Performance Analysis Of Virtual Machines Based On The Class Parameter Augmented Bayesian Network

With the development of information technology,massive data has been used in the fields of scientific research,engineering practice and financial industry.However,the computing power to handle the data is far beyond that provided by the traditional computer architectures.Thus,cloud computing composes computer clusters by aggregating a large number of inexpensive computers,and then uses virtualization technology and distributed system software to achieve complex computing power.Cloud computing providers,such as Amazon Cloud,Alibaba Cloud,etc.,usually provide cloud services to users in the form of virtual machines.The cloud providers offer the specific virtual machines for users to rent and then pay.In this virtual resource usage mode,it is important for providers to predict the performance of virtual machines and find a properly configured virtual machine according to the specified performance requirements,thus to help make the purchase decision.Therefore,based on the data collected from the virtual machines,this thesis builds a model to analyze the relationship between virtual machine performance and configuration to assist decision.making.Specifically,the main work of this thesis is as follows:(1)For the performance prediction of virtual machine,this thesis proposes a Bayesian network model with classification parameters.Firstly,random forest classification algorithm is used to classify the virtual machine,and then a Bayesian network model is constructed according to the classification result and corresponding performance,so as to realize the virtual machine performance prediction under arbitrary feature configuration.(2)For the virtual machine configuration decision problem,this thesis improves the random forest algorithm based on the proposed model.Given a random performance value for a virtual machine,the corresponding values are assigned in the Bayesian network,and then a set of reasonable virtual machine configurations could be predicted.(3)This thesis uses the public data of online task tracking of Alibaba Cluster Tracking Program as the experimental data set.Experimental results show the effectiveness and accuracy of the proposed method.Based on the method proposed in this thesis,a prototype system based on our proposed model is designed and implemented.