A Distributed Population Management Approach for Mobile Agent Systems

Abstract

Many mobile agent systems rely on a population of software agents roaming a network of nodes (visitation sites) performing various tasks, such as monitoring security or measuring connectivity. For these systems, the number of agents in the population is critical for maintaining desired visitation rates throughout the network; however, the population distribution may change dramatically in reaction to an event, such as issues within the network or adversarial activity. As a result, population management is needed to ensure the number of agents in a system is available to achieve the system objectives. Although centralized management can be used to maintain agent populations, these approaches are not resilient to failure or scalable to large systems. This paper introduces a novel approach for managing the population of agents by governing the death and birth of agents through the distributed examination of the expected visitation rates. Nodes in the network individually monitor local visitation rates to determine if new agents should be created or destroyed, while queue management is used to distribute agents and dampen population oscillation (cyclical death and rebirth of all agents). Since these two population management components are available at every node, the agent population is maintained in a decentralized fashion. Experimental results demonstrate the proposed population management approach can appropriately manage an agent population under various conditions including sudden agent loss and attack situations.