Simulating Software Agent Colonies in Large Scale Distributed Artificial Intelligence (DAI) Networks

Wide area networks have a large number of computers capable of providing a wide range of services. Due to their low cost and high performance, autonomous mobile Agents are being used to access information sources in distributed, wide-area, networks. Architectures such as active networks also employ intelligent, mobile network processes. However, there are no analytic models for the performance of such systems and as a consequence, simulation is required to test the efficacy of their performance and policies. Assessment of performance and policy requires visualization techniques that depict the state of the Agent colony and its demand on network resources. We sketch the design for a large scale, DAI simulation and analyze its implementation with respect to time and resource requirements. An Agent colony is simulated using C over CSIM. C over CSIM achieves the clarity and logic of Agent based simulation for network services, as well as ease of modeling and reuse of model entities. We also present a Java based builder to create the simulation objects under Object Orientation, and a Visualization post processor to alleviate the inherent difficulty of visualizing a Distributed Artificial Intelligence (DAI) system. Decomposition is applied to partition the search domain into an Agent-oriented view and a Networkoriented view. The individual views are then abstracted to provide summary views to the user, who is given control over the degree of abstraction.