David,
I read through the two references you provided which do cover the
issues of programming Agents. Agents usually are composed of objects.
That's certainly a truism. Jim White's General Magic white paper
Mobile Agents covers two areas. The programming of Agents and Agent
infrastructures. It covers the application General Magic's offerings
in detail.
General Magic's Telescript to myself appears to be an infrastructure
layer I would separate into a generalized runtime layer. Programming
Agents themselves is specialized to individual applications domains, a
separate layer. The layering of responsibilities being the separation
of concerns that always happens as new network ideas mature.
Steven Versteeg thesis cover a similar set of issues but in a more
generic way as he states in the Abstract. He does call out and cover
those common issues of distributed Agent applications. Those are a
catalog of the requirements that need to be satisfied for any complete
runtime structure supporting Agents. He also cites as I did Agha's
work on Actors and more.
A practical example of an Agent infrastructure is Cougaar.org.
Cougaar is a JAVA based distributed Agent layer. It is Open Source
and originally began life a DARPA program contracted to BBN to solve
logistics problems. Applications examples can be found at
http://cougaar.org/docman/?group_id=7
That said Cougaar is written in JAVA and uses the JVM as its runtime.
I might ask can Agents be written in Eiffel and run on the JVM? I've
seen the Eiffel# work but since .Net is not ubiquitous it does not
appear to be a way to go. JVM's are available for many devices. It
like .Net is not a network standard or a really Open standard.
When I started this thread I noted the Eiffel features that appeared
to help create and test distributed objects. I still see the need for
a distributed runtime environment that supports a high level of
software quality.
The Eiffel OO language approach offers a way to program Agents in the
upper layer. If married to a runtime that does by analogy the
automation of the lower layer like Eiffel's automated memory
management I think it would be a big step forward.
John S. Wolter
.
