An autonomous dynamic load balancing algorithm on distributed systems

Computational Grids have become popular computing platforms. However, to. speed up applications on computational Grids, it is necessary to develop an efficient resource management system. Although the scale of computational Grids has expanded, as has research on scheduling for computational Grids, all existing scheduling algorithms have problems, In, particular, a really efficient scheduling algorithm should take into account heterogeneities of networks and resources, scalability, fault tolerance, and other elements. We have described a new dynamic load-balancing algorithm using agents as a first step in the development a high-performance scheduler. The agents of the algorithm work autonomously and solve the problems of scalability and fault tolerance. Moreover, the use of migrating applications solves network heterogeneity and resources problems originated from heterogeneity. We showed in a scheduling simulation that this load-balancing algorithm mean response time is shorter than that of a sender-initiated algorithm with threshold location policy.