Lists Home |
Date Index |
> Right, premature optimization is indeed the root of
> much evil. But after 6 or so years of experience with
> XML, it doesn't seem appropriate to label concern
> about XML's inefficiency "premature."
True....but as others have pointed out, processing of XML can easily
not be the primary bottleneck. In nearly 30 years in this business,
I have come to the conclusion that you rarely know where the
performance sticking point will be....till you get there. My point
was not to optimize till you have (and can identify) the source of
the problems. Then, and only then, should you look at optimization.
So sure...if you implement an XML based solution, and it doesn't
perform, THEN you should look at optimization alternatives and
strategies, which may or may not involve Binary XML encodings,
elimination of XML altogether (in favour of a custom protocol),
refactoring and whatever.
> One big reason for XML prototypes to fail to scale up
> to production systems, AfAIK, is performance (or the
> cost of buying enough hardware to overcome performance
> problems). So, I think many organizations are at the
> stage of buying into XML's (broadly defined to include
> infoset-oriented technologies such as XPath, XSLT,
> XQuery, SOAP) mindshare, network effect, ease of
> implementation, etc. but are blocked by the
> optimization issues.
I did not mean to infer that you should not design your apps to be
performant. It's a big grey area as to when do you cross the line
from good design practice (which considers performance issues) to
indulging in premature optimization.
> Good point! Still, you can't protect people from
> their own folly.
True. Common sense isn't. Maybe it's a problem with peopleware and
not XMLware? ;-)
> There will be some people in the
> situation David Megginson described where reducing XML
> parsing time to zero won't significantly affect
> overall application performance, but they will use
> performance optimized techniques anyway.
Which would seem to validate my point, given that often you don't
know this in advance. Prototypes can be a good thing when it comes
to performance testing.
> One can only hope that Father Darwin will weed out that meme in the
> long run :-)
I prefer to help the good Father along. Ready, Fire, Aim! ;-)