Energy efficient dynamic resource management in cloud computing based on logistic regression model and median absolute deviation

Abstract

The unprecedented trend of using public cloud computing services by increasing number of customers motivates cloud services providers to optimize their resources usage and management to the limit. This is including managing cloud user’s virtual machines (VM) running on one or more of the thousands of hosting servers or physical machines (PMs) of the cloud datacenters. The cloud service providers are mainly concerned on answering the two main questions that dramatically impact their infrastructure usage and utilization; Where to initially place the VMs and where to move them in case we need to move them. Along with the VM consolidation technique, VMs migration will help in protecting the physical servers from being overloaded or reduce the number of active physical servers for better resources utilization and energy saving. Efficiently detecting overloaded servers will help in improving the cloud system performance and reduce the total operational costs which will provide competitiveness for the cloud provider in the market. In this work, we are proposing a general host overloading detection algorithm based on logistic regression model and median absolute derivation. The proposed algorithm is scalable and can be used with any VM placement and migration algorithms. An extensive evaluation procedure is used with dynamic workload to proof the efficiency of the proposed algorithm. The archived results show that the proposed algorithm outperforms all other known host status prediction techniques.