IKENGA: Infeasibility Knowledge-Enhanced Genetic Algorithm for Virtual Network Embedding

Network function virtualization (NFV) is a promising technology that enhances the flexibility and efficiency of network management, enabling multiple network services from users to concurrently share the resources of the underlying infrastructure. A critical challenge in NFV lies in the efficient deployment of user services onto the infrastructure while adhering to diverse resource constraints and service requirements. This task, referred to the virtual network embedding (VNE) problem, is essential for optimizing network performance. However, existing approaches struggle to effectively address the intricate constraints of VNE, often resulting in suboptimal solutions that undermine resource utilization and overall network performance. Therefore, in this paper, we propose an I nfeasibility Knowledge-ENhanced Genetic Algorithm (IKENGA) for VNE. Specifically, IKENGA pioneers the handling of infeasible solutions through a penalty-based fitness function that evaluates both feasible and infeasible solutions and a selective infeasible solution repair mechanism, significantly increasing the number of feasible solutions. Additionally, we integrate deep reinforcement learning into the genetic algorithm’s initialization process to facilitate environment-aware decision-making, thereby enhancing solution quality. Furthermore, we propose an adaptive pruning method to optimize the link embedding process. Extensive experimental evaluations demonstrate the effectiveness of our proposed IKENGA and the efficiency of crucial modules.