Cost-Optimized Predictive Autoscaling of Cloud Resources Using Game-Theoretic Queuing Theory

Efficient autoscaling of cloud resources is essential for ensuring service availability while minimizing operational costs. This paper presents a cost-optimized predictive autoscaling framework that combines queuing theory and game theory to make intelligent scaling decisions. The framework models incoming workloads using an M/M/c queuing system, predicting response times and system load to dynamically estimate resource demand. A non-cooperative game-theoretic model is then applied to balance the trade-off between performance (e.g., response time, queue length) and cost (e.g., infrastructure expenses). By continuously adjusting the number of active instances based on traffic intensity, waiting probabilities, and service constraints, the approach ensures SLA compliance while minimizing over-provisioning. Experimental results show that this predictive autoscaling method reduces cloud costs by 30% while maintaining optimal performance (e.g., response times under 100ms) when compared to traditional reactive scaling approaches.