[xsl] Re: Re: Re: RE: Re: lookup-table thoughts (was Re: matching multiple times, outputting once?

Subject: [xsl] Re: Re: Re: RE: Re: lookup-table thoughts (was Re: matching multiple times, outputting once?
From: Dimitre Novatchev <dnovatchev@xxxxxxxxx>
Date: Thu, 8 Nov 2001 20:17:45 -0800 (PST)
Jeff Kenton <jkenton at datapower dot com> wrote:

> .. From: Dimitre Novatchev <dnovatchev@xxxxxxxxx>
> ..
> .. Sometimes ago I asked the group whether it would not be appropriate to
> .. this extreme case of capability to be processed in parallel by introducing a
> .. XSLT instruction (e.g. xsl:parallel) that would give the XSLT processor a hint
> .. try to multi-process the children of xsl:parallel. Compare this to the current
> .. version of the language, where any two content-producing siblings can in theory
> .. parallelised, but which is never done, partly because there's no clear
> .. which of many possible alternatives is worth parallelizing.
> ..
> .. I still think an explicit hint is a necessary and a very useful feature.
> ..
> Most people (current company excluded, of course) do a bad job of understanding
> parallelism and providing hints about it.  It's hard to implement, but still
> probably better to have your XSLT processor do its own parallelization

This statement only shows that my message was not well understood.

Its essence was that ***when using a DVC algorithm*** it is perfectly clear that
going parallel will be very useful. It is exactly in this case, in which the
programmer would achieve very high efficiency, if he/she is able to indicate this
fact to the XSLT processor. 

By definition (and as can be seen in other languages -- see for example Transact
SQL)the ability to specify a hint does not mean by itself that transformations will
be run more efficiently -- it is the programmer's experience and judicious use of
hints that can make a big difference in performance.

A DVC algorithm implementation is one clear and useful usecase for xsl:parallel,
maybe there exist other such cases. In all such cases going parallel will result in
immensely improved efficiency of the transformation.

Dimitre Novatchev.

Do You Yahoo!?
Find a job, post your resume.

 XSL-List info and archive:  http://www.mulberrytech.com/xsl/xsl-list

Current Thread