Re: XML::Saxon::XSLT2

Hi,

I am really not experienced in measuring performance. Here is just a first
attempt using the xml and xslt I am currently dealing with. If you are only
interested in the result see last paragraph.

If you want me to upload the actual files I can do so, maybe you also have
ideas for a better methodology.

Methodology for now is compare saxon via command line and through
lnline::Java.

I take a random real word xml file of mine with 10 MB size and the last
xslt2 transformation I wrote and run it with performance report on saxon
(e.g. -t -repeat:10)

Saxon reports: Average execution time over 10 runs: 6960ms (with xsl:message
see below)

Saxon reports: Average execution time over 2 runs: 5265ms (without
xsl:message see below)

I guess I have to add some time for the stylesheet compilation time: 1360
miliseconds (first) and 31 miliseconds (second) for the run without
xsl:message.

Now I run the same transformation (twice manually) thru XML::Saxon::XSLT2
(via Inline) and I get

Time taken was 26 wallclock secs ( 6.92 usr 1.33 sys + 0.00 cusr 0.00 csys=
8.25 CPU) seconds. (with xsl:message, but not printed to STDOUT)

Time taken was 24 wallclock secs ( 6.97 usr 1.27 sys + 0.00 cusr 0.00 csys=
8.23 CPU) seconds. (without xsl:message)

Now the whole thing is not really comparable, because via Saxon I output a
lot of <xsl:message>s, while via Inline I don't print them on the screen, so
I change that and uncomment my xsl:message and run everything again. Results
above. Also I should probably use perl's benchmark on both runs to get more
comparable input.

If I do that and run Saxon through a perl script via a system call and
benchmark it, it shows that saxon is considerably faster:
8-9 wallclock seconds indicating that Inline::Java is slightly less than 3
times slower (with these xml and xslt files). That's still faster than I
expected and for unifying my environment (everything practically in perl) I
am gladly paying that price in performance.

So far for now.

On Mon, Aug 23, 2010 at 10:06 AM, Tod Harter <
[email protected]> wrote:

> Yeah, there's nothing slow or buggy about Java XML processing. Nothing
> really slow or buggy about Java stuff in general, it is just not nearly as
> easy to slap something together quickly like it is in perl. Once it works it
> works well. OTOH I too would look at Inline::Java warily. TBH if you're
> going to be going that way anyhow it probably makes more sense to just write
> a java application. If the main goal is to run stuff through XSLT perl isn't
> adding a whole lot and there are extensive Java libraries and frameworks for
> that.
>
> On Mon, Aug 23, 2010 at 8:18 AM, Robin Berjon <[email protected]> wrote:
>
>> On Aug 23, 2010, at 14:11 , Thomas Krichel wrote:
>> >  Robin Berjon writes
>> >> Well, I can't speak for Toby but if you look at the source you'll
>> >> see that it relies on Inline::Java to do its magic. That's neat and
>> >> all but it's also a little bit scary — I don't know that
>> >> Inline::Java has seen that much real world usage.
>> >
>> >  Wouldn't that also make it slow? When I think JAVA I think
>> >  slow and buggy...
>>
>> The only way to know for sure is to benchmark it. I would *guess* that
>> crossing the Perl/Java boundary will come at a cost, but that what happens
>> inside of Saxon itself would be fast. Saxon is a really good implementation,
>> it's really fast.
>>
>> --
>> Robin Berjon - http://berjon.com/
>>
>>
>>
>> _______________________________________________
>> Perl-XML mailing list
>> [email protected]
>> To unsubscribe: http://listserv.ActiveState.com/mailman/mysubs
>>
>
>
>
> --
> The Wise adapts himself to the world. The Fool adapts the world to himself.
> Therefore, all progress depends on the Fool.
>
> _______________________________________________
> Perl-XML mailing list
> [email protected]
> To unsubscribe: http://listserv.ActiveState.com/mailman/mysubs
>
>

_______________________________________________
Perl-XML mailing list
[email protected]
To unsubscribe: http://listserv.ActiveState.com/mailman/mysubs