Dynamic Investment and Retirement Operations of Cloud Computing Service Management

Abstract

The emerging market of cloud computing encourages cloud service providers (CSPs) to deliver efficient and effective services on the basis of appropriate provisioning of service resources. Overprovisioning and underprovisioning of service capacity lead to the wastage of cloud resources and degradation in service performance, respectively, causing unnecessary mismatching costs. In this paper, we study the cloud service management problem to find optimal dynamic investment and retirement strategies over the complete life cycle of a cloud service product involving growth, maturity, and declining stages. We extend the traditional dynamic inventory model to the declining stage by showing that time-dependent (st, St) policies remain optimal, and propose a near-optimal Adaptive-Approximation-Algorithm-Based policy with the guaranteed performance of 3. We also design a data-driven framework for the cloud service management problem based on the auxiliary data predictive of the demand and utilize predictive prescriptive methods to operate the online investment and retirement strategies. Numerical experiments show that predictive prescriptive methods significantly improve the effectiveness of cloud service management by reducing costs and improving the coefficient of prescriptiveness.

Keywords: cloud computing, service resource management, life cycle management, predictive prescription, (s, S) policy

JEL Classification: C61, M15