Using Neural Networks for Packet Routing

Abstract

Packet Routing has been and is the core of any Networking Node and how it interacts with the neighboring network nodes and it also governs how traffic is routed through these devices. Traditional routing algorithms rely on configured peering rules which govern with which routing nodes each can talk to and which networks it can learn and forward traffic to. This mechanism also protects these nodes from learning rogue networks and forwarding packets to unintended destinations. Some of these approaches suffer from fundamental limitations. Choosing an appropriate data structure is not a straightforward task and often hard to estimate. In this paper I will expand on the idea suggested by the paper A Deep Reinforcement Learning Approach for Adaptive Traffic Routing in Next – gen networks by suggesting next gen algorithms which can expand the concepts of Neural Nets to both Internal and External Routing domains which takes care of machine learning based packet routing in both enterprise and public networks and internet. In this paper we will develop a methodology to use Graph Neural Networks, Deep Reinforcement Learning, Recurrent Neural Networks and Autoencoders and Generative Adversarial Networks to create a system which can learn paths and forward traffic and also maintain the quality of service of those paths