大型分布式计算中的分级节能调度

Hierarchical Scheduling of Large Scale Distributed Computation

随着云计算的快速发展,大型分布式计算被广泛应用。但是,其运行时的巨大能量消耗已经成为应用推广的难题。目前的节能研究主要提出通过调度来减少服务器的运行数量以节能,而没有考虑网络的能耗。提出的分级调度算法HAS(Hierarchical Scheduling Algorithm)针对各计算节点间可能出现任务调度的情况,以DMNS(Dynamic Maxi-mum Node Sorting)调度方法将这些应用尽量分配到连接到同一级交换机的服务器中,然后,将应用数量少的计算节点上的任务转移到还能增加任务的节点,从而减少节点的数量。同时,调度时选择的是较少的数据交换量和较短的交换路径,以节约网络能耗。HAS算法的复杂度较好,且其稳定性也通过计算仿真得到验证。通过仿真数据对比表明,HAS比目前的其它方法更优。

With the rapid development of cloud computing,large scale distributed computation is used widely.However,recently,energy consumption of such distributed systems has become problem for further application.Existing methods for saving energy are developed mainly by decreasing the amount of running servers.However,these approaches do not consider the energy cost on network devices.This paper proposed a hierarchical scheduling algorithm.Our algorithm employs a dynamic maximum node sorting（DMNS） method to optimize the assignment of applications on servers which are connected to a switch in the same level.Secondly,we transfered the applications on the nodes with low load to the nodes which can handle more application in order to reduce the number of nodes.In addition,we chose the transfer path which bears less capacity of data exchange and less length which helps to reduce the energy consumption of network.As a result,both the running servers and the data transfer can be greatly reduced.The time complexity of HSA is satisfactory,and its stability is verified through simulations.Experimental results show that the performance of HSA outperforms existing methods.