Load Balancing Strategy in Grid Environment

In the provision of a Grid service, a provider may have heterogeneous clusters of resources offering a variety of services to widely distributed user communities. Within such a provision of services, it will be desirable that the clusters will be hosted in a cost effective manner. Hence, an efficient structure of the available resources should be decided upon these clusters. A static structure, adopted in classical distributed systems, where a single master node controls all resources and decides where incoming jobs should be executed, is not efficient for Grid computing. For this purpose, we propose a dynamic tree-based model to represent Grid architecture in order to manage workload. This model is characterized by three main features: (i) it is hierarchical; (ii) it supports heterogeneity and scalability; and, (iii) it is totally independent from any physical Grid architecture. Over the proposed model, we develop a load balancing strategy suitable for large scale, dynamic and heterogeneous environments. The proposed strategy is based on a neighbourhood load balancing whose goal is to decrease the amount of messages exchanged between Grid resources. As a consequence, the communication overhead induced by task transfer and workload information flow is reduced, leading to a high improvement in the global throughput of a Grid. The first experiment results of our strategy are very promising. In effect, we have obtained a significant improvement of the mean response time with a reduction of the communication cost.