Optimizing Cloud Resource Management with Advanced Scheduling Algorithms and Reinforcement Learning

Objectives: Effective scheduling of cloud resources plays a critical role in maintaining performance and minimizing operational delays, especially in large-scale environments. This study aims to improve task scheduling efficiency by exploring intelligent algorithms that adapt well under dynamic workloads. Methods: Using the CloudSimPy simulation framework, we implemented and evaluated four scheduling algorithms—First-Fit, Random, Weighted-Based, and Reinforcement Learning (RL). The performance was benchmarked using real workload data from the Alibaba Cluster Trace 2017. We then compared our results with existing scheduling outcomes reported by Fengcun Li, particularly his use of the Tetris and First-Fit algorithms. Findings: The results highlight the significant advantage of Weighted-Based and Reinforcement Learning strategies. RL achieved the highest gains with up to 50% improvement in job completion time and 44% reduction in average slowdown. Weighted-Based scheduling also delivered strong performance while maintaining lower computational overhead. Compared to Fengcun Li’s reported performance of Tetris and First-Fit, both RL and Weighted-Based methods showed more consistent and adaptive scheduling under real cloud conditions. Novelty: By replicating a real workload simulation environment in CloudSimPy and directly comparing our results with previously reported methods, this study offers a practical and grounded perspective on next-generation cloud scheduling. It demonstrates the clear benefits of incorporating AI-driven and weighted decision-making strategies in real-world job scheduling. Keywords: Job Scheduling, Makespan Optimization, Reinforcement Learning, Scheduling Algorithms, Cloud Computing