Research On Anomaly Detection And Dynamic Migration Of Virtual Machines In Cloud

With the further development of cloud computing technology,the advantages of customizable service and scale economy of service in cloud platform are becoming more and more prominent.However,the following problems still exist: uneven load of physical hosts leads to resource competition behavior of deployed virtual machines,which leads to abnormal or even downtime phenomenon of running virtual machines.Therefore,this thesis takes virtual machine as the research object,studies from two aspects of anomaly detection and dynamic migration,establishes the anomaly detection model,proposes a many-objective cloud virtual machine migration optimization model and improves the memetic algorithm.The main work is as follows:(1)This thesis proposes a combinatorial learning model,which integrates Long Short-Term Memory(LSTM)and Gaussian-Bernoulli Restricted Boltzmann Machine(GRBM)to generate an end-to-end deep learning network LsGrbmAd,and applies it to the anomaly detection of virtual machines in cloud.First,in the model training stage,LSTM is used to capture the temporal characteristics of cloud virtual machine performance indicators.In addition,Dropout is used to prevent over-fitting of training data.The free energy obtained from GRBM is fitted,and a baseline model with parameter ? is proposed.Then,in the test phase,the free energy obtained by cloud virtual machine performance indicators data is compared with the baseline model,showing how the LsGrbmAd model detects anomalies.Finally,through experiments,the results of different values of parameter ? are compared and analyzed,and the best parameter reference model is selected.Compared with other anomaly detection mechanisms,the experiment results show that LsGrbmAd model has obvious advantages and application potential for improving cloud virtual machine anomaly detection on precision,1 and accuracy.(2)This thesis proposes a many-objective cloud virtual machine migration optimization model and improved memetic algorithm.First,in this thesis,five optimization objectives,including communication energy consumption between virtual machines,the stability of service quality,the degradation of service quality caused by migration,load balancing and migration cost,are studied,and an optimization model of cloud virtual machine migration is proposed.Second,the algorithm only needs to input all virtual machines in the physical host where the abnormal virtual machine occurs,which optimizes the time and space complexity of the algorithm.Then,a probabilistic local search strategy is proposed to guide the evolutionary process.Finally,the experimental results show that IMA has certain advantages in the diversity of solution set and convergence speed of allocation scheme,and it also realizes the compromise optimization of many objectives.This paper has 18 figures,8 tables and 95 references.