DL-ViNE: reinforcement learning algorithm for efficient virtual network embedding under direct-link constraints
Résumé
The Fifth and Sixth Generation (5G/6G) of mobile networks aim to support diverse industrial and service applications, each with specific Quality of Service (QoS) and resource requirements. Network Slicing has emerged as a critical paradigm to meet these demands by creating multiple Virtual Networks (VNs) on a shared physical infrastructure. This process, known as Virtual Network Embedding (VNE), involves efficiently mapping virtual nodes and links onto physical resources to maximize utilization and operator benefits. With the growing adoption of Kubernetes as the leading container orchestration platform, most modern infrastructures, whether cloud-based or on-premises, are hosted on Kubernetes clusters. Kubernetes enforces direct communication between pods, necessitating a direct-link mapping approach in the VNE process. However, existing solutions primarily focus on mapping virtual links to physical paths, leaving the direct-link scenario unexplored. In this paper, we introduce DL-ViNE (Direct Link Virtual Network Embedding), a novel reinforcement learning-based algorithm designed to address the VNE problem under direct-link constraints. Our approach enhances slice acceptance while meeting the unique demands of Kubernetes-hosted infrastructures.
| Origine | Fichiers produits par l'(les) auteur(s) |
|---|---|
| Licence |