Principal Investigator:
Jeff Bradshaw
Research Category:
Software Agents

Project Description:
KAoS is a collection of componentized agent services compatible with several popular agent frameworks, including Nomads, the DARPA CoABS Grid, the DARPA ALP/Ultra*Log Cougaar framework, CORBA, and Voyager. The adaptability of KAoS is due in large part to its pluggable infrastructure based on Sun's Java Agent Services (JAS). For additional descriptions and perspectives on KAoS, the reader is referred to the publications page.. While initially oriented to the dynamic and complex requirements of software agent applications, KAoS services are also being adapted to general-purpose grid computing and Web services environments as well.

KAoS domain services provide the capability for groups of agents, people, resources, and other entities to be structured into organizations of agent domains and subdomains to facilitate agent-agent collaboration and external policy administration. Domains may represent any sort of group imaginable, from potentially complex organizational structures to administrative units to dynamic task-oriented teams with continually changing membership. A given domain can extend across host boundaries and, conversely, multiple domains can exist concurrently on the same host. Domains may be nested indefinitely and, depending on whether policy allows, entities may become members of more than one domain at a time.

KAoS policy services allow for the specification, management, conflict resolution, and enforcement of policies within domains. Policies are represented in DAML (DARPA Agent Markup Language) as ontologies. The KAoS Policy Ontologies (KPO) distinguish between authorizations (i.e., constraints that permit or forbid some action) and obligations (i.e., constraints that require some action to be performed, or else serve to waive such a requirement). Through various property restrictions in the action type, a given policy can be variously scoped, for example, either to individual agents, to agents of a given class, to agents belonging to a particular group, or to agents running in a given physical place or computational environment (e.g., host, VM).

Some of the important features of KAoS are worth noting. First, the approach does not assume that we are dealing with a homogeneous set of agents that have been designed in advance to work with KAoS services. Rather the goal is to be able to have KAoS services work with arbitrarily written agents after the fact through support being added transparently at the platform level. Second, insofar as possible the KAoS framework supports dynamic runtime policy changes, and not merely static configurations determined in advance. Third, the framework is extensible to a variety of execution platforms that might be simultaneously running with different enforcement mechanisms--initially agent platforms implemented in Java and Aroma--but in principle any platform for which policy enforcement mechanisms may be written. For example, we are now extending KAoS to work in conjunction with the Globus Toolkit version 3. Fourth, the KAoS framework is intended to be robust and adaptable in continuing to manage and enforce policy in the face of attack or failure of any combination of components. Finally, we address the need for easy-to-use policy-based administration tools capable of containing domain knowledge and conceptual abstractions that let application designers focus their attention more on high-level policy intent than on implementation details. Such tools require sophisticated graphical user interfaces for monitoring, visualizing, and dynamically modifying policies at runtime.

Key Personnel:
Maggie Breedy
Larry Bunch
Renia Jeffers
Matt Johnson
Shri Kulkarni
James Lott
Niranjan Suri
Andrzej Uszok

Sponsors:
Office of Naval Research
Air Force Office of Scientific Research
Army Research Institute
DARPA