On Mon, Jan 17, 2011 at 22:27, Andreas Veithen <[email protected]> wrote: > And here is the answer to a question asked by Amila about this: > >> Have you measure the performance factor? >> One of the problems I see with the current DOOM implementations is it is >> very in efficient in performance wise. > > Eventually the only real performance test is to run some scenarios on > Rampart with the standard Axiom implementation and then to rerun the > same scenarios with the Axiom+DOM implementation built on top of DDOM. > The Axiom implementation is work in progress and there is not enough > of it yet to run Axis2. There are also some cleanups and changes in > Axis2, Axiom and Rampart that are required to make this work. > > There are actually three ways DDOM can potentially improve the overall > Rampart performance: > * It avoids conversions between Axiom and DOM. > * A couple of months ago I did some performance tests with CXF (with > Sun's SAAJ implementation vs. the SAAJ implementation built with DDOM) > and Dennis Sosnoski's signature and encryption scenarios. For these > scenarios, deferred building is not relevant because WSS4J always > reads the entire message. Nevertheless the results suggested a slight > performance improvement (although it is difficult to isolate the > contribution of the object model because the timings are dominated by > the cryptography stuff). Since Sun's SAAJ implementation extends > Xerces, this would mean that DDOM is as least as good (if not better) > than Xerces performance-wise. Therefore it is probably better than > DOOM. > * There is some particularly inefficient code in xmlsec, namely the > part where it extracts the base64 encoded content, decrypts it and > replaces it with the parsed XML. That is due to some intrinsic > limitations in the JAXP and DOM APIs. This could be avoided by taking > advantage of some of the advanced features in DDOM. Since only a > particular piece of code in xmlsec has this issue, this can be > achieved without the need to rewrite the entire lib and without > dropping support for standard DOM. > > Andreas >
There is actually a fourth way that it can improve performance (when ADB is used as databinding): the DDOM architecture is such that it has a natural solution for the problem described in [1]. Andreas [1] http://ws.apache.org/axiom/userguide/ch05.html#d0e1153 --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
