Re: [xsl] FO profiling

[Michael Kay <mike@xxxxxxxxxxxx>]

When measuring Saxon performance, we generally find that running for about a
minute is enough to eliminate any Java warm-up effects. Of course the results
still vary from run to run, but I've always assumed that is random fluctuation
(e.g. due to garbage collection) that will remain random regardless how long
you run for. An hour seems quite excessive.

Michael Kay
Saxonica


On 3 Oct 2013, at 11:46, Jesper Tverskov wrote:

> Hi list
>
> For FOP I have made a crude system of a bat-file with a stop-watch
> (only centi-seconds are available) calling 1000 bat-files doing the
> same XSL transformation 1000 times to get an ok average with ok
> variation for my purpose.
>
> Each test takes an hour with the stylesheet I'm testing. E.g.:
> Replacing some SVG in the FO with an fo:inline element containing an
> fo:ruler element, reduced processing time 11%. It takes two tests, 2
> hours, to determine that!
>
> I'm wondering what other people do when profiling FOP. It would be
> nice to see a detailed list of hotspots like when using the XSLT
> profiler in Oxygen. I'm not much for doing the detailed time stamping
> of the FO output myself, if products doing it in a more professional
> and user-friendly way exist.
>
> I don't see any switch in the FOP documentation, I can turn on/off,
> but maybe I haven't looked long enough?
>
> I'm wondering if the commercial FO processors have such profiling
> build in? I'm also surprised about how many times you need to do the
> same test to get a reliable result. My thousand times are probably far
> from enough for more serious testing.
>
> Regards
> Jesper