An efficient multi-agent computationnal model for massively distribution of independent and heterogeneous tasks

This paper aims to present a new model of platform for massively distribution and execution of heterogeneous and independent tasks in an heterogeneous distributed system. This model is built around a multi-agent system middleware. In order to uniformly perform the tasks, the proposed model is based on an abstraction layer which represents separately the task, its metadata and its data using the object-oriented approach. The platform contains five modules allowing for each task, its configuration, production, assignment, execution and its following-up. The main module of this platform is the assignment one that uses an efficient load balancing procedure based on aspect oriented approach. This assignment procedure uses the task metadata, the communication latency and nodes performance to associate the appropriate executor agent to the producer one. To avoid the mediator agent overloading, the task data exchange is established only between the producer and executor agents. The model presents a deployment and caching system of the task code to strongly reduce the amount of data exchanged between the agents of the platform. These properties obviously prove the efficiency of the proposed model.