一种基于网格距离的资源调度

Improving Scheduling for Grid Resource Management

在网格资源调度中,当任务选择与自己距离较近、可用带宽较大的资源时,调度体现更好的特性。文章就距离、可用带宽并结合资源的使用费用,提出了网格距离的概念,实现网格资源选择中对上述因素的约束。定义了网格资源和应用的模型,在该模型上完成网格距离的计算,提出了资源选择算法。仿真实验表明,调度在通信开销、稳定性、任务完成时间以及任务执行的失败率等方面都得到了改善,同时促进整个网格系统资源交易的吞吐量。

Purpose. There exist already many algorithms on scheduling for grid resource management. The full paper begins with a discussion of a number of these algorithms mentioned in Refs. 1 through 5. We present our scheduling algorithm which we hope is better than those discussed. In the full paper, we explain our proposed algorithm in detail ; here we just give a briefing. Our scheduling algorithm is based on the concept of ＂grid distance＂ proposed by us. The explanation in the full paper centers around three topics： （1） resource selection; （2） grid distance; it is worth mentioning that, in the mathematical expression for grid distance, there are three coefficients α,β and x and that α＋β＋x= 1; we can take physical distance, bandwidth and cost into optimal consideration by selecting suitable values of α, β and x; （3） scheduling algorithm based on grid distance. Finally we performed simulations to compare the performance of our algorithm with that of the None Minimisation algorithm of Ref. 5 by Buyya and Giddy. For our algorithm, we used three different combinations of parameters α,β and x; （1） α= 1, β and x= 0 ;（2） α=0.5, β=0.3, x=0.2; （3） α=0.7,β=0.2,x=0.1. The average costs of our algorithm are 4 400 G $ （combination 1）, 3 980 G $ （combination 2） and 4 370 G $ （combination 3） respectively. The average cost for None Minimisation algorithm is 4570G $. Thus the second combination of parameters （α =0.5, β=0. 3,x=0. 2） is the best.