云计算平台下基于近似ε-约束的多目标作业调度优化算法  

MOJSP optimization algorithm based on approximate ε-constraint in cloud computing platform

作　　者：高燕[1] 陈小辉[1] 任勇军[2] 

机构地区：[1]榆林学院信息工程学院,陕西榆林719000 [2]南京航空航天大学计算机科学与技术学院,南京210016

出　　处：《计算机应用研究》2016年第3期711-715,共5页Application Research of Computers

基　　金：国家自然科学基金资助项目(61170035);中国博士后科学基金资助项目(2012M512008);陕西省教育厅科学研究计划项目(12JK0932)

摘　　要：针对云计算中平台主机之间工作负载分布的作业调度问题,提出了一种基于近似ε-约束的优化算法。将作业调度问题建模为一个数学决策模型,求出模型的可行工作调度集,利用ε-约束算法获得每个单目标模型的帕累托前沿,从而优化作业的总平均等待时间、最长工作调度中作业的平均等待时间(如调度跨度)和所需主机数目。实验通过建立实例将算法与传统的加权和(WS)算法进行比较,实验结果显示,算法需要更少的平均等待时间和主机数目,找到的非支配解平均数比WS算法多77.8%,表明算法更具多样化,适合用于解决云计算环境下的大规模多目标作业调度问题。For the issue that optimizing job scheduling between the platform host workload distribution in cloud computing,this paper proposed an optimal algorithm based on approximate ε-constraint to solve the homogeneous cloud computing platform multi-objective job scheduling problem. Firstly,it modeled the job scheduling problem as a mathematical decision model.Then,it found the feasible set of job scheduling model. Finally,it used ε-bound algorithm to obtain the Pareto frontier of each single object model,in order to optimize the total average waiting time for a job,the longest job scheduling the average waiting time and the number of jobs required by the host（ eg span scheduling）. It created an instance to compare proposed with traditional weighted sum（ WS） algorithm,the results show that proposed method needs less average waiting time and hosts numbers than WS algorithm,and it is superior to WS method with 77. 8% increasing the number of non-dominated solution. Experimental results indicate that proposed algorithm has more diversity,which is suitable for settling large-scale multi-objective scheduling problem in cloud computing.

关 键 词：云计算平台 多目标作业调度 ε-约束 优化作业 帕累托前沿 加权和算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]