[Bese-devel] performance

[Henrik Hjelte]

On ons, 2005-11-23 at 03:11 +0200, Alex Mizrahi wrote:
> slower. as i understand it does lengthy initialization. further using
> links appears to be faster.

So the test is of limited value if it depends on what the page "does". A
slow database for example will make any language look slow. And timings
given in seconds say little if you don't know what hardware it runs on.

I do think people are interested in measuring the performance, at least
I am. Here are some thoughts:

Ucw "overhead" is really what happens before and after the application
code, so in order to measure this have a simple page that shows a "Hello
world" message. This could be compared as a percentage to timings for
downloading a static file from apache in order to abstract away hardware
differences. 

This will of course make ucw seem slow, and static html is no real
alternative to ucw. Maybe one could have a PHP (or Java or whatever)
solution to the problem of rendering "Hello world", then we could
compare ucw to the overhead of another framework which also connects to
apache. 

The "Hello world" test with html/PHP/UCW could be used to:
* Compare the performance of ucw under different Lisp implementations.
* Compare hardware.
* IF UCW needs optimizing there is a simple test that can be used with a
profiler such as the two in sbcl.

Of course it would be nice to be able to compare a more advanced
application, but that is also a much more advanced problem. I think
there is a big risk that it becomes misleading, and I personally don't
have the need to convince myself or a boss that ucw or lisp is
superior. 

Any objections or thoughts?  

I don't know PHP, but if someone makes that part I can make a script of
the rest. I didn't know of the ab program, but it seems good for the
simple test.

/Henrik Hjelte