Advances in Distributed Scheduling Algorithms: A Three-Layer Architecture Integrating Deep Reinforcement Learning and Energy Optimization (2023-2024)

Abstract

This paper reviews the developments of recent distributed scheduling algorithms across cloud computing, energy systems, manufacturing, and quantum computing areas, and proposes a new three-layer architecture based on deep reinforcement learning and energy optimization strategies. Using a thorough reading of works from between 23 and early 2024, we illustrate major advancements in both energy-efficient scheduling and the integration of deep learning, with newer algorithms realizing up to 27.8% energy savings and up to 40% acceleration in the training processes of a distribute network. We present an energy-efficient architecture that is realized via containerized microservices on a Kubernetes orchestration engine, achieving 30% decreased energy consumption while attaining sub-50ms response times in the 99th percentile and resource utilization above 90%. The method that balances statistical validation with real world validation across 1000 node deployments leads to both theoretical contributions in algorithm designs and their practical implementations in production scenarios, with directions in quantum computing and better AI capabilities being the main draw of where to improve next.