Demand Response in Data Centers Through Energy-Efficient Scheduling and Simple Incentivization

Abstract

To cope with the explosive growth of the Internet, data centers, which primarily serve as its cloud computing infrastructures, are also growing at a fast rate. These data centers typically serve both real-time delay-sensitive jobs and deferrable but deadline-oriented jobs. They exhibit two fundamental properties, which make them particularly suitable for implementing a Demand Response (DR) strategy. First, they are big consumers of electricity, and second, they deal with time-flexible loads. We consider a data center where the various diverse services are deployed in different tiers to capture some of the intricacies of real operations focusing on deferrable jobs. To tackle resource provisioning for deferrable jobs, we leverage Receding Horizon Control (RHC) based scheduling. We formulate and solve the optimal server provisioning problem in two different ways, namely, minimization of operational cost and joint minimization of capital and operational cost, for server provisioning at different tiers, based on a discrete-time model. We then propose and evaluate a simple and effective mechanism to facilitate more renewable integration into the grid through providing monetary incentives on deadline deferral. We show that our proposed method achieves significantly good performance with respect to various important performance parameters of the system.