An Efficient VM-Level Scaling Scheme in an IaaS Cloud Computing System: A Queueing Theory Approach

Abstract

Cloud computing is becoming an effective and efficient way of computing resources and computing service integration. Through centralized management of resources and services, cloud computing delivers hosted services over the internet, such that access to shared hardware, software, applications, information, and all resources is elastically provided to the consumer on-demand. The main enabling technology for cloud computing is virtualization. Virtualization software creates a temporarily simulated or extended version of computing and network resources. The objectives of virtualization are as follows: first, to fully utilize the shared resources by applying partitioning and time-sharing; second, to centralize resource management; third, to enhance cloud data center agility and provide the required scalability and elasticity for on-demand capabilities; fourth, to improve testing and running software diagnostics on different operating platforms; and fifth, to improve the portability of applications and workload migration capabilities. One of the key features of cloud computing is elasticity. It enables users to create and remove virtual computing resources dynamically according to the changing demand, but it is not easy to make a decision regarding the right amount of resources. Indeed, proper provisioning of the resources to applications is an important issue in IaaS cloud computing. Most web applications encounter large and fluctuating task requests. In predictable situations, the resources can be provisioned in advance through capacity planning techniques. But in case of unplanned and spike requests, it would be desirable to automatically scale the resources, called auto-scaling, which adjusts the resources allocated to applications based on its need at any given time. This would free the user from the burden of deciding how many resources are necessary each time. In this work, we propose an analytical and efficient VM-level scaling scheme by modeling each VM in a data center as an M/M/1 processor sharing queue. Our proposed VM-level scaling scheme is validated via a numerical experiment.