Data Center Job Scheduling and Energy Management Under Uncertain Environments

Data centers have become crucial infrastructure in the digital age, leading to a significant increase in energy consumption. Job scheduling stands out as an effective method to regulate the data center energy consumption by delaying job execution within resource availability and quality of service constraints. However, aleatoric uncertainties associated with incoming job information and real-time electricity market prices, and epistemic uncertainties inherent in the learning environment jointly present unique challenges for efficient job scheduling schemes. To tackle multiple types of uncertainties, we propose an efficient risk-aware job scheduling method for data centers in uncertain environments. Firstly, we formulate the data center job scheduling problem within a Markov framework incorporating job heterogeneity. To capture epistemic and aleatoric uncertainties, the policy function is reconstructed by integrating state-action value distributions with efficient exploration based on enhanced distributional reinforcement learning. Furthermore, to account for the risk preferences in data center decision-making, we include consideration of Conditional Value at Risk in the model. Numerical simulation results demonstrate that the proposed strategy can rapidly adapt to uncertain environments and help data centers make risk-aware job scheduling decisions.