Second Chance works out better: Saving more for data center operator in open energy market

Abstract

Geographical load balancing (GLB) is a promising technique to reduce power cost for cloud service providers (CSPs). To fully exploit the potential of GLB, Camacho et al. [1] advocate broker-assisted GLB where CSPs first bid in deregulated electricity markets and then balance their workloads accordingly. In this paper, we further explore along this line and propose the idea of Second Chance, which explores the design space in sequential bidding into sequential geographical markets. We formulate the optimal sequential bidding and GLB problem as a Markov Decision Process (MDP) problem. To solve this problem, however, faces the curse of dimensionality commonly encountered in MDP approach. To tackle this challenge, we first establish an optimality criterion for the problem and derive the structure of cost-to-go function. Then we analytically characterize the optimal action. Real-world trace-driven evaluation shows that the electricity cost can be reduced by more than 10% by jointly using GLB and Second Chance.