Workload Transfer Strategy of Urban Neighboring Data Centers With Market Power in Local Electricity Market

Abstract

This paper investigates how to minimize the operational cost of cloud service provider (CSP) that operates urban neighboring data centers (DCs) in the same electricity market and can conduct workload transfer among DCs. Due to the substantial electricity demand of DCs, their market power which can have impact on the locational marginal prices (LMPs) of the electricity market should be taken into consideration. We formulate a bilinear bilevel problem which regards the CSP as a price maker and explores cost-minimizing workload transfer strategies. The upper level is the operational cost minimization problem of CSP and the lower level corresponds to the economic dispatch problem of independent system operator (ISO) of electricity market which determines the electricity prices. It is challenging to directly solve the bilevel problem with bilinear term in the objective function. Hence, we first reformulate the original problem into a single level problem and then based on the property of the problem we develop a polytope cutting algorithm that attains the global optimal solution. The proposed algorithm solves linear optimizations iteratively by cutting the non-convex polytope feasible set into convex sets. In addition, considering the varying communication environment in practice, we analyze the impact of transfer price uncertainty on total cost of CSP, and show that the expected cost surprisingly decreases with the increasing uncertainty. Simulations based on the standard IEEE test cases show that the cost of CSP is significantly reduced and a win-win result for both the CSP and independent system operator (ISO) is possible.