Energy-Aware Data Center Job Scheduling Scheme under Uncertain Environment

As a pivotal industrial infrastructure in the digital age, data centers are facing rapidly increasing energy consumption and costs. Leveraging the operational nature of data centers, job scheduling can be employed to adjust job execution times, providing temporal flexibility in energy consumption. However, the inherent uncertainties associated with job arrivals and electricity prices pose significant challenges to effective job scheduling. In this paper, we propose a data center job scheduling scheme based on distributional reinforcement learning to capture the uncertainty of job arrivals and prices for a better operation strategy. Firstly, we formulate the job scheduling problem within the Markov decision process framework. To solve the problem of uncertainty faced by the data center, we introduce a scheduling model based on Quantile Regression Deep Q-Network. Finally, the simulation results demonstrate that the proposed method can adaptively optimize the data center job scheduling strategy.