| This thesis describes a research agenda that applies principles from Organization Theory to Multiagent Systems. The need for a large-scale multiagent system in a dynamic environment motivates an architecture with localized reconfigurations for bottlenecks, for loss of resources, and for domain-dependent needs. The first hypothesis of this research is that distributed services allocated according to the domain-dependent levels of a multiagent organization will scale better than existing methods for distributed resource management in a dynamic environment. Given a system implemented with this hypothesis in mind, its scalability in reorganizing a multiagent system will be demonstrated using contests designed to compare its performance with traditional approaches. A second hypothesis is that rational reorganizations can be effectively and efficiently achieved using decision-theoretic models of a small set of organizational structural dimensions. Rationality in reorganizing a multiagent system will be demonstrated using simulations designed to verify intuitive responses to induced loads and faults. This paper describes the design, implementation, evaluation, and analysis of optimality for a scalable and rational multiagent organization architecture. |
