sn> My tests comparing the two methods showed no net difference.  Each
sn> method would slightly better the other in some tests, but in no
sn> test was either better by more than 10%, if I recall.

After originally posting the question today, I did run a short series
of tests (that I already had laying around) where that was all I
changed.  For typical SOAP stuff, by which I mean lost of relatively
small XML nodes, both modes scale pretty linearly.  The SAXSource
stuff mostly beat the DocumentBuilder stuff, but it was consistent
with your findings and was only a few percent better.

One aspect of the results surprised me. 

Some of the test cases I ran had only a half dozen or so XML nodes in
the response, but one of those was a String parameter.  It's actually
a CDATA block containing a non-SOAP blob of XML.  The times for those
test degraded much faster than linear with SAXSource.  By the time I
got up to a payload size of a couple megabytes, it was running 25 time
slower than DocumentBuilder for the same data (that's 25 times, not
25%).

I didn't have any time to look into, but I obviously suspect some
pretty inefficient String handling somewhere.  (The XML parser was
Xerces 2.4.)
-- 
[EMAIL PROTECTED] (WJCarpenter)    PGP 0x91865119
38 95 1B 69 C9 C6 3D 25    73 46 32 04 69 D6 ED F3

Reply via email to