Phone Discussion with Marshall Brinn
Frank Manola
Object Services and Consulting
March 2, 1999
On March 2, 1999, I had a 45-minute phone discussion with Marshall Brinn,
BBN/GTE, ALPINE Technical Director, on CoABS-ALP technology interchange.
I had previously exchanged emails with him introducing the CoABS-ALP technology
interchange activity, and sending him the URL of Craig Thompson's list
of ALP-CoABS
common architectural issues. He asked about the specific goals
of the interchange activity (i.e., were we after TIES and actual technology
development, white papers, or what). I explained what I understood
the goals of the program were, and he seemed pleased that the goals involved
actual technology development.
Marshall described ALP's architecture as having been developed relatively
independently of mainstream agent work. He felt that addressing agent
technology in a standard way, and addressing "classic" agent problems,
were two elements that ALP was potentially missing. He welcomed the
idea of CoABS looking at the way ALP does things, and making suggestions
for extensions/changes, since in many cases ALP was not married to a particular
way of doing things.
Marshall also had some initial suggestions for specific issues in ALP
that he'd be interested in seeing CoABS help with:
-
ALP currently has no dynamic name service, or advertisement and discovery
mechanism; he'd like to see some joint work in this area.
-
ALP plugins are currently static (not mobile); he'd like to see some
work on how to move plugins between nodes for load balancing, and to place
them near data sources, etc. to reduce communications.
-
ALP currently has no real mechanism for guaranteeing convergence or stability
of planning, or of determining completion, or when the plan is "good enough";
he felt that cutting-adge agent technology could help here.
-
ALP system monitoring is also an issue, e.g., agents crashing, communication
links going down, obtaining qualities of service (e.g., by altering communications
protocols); he would be interested in tools, user interfaces, protocols
(e.g., along the lines of SNMP applied to distributed agents) or infrastructure
developments that would help in this area.
I indicated that CoABS had work that was potentially relevant to all these
issues. I also noted that there were additional aspects of these
issues that could be considered as well. For example:
-
There were "logical level" issues relating to both mobility and system
monitoring. For example, you might want to migrate a plugin to a
cluster because the corresponding organization had acquired new capabilities
or responsiblities, and the cluster's behavior needed to be updated to
reflect that.
-
There were "physical level" issues relating to advertisement and discovery
mechanisms. For example, the CoABS NEO TIE involves the idea that
you can use discovery as a fault-tolerance mechanism, to deal with the
situation where an agent crashes by finding another with similar capabilities.
-
Dealing with convergence of planning involves some separation between the
idea of planning as creating a specific logistics plan (whose completion
can be bounded) and the idea of controlling the ongoing activities of an
organization whose "business" is logistics (like DLA), in which the overall
activities aren't really "completed" (only individual tasks within the
activities).
I also mentioned that CoABS work on Agent Communication Languages might
be relevant to extending ALP's directive vocabulary.
Marshall then asked what ALP could do for CoABS, and I gave him the
general list I'd included in the Design Document, namely:
CoABS should look at the cluster model--the ability to extend/specialize
agents with plugins
CoABS should look at ALP's penalty function cost model
access/experience with realistic plan representations, external systems,
planning process
possibly provide a testbed for experiments with CoABS technology using
more realistic scenarios and scales
provide insights on Grid requirements
Marshall expressed some concern about the amount of additional work that
might be involved in this activity, since no one there had much spare time.
I indicated that I would try to minimize that by working up ideas as much
as possible for them to review, rather than having them spend too much
effort on developing the ideas themselves. Marshall suggested that
we continue interacting via phone calls and email as long as this remained
productive, and arrange face-to-face meetings when they became strictly
necessary (easy to arrange, since our offices are relatively close).
Later that same day, I sent Marshall a list of technical questions on
ALP that I had compiled while reading their documentation, along with the
URL to the initial draft of our ALP-CoABS
Technology Integration Design Document (since Jim Hendler had sent
the URL to Brian Kettler).