Scalable optimization for congestion-aware NFV deployment

• We proposed a novel NFV model that balances cost, workload, and resource use. • We modeled server reliability to guide VNF placement and reduce congestion. • We linearized the non-convex MINLP using dynamic programming for scalability. • We developed a cutting-plane heuristic to accelerate convergence. • We validated our approach on real topologies, outperforming baseline solutions. This paper introduces a novel optimization framework for Network Functions Virtualization (NFV) that addresses the efficient implementation of end-to-end service requests in physical networks. Our approach characterizes each server node by a reliability function reflecting its computational load, which aids in balancing workloads and mitigating congestion. By optimizing the reliability metric along the route, our approach ensures robust end-to-end service quality. We formulate the NFV deployment problem as a non-convex mixed-integer non-linear programming (MINLP) model aimed at minimizing both deployment and operational costs while maximizing resource utilization, addressing also per-node installation conflicts and inter-VNF incompatibilies. Given the NP-hard nature of the problem, we develop efficient linearization techniques and bounding schemes, using also dynamic programming, to convert the formulation into a tractable mixed-integer linear programming (MILP) model. Additionally, a cutting-plane-based heuristic with a warm-start strategy is proposed to further accelerate convergence. Experimental evaluations on real-world network topologies demonstrate that our framework offers scalable and cost-effective solutions compared to existing approaches.