=========================================================================

  P e r f o r m a n c e     T e s t s    f o r    C 1 4 n
  
=========================================================================

Here are some performance numbers that demonstrate

 a) how subtree based canonicalization costs almost same as XML serialization
 b) how nodeset based canonicalization is really bad for performance


Consider the following four algorithms

  * Algorithm A : Plain serialization
  * Algorithm B:  The very efficient subtree based C14N
  * Algorithm C: The moderately efficient nodeset based C14N, which does not expand out namespace nodes (the one mentioned in exclusive C14n spec, that Thomas pointed out)
  * Algorithm D: The extremely inefficient nodeset based C14N which expands all namespace nodes. (the one mentioned in inclusive C14N spec)


Algorithm A, B and C are available in JDK 1.6, and that is what I have used to demonstrate the performance (with permission from Sean Mullan)
The JDK 1.6 tries to use the subtree based code if possible. To make it use the subtree based algorithm, I just assign an ID to the subtree, and then create a Reference to this ID. But to make it use the nodeset based algorithm, I use the same reference to that ID, but then I add a Xpath Filter transform with an expression 1=1. This expression always evaluates to true, so this is exactly same as signing the subtree.


We will run these algorithms on these three xml files.

   * 5k_few_nodes.xml:  This is a file with very few nodes. There is just one very large text node
   * 5k_many_nodes.xml:  This is a file with many nodes, each node is very small
   * 5k_many_nodes_namespaces.xml: This is a file with many nodes, it also has many namespace nodes


Here are the numbers on my machine

----------------------------------------------------------------------------------------------
	                          Algorithm A |   Algorithm B   |  Algorithm C    |  Algorithm D  |
                             (serialize)  |  (subtree c14n) | (nodeset c14n)  |  (original)   |
-----------------------------------------------------------------------------------------------

5k_few_nodes.xml             |    3.0ms   |     4.0ms       |     6.7ms       |               |

5k_many_nodes.xml            |    4.1ms   |     4.3ms       |    21.5ms       |               |
	
5k_many_nodes_namespaces.xml |    5.0ms   |     5.4ms       |   164.0ms       |               |

-----------------------------------------------------------------------------------------------
	


Compiling the code
==================
You only need JDK 1.6, no additional jars. Compile like this

javac w3c/xmlsec/TestSignPerf.java



Running the code
================

Run like this

java -server -cp . w3c.xmlsec.TestSigPerf 5k_many_nodes_namespaces.xml 1000 signWithNodeset

Arguments
  -  First argument is file name to be signed. The code looks for a soap:Body element and signs it
  -  Second argument is number of iterations. the code does an initial run to load the classes, and then it runs
      these many number of iterations - give a large value like 100, so the JITC optimizes the code paths
  - Third argument is the "mode" , it can be "serialize", "sign" or "signWithNodeset" corresponding to 
       Algorithm A, B and C