PLANAR: A Machine Learning Based Approach for Robust Network Resource Placement and Routing

The convergence of Software-Defined Networking (SDN) and Network Function Virtualization (NFV) has transformed wide area network operations. A critical challenge in these networks is the strategic deployment of Virtual Network Functions (VNFs) to optimize performance, resource utilization, and cost, amidst varying and unpredictable traffic patterns. Traditional methods often assume a single, static traffic pattern, but this approach is impractical due to the dynamic nature of network traffic. This paper presents a robust machine learningbased approach (PLANAR) for VNF placement and routing that accommodates multiple traffic patterns, providing worst-case performance guarantees. Our gradient descent-based stochastic optimization algorithm efficiently scales to handle hundreds of traffic patterns, significantly enhancing network resilience and resource management. The technique is effective and scalable and works well on a wide variety of network topologies.