Online virtual machine migration for renewable energy usage maximization in geographically distributed cloud data centers

Energy consumption and its associated costs represent a huge part of cloud providers' operational costs. In this study, we explore how much energy cost savings can be made knowing the future level of renewable energy (solar/wind) available in data centers. Since renewable energy sources have intermittent nature, we take advantage of migrating virtual machines to the nearby data centers with excess renewable energy. In particular, we first devise an optimal offline algorithm with full future knowledge of renewable level in the system. Since in practice, accessing long-term and exact future knowledge of renewable energy level is not feasible, we propose two online deterministic algorithms, one with no future knowledge called deterministic and one with limited knowledge of the future renewable availability called future-aware. We show that the deterministic and future-aware algorithms are 1+1/s and 1+1/s−ω/s.T_m competitive in comparison to the optimal offline algorithm, respectively, where s is the network to the brown energy cost, ω is the look-ahead window-size, and T_m is the migration time. The effectiveness of the proposed algorithms is analyzed through extensive simulation studies using real-world traces of meteorological data and Google cluster workload.