XML 2004 Conference Diary, Day Three

by Edd Dumbill

Previous entries: day one, day two.



Hotel lounge. Wednesday, 10am



Lurking in the lounge after breakfast, I was greeted by Robin
Cover, Bede himself. The last time I met him was at XML Europe in
Paris four and a half years ago, at which point I was in considerable
awe. Cover has done an amazing job of chronicling XML and SGML over
the years, and it was great to meet somebody who puts my history as an
XML commentator into its rightful context. That is, I'm still the new
boy.



But maybe not quite so new as I used to feel. When I first arrived
in the XML world I felt, along with Simon St.Laurent, somewhat of an
incomer and a young turk. It's taken a depressingly short time for us
to become part of the establishment itself, of course. Instead of
angry young men we're a bit more the grumpy old ones.



Meeting rooms. 11am



Mike Champion is talking about XML and pain. Specifically, is the
pain worth the gain? Various facts that using XML for RPC is about
ten times slower than binary method such as Java RMI. He reckons the
bottlenecks for XML processing include well-formedness chechking,
Unicode conversion, character-by-character processing and node object
construction. And that's before you even hit schema validation.
The issue isn't text formats against binary formats, he says, but a
matter of XML's particular constraints.



Not everyone's suffering the pain, says Champion, but some
significant ones are: wireless industry, enterprise transaction
processing, and users of SOAP-based messaging. So how to relieve some
the pain for these folks? Well, Moore's law doesn't apply to
cellphone batteries or wireless bandwidth, so we can count out just
sittng and waiting. Compression's out, because it doesn't work for
small documents and adds a considerable processing overhead. Maybe
just write better code? Well, while XML parsers can get much faster,
that doesn't help bandwidth reduction anyway.



Other possible solutions include hardware acceleration, format
simplifications (e.g. SOAP's forbidding of DTDs), and binary XML
serialisations. The latter is the main avenue for investigation among
corporations and the W3C right now. There seems to be no obvious
solution with binary formats; often the most efficient solutions seem
to place higher constraints such as the requirement to have shared
schemas.



Where does Champion sit on this? He says "I come up firmly on the
side of waffling, I'm afraid." Convenience always comes at a
performance cost, and convenience always wins.
Unfortunately this means that there is real pain in niches that XML
really does have application to. The dilemma is that is XML is needed
in these areas, but the overhead is unacceptable.



Champion did make some recommendations, including not to deny that
there's a problem, and to take advantage of appropriate tools for the
job. Finally, beware of premature standardisation: "Leave evolution
to Darwin, not Berners-Lee".




Incendiaries, scammed drink tokens and hangover cures...