Hedging Against Uncertainty: A Tale of Internet Data Center Operations Under Smart Grid Environment

Internet Data Center (IDC) supports the reliable operations of many important online services. As the demand of Internet services and cloud computing keep increasing in recent years, the power usage associated with IDC operations had been surging significantly. Such mass power consumption has brought heavy burden on IDC operators. Recently there are extensive research on power management for IDCs. However, one important challenge faced by IDC operators has been overlooked. How to handle the uncertainties in IDC operations is a challenging task. The uncertainties come from both the dynamic workload and time-varying electricity prices. In this paper, we systematically investigate the problem of minimizing the operation risk of IDCs against those uncertainties at the same time guaranteeing quality of service under deregulated electricity market environment. We propose a novel hedging scheme and model the operation risk minimization problem as a bilevel programming. We also design an optimal hedging algorithm. We conduct extensive evaluations based on real-life workload data from Google and electricity price data from deregulated electricity market for multiple IDC locations. Results show that our scheme can significantly reduce the operation risk by countering the uncertainties.

I. Introduction

The past decade witnessed tremendous growth of online applications and services, including e-mail, blogging, social networking, entertainment, and e-commerce. Together with the recent trend of cloud computing, more and more data and computation are migrated to or hosted on a certain kind of physical infrastructure called Internet Data Center (IDC). IDC has many benefits including ease of management for end users and low amortized cost of ownership. For example, it uses geographical distribution and replication to improve data reliability [1]. Further, IDC operators such as Google, Microsoft, and Amazon.com, etc., can use economics of scale and statistical multiplexing to amortize the total cost of ownership over a large number of machines and users [2]–[5].