DL-ViNE: Reinforcement Learning Algorithm for Efficient Virtual Network Embedding Under Direct-Link Constraints

The Fifth and Sixth Generation (5G/6G) networks aim to support diverse applications with specific QoS and resource needs. Network Slicing has emerged as a key paradigm to meet these demands by creating multiple Virtual Networks (VNs) over shared physical infrastructure. This process, known as Virtual Network Embedding (VNE), maps virtual nodes and links to physical resources. With Kubernetes becoming the dominant orchestration platform, most infrastructures now rely on Kubernetes clusters, which enforce direct pod-to-pod communication, necessitating a direct-link approach to VNE. However, most existing methods focus on path-based link mapping. In this paper, we present DL-ViNE, a Reinforcement Learning(RL)-based algorithm that improves slice acceptance while addressing the specific constraints of Kubernetes-hosted infrastructures.