企业多云时序数据实时监测方案研究与实现  

Research and Implementation of Real-time Monitoring Plan for Enterprise Multi-cloud Time Series Data

作　　者：程学林[1] 郑佳卉 蒋烁淼 贝毅君[1] CHENG Xue-lin;ZHENG Jia-hui;JIANG Shuo-miao;BEI Yi-jun(School of Software Technology,Zhejiang University,Hangzhou 310027,China;Shanghai Zhuyun Information Technology Co.,Ltd,Shanghai 200120,China)

机构地区：[1]浙江大学软件学院,杭州310027 [2]上海驻云信息科技有限公司,上海200120

出　　处：《小型微型计算机系统》2023年第1期155-162,共8页Journal of Chinese Computer Systems

基　　金：浙江省公益技术应用研究项目(LGG21F020004)资助;浙大软件学院-驻云科技联合创新实验室项目(202003)资助;宁波市自然基金项目(202003N4317)资助。

摘　　要：在企业的IT架构不断上云、多云趋势明显的今天,云资源在系统的开发、运行和维护中扮演着举足轻重的角色.云资源在运行过程中每时每刻都能产生海量时序数据,然而当前市场上的云监测服务在多云实时监测、异常检测方面都存在一定局限性.针对上述问题提出一种基于时序数据的企业多云实时监测方案MCloudMonitor,围绕多云资源运行产生的海量时序数据提供时序数据存储和实时流处理服务,基于分层时间记忆网络来设计实时在线异常检测算法并将其整合进所实现的系统之中.除此之外,借助测试来评估方案的实现效果,证明基于该方案实现的系统能够帮助运维人员从企业的视角进行便捷的实时多云监测,并且能够准确地进行实时异常检测.Today, when the IT architecture of enterprises is constantly upon the cloud and the trend of multi-cloud is obvious, cloud resources play a pivotal role in the development, operation and maintenance of a system.Cloud resources can generate massive amounts of time series data at all times during operation.However, the current cloud monitoring services in the market have somelimitations in real-time monitoring and anomaly detection of multipleclouds.In response to the above problems, an enterprise multi-cloud real-time monitoring solution based on time series data(MCloudMonitor) is proposed, which provides time series data storage and real-time stream processing services around the massive time series data, which generated by the operation of multi-cloud resources, and designs real-time online anomaly detection algorithm based on Hierarchical Temporal Memory Network and integrate it into the implemented system.In addition, with the help of testing to evaluate the implementation effect of the solution, it is proved that the system can help operation and maintenance personnel to conduct convenient real-time multi-cloud monitoring from the perspective of the enterprise, and can accurately perform real-time anomaly detection.

关 键 词：多云 时序数据 实时监测 流处理 异常检测算法 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]