In this paper, we present a load balancing scheme that grid sites are clustered into regional grids around a set of well-known broker sites in terms of network transfer delay and the broker sites are organized in a fully decentralized fashion. Based on above scheme, we propose a hybrid policy to the job scheduling and load balancing in a computational grid which integrates static and dynamic approaches to make load distribution and redistribution driven by performance benefit jobs can gain. The hybrid policy also achieves a balance between the inherent efficiency of centralized approach, and the autonomy, load balancing and fault tolerant features offered by distributed approach. It is naturally suitable for large-scale grids. The main goal of our scheme is to improve the average response times of jobs. Our load balancing scheme is evaluated by simulations. We present some results that verify the effectiveness of our scheme.
展开▼
