Lyapunov-Driven Optimization for Energy-Cost Minimization in Data Center With Renewable Energy and Time-Varying Electricity Prices

Abstract

Data centers are increasingly adopting renewable energy sources to mitigate environmental impact and reduce operational costs. However, effectively optimizing energy costs remains challenging due to unpredictable workloads and fluctuating renewable energy availability. This article introduces LOECM, a Lyapunov-driven online scheduling algorithm designed to minimize energy cost without relying on future information. LOECM is formulated as an online stochastic optimization problem and leverages Lyapunov optimization techniques to balance immediate cost minimization against task queue stability. By exploiting the inherent structure of the optimization problem, we design an efficient solver with constant time and space complexity ($\mathcal {O}(1)$). Additionally, we propose the LOECM scheduling policy to dynamically manage workloads and node activation based on a covering subset strategy. To evaluate the practicality and effectiveness of our approach, we implement a prototype system called LOECMHadoop, integrating the LOECM scheduler into Hadoop—a widely-used data processing platform—across a 10-node cluster. Experimental results show that our power model has an average deviation of approximately 5%, and LOECM achieves over 20% electricity cost savings compared without our policy on the prototype system.