Energy and Performance Management of Green Data Centers: A Profit Maximization Approach

TLDR

Existing research has focused on reducing data center energy costs, but no study has examined the trade‑off between minimizing energy expenditure and maximizing revenue for cloud services. This paper proposes a systematic profit‑maximization framework for green data centers, addressing both cases with and without behind‑the‑meter renewable generators. The framework models revenue minus cost, incorporates service‑level agreements, renewable power availability, stochastic workloads, and formulates convex optimization problems for practical implementation. Simulations show the proposed strategy outperforms two recent energy‑performance management algorithms.

Abstract

While a large body of work has recently focused on reducing data center's energy expenses, there exists no prior work on investigating the trade-off between minimizing data center's energy expenditure and maximizing their revenue for various Internet and cloud computing services that they may offer. In this paper, we seek to tackle this shortcoming by proposing a systematic approach to maximize green data center's profit, i.e., revenue minus cost. In this regard, we explicitly take into account practical service-level agreements (SLAs) that currently exist between data centers and their customers. Our model also incorporates various other factors such as availability of local renewable power generation at data centers and the stochastic nature of data centers' workload. Furthermore, we propose a novel optimization-based profit maximization strategy for data centers for two different cases, without and with behind-the-meter renewable generators. We show that the formulated optimization problems in both cases are convex programs; therefore, they are tractable and appropriate for practical implementation. Using various experimental data and via computer simulations, we assess the performance of the proposed optimization-based profit maximization strategy and show that it significantly outperforms two comparable energy and performance management algorithms that are recently proposed in the literature.