Lists Home |
Date Index |
I was actually pretty surprised to see how little difference there was
between Piccolo and Xerces using the Sun JVM. This used to be much
worse, based on my past tests (including the ones in the JavaWorld
article you linked - thanks for the plug, Rick!).
I don't see any reason to expect that multiple concurrent parsers and
pooling would make a difference. The tests reuse a single parser (and a
single instance of the XBIS reader), and data is read into memory ahead
of time so there's no I/O involved in the timings. That seems to
eliminate any possible gains from running multiple copies simultaneously
(and incidentally, it also actually weakens the XBIS advantage of
reduced document size for these tests, as opposed to real-world
applications where I/O is a factor).
As for the settings issue that Christian raised, I haven't looked into
whether any parser settings would improve performance. The documents are
all deliberately kept pretty simple, though (no external entities or
such), and with validation off there doesn't seem to be a lot else to
tune. Everything is in the download from the Sourceforge site except the
restricted-license xml.xml file (modified from the W3C original to
eliminate an external DTD, hence not redistributable), so it's pretty
easy to try it out for yourself if you'd like to experiment.
Rick Jelliffe wrote:
>Good to see Dennis using Piccolo as well as Xerces. The tests confirm
>still one of the worst-performing XML parsers* (though, of course,
>It would be useful to also have a test of scalability: what the rates
>are for multiple
>concurrent parsers, with pooling. Or do you think the results should scale?
>Christian Nentwich wrote:
>>this is really cool, finally some hard data. One thing that would make
>>your presentation even more convincing is if you dumped the property
>>and feature values set in Xerces at the time of running into your
>>This would help to convince the reader that you didn't overlook
>>anything, e.g. leave some kind of whitespace normalisation turned on,
>>didn't leave any schema validation turned on, etc.