摘要
虚拟机(VM)负载预测对提高云数据中心的资源利用率及用户服务质量起着至关重要的作用。然而现有的预测方法通常只考虑单一负载类型,在真实的云环境中,要么难以保障预测精度,要么因为需要同时建立多个预测模型而产生庞大的训练和预测时间开销。针对现有预测方法无法有效兼顾多种类型负载场景下预测精度和时间开销的问题,提出了一种基于卷积神经网络(CNN)的多类型负载联合预测方法(TSF),能自动化构建并提取关键训练样本,并充分挖掘其中潜在的时序特征和空间特征,从而在考虑多种虚拟机负载情况下,能有效降低训练和预测时间成本,同时提高预测精度。
Virtual machine(VM)load prediction has always played a vital role in improving cloud data center resource utilization and user service quality.However,existing prediction approaches usually only consider a single type of VM load.In real cloud environment,these approaches are either hard to guarantee the prediction accuracy,or because of the need to establish multiple prediction models simultaneously,they will generate much training and prediction time overhead.Therefore,in view of the fact that the existing prediction approaches cannot effectively balance the prediction accuracy and time overhead in scenarios involving multiple types of loads,a multi-type load joint forecasting method(TSF)is proposed based on convolutional neural network(CNN).It automatically constructs and extracts the key training sample,and fully learns the potential temporal and spatial characteristics among them,such that it can effectively reduce training and prediction time consumption,and improves the prediction precision while taking into account the multiple types of VM loads.
作者
余显
李振宇
张广兴
谢高岗
Yu Xian;Li Zhenyu;Zhang Guangxing;Xie Gaogang(Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100190;University of Chinese Academy of Sciences, Beijing 100190;Computer Network Information Center, Chinese Academy of Sciences, Beijing 100190)
出处
《高技术通讯》
EI
CAS
北大核心
2020年第9期884-892,共9页
Chinese High Technology Letters
基金
国家重点研发计划(2018YFB1800201)
国家自然科学基金(61802366)资助项目。
