[jdom-interest] Performance benchmark results

[Dennis Sosnoski]

Because of the size of the documents I was using I needed to run multiple passes
on each document to get accurate results - running on a slower system would help
with this! If I can find some documents in the 1-2M byte range it'd be
interesting to see timings for a single pass on those, to.

Let me know what you find out with the other parsers. The only other one I was
aware of for SAX2 (aside from Crimson, which didn't show much difference from
Xerces in my early tests) is the updated AElfred parser. I think that one's
under the GPL, so incorporating it into applications may cause problems.

  - Dennis

philip.nelson at omniresources.com wrote:

> > I'll
> > probably either disable the run all models mode or require it
> > to be specifically
> > requested for the next iteration of the benchmark, rather
> > than have it be the default
> > as now.
>
> Makes sense, I'll use it that way.  Did you do much testing where you ran
> once or twice?  I would think repeatedly parsing the same document would not
> be a great test, particularly depending on how the parser does or does not
> do string interning.  Maybe the "priming" should be done with a separate
> small document that has at least one of each xml node type but names
> unlikely to be in a real document.  I suppose for the use case of parsing
> config files, the current method *is* appropriate.  I don't know, just
> something to think about.  In my normal useage, the documents types are
> similar but the content varies.
>
> Fact: benchmarking is really hard to do well.
>
> I may also give it a go with some other parsers, just as a point of
> reference. If you haven't guessed it, performance is a big issue for me :-)