RL-Based Smart Scaling for SLA-Adherent Cloud Resource Management

Cloud computing environments require intelligent resource management to handle dynamic workloads while maintaining strict Service Level Agreements (SLAs). Traditional auto-scaling methods rely on static thresholds and reactive rules, which often fail to respond effectively to sudden workload fluctuations. This can lead to under-provisioning, causing SLA violations, or over-provisioning, resulting in increased operational costs. To address these challenges, this project proposes a Reinforcement Learning (RL)-based smart scaling framework for adaptive cloud resource management. An autonomous RL agent continuously monitors system metrics such as CPU utilization, request latency, and workload intensity. Based on real-time observations, the agent dynamically performs scaling actions to adjust resource capacity. The reward mechanism is designed to balance SLA adherence with cost optimization. Through continuous interaction with the cloud environment, the agent learns an efficient scaling strategy over time. Experimental evaluation shows that the proposed approach improves system stability, reduces SLA violations, and enhances overall cost efficiency compared to traditional rule-based scaling methods.