Cooperating Mobile Agents for Mapping Networks

Contemporary computer networks are heterogeneous; even a single network consists of many kinds of processors and communications channels. But few programming tools embrace, or even acknowledge, this complexity. New methods and approaches are required if next-generation networks are to be configured, administered and utilized to their full potentials. The growing field of mobile agents research seeks to address problems in this domain. In this paper we describe a strategy for mapping a network using a collection of cooperating mobile agents. We present results from a simulation of such a system and discuss the relationship between diversity of the agent population and overall efficiency of the system. Knowledge of the topology of a computer network is a prerequisite for all higherorder interactions between nodes on that network. In current systems, routing maps are usually generated in a centralized (and often human-mediated) manner. Our mobile-agents approach, in contrast, is highly distributed and decentralized with agents spread across the network working to accumulate connectivity information. The agents in our system move around the network, discover topology information, and carry with them the information they have gathered as they explore. Our agents also collaborate; when they meet on a node they share information with each other, so an individual agent can acquire knowledge about parts of the network that it has never visited. We apply three different types of agent algorithms to a mapping task on a static network and comparing their effectiveness. We find that agent cooperation greatly improves system performance. Furthermore, we find that diversity of behavior between collaborating agents also improves the efficiency of the system as a whole. Agents that consistently exhibit identical behaviors duplicate one another’s efforts, to the detriment of overall performance. As agents interact with one another – and so have the potential to become ”like” their peers by learning from them – preserving behavioral diversity becomes a major challenge. Efficient division of labor in the absence of centralized control is a subtle, important problem.