Architecture and performance evaluation of a massive multi-agent system

Abstract

Multi-agent systems can be viewed as systems in which large numbers of agents process jobs by interacting with each other. We have been applying such a multi-agent system to electronic commerce (EC) on the Internet since December 1997. Since the server on the system has to host thousands of agents that process jobs in a disorderly manner, the mechanism for controlling the memory and CPU occupied by agents is very important. Moreover, the system can be extended to include multiple servers connected via a local area network (LAN) or the Internet. We have used mobile agent technology to extend such systems, and therefore the control mechanism supports agent mobility. In this paper, we describe a mechanism for controlling the use of memory and the CPU by thousands of agents. The results of a performance evaluation are also given.