About RDS

Books and Papers

IT Conversations






Would you like to receive a weekly digest of this weblog via email? Sign up to receive my free IT Strategy Letter.



Doug Kaye's thoughts on web services, web hosting and managed services.

Gartner or ZapThink? The analysts at the small web-services specialist firm ZapThink said that systems integrators "will need to alter their business models to survive" due to "the appearance of more applications based on Web services." 10 days later, the relative behemoth Gartner Dataquest said just the opposite: They say the trends in web services "suggests a strong pick up in demand for developers and integrators with skills using XML, SOAP, .NET and Java."

Just proves there's more art than science in the analysis business. Let's see, if each of them were right half the time...
Posted Tuesday, June 03, 2003 6:50:58 PM   

Spread for Web-Services Messaging? Speaking of messaging protocols, Brian Behlendorf of Collab.net suggested I take a look at Spread, a language-independent open-source messaging service, which Collab.net uses within its own products. Brian wondered if it might be a contender for a web-services messaging protocol. I hadn't heard of it, and since Brian's recommendations aren't to be taken lightly, I checked it out.

On first blush, Spread appears to be somewhat tightly coupled. It was initially developed to help share SSL session keys between load-balanced SSL servers, which illustrates its performance-centric bias. (Spread is implemented directly over UDP instead of TCP.) As is, Spread identifies messaging endpoints by <IP-address>:<port>, so it must be used with both an abstraction layer and mechanisms for delayed binding in order to support loosely coupled SOAs.

Spread is designed to handle multiple messages between pairs of hosts or between a host and a broadcast or multicast group, and the protocol is session oriented. I wonder how well this would support some of the web-services message-exchange patterns (MEPs) such as fire-and-forget. Some semblance of a session is required in order to guarantee delivery and to be notified of failure, but in loosely coupled SOAs those sessions aren't between the ultimate endpoints--they're between endpoints and their local messaging services. Loosely coupled systems must be able to send and receive messages even when the other endpoint is unreachable. I wasn't able to determine whether Spread could handle such scenarios.

Certainly a package like Spread can be used as a messaging system between heterogeneous tightly coupled systems. But can it play a role in more loosely coupled web services? What do you think?
Posted Tuesday, June 03, 2003 1:07:09 AM   

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   



Current Weblogs

Web Hosting Strategies
Web Services Strategies
Noise (personal)
Blogarithms (all)
(more info)




Click below for single-day archives of Blogarithms weblogs.

June 2003
Sun Mon Tue Wed Thu Fri Sat
1 2 3 4 5 6 7
8 9 10 11 12 13 14
15 16 17 18 19 20 21
22 23 24 25 26 27 28
29 30          
May   Jul

Click to see the XML version of this web page.


All content on this web site is governed by a Creative Commons License.
©2001-2003 Doug Kaye and RDS Strategies LLC (