Tuning performance for small XML documents

[Petr Kuzel <Pe...@netbeans.com>]

Hello,

we spotted startup time regression in our application after we had
integrated Xerces 1.2.3 parser instead of Crimson one. The application
reads during startup tons of small (<2k) XML documents.

A profiling tool marked UTF8Reader constructor as expensive
"method" that takes many% of the application startup time.
Looking at it I tried a simple experiment with utils.UTF8DataChunk.
I have changed naively CHUNK_SHIFT to 11. But there is a hidden dependency
resulting in:

org.xml.sax.SAXException: Internal Error: fPreviousChunk == NULL
java.lang.RuntimeException: Internal Error: fPreviousChunk == NULL
        at org.apache.xerces.framework.XMLParser.parse(XMLParser.java:969)

I have been looking for a documentation because sources are full of some 
literal constants and very hard readable, but failed. 

May be that this could be driven by a SAX property so applications handling
tons of small documents would set the property to a reasonable small
value decreasing array allocation penalty.

Does anybody know a workaround?

  Cc.

-- 
<address>
<a href="mailto:pkuzel@netbeans.com">Petr Kuzel</a>, Sun Microsystems
: <a href="http://www.sun.com/forte/ffj/ie/">Forte Tools</a>
: XML and <a href="http://jini.netbeans.org/">Jini</a> modules</address>