The Hunt for Red October^W^W ActionWebService

We are looking at possibly doing a SOAP API based on Rails, PHP, or the elusive ???, and it all depends on speed. We figure that one of the largest hang-ups would be parsing the XML. So yesterday, I started hunting through ActionWebService trying to answer the following: just how quickly can ruby parse SOAP requests?

I first found myself in the ActionWebService source code, where I quickly discovered that there were lots of SOAP::* calls, meaning ActionWebService uses the SOAP module that is in the Ruby standard library.

So, into the SOAP source code - let's look at SOAP::Parser. Right off the top we find it includes xsd/xmlparser, and it looks likes the SOAP parser is set up with an event-handler parser - which is certainly a good sign. So let's look at XSD::XMLParser, where things get interesting.

Finally, in xsd/xmlparser.rb we find this:

# Try to load XML processor.
loaded = false
[
  'xsd/xmlparser/xmlscanner',
  'xsd/xmlparser/xmlparser',
  'xsd/xmlparser/rexmlparser',
].each do |lib|
  begin
    require lib
    loaded = true
    break
  rescue LoadError
  end
end
unless loaded
  raise RuntimeError.new("XML processor module not found.")
end

Thus, Ruby will use three different libraries, in order: xmlscan, XMLParser, and REXML. I had already been investigating XML parsers, but had not heard of xmlscan. Well, xmlscan is The fastest XML parser written in 100% pure Ruby. Odd, I think: why not prefer XMLParser, the expat-based library, over a Ruby-based one? Must be faster, right? Let's find out.

If you download xmlscan you find it includes a file samples/xmlbench.rb - a simple benchmarking program. If you run it with a bunch of libraries installed, it gives you results. How novel! I ran it against a 280kb XML file:

$ ruby samples/xmlbench.rb ../test.xml
                                    user     system      total        real
** File **
XMLScan::XMLScanner             0.390000   0.040000   0.430000 (  0.429473)
XMLScan::XMLParser              0.440000   0.030000   0.470000 (  0.473714)
XMLScan::XMLNamespace           0.490000   0.060000   0.550000 (  0.549025)
XMLScan::REXML                  0.940000   0.120000   1.060000 (  1.053696)
REXML::Document.parse_stream    0.810000   0.120000   0.930000 (  0.925994)
REXML::Document.new             1.280000   0.130000   1.410000 (  1.414166)
XMLParser                       0.040000   0.010000   0.050000 (  0.036426)

Well, it seems XMLParser knocked the socks off the competition. Tell me then, why would the XSD parser use xmlscan over XMLParser? I am not sure. But it looks like I will be uninstalling xmlscan to make sure to maximize SOAP performance.