But still it moves. Or, as Jim Webber puts it, the emperors second set of new clothes is gone.

I find this whole thing vaguely sad. Sad because I am trying to get WS-RF based long haul deployment working -the fact that convergence is on the road map is not an immediate excuse to kill what I am doing. It does, however, justify the fact that my notification mechanism was extensible, because WS-Notification was always one of many, and soon, perhaps, none of many.

I'm actually going to cause trouble here by observing that one aspect of WSRF that does work for me is having a notion of state and idempotent read operations that access it. It's just a shame that in WSRF that state isn't RESTy, and it's another shame that the GetMultipleResourceProperties operation punts on whether the bulk read is atomic or no'. FWIW, the HP implementation of the CDDLM deployment API only implements the single property GetResourceProperty operation, doesnt support any writeable properties, and absolutely never does a 1:1 mapping between things at the end of Endpointers and operations. As for managed lifetimes, well, anyone who uses Java RMI knows distributed Garbage Collection is still a research area. If you want classic distributed objects, then you probably do have to resort to leasing to manage their lifespan. The question is, do you really need distributed objects? Or, if you do have remote things with their own state, what is the best way to represent them and communicate with them. Clearly, anything that looks like remote method calls is wrong...

On the future of the converged specification, I will have to wait and see. Within the GGF organisation, I am focused on deployment. Iteration one of my specification (and the working implementation) was pure SOAP, demonstrated back in 2004. Moving to WSRF has cost me a year. If I have to move to something else, which now seems inevitable, its going to cost me more time. And meanwhile there is Savas, now with his @microsoft.com colleagues, pushing for an an evolutionary approach, which still seems to think that UDDI isnt extinct. For notifications, I will probably focus on XMPP ahead of WS-N.

Despite the ongoing fuss, convergence of management protocols is a good thing for people that need to manage boxes (real and virtual), and applications on them. That does not mean that anyone writing a distributed application deployed across a cluster or grid should need to know or care what those management protocols are. That is someone else's problem. All you need to do is write manageable JMX objects, deployable components, write your .sf or .cdl deployment descriptor and push it out there. A case in point is that the primary use case and demonstration for our CDL deployment work is the two tier web server, in which one host is configured with the database, and a number of others configured with the app server front end. I think we can assume that classic HTTP is going to be the public facing protocol there.