虚拟CPU负载预测算法性能评估  

Performance Evaluation of Virtual CPU Load Prediction Methods

作　　者：万成威 王霞[1] 王猛[1] WAN Chengwei;WANG Xia;WANG Meng(Beijing Aerospace Control Centre,Beijing 100094,China)

机构地区：[1]北京航天飞行控制中心,北京100094

出　　处：《电讯技术》2022年第4期445-449,共5页Telecommunication Engineering

摘　　要：准确的虚拟CPU负载预测是提高虚拟机CPU调度性能的重要前提,然而,虚拟机操作系统环境下,虚拟CPU负载预测方法需要尽可能简单、有效。针对虚拟机CPU调度应用场景,以实际CPU负载为研究对象,选取五种简单的时间序列预测算法,详细评估其虚拟CPU负载预测性能,为虚拟机CPU调度的实现提供了研究基础。结果表明,平均移动法、线性递减加权平均移动法、一次指数平滑法等具有良好的CPU负载预测性能,并基于反馈控制思想改进了一次指数平滑法,进一步提高了其CPU负载预测性能。The accurrate virtual CPU(vCPU)load prediction is an important prerequisite to improve the performance of CPU scheduling of the virtual machine.However,in the operation system of virtual machine environment,the vCPU scheduling method should be as simple and effective as possible.The vCPU load prediction performance of five typical methods in the time series analysis area is fully tested with the actual CPU load datasets for the application of CPU scheduling in the virtual machine environmnet.The result shows that the methods of moving average,linear weighted moving average and exponential smoothing are fit for vCPU load prediction.Based on the feedback control principle,the exponential smoothing method is improved and the prediction accuracy is further improved.

关 键 词：虚拟化云计算 虚拟CPU调度 CPU负载预测 平均移动 指数平滑 

分 类 号：TP302.7[自动化与计算机技术—计算机系统结构]