Here are the results of my comparison of various SAX parsers and the XML Pull Parser, a non-SAX parser. Tests were done on a AMD Athlon/1.8Ghz with 512MB RAM running Windows XP and Java 1.4.2 The SOAP test documents were obtained from Aleksander Slominski's site.

Contents

• Summary
• Testing Methodology
• Downloads
• Benchmark Results
• SAXBench configuration

Summary

Below is a description of the parsers I tested. Conformance reports for some of these are available at http://cafeconleche.org/SAXTest, though that page has not yet been updated for Piccolo 1.04. The latest release brings Piccolo into compliance with almost all of the XML 1.0 rules.

Testing Methodology

These performance benchmarks attempt to predict performance in a server environment. In such an environment, the parser will typically be instantiated once and then run many times, often on similar documents. In such a scenario, Java's Just-In-Time (JIT) compiler will compile the code early on, drastically improving performance for all but the first few parse operations. For this reason, SAXBench parses a document continuously for five seconds before starting to measure performance.

Most of the test documents are also read into memory first, to avoid I/O variability from the performance results. I have specifically chosen to leave the documents in a byte array and let the parsers do the character decoding rather than pre-read the documents into a String. Some parsers have optimized character decoding over Java's default decoding; I feel the decoding should be included in the results, because the input source in a server environment will often be an InputStream, which requires character decoding.

Downloads

Download SAXBench source

Performance Benchmarks

Parser Options

Test Options