>From: "Adam Winer" <[EMAIL PROTECTED]>
>
> On 4/18/06, Gary VanMatre <[EMAIL PROTECTED]>wrote:
> >
> > >From: "Martin Marinschek" <[EMAIL PROTECTED]>
> >
> > >
> > > Facelets are faster - the JSP overhead goes with them. Adam quoted 14%
> > > speed gains by using Facelets.
> > >
> >
> > Is that metric a comparison of the time it takes to compile a JSP versus
> > parsing the XML document or is that averaged out over several invocations?
>
> This was averaged out over, say, 1500 invocations, and after a warmup
> invocation so that compilation and parsing are not part of the
> problem. I had some more recent benchmarks were I got even larger
> performance improvements. Plus, because Facelets instances are
> reused, they provide a *very* handy place to cache data structures,
> wh ich I' m looking at taking advantage of to improve the performance of
> component assembly.
>
> If you were just talking the first request for a given page, and
> comparing compilation time in JSP versus Facelets parsing, I'm
> guessing the difference is more like 90%. I've never measured it, but
> "godawful slow versus blazingly fast" is a handy summary. ;)
>
>
> On 4/18/06, Gary VanMatre <[EMAIL PROTECTED]>wrote:
> >
> > >From: "Martin Marinschek" <[EMAIL PROTECTED]>
> >
> > >
> > > Facelets are faster - the JSP overhead goes with them. Adam quoted 14%
> > > speed gains by using Facelets.
> > >
> >
> > Is that metric a comparison of the time it takes to compile a JSP versus
> > parsing the XML document or is that averaged out over several invocations?
>
> This was averaged out over, say, 1500 invocations, and after a warmup
> invocation so that compilation and parsing are not part of the
> problem. I had some more recent benchmarks were I got even larger
> performance improvements. Plus, because Facelets instances are
> reused, they provide a *very* handy place to cache data structures,
> wh ich I' m looking at taking advantage of to improve the performance of
> component assembly.
>
> If you were just talking the first request for a given page, and
> comparing compilation time in JSP versus Facelets parsing, I'm
> guessing the difference is more like 90%. I've never measured it, but
> "godawful slow versus blazingly fast" is a handy summary. ;)
>
Wow, That's outstanding! I've not tried to profile Clay but I'm pretty sure it's
not in the same ball park. It has allot of caching but the symbols and inheritance
resolution take some processing.
I've wondered why the JSP compiler was not retooled to build a component tree?
That would seem to be the place to take the rendered output of a non-faces
JSP tags and stuff it into a verbatim component versus the reprocessing of the
rendered markup looking for markers. I have to believe that a retooled faces
jsp compiler would perform better since there wouldn't be any processing of
metadata. The logic would be hard wired to build a specific component tree
within a servlet. But, at the same time, I'm sure that if I looked at the code gen
for JSP, it would scare the pants off me and I probably wouldn't be able to sleep
for weeks :--)
> -- Adam
Gary
