Evaluation of Distributed and Centralized Agent Location Mechanisms

Abstract

With the proliferation of Multi-Agent Systems (MAS), there appears a need to provide effective, robust and scalable agent location mechanisms. Previous studies have presented several solutions to the agent location problem. The majority of those solutions used a centralized approach, however a few distributed solutions were presented as well. Yet, proposed solutions were not compared systematically, and therefore it remains an open question which mechanism should be used, and for which specific MAS parameters. In particular, it is important to compare performance of centralized and distributed solutions, as they are expected to exhibit the most diverse behaviors. This is precisely what we do in this research. We suggest that a distributed peer-to-peer approach [6] to agent location had many advantages over centralized middleagent location mechanisms [1],[2],[4]. Via a series of experiments, we show that, in large-scale, overloaded MAS, a distributed approach is more efficient than a centralized one. We also show that through careful planning of the connection model among the agents, one can manage the communication overhead of a distributed location mechanism.