[XSL-LIST Mailing List Archive Home] [By Thread] [By Date]

Re: [xsl] measuring bulk performance & turn around times of XSL t ransformations? ideas for: XML to XML, XML to HTML, XML to FO (then to P D F)

---

---

On Wednesday 28 April 2004 4:16 pm, you wrote:
> Of course Kevin, but once it is loaded and given large input, the
> transformation times were quite useful in telling me what coding
> technique was the fastest and which had the smallest use of memory. I
> obtained the memory-in-use information from the available performance
> counters.

Ok. For larger stuff you maybe fairly safe doing that. Although I would be
careful about how you handle input/output if there is much, it can have a
drastic effect on measurements. For our stuff we only do memory to memory
transforms to limit this. Most of this is covered in the Sarvega benchmark
that you mentioned. You can get this from www.sarvega.com along with the
last result set. If you just want the test driver email me off list and I
will send you a copy.

> This doesn't mean that I wouldn't be interested in setting up a much more
> reliable performance testing application or tool - like you're talking
> about, but most of the time, a developer's time is very limited :-)

Sure. I see a lot of transforms in the <10ms range where these things tend
to be more important.

On the other questions about measuring the stylesheet code efficiency, if
you are on a recent Pentium you can use some hardware counters to get very
accurate & cheap timing. I used to do this by hooking into the processors
function extension mechanism and inserting timing code around what I was
interested in. Better still tends to be having an XSLT profiler built into
or on top of the processor. I don't know much about the tools for MSXML to
do this, for obvious reasons, but I think there are some out there. Putting
a decent profiler in the Sarvega processor was one of the most useful
things we have done, it has saved hours of work.

Hope that helps,
Kev.

---