These are the old pages from the weblog as they were published at Cornell. Visit www.allthingsdistributed.com for up-to-date entries.

April 27, 2004

ws-no-messagedelivery

[from the you-got-to-be-kiddin'-me category]

I have been pretty forgiving about the plethora of ws-* specs, even if it is easy to make fun of it, or question the reasoning about some of the specs. But I didn't see this one coming. And I guess I am not the only one. There will be some places around the world tonight were some guys are having a beer saying 'he, did you see that ws-md spec from ...' with a responses like 'yeh, what the heck have these guys been smoking?' and 'is this payback for ignoring the ws-caf stuff?'.

Someone sometime, preferably over a beer, will have to explain to me which problems this spec aims to solve that couldn't be achieved by influencing or building on one of the older specs.

By the way. Anyone who comes up with the idea of Abstract Message Delivery Properties (AMDP) and standardized Message Exchange Patterns (MEPs) and ignores multicast does not get any bonus points.

Posted by Werner Vogels at April 27, 2004 08:42 PM
TrackBacks

Comments

What? You don't see a need for a WSDL centric, typed notion of an endpoint reference? And the AMDPs, it's a reification of the message - giving us a way to finally refer to individual messages and correlate them with other messages in an exchange.

Granted it doesn't deal with multicast, but pretty much nothing in WS does at all.

Are you saying we should do all this in an application specific way, reinventing the wheel again and again?

Sorry, I just don't see your argument. . .

Posted by: Hal on April 28, 2004 11:29 AM

Hal, what I was refering to is that this style of endpoint referencing replaces the ws-addressing style of endpoint references. And given that some other specs (reliable messaging, eventing) rely on wsa ERs it invalidates those specs also.

Was there really no way of achieving the same result by starting with ws-addressing, either by building upon it or by modifying/extending it?

Can we now also expect new specs that do reliability and event delivery using ws-messagedelivery?

Posted by: Werner on April 28, 2004 11:43 AM

Asside from the IPR issues, I think the WS-MD model is simpler. I know the spec is huge compared to WS-A. . . But it's easy to put down in a 4 page summary (hopefully coming out soon).

The issue regarding WS-A, I believe, is whether you want to be WSDL centric or not. The model is cleaner.

As to the last issue, I believe the answer is "yes".

Posted by: Hal on April 28, 2004 02:56 PM

Hal, I think that "WSDL centric" understates the case for MessageDelivery. In my view, MD is better because it is entirely based on the web services model. As a bi-product of this design decision, MD is derived from WSDL.

This is also one reason it is not derived from WS-Addressing. WS-Addressing defines, for lack of a better term, a poor man's IOR. And we all know that web services aren't distributed objects. Since WS-Addressing is flawed and departs from the web services model, it is an inappropriate starting point.

Aside from the fact that WS-MD is a superior solution for the web services environment, none of the open standards are able to reference the proprietary product specs from IBM and Microsoft. So from a standards-development standpoint, this starts the process of filling a large hole.

Greg

Posted by: greg on May 3, 2004 11:27 AM