Enhancing Routing in SD-EONs through Reinforcement Learning: A Comparative Analysis

TL;DR

This paper compares reinforcement learning algorithms for routing in SD-EONs, showing Q-learning significantly reduces blocking probability compared to traditional methods, especially under varying traffic conditions.

Contribution

It introduces a reinforcement learning framework for SD-EON routing and demonstrates its superiority over traditional algorithms through extensive comparison.

Findings

Q-learning reduces blocking probability by up to 58.8% over KSP-FF.

Q-learning achieves up to 81.9% BP reduction over SPF-FF at low traffic.

Reinforcement learning improves network performance in dynamic environments.

Abstract

This paper presents an optimization framework for routing in software-defined elastic optical networks using reinforcement learning algorithms. We specifically implement and compare the epsilon-greedy bandit, upper confidence bound (UCB) bandit, and Q-learning algorithms to traditional methods such as K-Shortest Paths with First-Fit core and spectrum assignment (KSP-FF) and Shortest Path with First-Fit (SPF-FF) algorithms. Our results show that Q-learning significantly outperforms traditional methods, achieving a reduction in blocking probability (BP) of up to 58.8% over KSP-FF, and 81.9% over SPF-FF under lower traffic volumes. For higher traffic volumes, Q-learning maintains superior performance with BP reductions of 41.9% over KSP-FF and 70.1% over SPF-FF. These findings demonstrate the efficacy of reinforcement learning in enhancing network performance and resource utilization in dynamic and complex environments.