Evaluation of Agent-Network Environment Mapping on Open-AI Gym for Q-Routing Algorithm

Abstract

The changes in network dynamics demands a routing algorithm that adapts intelligently with the changing requirements and parameters. In this regard, an efficient routing mechanism plays an essential role in supporting such requirements of dynamic and QoS-aware network services. This paper has introduced a self-learning intelligent approach to route selection in the network. A Q-Routing approach is designed based on a reinforcement learning algorithm to provide reliable and stable packet transmission for different network services with minimal delay and low routing overhead. The novelty of the proposed work is that a new customized environment for the network, namely Net-AI-Gym, has been integrated into Open-AI Gym. Besides, the proposed Q-routing with Net-AI-Gym offers optimization in exploring the path to support multi-QoS aware services in the different networking applications. The performance assessment of the NET-AI Gym is carried out with less, medium, and a high number of nodes. Also, the results of the proposed system are compared with the existing rule-based method. The study outcome shows the Net-AI-Gym’s potential that effectively supports the varied scale of nodes in the network. Apart from this, the proposed Q-routing approach outperforms the rule-based routing technique regarding episodes vs. Rewards and path length.