Adaptive queueing policy configuration on switch using machine learning methods

Recieved: 12/28/2023

Accepted: 03/01/2024

Published: 06/26/2024

Keywords: traffic load balancing, machine learning methods, DQN, switch, reinforcement learning

To cite this article

Timoshkin M.O., Stepanov E.P. Adaptive queueing policy configuration on switch using machine learning methods. // Moscow University Journal. Series 15. Computational Mathematics and Cybernetics. 2024. N 3, p.73-84 https://doi.org/10.55959/MSU/0137-0782-15-2024-47-3-73-84.

This work is licensed under a Сreative Commons Atribiution - NonCommercial 4.0 International (CC BY-NC 4.0)

Timoshkin M.O. (Lomonosov Moscow State University (MSU))
Stepanov E.P. (Lomonosov Moscow State University (MSU))

Abstract

The traffic load balancing problem is relevant in modern networks that have many alternative routes between any pair of nodes. Balancing provides a uniform load of network resources. The paper proposes a method for adaptive queueing policy configuration on a switch to get a uniform queue load on the switch output ports. Because modern applications require the data transmission delay to be around milliseconds, reinforcement learning method DQN was applied to solve the problem. An experimental study demonstrated the convergence of the proposed method during the training to uniform queue load at output ports.