Optimizing Resource Allocation in Cloud Environments Using Reinforcement Learning

Traditional cloud resource allocation techniques, which rely on static, rule-based systems, are inefficient and lead to greater costs and lower performance due to issues like over- and under-provisioning. The study presents a novel way to dynamically optimize resource allocation in real-time using reinforcement learning (RL) algorithms, such as Deep Q-Networks (DQN), Proximal Policy Optimization (PPO), and Actor-Critic approaches. The RL-based system overcomes the limitations of traditional techniques by adjusting to shifting workloads and learning from the past. The average resource usage efficiencies of DQN, PPO, and Actor-Critic are 85.7%, 88.1%, and 87.4%, respectively, compared to 72.3% for traditional systems. The results demonstrate that RL approaches perform much better than traditional systems. The RL techniques also enhanced system performance and decreased expenses by as much as 29.4%; PPO had the highest results in response time (270 ms), throughput (190 requests/second), and user satisfaction (88%).