Abstract：Accurately quantifying the dependencies among properties and performance of virtual machines(VMs)as well as predicting their performance under specific configurations is the basis for achieving fine-grained VM resources allocation.However,although the traditional Bayesian Network(BN)method can represent the uncertain dependences among properties and performance of the VMs,when a group of VM properties do not appear in the dataset,BN cannot predict the corresponding VM′s performance.In addition,when there are too many VM property values,the combination of VM performance may explode,which will increase the difficulty of VM performance prediction.Thus,we propose a Class Parameter Augment Bayesian Network model(CBN).The CBN model first uses random forest(RF)algorithm to classify the VM properties,and then builds CBN according to the classification results and the corresponding performance values,so as to achieve the accurate prediction of VM performance under arbitrary VM configuration.At the same time,CBN reduces the difficulty of performance prediction caused by combination explosion.The experimental results show the effectiveness and accurateness of our proposed method.
尚聪聪,郝佳,张彬彬,岳昆. 使用带分类参数贝叶斯网的虚拟机性能预测[J]. 小型微型计算机系统, 2019, 40(7): 1416-1422.
SHANG Congcong,HAO Jia,ZHANG Bin-bin,YUE Kun. Performance Prediction of Virtual Machines Via the Class Parameter Augmented Bayesian Network. Journal of Chinese Computer Systems, 2019, 40(7): 1416-1422.