Scalable multi-agent systems

The multi-agent research community is currently faced with a paradox. While promoting the use of agents as the silver bullet for various software engineering problems, it faces difficulties in presenting successful deployments. Despite the countless multi-agent prototypes that have been developed, the number of actually deployed and in use MAS is at best very small [1]. One reason for the noticeable absence of deployed multi-agent systems is their inability to scale. This paper reports about the use of a CORBA/Java middle-ware layer called DICE, which enables transparent access to the resources of different physical machines. Using this layer it becomes possible to build large multi-agent systems that require large numbers of concurrent threads and significant memory resources. Using DICE it becomes possible to build large agent societies consisting of complex agents defined in JESS.