GMT

    [ https://issues.apache.org/jira/browse/JCR-3721?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=13895845#comment-13895845

Maxim Zinal commented on JCR-3721:

One more comment: I was able to (mostly) solve our performance problems by removing path hierarchy
constraint from all our application's XPath queries. Queries without path hierarchy constraing
(but with additional property constraints, which we had to add in order to get the same results
as before) run 10-15 times (yes, 1000-1500 percent) faster.

Our test (and production) database contain ~2 500 000 nodes with total size ~350 Mbytes, and
lucene filesystem indexes occupy ~1 Gb space, if that matters.

IMHO that's not a very good result for a JCR system, which is hierarchy-based by it's nature.

> Slow and actively called NodeId.toString()
> ------------------------------------------
>                 Key: JCR-3721
>                 URL: https://issues.apache.org/jira/browse/JCR-3721
>             Project: Jackrabbit Content Repository
>          Issue Type: Bug
>          Components: jackrabbit-core
>    Affects Versions: 2.6.5, 2.7
>         Environment: Debian/GNU Linux 7.3 / Oracle JDK 7 / Apache Tomcat 7.0;
> Windows Server 2008 / IBM WebSphere AppServer 7.0
>            Reporter: Maxim Zinal
>         Attachments: NodeIdToString.patch
> I performed some JackRabbit profiling while trying to investigate the reason of low performance
of our application.
> The mostly interesting thing I've found is that NodeId.toString() method is heavily used
for hierarchy-based XPath queries, and it performs really bad.
> This are the numbers for my test application:
>  - Total CPU time: 879 178 msec
>  - CPU time in NodeId.toString(), including subcalls: 223 705 msec
> A quick check against NodeId.toString() implementation shows that it is based on UUID.toString(),
which itself is very ineffective in both in Oracle and IBM JDK.
> I've wrote a quick replacement for this method, and my measurements show that overall
performance became significantly better for our case.
> Hope that this will help to improve JackRabbit performance for similiar applications.
> P.S. Another interesting thing I've found is that a lot of time is spent inside log4j.Category.getEffectiveLevel()
method - I suspect this is caused by numerous log.debug() calls without proper isDebugEnabled()

This message was sent by Atlassian JIRA

View raw message