Organizational Patterns for Data Management in Large-Scale Distributed Multiagent Systems

Abstract

One of the important advantages of a distributed multi-agent system comes from the parallelism that can be realized by distributing data over a network of agents. A properly designed distributed agent system can scale to handle arbitrarily large data sets by adding more computers, network bandwidth, and storage. However, this scalability comes at a price. In this paper, we show how a naive approach to data management in a distributed agent system unnecessarily limits its scalability and present our approach to data management and visualization using the Cougaar agent architecture. Finally, we examine the implementation of a distributed MAS realized as three systems of increasing scale and analyze how these principles were and were not followed in an actual, functional system.