Intelligent system for predict performance degradation of virtual machines in cloud environment

In this article the information-extreme intellectual technologies of analyzing and synthesis of the forecasting system are researched. In this case the authors analyzed the degradation of the virtual machines owing to their interference on a common physical infrastructure. The authors proposed the approaches of the formation the input of the mathematical description which based on the cluster-analysis of the performance and resource usage metrics of the virtual machines. Considered feature set for recognize a condition of performance degradation includes the amount of allocated to host virtual machines from each resource consumption class, the amount of available CPU, RAM and disk space and network channel. The algorithms are based on adaptive binary coding of feature vectors and optimization of geometrical parameters of feature space partition into classes equivalence to maximize the information ability of system intended to predict functional state of the computing environment. The modified information criterion for estimate efficiency of machine learning is expressed in terms of false omission rate and positive predictive value. The physical modeling of proposed algorithms are implemented by the example of cloud services from Google.