Research On Node Anomaly Detection Method In Virtual Computing Environment

With the rapid development of cloud computing,the scale of cloud platform has expanded dramatically.Because of resource competition and software recession,virtual machine may behave abnormally during operating.And it will not only affect the quality of service in cloud platform,but also it will cause the loss of users and other serious consequences.Therefore,the research on node anomaly detection under virtual computing environment is important for improving the stability of cloud platform.In this paper,we research about cluster anomaly detection method for cluster anomaly detection in virtual computing environment.We also analyze and summarize the advantages and disadvantages of existing node anomaly detection methods.On this basis,we mainly study the anomaly detection method of single node and the anomaly method for multiple homogeneous nodes aiming at the characteristics of virtual computing environment.We solved the problems of real-time anomaly detection of virtual machine and the method that we proposed got high accuracy rate and low false positives rate.The content of the paper mainly includes the following several aspects:(1)We proposed a single node anomaly detection framework based on combined clustering aiming at the problem of low accuracy and high false positive rate of node anomaly detection method based on single clustering.We improved the subspace clustering algorithm and density clustering algorithm to meet the requirements of data stream clustering.We use the two improved algorithms as the based clustering algorithm to generate cluster members,then we design choice method based on the degree of clustering difference for selecting members,and finally we design useful consensus function to achieve the integration of clustering members.The experimental result show that the proposed model has better results than single clustering.(2)We proposed a anomaly detection based on context for multiple node anomaly detection.This method combines context information among nodes and the historical information of single node for detection in isomorphic distributed computing systems.The experimental results show that the proposed method is superior of the existing methods in terms of accuracy and recall.