Deep Reinforcement Learning-Based Adaptive Resource Scheduling for Modern Operating Systems

Abstract

With the rapid growth of cloud computing and data-intensive applications, traditional operating system resource scheduling mechanisms face significant challenges in handling dynamic workloads while meeting multiple quality-of-service (QoS) objectives. This paper proposes a deep reinforcement learning (DRL) based adaptive resource scheduling framework that intelligently optimizes joint scheduling decisions across CPU allocation, memory quotas, I/O priorities, bandwidth control, and task placement. Our approach employs an actor-critic architecture with temporal state encoding and cross-resource dependency modeling to capture the complex interactions among system resources. Experimental results demonstrate that the proposed method achieves significant improvements in system throughput (23.5%), reduces average response time by 31.2%, and decreases tail latency by 28.7% compared to conventional heuristic-based schedulers, while maintaining competitive energy efficiency.