Optimization of Energy Consumption in Green Data Centers Using Multi-Agent Reinforcement Learning

Abstract

Data centers are now an important infrastructure to the digital economy; yet, they also consume the most electricity in the information technology industry. The swift development of cloud computing, artificial intelligence, and the massive processing of data has greatly raised the level of energy consumption, resulting in a decrease in operational expenditures and environmental issues. The goal of green data centers is to improve such cases by designing systems that are resource-efficient and by using smart resources in managing the system. Nevertheless, it is a complex issue to optimize energy consumption since the workload changes dynamically, there exist thermal constraints, and subsystems of computing and cooling are interdependent. In this paper, a Multi-Agent Reinforcement Learning (MARL)-based model will be suggested to optimize the green data center energy consumption. The suggested model implements several smart agents that handle the workload scheduling, server control, and cooling. These agents engage with the environment and learn optimum policies through the reinforcement learning methodologies and implement adaptive and real-time decisions in different operational environments. The workload datasets are realistic, and they are used to determine the effectiveness of the proposed approach in a simulated data center environment. The findings suggest that the MARL framework can attain a maximum of 29% of energy savings and 83% of better resource utilization, compared to the traditional rule-based, DVFS-based, and single-agent reinforcement learning algorithms. The model also has a low rate of SLA violation, which guarantees a good service performance. In general, the results have supported that multi-agent learning can offer an efficient, scalable, and sustainable method of energy optimization in a modern data center, and that it has a future role in cloud and edge computing systems.