Re: XSLTPROC performance

On 21/12/2007, Colin Paul Adams <colin@xxxxxxxxxxxxxxxxxx> wrote:
> The cost of the JVM start-up cannot be eliminated, but it can be
> amortised. For a fair comparison, you need to know what this overhead
> is, so you can make sensible judgements.

I don't think it needs to be amortised... If you start the timing at
the correct point and ignore the first few runs then I'd say you can
safely forget about JVM startup and warm-up for the remaining runs.
The only thing then is when the garbage collection runs, but you can
mitigate that by performing enough runs and removing any outliers -
leaving you with a reasonble comparison time.

You'd only need to worry about amortising JVM startup time if you
started timing at the beginning, run the transformation a number of
times, and then stopped timing at the end of the last transform,
dividing the overall time by the number of transforms...


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