[ZPT] Speed of ZPT in Jython vs Cpython.

[Alan Kennedy]

Greetings,

I did manage to get some of ZPT working in Jython with a 
couple of trivial changes. I realise now that a lot of code 
in the current ZPT implementation is necessary for 
integration with Zope. Understandable really :-)

So what I did was put a stopwatch on the compilation and 
interpretation of the HTML TAL/tests/input/test01.html, 
parsed using the HTMLParser, and evaluated using the 
TALInterpreter.

The figures for Jython were a little disappointing. I tried 
two different JVMs, and both of them were orders of 
magnitude slower than CPython. The timings in milliseconds 
are, as recorded on a 1.2Ghz Athlon,

                                Jython            Jython
               Cpython         (JView)         (JDK1.3.1)
---------------------------------------------------------
Compilation         50             270               350
Interpretation      10              95               190

I tried Microsoft's Jview because it is reputed to be the 
fastest JVM around. And the figures do bear out that it is 
comparatively faster than jdk1.3.1.

These sort of performance figures are a little worrying, 
because it looks like evaulation of even a simple TAL page 
will take a lot of CPU. So, for reasonably complex pages, I 
could probably expect to turn out 2 - 3 pages a second on 
machine with mid-range speed.

Hmm, it's not going to work that well in a web server 
environment :-(

Are there any quick hits to speeding up ZPT in Jython? 
Suggestions?

I did think of trying to write the whole thing in pure 
Java. I'm not yet sure how much work this would be, and I'm 
loath to commit to that until I know what I'm letting 
myself in for.

But now that I know the structure of ZPT a little better, 
it seems to me that I can have a reasonably complete XML 
TAL engine by translating the following files:

TALParser.py
TALGenerator.py
TALInterpreter.py
XMLParser.py
DummyEngine.py

Also, it appears to me that DummyEngine is really not so 
dummy, in that it evaluates the full range of Python 
expressions. It's only when you get into path based 
expression syntaxes, security models, etc, that something 
more sophisticated is needed. Is that correct?

Cheers,

Alan.



---------------------------------------------
This message was sent using WebMail by CyberGate.
http://www.gate.net/webmail