Dynamic Resource Allocation Using Reinforcement Learning in Edge-Cloud Environments

Edge-cloud computing has emerged as a powerful pattern to support delay-sensitive and resource-intensive applications in modern distributed systems. However, dynamic and unexpected workloads in this environment face significant challenges for efficient resource allocation. This paper proposed a Multi-Agent Reinforcement Learning (MARL) approach to optimize resource management in edge-cloud architectures. Specifically, three methodologies are introduced to address diverse real-world scenarios. Deep Q-Network (DQN) is used to learn resource selection. Proximal Policy Optimization (PPO) supports continuous control for adaptive scaling and ensuring consistent performance under fluctuating workloads. Finally, they introduced a MARL method for decentralized agents operating collaboratively, each optimizing local efficiency, minimizing network delay, and reducing forwarding costs. The MARL models adapt more effectively to dynamic conditions than traditional static and heuristic-based approaches. Experimental results show an average of 20 % improvement in task completion efficiency and a 15 % reduction in resource wastage, confirming the effectiveness of MARL in edge-cloud resource orchestration.