Confusion Over Messaging? I think Phil Wainewright and John McDowall may be comparing apples to oranges, and I may have made it worse rather than better.
There are two questions: (1) Where should we look for long-term messaging standards, and (2) what should we do in the meantime?
In his section entitled "The Environment," John considers the latter question. As CTO for Grand Central Communications, that's the world in which he lives. John points out that the parties to a system may be quite asymmetrical. They'll have different levels of sophistication and certainly different technologies. They can't assume JMS (Java), and virtually no one in his world runs ebMS. No universal messaging standard exists, so by definition (as John suggests) today's heterogeneous systems must be loosely coupled--not just in the sense of asynchronicity, but in terms of messaging at all layers. For now, web-services networks like Grand Central Communications play an important role in mediating disparate protocols, including messaging.
OTOH, Phil (and to some extent, I as well) were speaking about the former issue: the longer-term adoption of a universal messaging protocol. I'm convinced that there will eventually be such a standard, just as TCP/IP beat out its competitors (e.g., XNS and IPX) and HTTP and SSL have become ubiquitous. My point was that for different reasons, neither JMS nor ebMS can survive to ubiquity. (JMS because it's not language independent, and ebMS because of the ebXML-versus-WS politics.) Eventually, however, one protocol will rise to the top, we'll all use it, and the messaging-arbitration role for web-services networks like Grand Central will disappear.
In his earlier section, Messaging Categories, John deals with an altogether separate issue: the different levels and attributes of messaging systems. I accept that there are levels such as reliable/guaranteed/transactional (or whatever they should be called). But his other criteria are orthoganal to these. For example, all three of John's categories (even TCP/IP for that mater) would likely provide ordering, one/once only, and best-effort delivery. Systems that don't (such as SMTP) wouldn't even meet the most basic criteria for "reliable." Furthermore, these criteria aren't hierarchical in the same way as John's first three categories.
Finally, John pitches the "minimalism" as loose coupling. That is to say that one should be able to link systems while requiring as little of them as possible. Again, that makes sense if there are no standards and if you're Grand Central. But as I wrote, I expect this problem will go away within 24-36 months, once we have a universally accepted messaging protocol.
Posted Tuesday, June 03, 2003 12:35:16 AM