Simple Perl XML Benchmark

This article presents a simple benchmark using various Perl XML modules (and some non-Perl solutions).

You will need a whole bunch of modules installed to test all of them. The test framework itself needs Getopt::Long, Benchmark, Text::Diff, XML::SemanticDiff and XML::Twig (and probably some more I have forgotten!). In order to run the XSLT examples you will need libxml2 install (which you will also need to you run all the XML::LibXML-based examples).

Remember that performance is just one (small) part of why you choose a solution: ease of use and power are also very important (for example the regexp based solutions are not generic and would not work for different XML files).

The tests

4 different tests are performed:

1. nothing: just parse the document and output it back,
2. extract: extract the content (text only) of all elements with a given name (message),
3. replace: prefix the content of all elements with a given name (message) with a text (including the element number), then output the document back,
4. complex: an element (process) has an attribute action used to perform an action on the element: delete, duplicate, change_tag (for a new, fixed, one), erase (the tags, not their content), prefix (with a new element) or add_att (add a fixed new element), the document is then output back as XML.

Not all modules are used for all tests, but hey... there's already a good number of them.

The size is computed by taking the process size (VmSize from proc/$$/status in an END block.

Conclusions

No conclusion yet... but you can draw your own ;--)

Just note that I am not advocating any of the tested solutions. They all work for the problems at hand (except that XML::Simple extracts properly all messages but does not return them in the right order), but that's all. Especially a number of them (mostly those based on regexps, but some of the SAX and XML::Parser ones too) depend on the original XML file being very simple: no entities, no non-7-bit-ascii characters, no comments, PI's, CDATA sections, no DTD, no namespaces, no > in attribute values (which is legal!), no nested message or process elements...

Contributors

I would like to thank Alberto Sim�es for contributing the XML::DT examples (in record time!) and Robin Berjon, Matt Sergeant and Barrie Slaymaker for their comments.

TODO

• comment the examples, especially the non-obvious ones
• add more!

Instructions

• download simple_benchmark.tar.gz,

• tar zxvf simple_benchmark.tar.gz
  
  perl run_all 

• if you want to test XML::SAX::PurePerl, you can add the --slow option to run_all, just be aware that it might take a while...

The test file (by default in test.xml) is generated by gen_benchmark, it size is about 3 Mb. You can tweak gen_benchmark to generate other sizes and other types of documents.

Note that you will need a whole bunch of modules and libraries installed to run all the tests:

• external libraries: expat, libxml2 and libxslt,
• XML modules: XML::DOM, XML::Filter::BufferText, XML::Filter::Dispatcher, XML::LibXML, XML::LibXML::SAX, XML::Parser, XML::Parser::Lite, XML::SAX::Base, XML::SAX::Expat, XML::SAX::Machine, XML::SAX::PurePerl, XML::SAX::Writer, XML::Simple, XML::TreeBuilder, XML::Twig,
• modules used by run_all: Getopt::Long, Benchmark, Text::Diff, Digest::MD5, XML::SemanticDiff, File::stat, Number::Format, File::ReadBackwards, Memoize, YAML, File::Slurp.

Results

Results on my machine:

Notes:
[1] In the extraction test XML::Simple returns all the messages, but not in the right order.
[2] The version of XML::SAX::PurePerl is the CVS version, which performs about 3 times faster than the CPAN one.

---

updated Mon Nov 15 14:18:58 2004

home

Copyright � 2003, Michel Rodriguez