• 1 Département d'Informatique [Oran]
• 2 Department of Computer Science

Most of the existing load balancing strategies were interested in distributed systems which were supposed to have homogeneous resources interconnected with homogeneous and fast networks. For Grid computing, these assumptions are not realistic because of heterogeneity, scalability and dynamicity characteristics. For these environments the load balancing problem is then a new challenge presently for which many research projects are under way. In this perspective, our contributions through this paper are two folds. First, we propose a distributed load balancing model which can represent any Grid topology into a forest structure. After that, we develop on this model, a load balancing strategy at two levels; its principal objectives : the reduction of average response time of tasks and their transferring cost. The proposed strategy is naturally distributed with a local decision, which allows the possibility of avoiding use of wide area communication network.