Re: XSLTPROC performance

On 21/12/2007, Colin Paul Adams <colin@xxxxxxxxxxxxxxxxxx> wrote:
> >>>>> "Abel" == Abel Braaksma <abel.online@xxxxxxxxx> writes:
>
>     Abel> Often, people measure the timing including the startup of
>     Abel> the JVM. If you use Saxon properly, you keep the classes and
>     Abel> the compiled stylesheets in memory. You'll see that this
>     Abel> will improve your performance a lot, especially when talking
>     Abel> about many small XML files.
>
> No it won't, not unless you run it a second time!
>
> So it depends.

If you use Saxon's directory transform feature from the command line,
or use Kernow, then the cost of the JVM startup and compiling the
stylesheet is incurred once, not once per input XML file (which is
what Abel was getting at).

Kernow also has a basic performance testing facility which will run a
transform five times, ignore the first two and give an average time of
the final three - ignoring the first two should remove any "warm up"
costs from the average (Ive quoted the defaults there, its all
configurable)


-- 
Andrew Welch
http://andrewjwelch.com
Kernow: http://kernowforsaxon.sf.net/