[XML-SIG] A bit o' challenge

Uche Ogbuji uche.ogbuji@fourthought.com
Thu, 03 May 2001 19:42:31 -0600

OK, so the conventional wisdom lately has been that the Java processors such 
as Xalan and Saxon cream 4XSLT for performance across the board.  Alexandre 
Fayolle said that he thought they were "orders of magnitude" faster.

Well, I know that one always does better in his own benchmarking, but I have 
been working with 4XSLT quite heavily in the time leading up to the 0.11 
release, and I'm having trouble crediting this impression.  4XSLT is to my 
observations (and measurements using the time command-line timer) a good 25% 
faster than Saxon and faster by an even greater proportion than Xalan for most 
small to medium tasks.

I have indeed noticed that on huge documents, such as the "Cemetary" benchmark 
(3MB source), Saxon 6.0.2 is up to 4 times faster than 4XSLT (similar for 
Xalan), but this is still not "orders of magnitude" faster, and this only 
seems to be true for the size and type of document I'd only expect to process 
in benchmarks.

Now one note: I *always* use cDomlette.  It is much faster than pDomlette, and 
that is why I've declared that I'll be working to make it the default in 
4Suite as of 0.11.1.  Once again, I encourage everyone to help shake out any 
remnant bugs in cDomlette.  See this posting for more info:


So here's the bit o' challenge.  I'm looking for regular-sized, real-world 
transforms in which Saxon or Xalan smoke 4XSLT.  If you have such test cases, 
and can reliably reproduce 4XSLT's lassitude using cDomlette, please send it 
my way so I can have a look (and maybe find the performance bugs that I'm too 
close to see).

I'm also interested, of course, in hearing positive reports about 4XSLT's 

So I say 4XSLT is competitive, and as far as I can tell, is usually faster 
than the Java processors (though we can't touch MSXML yet).

P.S. What got me starting to ponder was DataPower's benchmark that showed 
4XSLT some 20 times slower than the group of Java processors.  The nonsense 
behind this I was able to grasp with one glance at their tortured "driver" for 
4XSLT.   I've made my complaints about their incompetence, but here's your 
chance to show I'm all wet.

Thanks, all.

Uche Ogbuji                               Principal Consultant
uche.ogbuji@fourthought.com               +1 303 583 9900 x 101
Fourthought, Inc.                         http://Fourthought.com 
4735 East Walnut St, Ste. C, Boulder, CO 80301-2537, USA
Software-engineering, knowledge-management, XML, CORBA, Linux, Python