Exploiting GNN and DRL for Online Service Provisioning Over Elastic Optical Networks

• Published in: 2024 International Conference on Optical Network Design and Modeling (ONDM) pp. 1 - 3
• Main Authors: Hernandez-Chulde, Carlos, Casellas, Ramon, Martinez, Ricardo, Vilalta, Ricard, Munoz, Raul
• Format: Conference Proceeding
• Language: English
• Published: IFIP 06-05-2024
• Subjects: Deep Reinforcement Learning; Elastic Optical Networks; Feature extraction; Heuristic algorithms; Network Function Virtualization; Network topology; Optical design; Optical fiber networks; Simulation; Training
• DOI: 10.23919/ONDM61578.2024.10582766

The dynamic provisioning of network services across Data Centers (DCs) interconnected by an Elastic Optical Network (EON) remains a challenging problem, as it requires the orchestration of computing resources in DCs and spectrum resources on optical links. This work proposes a solution based on Deep Reinforcement Learning (DRL) and Graph Neural Networks (GNNs) for autonomous provisioning of network services over inter-DC EONs. The proposed solution includes a GNN to extract the topological features of the inter-DC EON network topology and a DRL agent for the simultaneous and efficient selection of the DCs to host the VNFs and the lightpaths supporting the connection between them. The experimental results show that the proposed solution achieves faster convergence and reward than another DRL-based solution in the training phase. Our proposed solution also outperforms a heuristic algorithm and the other DRL solution for provisioning network services on an inter-DC EON in terms of service blocking rate. Our proposed solution also reduces the service blocking rate compared to a heuristic algorithm and the other DRL solution.