RE: Causality/Correlation

I just want to say on the record that I am in favor of the recent 
AMG work in the area of modeling the various message patterns with 
the abstract concepts of the correlation id and message reference. 
I think that these changes correct what was a major deficiency in 
the AM. 

The concrete implementation of these abstractions can be done
in various ways, and the AM (correctly) does not comment on these.
In some cases it may be all handled by the transport binding, as 
can be done easily in the case of simple request-response over HTTP.

Its real value, I think, is that it extends the AM to accurately 
model the semantics of other messaging patterns, such as 
publish/subscribe and one/many (request/reply), which we have 
requirements to support. And it also simplifies the AM by removing 
the Data/UnitData distinction.

I suspect that we may end up specifying optional protocol elements 
to implement these abstractions in a standardized fashion. They may 
be unneeded (redundant) for certain bindings, but with them the 
protocol could work well over multiple intermediaries with 
independence from particular transport bindings.

  Randy Waldrop
  webMethods, Inc.

Received on Thursday, 29 March 2001 15:17:22 UTC