云计算环境下基于强化学习的虚拟机资源调度  

The virtual machine resource scheduling based on reinforcement learning in cloud computing

作　　者：亢中苗 汪莹 张珮明 陶志强[2] 李家樑 KANG Zhongmiao;WANG Ying;ZHANG Peiming;TAO Zhiqiang;LI Jialiang(Electric Power Dispatching Control Center of Guangdong Power Grid Co.,Ltd,Guangzhou 510600,China;Guangdong Planning and Designing Institute of Telecommunications Co.LTD,Guangzhou 510630,China)

机构地区：[1]广东电网有限责任公司电力调度控制中心,广州510600 [2]广东省电信规划设计院有限公司,广州510630

出　　处：《自动化与仪器仪表》2020年第10期68-72,76,共6页Automation & Instrumentation

基　　金：广东电网有限责任公司科技项目资助(No.GDKJXM20162500(036000KK52160032)。

摘　　要：针对云计算环境下多用户多资源虚拟机调度的延迟问题,为了优化虚拟机的任务完成时间,提出一种基于强化学习的虚拟机资源调度方法。首先以虚拟机的任务完成时间为优化目标建立基于延迟的虚拟机资源调度模型,然后通过定义虚拟机配置数组来将多资源多类的虚拟机资源调度问题转化为单维的虚拟机调度决策问题,接着提出面向延迟的虚拟机资源调度强化学习算法,将任务完成时间作为奖赏函数,通过贪婪行为策略选择最优的调度策略来达到最大奖励,从而获得最优的平均任务完成时间。仿真结果表明,与现有算法相比,该方法在提高虚拟机资源利用率与减少任务的延迟方面性能显著。A reinforcement learning based virtual machine scheduling approach is proposed to solve the delay minimization problem in cloud computing with multiple users and multiple resources.A virtual machine scheduling model is established,with the aim at optimizing the job completion time of virtual machines.Then,the multi-resource multi-class virtual machine scheduling problem is transformed to a single-dimension virtual machine scheduling problem by defining an array for the virtual machine configurations.A delay-oriented reinforcement learning virtual machine scheduling algorithm is proposed to solve the problem,which takes job completion time as the reward function,The optimal scheduling strategy is selected greedily to approach the maximum reward,as well as the average job completion time.Simulation results show that,the proposed algorithm outperforms the existing algorithms in increasing the resource utilization of virtual machines and reducing the delay for jobs.

关 键 词：云计算 虚拟机调度 强化学习 优化模型 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]