Adaptive Multi-Tenant Resource Scheduling in Cloud Computing via Reinforcement Learning

Abstract

This paper proposes an adaptive resource allocation algorithm that integrates reinforcement learning to address the challenges of dynamic resource demands and complex multi-objective constraints in cloud computing environments. The method abstracts resource supply and demand states in the cloud into multidimensional state vectors and employs joint modeling of policy and value networks to enable dynamic decision-making. A multi-objective reward function is designed to balance potential conflicts among key metrics, including resource utilization, energy efficiency, latency, and fairness. To enhance model representation and adaptability, a hierarchical state encoding mechanism and a residual gating-based policy updating approach are introduced, ensuring stability of scheduling in complex environments. Experiments comparing several public reinforcement learning scheduling models verify the significant advantages of the proposed method in multi-objective optimization. Results show that the method effectively improves resource utilization, reduces average task latency, enhances overall energy efficiency, and achieves higher fairness in multi-tenant sharing conditions. Overall, this study demonstrates that reinforcement learning-driven adaptive resource scheduling performs well in complex and dynamic cloud scenarios and provides technical support for building efficient, green, and balanced resource management systems.