PtrNet-RSA: A Pointer Network-Based QoT-Aware Routing and Spectrum Assignment Scheme in Elastic Optical Networks
Yuansen Cheng, Shifeng Ding, Yingjie Shao, Chun-Kit Chan
Abstract
To enable flexible service provisioning in elastic optical networks (EONs), quality of transmission (QoT) estimation and dynamic routing and spectrum assignment (RSA) are critical network management tasks. Recently, machine learning approaches have been extensively investigated to address these tasks individually. We propose to take an integrative approach to design a Pointer Network (PtrNet)-based QoT-aware RSA (PtrNet-RSA) scheme to optimize the blocking probability and the generalized signal-to-noise ratio (GSNR) for the lightpath allocation in EONs. Given the profiles of the physical network and the service requests, the proposed scheme can generate high-GSNR lightpaths without pre-calculated candidate paths. Reinforcement learning is employed to train the model while the parameters of the PtrNets are optimized by utilizing the actor-critic algorithm. Extensive simulations have been conducted in EON environments based on the Gaussian Noise (GN) model with dynamic traffic. The results show that the proposed PtrNet-RSA scheme consistently outperforms the benchmarks with various network topologies, significantly reducing the blocking probability while guaranteeing a good lightpath QoT.