Re: XSLTPROC performance

>>>>> "Abel" == Abel Braaksma <abel.online@xxxxxxxxx> writes:

    Abel> I think what Colin tries to get at is that on most "real world"
    Abel> systems, you'll have to start the JVM. Depending on the type

Well, on some. 

    Abel> of system you'll have to start it once (a servlet which
    Abel> stays in memory) or you have to start it each time again (a
    Abel> workflow management system running every now and then a
    Abel> couple of XSLT).

    Abel> It's negligible in the case of the servlet, as it is part of
    Abel> the overall startup time. But it is not negligible if your
    Abel> system will have to deal with this startup time. In which
    Abel> case I agree with Colin that you shouldn't ignore it.

    Abel> What's fair or not depends on the context. Some people don't

Yes. Precisely.
That is why I was arguing that for a set of comparison benchmarks you
need to give lots of data. And in the case of a Java-implemented
processor, that includes the JVM start-up time for workloads where it
is necessary (in the case of a web-server application, it might be
reasonable to ignore it completely, even in a benchmark, but I
wouldn't advocate doing so even then).

    Abel> like the startup time of Diesel because they only do

Hey. We're talking XSLT here, not DSSSL!
:-)
-- 
Colin Adams
Preston Lancashire