Optimal Power Scheduling of Data Centers with Deferrable Computation Requests

Data centers, as huge power consumers in distribution power networks, are believed to have significant operational flexibility with dispatchable computation requests and indoor thermal inertia. This flexibility can be utilized to arbitrage predictable time-varying electricity price or provide regulation services in power market. This paper studies the optimal power scheduling of a data center with deferrable computation requests and controllable air conditioning. The objective is to utilize the data center's flexibility to reduce its total operational costs (including both the electricity costs and server degradation) with time-varying electricity prices in a wholesale power market. A queue model is designed to describe the constraints of computation requests and a linear model is utilized to approximate the operation of air conditioning and thermal dynamics of the data center. The proposed model is a mixed integer quadratic programming that can be efficiently solved by the branch-and-bound algorithm in off-the-shelf solvers. Numerical experiments based on operation data of a real-world data center are conducted to validate the proposed method.