Towards a New Micro Agents Middleware for Massively Distributed Systems

In the field of multi-agent systems, there are several middlewares that can facilitate the creation of Multi Agents Systems (MAS) applications. These middlewares offer programming frameworks strongly coupling the learning models of agents and their communication patterns. In most cases, when the number of agents increase, the communication models used heavily decrease the performance of these middlewares. In this paper, we present a novel model of micro agents middleware for massively distributed systems. This model is FIPA [1] compliant, and it is based on the principle of decoupling two aspects of the agent: the cognitive and the communication aspect. With this middleware, it is possible to create and deploy heterogeneous agents operating in distributed containers. To communicate with each other, agents of the platform can use polymorphic communication models adapted to their context. The main idea is to allow the agent to use any messaging system that has proven its performance in the field of High Performance Computing (HPC).