云环境下虚拟机监控的研究与实现

发布时间：2018-04-15 06:30

本文选题：云监控 + 虚拟机　；参考：《大连理工大学》2016年硕士论文

【摘要】：如今,云计算已经成为最广泛的互联网服务模式。对云平台的资源进行有效、及时、高效、低开销的监控,是保证云计算服务质量的关键因素,同时为后续系统作业管理、负载管理和均衡等工作提供依据。云平台中虚拟节点的资源监控包括：虚拟机运行状态的获取、虚拟机异常的分析和故障的预测。本文提出一种基于聚类的异常检测算法,并基于该算法设计和实现了云环境下的虚拟机异常检测系统,该系统的主要功能是对云平台中虚拟机节点的运行状态进行监控以及故障预警。基于聚类的异常检测算法分为两部分：基于聚类的建模方法和基于非参数CUSUM的异常分析方法。第一部分利用k-means和k-modes两种聚类方法分别建模。首先输入训练数据并指定聚类中心；然后用两种算法分别对虚拟机状态建模,得出结果并对结果做出修正；最后根据建模结果将虚拟机状态分为三类：正常、异常、故障。由于本文采集的数据均为数值类型,因此两种算法中k-means效果较好。第二部分对划分为异常的数据进行处理。利用CUSUM算法,当系统发现虚拟机状态异常时,增大采集频率,并对异常数据进行累计,达到预警门限时发出预警。在Hadoop和Spark平台上实现云环境下虚拟机监控系统。系统采用集中式监控体系结构,对主从节点的虚拟机进行设计。从节点的功能是对虚拟机运行状态的数据进行采集；将采集到的数据通过Kafka消息系统发送给主节点并存入Rsdis数据库中。主节点通过消息系统接收检测数据,并利用相关算法对异常分析和故障预警,同时主节点具有用户接口,供用户查看虚拟机运行状态以及具体报警信息。实验结果表明,Spark平台下的监控系统能实现预期功能,而Hadoop平台下时效性稍差一些。
[Abstract]:Today, cloud computing has become the most extensive Internet service model.The monitoring of cloud platform resources is effective, timely, efficient and low cost, which is the key factor to ensure the quality of cloud computing service. It also provides the basis for the following work such as job management, load management and balance.The resource monitoring of virtual nodes in cloud platform includes the acquisition of virtual machine running state, the analysis of virtual machine anomaly and the prediction of fault.This paper presents an anomaly detection algorithm based on clustering, and designs and implements a virtual machine anomaly detection system based on this algorithm.The main function of the system is to monitor the running state of the virtual machine node in the cloud platform and to warn the failure.The algorithm of anomaly detection based on clustering is divided into two parts: modeling method based on clustering and anomaly analysis method based on nonparametric CUSUM.In the first part, two clustering methods, k-means and k-modes, are used to model the model.First input the training data and specify the clustering center; then use two algorithms to model the state of the virtual machine get the results and make a correction. Finally according to the modeling results the virtual machine state can be divided into three categories: normal abnormal fault.Because the data collected in this paper are of numerical type, k-means is effective in the two algorithms.The second part deals with the data divided into anomalies.Using CUSUM algorithm, when the system finds the abnormal state of the virtual machine, it increases the acquisition frequency, accumulates the abnormal data, and issues an early warning when the warning threshold is reached.The virtual machine monitoring system under cloud environment is implemented on Hadoop and Spark platform.The system adopts centralized monitoring architecture to design the virtual machine of master-slave node.The function of slave node is to collect the data of virtual machine running state, and send the collected data to the master node through Kafka message system and store the data in Rsdis database.The primary node receives the detection data through the message system, and uses the related algorithms to analyze the anomaly and the fault early warning. At the same time, the primary node has a user interface for the user to view the running status of the virtual machine and the specific alarm information.The experimental results show that the monitoring system based on Spark platform can achieve the expected function, but the timeliness of the system under Hadoop platform is slightly worse.
【学位授予单位】：大连理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP302

【参考文献】