An Online Algorithm for Data Center Demand Response
Shahab Bahrami, Yu Christine Chen, Vincent W. S. Wong · 2019
Data centers often support a range of delay-tolerant workloads with adjustable execution time under a prespecified service level agreement. This potential for workload management has motivated utility companies to deploy demand response programs to encourage data centers toward shifting workload execution away from peak load periods. In this paper, we focus on data centers' demand response considering the uncertainties in the arrival rates of the workloads, local renewable generation, and time-varying electricity prices. The centralized workload scheduling is shown to be a convex optimization problem. We deploy an online convex optimization framework to solve the centralized problem without any knowledge of the stochastic process that uncertain parameters follow. It also enables us to design a decentralized algorithm to address the high computational complexity of the centralized approach as well as the data centers' coupled decision making under the real-time pricing scheme. We perform extensive simulations to demonstrate the lower running time of the decentralized algorithm compared to the centralized approach. Data center demand response benefits the utility company by 12.4% reduction in the peak load demand. It also benefits a data center by 12.2% reduction in the average daily cost.