Research On Anomaly Detection And Recovery Of Virtual Machine In Cloud

As the integrator of Internet technology,cloud computing has great advantages in distributed computing,data storage and other fields.However,with the popularization of cloud computing,the problem of low resource utilization of servers in the cloud has gradually become more prominent.For this reason,some cloud service providers have adopted a resource overbooking strategy,which can improve the utilization of server resources in the cloud.However,it may lead to virtual machine exception,which is caused by the peak value of virtual machine using a certain resource in the same virtual resource pool in the same time period,which shows that the virtual machine in the same virtual resource pool has resource preemption.Since this kind of anomaly is different from intrusion anomaly and fault anomaly,it is necessary to design corresponding detection methods and recovery strategies for abnormal virtual machine according to the causes of this kind of anomaly.The main research contents are as follows:(1)An improved C-t-SNE(t-SNE Based on Classification selection,C-t-SNE)dimension reduction algorithm adapted to cloud is proposed.The algorithm constructs the original space and the corresponding projection space.According to the relationship between the cloud virtual machine and the task performed,the comparison data set is extracted for each cloud virtual machine to reduce the number of comparisons between the data in the dimensionality reduction process to achieve Cloud virtual machine data quickly reduces dimensionality while preserving the association between virtual machines.(2)A Local Outlier Factor Based on Density Space(LOFBDS)algorithm is proposed.LOFBDS algorithm refers to DBSCAN(Density Based Spatial Clustering of Aapplications with Noise,DBSCAN)algorithm,integrates the properties of cloud virtual machine in density space into LOF(Local Outlier Factor,LOF)algorithm,and proposes judgments on cloud virtual machines Rules to optimize the detection process of normal cloud virtual machines and improve detection efficiency.(3)Proposed Three-weighted Flower Pollination Algorithm(TFPA).In view of the differences between virtual resource pools and the different demands of virtual machine tasks on resources,the paper puts forward the differences weight ?,over allocation weight ? and over resource demand weight ?.The mapping process from the solution obtained by the algorithm to the virtual resource pool is added,and the calculation time of the algorithm is reduced through the concept of feasible solution step size.This paper consists of 16 figures,7 tables and 52 references.