RE: XSLTMark 1.2.1, first comprehensive XSLT benchmark

[Kay Michael <Michael dot Kay at icl dot com>]

> Standard remark: Java performance depends on JDK. Have you considered
> moving from 1.1.8 to 1.3? Should make quite a difference. 

Absolutely. Allow me to republish some figures I put out on the Saxon
mailing list yesterday. These are all running the same transformation, but
in different environments, and the total time varies by a factor of 15.
These tests are with a fairly large source file: I originally thought there
must be a problem with the Microsoft JVM results, but I've been
investigating further, and I think it's simply that the JDK 1.3 hotspot
compiler really gets a chance to cruise along if you give it a big chunk of
work to do.

Here are some execution times for a 2.75Mb source file (all times in ms)

Machine A (Pentium III, Win 98, 128 Mb)

Saxon 6.0, SUN JDK 1.3          10050
Saxon 5.5.1, SUN JDK 1.3        20160
Saxon 6.0, MS Java              38400
Saxon 5.5.1, MS Java            62780
xt, MS Java                     21000 approx

Machine B (Pentium II, Win NT, 64 Mb)

Saxon 6.0, SUN JDK 1.3          23694
Saxon 5.5.1, SUN JDK 1.3        49701
Saxon 6.0, MS Java              95778
Saxon 5.5.1, MS Java           148173 
xt, MS Java                    115000 approx 

I will try to do find time to do my own tests with the XSLTMark benchmark,
though I don't want to start getting into a "faster than thou" slanging
match with anyone.

Mike Kay


 XSL-List info and archive:  http://www.mulberrytech.com/xsl/xsl-list