Efficient agent communication in multi-agent systems

Abstract

In open multi-agent systems, agents are mobile and may leave or enter the system. This dynamicity results in two closely related agent communication problems, namely, efficient message passing and service agent discovery. This paper describes how these problems are addressed in the Actor Architecture (AA). Agents in AA obey the operational semantics of actors, and the architecture is designed to support large-scale open multi-agent systems. Efficient message passing is facilitated by the use of dynamic names: a part of the mobile agent name is a function of the platform that currently hosts the agent. To facilitate service agent discovery, middle agents support application agentoriented matchmaking and brokering services. The middle agents may accept search objects to enable customization of searches; this reduces communication overhead in discovering service agents when the matching criteria are complex. The use of mobile search objects creates a security threat, as codes developed by different groups may be moved to the same middle agent. This threat is mitigated by restricting which operations a migrated object is allowed to perform. We describes an empirical evaluation of these ideas using a large scale multi-agent UAV (Unmanned Aerial Vehicle) simulation that was developed using AA. © Springer-Verlag Berlin Heidelberg 2005.