Agent Technology White Paper and RFP Roadmap
Draft .04
March 14, 2000
OMG document internet/00-03-08
a previous version of this document appears as Chapter 9 in the Agent
Technology Green Paper
document editor: Craig
Thompson
Send your comments and contributions to the
document editor.
Contents
Document Revision History
-
Draft .01 - 11/18/99 - initial document reflects minutes
of OMG Cambridge meeting. Craig Thompson is initial editor until
further notice.
-
Draft .02 - 01/04/00 - some lite cleanup before OMG
Mesa meeting
-
Draft .03 - 03/05/00 - document updated to add additional
other
potential RFPs from emerging DARPA CoABS roadmap
-
Draft .04 - 03/15/00 - document modified fairly substantially
to reflect discussion at OMG Denver meeting: (a) document restructured
to emphasize initial RFP candidates and deemphasize for now the lower priority
other candidates. (b) new material added to Agent
Discovery RFP description and several other candidate RFP descriptions
(c) backed off for now from aggressive roadmap
and timetable. (d) section on other
potential RFPs reorganized. (e) added links to other relevant
documents including OMG and FIPA specs and Agent WG meetings [caveat:
did not check all the links.], (f) numerous small
changes.
Executive
Summary
The purpose of this white paper (in an early draft
state) is to provide OMG with a rationale (policy, motivation) for a proposed
series of Agent Technology RFPs. The list of RFPs is culled from
material covered in the Agent
Technology Green Paper and discussed at OMG Agent Working Group meetings.
The rationale provides reasons to order RFPs the way we recommend below.
At the current time, we believe it is premature
to immediately start issuing RFPs without first identifying and agreeing
about the most promising candidates, fleshing out their descriptions, and
understanding how they complement each other and also their relationship
to the rest of OMG object technology. When we do begin to issue
RFPs (perhaps towards the end of 2000), we recognize that experience with
initial RFPs may lead to later revision of this document including revising
the list of RFPs or their ordering so we have not spent much time ordering
later RFPs, rather providing criteria to select the initial core.
Candidate Agent
Technology RFPs
All of these RFP candidate descriptions need more work. Here they
are only sketched at the level we talked about them during the Boston meeting.
Some additional work is needed to insure that
(a) this is the right list of RFPs and we are not missing something
important
(b) the list takes into account other relevant OMG work and avoids
duplication, and
(c) we are not duplicating work done on agent standards elsewhere (e.g.,
FIPA)
RFP Format
OMG RFPs follow a generic
RFP template. Most of an RFP is boilerplate indicating how to
respond. Section 6 Specific Requirements on Proposals generally specializes
the RFP to a particular subject. Subsections of Section 6 are listed
below. This is the template we'd need to complete for each RFP:
-
6.1 Problem Statement
-
6.2 Scope of Proposals Sought
-
6.3 Relationship to Existing OMG Specifications
-
6.4 Related Documents and Standards
-
6.5 Mandatory Requirements
-
6.6 Optional Requirements
-
6.7 Issues to be Discussed
-
6.8 Evaluation Criteria
-
6.9 Other Information Unique to this RFP
For reference, here is a list of OMG
Technology Adoptions. These are responses to RFPs.
Initial Cut at Higher
Priority RFP Candidates
Agent identity
-
Responsibility for writeup: David
Levine, IBM
-
References
-
Notes
-
Persistent identity over time - identity belongs to identity authority
-
One unique id—or possibly many under various circumstances (see FIPA work)
-
Insuring identity
-
Authentication
-
retiring identity (and shunning, time-bound)
-
GC issues
-
Issues
-
How does this play in existing agent platforms?
-
What is the relationship of agent identity to agent security?
-
What is the relationship of agent identity to object identity?
-
Does mobility create additional identity problems? what about versioning,
replication, proxies, etc.
Message transport
-
Responsibility for writeup: Frank
McCabe, Fujitsu
-
References
-
Notes
-
Durable, reliable; send/receive
-
Might need to build a transport gateway
-
current OMG architecture handles this.
-
Concrete realization of FIPA transport
-
Use event or messaging style?
-
IIOP, events, CORBA, msg, Java msg, com/OLE
-
Message source and value types
-
FIPA role, OMG role, gateways
Agent Discovery and Matchmaking
-
Responsibility for writeup: Craig
Thompson, OBJS
-
Analysis
-
The document Agent
Discovery and Registration Service - Notes is an analysis of the needs
of agent systems for a Discovery (aka Trader) service. Our conclusion
is that the OMG
ECDTF Resource Registration and Discovery RFP in progress can meet
most of our needs. In fact, that RFP is more general in that it calls
for a general purpose resource description represented in XML that could
include XML-encoded descriptions of agents. We do have the following
concerns regarding whether the ECDTF RFP will meet the needs of agent technology:
-
A tricky part for this RFP is that it essentially includes the entire Ontology
problem as well, in that anything can be encoded and traded. It is
not clear that we need an additional RFP a this time in the agent discovery
area but we might well need a richer ontology specification than the one
ECDTF comes up with (remains to be seen).
-
It is not clear how rich the query/matchmaking capability will be - it
could vary from something like LDAP, to the sort of property matching algorithm
used in the currently adopted OMG
Trader specification, to a query language like XQL (a sort of SQL for
XML), all the way to including Prolog like inference and pattern matching.
-
It is not clear if multiple matchmaking algorithms will be permitted to
correspond to different kinds of resources.
-
It is not clear how features, constraints, and access points will be used.
Action: contact RFP responders and alert them to Agent WG issues.
-
References
Agent Communication Language
Ontology
-
Responsibility for writeup: volunteer needed
-
References
-
Notes
-
Issues
-
There are a lot of ontologies out there; should we support any or all of
them?
-
XML (and XML Schema) and RDF (and XQL, etc.) are proposed by W3C to provide
ontology representation and tools. What else is needed? Look
to the DARPA DAML program for answers.
-
How is resource description as required by the OMG
ECDTF Resource Registration and Discovery RFP in progress related to
ontologies? It is essentially the same problem. The RFP requires
XML descriptions of resources.
-
How is UML XML (UML externalized in XML) related? This is an issue
as UML provides a rich way to represent several kinds of semantics including
behavioral specifications and constraints.
-
How are ontologies related to the security or policy idea of enclaves or
domains?
Content Language
-
Responsibility for writeup: volunteer needed
-
References
-
Notes
-
focus on reflective aspect
-
CORBA-specific content language profile
-
Issues
-
should this be lumped into ACL RFP
Agent security ("Trust")
-
Responsibility for writeup: volunteer needed
-
References
-
Notes:
-
Orthgonalized; similar to OO security but with some twists
-
Interop Security RFP
-
FIPA- systematically secure…across transports
-
In a touchy-feely sort of way, some agent-only issues are:
-
Who agents tell what to
-
Whether they are trustworthy or not
-
Who do you trust and who don’t you trust
-
Issues:
-
Sender Identification
-
Tampering
-
Encryption
Agent/object mobility (triaged - consider
requirements sooner so as not to preclude but RFP later)
-
References
-
Issues
-
is agent mobility different than object mobility?
-
are there several different models for agent mobility, each with different
utility?
-
should agent mobility be orthogonal to cognitive agent services, agent
security services,
-
how is agent mobility related to asynchronous
messaging, security
UML profile for agents & ACL &
agent platforms
-
References
-
Issues
-
What UML extensions are needed to support agents?
-
How does UML relate to ACL and ontologies?
-
How should agent behavior be represented?
-
How many of these extensions will the OO UML 2.0 decide to support anyway?
For those agents requiements outside of UML 2.0, perhaps separate agent,
ACL, and agent platform profiles will be useful.
Other possible RFP areas that are out
of scope for the near term
Listed below are a number of potentially valid RFP topics some of which
are candidates for higher priority but doing the work to refine them into
RFPs awaits champions for each. If someone is interested, please
come to OMG Agent WG meetings and participate or send descriptions to the
document
editor.
-
agent-software interface - see
FIPA97 Agent-Software
Integration - a standard way in which non-agent based software can
be integrated into a FIPA agent platform.
-
human-agent interface - see FIPA98
Human/Agent Interaction
-
multi modal agent-based user interface framework
-
GUI, speech and/or other natural language, gesture
-
dialog management
-
Grid and Grid Services - the agent grid is described in the Agent
Technology Green Paper (also Kettler's
presentation and OBJS
grid RFI response in Agent WG Mtg #6) as an interoperability framework
allowing heterogeneous agent systems to interoperate more easily.
It is itself a kind of minimal agent system.
-
information management framework - information agents for extracting
content from data sources, mediating and aggregating, and routing content
to customers. May involve optimization and also connecting in new
services to process the data.
-
team formation
-
cognitive agent mechanisms and services
-
contracts
-
constraint language
-
unification service
-
planning service
-
scheduling service - see OMG
Dynamic Scheduling RFP
-
negotiation service - see OMG
ECDTF Negotiation Service
-
reputation service - using digital certificates - better business bureau
for agents
-
incentive management
-
uncertainty management
-
conflict resolution service
-
list others here ...
Criteria for Sequencing
RFPs
There is more material here than can be considered in one RFP so the RFPs
must be sequenced according to some adoption roadmap.
The center of interest in the Agent Working Group is in multi agent
systems that are used in application development in robust ways
in distributed environments that operate in LANS, WANs and over
the web. For the present, we are less focused on: individual
agents, desktop agents, collaborative filtering, and agent-user interfaces.
Criteria that can affect the ordering of RFPs are as follows:
-
ACL communication - we need standard ways that agents can communicate
with each other and standard ways for them to talk about domain entities
in different domains. RFP areas that directly provide help here are
ACL,
Ontology, and Content Language.
-
interoperability - agent systems will come in various forms and
technology that allows agent and agent systems to interoperate will be
immediately useful. RFP areas that provide direct help in interoperability
are: agent identity, agent registry/discovery, agent transport,
and possibly agent grid, and agent-software interface.
-
security - if agent technology becomes widely popular, it will need
security models before people trust agent-based applications. So
we need to ask what extensions to the OMG security service are needed and
must not be precluded in agent system development
-
mobility - if agents are to realize the claim of mobility, then
mobility services should be in place. We view mobility to be largely
orthogonal to agents (or objects) and only need to insure that earlier
RFPs do not preclude adding mobility later in a consistent way.
-
distributed, robust, large-scale - we believe these are general
desirable criteria for agent systems as well as object systems. RFPs
should not preclude deployment of agents in such industrial strength environments.
Another kind of criteria involves market driven need. Which of the
RFPs are most needed to create new markets in the electronic commerce,
manufacturing, telecom, and other domains?
Perhaps the most important criterion will be which of the RFPs are industry
technology providers most interested in working on first?
Initial Roadmap
We recommend that the first two bundles of RFPs be:
-
Agent Interoperability RFP - covering agent identity, agent transport,
and agent registration and discovery
-
Agent Communication RFP - covering agent communication language,
ontology, and content language
The first bundle is in several ways orthogonal to the second, at least
insofar as different ACLs could be accommodated by the same Agent Interoperability
services. Also, within these bundles, the services should be separately
specified as they are separable but they must work closely together so
they are included in the same bundle. We currently seem to be on
the vector to work on the first bundle first and then the second but we
can decide about overlap at upcoming meetings.
Next on the list should be Agent Security RFP when we decide
what aspects of security are special to agents and go beyond object security.
Other RFPs will be considered at a later date: these will likely
include some from the deferred list - mobility, teams, information
management framework, etc.
IDL (if any) and UML extensions needed by agents will be requested as
part of every agent RFP.
Timetable
The timetable is not yet set. We first need to work on fleshing out
the RFP descriptions above. When those are more stable, we can agre
to issue RFPs. This will likely happen at OMG Orlando in December
2000 or thereafter. It depends considerably on how much help we get
writing RFPs and whether industry groups clamor to respond to them.
Issues with
current document
-
is there a good reason why this initial list of RFPs is not architecturally
derived from the FIPA architecture? should it more directly cover
things like agent management, agent wrappers for objects, ...
-
agent interface to object services - for instance, agent persistence, agent
transactions, ... - should these be considered or are these just ordinary
object services that agent systems make use of?
-
should the focus be on agent systems and how to provide gateways that show
how at least OMG solutions like IIOP can help but permitting other kinds
of implementations -or- are these just going to be OMG agents (a much more
rarified market)
-
should the interoperability and communication bundles be overlapped or
sequenced?
-
add other issue here ...