Congestion Control: A Renaissance with Machine Learning
Wenting Wei, Huaxi Gu, Baochun Li
Abstract
In the past several decades, it has been well known that the Transmission Control Protocol (TCP), even with its modern variants such as CUBIC, may not perform optimally when available bottleneck bandwidth needs to be fully utilized, yet without unnecessarily increasing the end-to-end latency. These observations have led to a recent resurgence of interest in the topic of redesigning congestion control protocols and replacing modern TCP variants using machine learning. In this article, we examine and compare some of the most prominent recent research results on the use of machine learning to redesign congestion control protocols, with an editorial commentary on potential research directions in the near-term future.