Research On Virtual Machine Anomaly Detection System Oriented IaaS

Infrastructure as a Service(IaaS) is an important model of cloud computing. Itregards computer hardware as virtual resource pool and provides users with access toon-demand, elastic and scalable service. With the development of cloud computing, alarge number of applications and data are deployed in the cloud, reliability of cloudcomputing faces enormous challenges. The study shows that main cause of systemfailure is due to the lack of anomaly detection and fault tolerance mechanisms. However,cloud computing itself has a large-scale, distributed, virtualization and highly dynamiccharacteristics. It leads to the traditional anomaly detection system can’t adapt to thecloud computing environment.This paper design an anomaly detection system based on principal componentmethod and Bayesian decision mechanism. By collecting virtual machine performanceindicators, and establish normal contour of the virtual machine in the semi-supervisedmode. Compare with the normal contour to identify the current status of virtual machine.At the same time, this mechanism can dynamically adjust the data collection strategiesbased on a virtual machine status. Also a variety of anomaly detection strategies areimplement.In this paper, we study the key issues in anomaly detection under IaaS.The main work accomplished in this paper is as follows:(1) Deeply study the characteristics of the IaaS model. Because of the dynamic ofvirtual machine, the performance indicators are proposed in this paper to stand thestatus of the virtual machine. And introduce the way to collect the performanceindicators in detail.(2) As the traditional data collection methods can’t dynamically adjust thefrequency of data collection, and can’t adapt to cloud computing environments. Thispaper proposes an adaptive data acquisition strategy, according to the status of thevirtual machine, dynamically select coarse-grained and fine-grained data collectionmethods to improve the accuracy of anomaly detection, but without increasing thevirtual machine performance.(3) The traditional anomaly detection algorithm can’t meet the cloud computingenvironment in the aspect of multi-object detection, data collection capacity, high datadimensions and high dynamic characteristics. In this paper design an anomaly detection method based on principal component method and Bayesian decision mechanism.Meanwhile, in order to improve the accuracy of anomaly detection, this paper alsoproposed anomaly detection strategy based on time anomaly detection method andbased on the time window of anomaly detection.(4) Based on the above research, this paper designs and implements a virtualmachine anomaly detection system under IaaS. And use open source technology to buildthe OpenStack cloud computing platform, by simulating abnormal injection approach toassess the effectiveness of the virtual machine anomaly detection system.In summary, this paper conduct a comprehensive, deep research about virtualmachine anomaly detection system under the IaaS. By improving anomaly detectionalgorithm, realize anomaly detection system, and verify its effectiveness.