Energy-Efficient Task Scheduling for Data Centers with Unstable Renewable Energy: A Robust Optimization Approach

In the cloud computing era, the green Internet data centers are being restructured to allow high utilization of renewable energy so that the system becomes more environment friendly and cost effective. In this paper, we focus on the energy-efficient task scheduling problem in an Internet data center system powered by both conventional and renewable energies. An electricity cost minimization framework is built to intelligently schedule different tasks from different users on geo-distributed computing nodes. In such system, the indeterminate and intermittent renewable energy imposes huge scheduling challenges. To deal with the uncertain nature of renewable energy, a new flexible uncertainty model is developed. Specifically, reference distributions are introduced according to predictions and field measurements, and uncertainty sets are defined afterwards to confine the unstable renewable energy generations. The model allows the renewable energy generation distributions to fluctuate around their reference distributions. Chance constraint approximations and a robust optimization approach are proposed to first transform and then solve the energy-efficient task scheduling problem. Simulation experiments based on real-world data evaluate two different task scheduling strategies. Impacts of different parameters such as computing node number, task size, task number, electricity price traces and renewable energy generation traces are evaluated which indeed provide us some insights on how to build an energy-efficient Internet data center.