Using a Performance-based Skeleton to Implement Divisible Load Applications on Grid Computing Environments

Abstract

Applications with divisible loads have such a rich source of parallelism that their parallelization can significantly reduce their total completion time on grid computing environments. However, it is a challenge for grid users, probably scientists and engineers, to develop their applications which can exploit the computing power of the grid. We propose a performance-based skeleton algorithm for implementing divisible load applications on grids. Following this skeleton, novice grid programmers can easily develop a high performance grid application. To examine the performance of programs developed by this approach, we apply this skeleton to implement three kinds of applications and conduct experiments on our grid test-bed. Experimental results show that programs implemented by this approach run more rapidly than those using conventional scheduling schemes.