Abstract

An innovative approach for cloud Autoscaling for optimal resource utilization is proposed by integrating an LSTM-enhanced Actor-Critic model. Out model effectively captures temporal variations in resource demand, leading to more accurate and responsive scaling by leveraging Long Short-Term Memory (LSTM) networks within the Actor-Critic framework. Without compromising system performance, a dynamic pricing mechanism is introduced that balances on-demand and spot instance costs which ensures optimal budget utilization. The experimental results demonstrate the significant improvement in resource utilization and system throughput by introducing the proposed hybrid scaling strategy combining vertical and horizontal scaling. The proactive scaling, reducing SLA violations, and enhancing overall system reliability is enabled by predictive capabilities of the LSTM model. The advantages of the proposed model over the static threshold based autoscaling methods is highlighted by the key performance metrics such as resource utilization, throughput, SLA compliance, and cost efficiency. Our model achieves an average resource utilization of 92.7% while reducing SLA violations by 63% is validated by the experiments demonstrates a substantial improvement over conventional technique. While maintain system performance and SLA compliance the incorporation of pricing mechanism optimizes cost management leads to a 12% reduction in the total operational expenses. To deliver a scalable, cost-effective, and high-performance cloud autoscaling solution the experimental findings outperformance the potential of combining LSTM-driven reinforcement learning to implement hybrid scaling combining with pricing strategy.