From: AAAI Technical Report FS-95-03. Compilation copyright © 1995, AAAI (www.aaai.org). All rights reserved.
Distributed
Soundar
Information Retrieval and Integration
Centralized
Agent-base System
Kumara,
through a
Nina Berry,
Goutam Satapathy,
Octavia
Camps,
Brian Sheldon and Tim Thomas
Intelligent Design and Diagnostics Research Lab
Department of Industrial and Manufacturing Engineering
207 HammondBuilding
The Pennsylvania State University
University Park, PA16802
(kumara, goutam, or berry @marie.iddr.ie.psu.edu)
Rapid changes have taken place in our world over the
ability to retrieve and manipulate data quickly bepast few years. Information technology is revolutioniztween the different software applications. The first
ing the way we live. Knowledgeis power and the ease
layer of hidden agency consists of the five supportof access to this knowledge makes the manipulation of
based agents: user interface, database, unity, priorthe data critical. Yet, if one can harness the vast inteity, and timephase. Each agent represents a specific
gration of data which flavors the users’ knowledgethe
task that assists in the integration of data and provides recommendations, based on the users objectives.
end result could somehowsatisfy the elusive quest for
complete knowledge integration. This paper overviews
Althoughthe user only has direct access to the user inthe Distributed Intelligent Architecture for Logistics
terface agent, the supervisor directly decides howthe
(DIAL) system, currently being developed as a project
current request should be handled by the underlying
for the U.S. Army Logistics Evaluation Agency. The
software agents.
The second hidden layer consists of application
DIALsystem is designed to integrate existing software
packages into an unified architecture to support the
agents acting as direct interfaces or wrappers to the inneeds of the end users. To ascertain what integration is
tegrated applications. These agents provide the comneeded in the DIALsystem three major questions were
munication needed between the applications and the
addressed (1) what task(s) will the users be facing,
other agents in the system. The complexity behind
what data will be needed by the user, and (3) how
each application agent depends on the complexity of
to integrate across a distributed platform of resources.
the attached application. All of the interface agents
The solution to these questions resides in the design
have similar front-ends that can interpret the stanand development of the centralized agent-based archidardized communication of the other agents. However,
tecture DIAL, where the software applications, data,
translating this into something understood by the apand users are all treated as software agent components
plications is solely dependent on the particular appliresiding in distributed computer systems.
cation. For example, if the user is using an application
that requires additional data files, this request must
Within the DIALsystem a software agent is defined
be routed through the corresponding application agent
as an entity capable of performinga specific single task,
before the file can be retrieved the support database
such as the retrieval and formatting of a data file. Curagent. It should also be noted that the application
rently the architecture is composedof agents residing
agents are responsible for understanding the formatin the following three categories: supervisor agent (deting of the files used by each of their respective applicides course of action to solve the current problem),
cations.
support agents (provides task-oriented support for the
A prototype of the DIALsystem has been created
user and applications), and application agents (interin
the SUNMicrosystem environment and builds on
faces to the software application). At the heart of the
the supplied ToolTalk facilities for inter-agent commuDIALsystem is the supervisor agent, which is responsinication. Through further experimentation with this
ble for deciding howthe users’ requests will be resolved
system
we hope to tackle those issues needed to furby the underlying agents. This centralized problemther
enhance
and improve the use of this multiple resolving approach provides the system with two main
sources environments. Furthermore, we believe that
advantages (1) multiple users problems can be easily
the concepts being discovered during this project are
integrated within a centralized decision area and (2)
generalizable
to other domains where the integration
reduces the overall complexities of internal agent inof
distributed
multiple information resources are reteractions found in decentralized problem-solving sysquired.
tems.
The hidden layers of agents within the remaining architecture collectively cooperate to improve the users
149