Summary
International Conference on Emerging Technologies for Communications
2022
Session Number:S8
Session:
Number:S8-6
AIMD Window Flow Control using Reinforcement Learning
Kento Otani, Shota Inoue, Keita Goto, Hiroyuki Ohsaki,
pp.-
Publication Date:2022/11/29
Online ISSN:2188-5079
DOI:10.34385/proc.72.S8-6
PDF download (546KB)
Summary:
Machine-learning-based traffic control techniques that do not require prior knowledge of the internal state of the network have been actively studied. Sacco et al. proposed Owl, a window flow control method that uses Q-learning, a type of reinforcement learning. Owl changes the window size additively (i.e., addition and subtraction) as an action of the sending host. We believe that AIMD-type window size adjustment is effective in window flow control schemes using reinforcement learning. This study proposes Q-learning-based AIMD window flow control (Q-AIMD) to change the congestion window in an AIMD fashion and clarifies its effectiveness through experiments.