Multiplexing of Backup VMs Based on Greedy Policy

Abstract

With the development of the newly proposed cloud computing, an increasing number of software systems are deployed in the cloud environment. Both researchers and engineers in academia and industry focus on seeking efficient virtual machine (VM) deployment and VM consolidation strategies. Increasing the utilization of physical or virtual resources which are directly providing services is mainly concerned, while overlooking the under-utilization of backup virtual resources. Being the valid solution of ensuring the high availability of cloud system, offering redundant backup resources is commonly used. However, excessive consumption of redundant backup instances brings about wasting of resources as well as raising capitalized cost of operation and maintenance of cloud infrastructures. This paper proposes the multiplexing methodology of backup virtual resources based on the restless multi-armed bandit model, aiming to increase the resource utilization and satisfy the guaranteed level of service availability in cloud system. With the known failure models and recovery models of VMs providing services, optimality and performance of the greedy multiplexing policy are also presented. Through the validation in simulation experiment, the proposed policy achieves the goal of extending traditional 1:1 backup provision to 1:M (M >> 1) between the backup VM and the VMs providing services. Thus, the utilization of backup resources can be significantly enhanced correspondingly. On the premise that the guaranteed level of service availability is no lower than 95%, the utilization enhancement of backup VMs can be raised to 90%.