A good post by David on the relationship between data format and protocol extensibility and evolvability. I’m not sure I totally agree with his conclusion – that “protocol designers […] shouldn’t have high hopes that they can regularly provide for compatible protocol evolution” – but the bulk of it makes sense. I also think he’s approaching it from too theoretical a POV, since there are practical considerations that change the dynamics he’s studying (as I mentioned to him).
However, he finishes with this whopper;
As an afterward, it may be worth pursuing trying to solve problems of protocol evolution by examining whether the use of a constrained protocol, ie HTTP, provides any greater evolvability for protocols
That kinda came out of left field, no? I don’t see how that follows from what he wrote. I agree, of course, but just because a constrained interface provides self-descriptive messaging, and self-descriptive messaging provides better evolvability characteristics than the alternative, since the semantics of the message are unambiguous. This is in constrast to an SOA style message, where the most important semantic – the operation – is actually purposefully removed from the message.
I’m looking forward to his next post, which I anticipate to have a title something like “Oh my, you mean HTTP and RDF are what I’ve been looking for all along?!”. 8-)
[Oops, I forgot to promote this to my “live” weblog last week. Here goes.]
Had a blast at the Middleware 2004 Program Committee meeting in Toronto this weekend. I got to catch up with my old dist-obj buddy Doug Lea, as well as meet some people whose work I have at least a passing familiarity with such as Roy Campbell, Stefan Tai, and Chris Gill. Unfortunately Werner couldn’t make it en corpo – it would have been good to talk face-to-face with him about REST, SOA, distributed objects, etc.., – but he was able to dial-in which was appreciated.
I hadn’t heard about this conference before being invited to be on the PC, so wasn’t exactly sure what to expect in terms of paper and organizational quality (though after seeing who was involved, I was less concerned 8-), but after this weekend, I’m confident it will be an exceptional conference. Please consider attending!
Tim writes;
In recent decades, he points out, good new technologies have first appeared in rough-and-ready form on the Internet, then migrated into the enterprise. […] But all the WS-* hullabaloo is trying to go the other way; […]
Yep, I’ve been pointing this out for years.
But why is this so? Mark’s observation are empirically correct, but how do you explain it? I believe the study of software architecture provides a hypothesis; Intranet based architectures are insufficiently constrained to provide the necessary architectural properties to manage an abundancy of trust boundaries. An intranet is a special case of the Internet in this way, and therefore architectural styles developed for the latter are not, in general, suitable for the former. But because the Internet is the general case, architectural styles developed for it are transferrable to the intranet.
I’ve been recently thinking about this in terms of “Fitness Landscapes”, which I learned about from one of Stuart Kaufmann‘s books a few years ago. But Christoper Alexander‘s work on architecture and pattern languages is also relevant. In fact, I bet that in most fields there’s some guru who’s made this same fundamental observation.