On Alternation and Information Sharing among Cooperating Autonomous Management Agents

In an environment of autonomous agents, it is advantageous for agents to function with as little information exchange as possible. Although zero information exchange is impractical, a small amount of information exchange is sufficient for coordinated management of two resources. At minimum, cooperating agents need to be able to operate alone for short periods. We explore the limits of coordinated execution for a two agent system for managing resources based upon performance, and conclude that the best performance is achieved if at least one agent has sufficient time undisturbed to manage its part. Counter intuitively, giving agents full knowledge of each others´ actions does not improve performance in our situation, and can even cause instability. These results suggest that coordination is more important than knowledge exchange in assuring convergence of networks of distributed agents.