Re: XSLT Performance (was BLANK)

[Rick Jelliffe <ricko@allette.com.au>]

From: Lynda VanVleet <lyndavv@earthlink.net>

> When talking with David Turner (Microsoft), he calls
> them "perceived performance issues."  Is performance of XSLT
implementations
> for largescale EAI/B2B integrations real or perceived?

As part of a testing program for Schematron 1.5 reference implementation, I
recently tested 12 different XSLT applications (see Screamatron Torture Test
at
 http://www.ascc.net/xml/schematron/1.5/#conform ).

I cannot provide timing figures for the MS product and the Oracle product.
But I found an 20:1 performance difference between different
implementations, from under 5 seconds to over 80 seconds.   Even disgarding
the  worst as anomalous and probably temporary, there was a 5:1 difference.

This indicates that one cannot generalize about XSLT implementations: they
vary so much.  By the benchmark, there are clearly several implementations
that are probably not suitable for use on Webservers. I note that I used
command-line versions, so this is akin to an inefficient CGI use; the
results may not be indicative of optimal performance with preloaded
stylesheets and used in servlets without Java initialization overhead.

Anyone who is installing a high-volume or speed-sensitive XSLT website is
well-advised to spend a day benchmarking XSLT implementations. It could
easily determine whether the project succeeds or fails.

Cheers
Rick Jelliffe

Disclaimer: The tests represent more a diary of what I found, and are not a
systematic test of XSLT features or performance, and that implementations
are improving every week so what was true two weeks ago for the test may
already be out-of date. And also that an XSLT processor that performed
poorly on my particular test may be perfectly acceptable for your uses. And
also that  inclusion in that test does not represent any endorsement, but
was just the PC-based shrinkwrapped and Java-based implementation that we
convenient at the time.  The main purpose of the test was not to test XSLT
implementations but to test my code for portability and develop workarounds.
The test results are not available for quotation in commercial material.