Performance bottleneck when marshalling array types.

[heiko.braun]

Radim Tlusty wrote:

" I have made deeper analyse of this issue using JBoss profiler and I have found the main bottleneck.

Most of time was spent in class org.jboss.xb.binding.Content in method handleContent(ContentHandler). There is used LinkedList (attribute content), but inside of method handleContent() is called 'content.get(i)' . This list is big for large documents and the access of linked list elements by index is of course slow.

For my test purpose I have only interchanged the LinkedList by ArrayList and my test case which took before 97s is now ready in 7s. Only for comparison: in JBoss 4.0.4.RC1 with JBoss-WS4EE it lasts 2s and in Sun Application server 9 it's 0,6s. So there is still place for improvement.

[heiko.braun]

Alex wrote:

The Iterator is now used to iterate through the LinkedList instance instead of index-based access.

[heiko.braun]

Heiko wrote:

What's the reason to stick with the linked list? Compared to the ArrayList replacement, the Iterator approach doesn't really have an impact on performance.

[heiko.braun]

Alex wrote:

ran my benchmarks for LinkedList vs ArrayList. Iteration using Iterator is not very different. But the population is. E.g. up to 5000 elements, ArrayList appears to be faster. But for 10'000 elements, LinkedList is significantly faster. I agree, ArrayList is preferable in general. Perhaps, it makes sense to switch

[heiko.braun]

Well it depends what the majority of usecases is closer to. When i look at my test data, the 10.000 item xml file is >1mb and thus it should better be transfered as a binary attachment in in the first place.

As far as i can tell from the forums and support cases arrays that size are really edge cases.

From my point of view we should switch to the ArrayList.

[aloubyansky]

Ok, switched to ArrayList.

[cbax007]

Where can I get this fix? What release of JBoss will it be incorporated into? I looked at the JIRA issue and i couldn't see what files you changed via the change log. I would like to get and test this fix as I am experiencing the same issue.

[aloubyansky]

I have updated the http://repository.jboss.com/jboss/jbossxb/snapshot/lib/jboss-xml-binding.jar to include the fix. But it probably will take some time to pick it up. Otherwise, it's in CVS HEAD in common sub-module of jboss-head.
It will be in the next JBoss release, i.e. 4.0.5.RC....
As to the version control tab in Jira, I have to clarify it myself.

[aloubyansky]

FYI

Modified: src/main/org/jboss/xb/binding/sunday/unmarshalling
SundayContentHandler.java XsdBinder.java
Log:
switched from LinkedList to ArrayList, related to JBXB-65

Modified: src/main/org/jboss/xb/binding AbstractMarshaller.java
Content.java DtdMarshaller.java
NamespaceRegistry.java ObjectModelBuilder.java
Log:
switched from LinkedList to ArrayList, related to JBXB-65