An Agent-based Architecture for Distributed Computing

James J. Nolan
Center for Image Analysis and
Laboratory for Collaborative Work Environments
Department of Computer Science
George Mason University

The application program is rapidly being decomposed from a stand-alone
program to a more loosely coupled set of services available on the
network.  This change presents many opportunities from a distributed
computing perspective.  "How do we discover services on the network?" and
"How do we compose heterogeneous services (e.g., from different vendors)
to solve a problem?" are examples of questions that quickly emerge.  We
have developed an agent-based architecture for distributed computing that
addresses such questions.  The architecture is motivated from the domain
of distributed imagery and geospatial computing and is called the
Agent-based Imagery and Geospatial processing Architecture (AIGA,
http://aiga.cs.gmu.edu).  Imagery and geospatial systems are used in the
intelligence gathering, cartography, and resource management domains,
among others.  These systems utilize low-level imagery and geospatial
services to answer high-level queries.  In production intensive
environments, it is typical for these systems to process hundreds of
images and geospatial data sets per day that each range from several
megabytes to several gigabytes in size.

This talk will present AIGA, which relies on a well-defined set of
low-level imagery and geospatial processing agents.  An RDF/XML based
approach has been used to build a unified basis for describing the agents;
the inter agent communication; describing the initial, intermediate and
resulting datasets available in the system; and presenting a query to the
system.  AIGA enables discovery of agents to solve a particular query and
provides an approach for application-specific agents to be easily
constructed and deployed by using a network of lower-level processing
agents.  By allowing these agents to be reused, we implement a simple yet
effective knowledge management approach.  The prototype system will be
discussed, which consists of approximately 100 imagery and geospatial
processing agents, a collaboration space, an agent supporting search and
discovery, data access agents, and client agents.  These agents are
mobile, operate in a task and/or data parallel fashion, and can support a
scheduling mechanism.

---------------------------------------------------------------------------------------------
Bio:  James J. Nolan is currently a Research Associate at the Center for
Image Analysis and Laboratory for Collaborative Work Environments at
George Mason University.  His research interests lie in the intersection
of image processing, geospatial computing, parallel and distributed
computing, and decision support systems.  His current research focus is on
developing a distributed agent architecture for intelligence analysis.
Prior to joining GMU, Jim was an engineer with The MITRE Corporation
supporting intelligence community projects as well as internal research
programs.  Jim's experience at MITRE ranged from developing processing
specifications for a mission to map the world (SRTM) to developing
middleware for massively distributed imagery intelligence applications.
Before this, Jim developed systems to integrate remotely sensed imagery
and geospatial data for terrain analysis at the US. Army Topographic
Engineering Center.  Before this, Jim was an analyst at the Naval Polar
Oceanography Center where he developed approaches to automate the process
of analyzing remotely sensed imagery for sea-ice analysis.  Jim received a
B.S. from the University of Maryland at College Park in 1991, an M.S. from
GMU in 1998, and expects to complete his Ph.D. at GMU in 2003.