Re: XML Protocols Shakedown

> Any reports of the XML Protocols Shakedown (yesterday?) up yet?

Let's see what I can recall, with a little help from my notes...

Apologies if I'm remembering things that they didn't really say or ommitting
something they considered important...

We started with self-introduction of the  panelists...

   o Dave Winer, Userland
     XML RPC designer, collaborated on SOAP
   o Henrik Frystyk Nielsen, Microsoft
     HTTP designer for too many years...
     co-author of SOAP
   o Noah Mendelsohn, Lotus/IBM
     collaborated on SOAP
     also representing IBM's interest in ebXML
     interested in ubiquity of basic ebusiness protocols, ala the fax machine
   o Ken MacLeod
     background in Unix systems design and lightweight protocols
     agnostic w.r.t. the present proposals
   o Henry Thompson,  University of Edinburgh
     experience with LISP RPC at Xerox PARC
   o Michael Condry, Sun Microsystems
     focussed on user requirements

I gave a little of my own background...  comp.lang.{python,perl,...}
background, FFI stuff, W3C Mobile Code workshop in '95, OMG/W3C workshop on
objects in the Web in '96, etc. And I acknowledged the contribution of Janet
Daly and Eric Prud'hommeaux in setting up the panel and recruiting the
panelists.

Then I gave some context... the WBXML and ICE submissions to W3C, the idea of
a W3C workshop on XML in protocols and distributed applications that didn't
quite happen, lunch at XML '99, XTech '99 and IETF 47 in March, leading up to
this WWW9 session, and keeping in mind that focussed discussion on how W3C
should
deploy its resources would happen the following week, so we should freel free
to stick to technical issues in this panel.

The each panelist had 5 to 10 minutes to give the ideas they think are most
exciting/relevant/important. We went in the opposite order:

Michael C. emphasized the user experience. He raised issues with basing new
protocols on HTTP by reference to the recent draft by Moore

On the use of HTTP as a Substrate for Other Protocols
4 May 2000

Michael also questioned the strategy of leaving security to layers to be
specified later, and emphasized the need for neutral governance of any spec
that is intended to be used in competitive B2B environments, and in
particular, the need to coordinate with the IETF.

Henry T. noted the analogy between XML and LISP:

 (foo a b c) ~= <foo><a/><b/><c/></foo>

and recalled that as soon as the folks at Xerox PARC had a LAN, they started
doing RPC by printing an expression, having it evaluated on another machine,
and returning the result. He then noted the benefits of and trend toward
declarative data formats over
and away from executable formats.

He then gave a fairly novel view of how XML should be used in distributed
applications, which is to shift focus away from the messages exchanged and
toward the audit trail of contributions of each party to the conversation. He
sketched a mechanism using HTTP, XLink, and XInclude (aka translclusion)
rather than new protocol infrastructure. I can't reconstruct the details, but
the audience received it with applause.


Ken M. explained the need he saw for a base protocol that does have

   * simplicity
   * flexibility
   * security

but left to higher layers

   * RPC and other message patterns
   * Quality of Service (QOS)

Noah M. gave two different use cases at very different design centers:
(1) a one-off stock quote fetch, or (2) high-volume ecommerce
relationship with high robustness requirements.

[Hmm... my notes say "RPC cf Bruce Nelson" but I don't recall
what that means. Same for "SOAP enc + method?"]

He explained a sort of "ying/yang" tension in this design space between
 simple vs. robust
 fast deployment vs. space-shuttle robustness
 self-describing vs. external schema

He suggested that W3C should work towards ubiquitous infrastructure support
for XML protocols, akin to the way that business peers can reasonably expect
each other to have fax machines.

My notes show a couple more use cases, though I'm not sure
it was Noah who gave them:

   * credit card validation [ala zip-zip machines]
   * ordering widgets

Henrik F. N. noted that a lot of XML protocols discussion related to which
programming model was best for building distributed applications, and
suggested this is a red herring, since lots of programming models are used in
the Web, and that fact isn't likely to change, nor should it.

But, he said, there's no good reason to put a centralized barrier of 2 years
of community scrutiny in front of people with new designs for distributed
applications.

He sketched a image of an oreo cookie, with TCP on the bottom and application
layer stuff on top, with SOAP providing the creamy filling in between.

Henrik said a success criterion for this work is that folks with new ideas can
implement/deploy them without re-inventing infrastructure like namespaces.

Dave W. encouraged us to focus on the user benefits. He observed some
essential characteristics of technologies to be deployed in the web: low-tech,
and no lock-in. He described the achieving a new feature of the web with
user-visible benefit: spell-checking service provided by 3500 unaffiliated
sites interoperating via a simple XML-RPCish API/protocol. He mentioned other
features that could be deployed this way: search and user registration.


Questions from the floor followed...

Q: Rather than RPC, shouldn't we use more intentional mechanisms? [or
something like that]

Dave W. answered that user-visible functionality like spell-checking was this
sort of mechanism.

Henry T. noted that some designs assume a human in the loop, but his
expectation of this design space is that these are system-to-system mechanisms
that should work without a human in the loop.

Q: How about BXXP? [er... my notes are pretty hopeless at this point...
something about "both layers" and HTTP/SOAP/BXXP... claims of whether BXXP is
simpler than other proposals on the table...]

Q [somebody from IBM]: I did a little hacking, and I discovered that
integration with DOM is essential.

Henry T. observed a requirement for separate application data structures and
xml markup styles, with transformation between them, citing the The Cambridge
Communiqué.

Other panelists [who?] observed that many applications have a requirement for
xml serialization but no constraints on the serialization other than that the
other party can "unmarshall" it, and in this case, a ubiquitous serialization
of common data structures would be a big help.

Q [somebody from Allaire]: based on experience with WDDX, it's clear to me
that we need a standard for

   * serialization of common data structures, including hash tables
   * common messaging patterns (e.g. request/reply)

but QOS and transactions should be left separate.

That's all I can recall from my notes.

--
Dan Connolly, W3C