[jdom-interest] Performance benchmark results

[graham glass]

i noticed that you're using JDK 1.2. i think the version
of the JDK that you're using could affect the relative
performance figures.

cheers,
graham

-----Original Message-----
From: jdom-interest-admin at jdom.org
[mailto:jdom-interest-admin at jdom.org]On Behalf Of
philip.nelson at omniresources.com
Sent: Friday, May 11, 2001 10:40 AM
To: dms at drizzle.com; jdom-interest at jdom.org
Cc: dms at sosnoski.com
Subject: RE: [jdom-interest] Performance benchmark results


> After struggling with some DSL connection problems I've finally got my
> benchmark program and results from an initial set of tests
> online. This
> compares the performance of Crimson and Xerces DOM representations, as
> well as JDOM, dom4j, and Electric XML. The main page is at
> http://www.sosnoski.com/opensrc/xmlbench/index.html
>

Thanks again for doing this. While the results are somewhat disapointing,
they do provide some context for improvement and with a few of the tests, a
concrete target.  I will take a look at your code.  The difference you
report here between xerces-d and JDOM is nothing like what my own tests have
shown.  The only real difference I know about is the size of the document
but unfortunately, generating a trace for a 100K doc is not practical.

IMHO, the view of JDOM being lightweight, though currently under stress, is
in fact because of the fact that a faster parser could be used, if one were
available.  JDOM can't ever be faster than the sax parser it's based on now
but a custom parser could be faster.  I think it's important to note that in
my analysis for which I posted graphs, JDOM accounted for around 30% of the
build time itself with the underlying parser accounting for the rest.  That
would indicate that there is only room for a fraction of a 30% performance
gain unless someway could be found to improve the parsers portion of the
time.  I need to find out why this is not reflected in your results.


> Please let me know if you see any errors or have any suggestsion for
> improvements in the tests. I'm planning to add an update in a
> couple of
> weeks with results using new versions of the code bases,
> small files in
> addition to the medium sized (100-200K) ones used for these tests, and
> some added tests.

Yes, the smaller ones, 5-20K are much more relavent to me personally though
RPC is not the reason.  My experience with serialization of small documents
has been very good and in the place I am using it, I don't have the option
of xml serialization.

I decided to download and run your tests against the almost current version
(which didn't make much difference) and the small document I was using
(sorry, but I can't release that document but I am just using it for
comparison).  The makeup of the document is reported in the results.  The
results are closer to my own findings.