Improving TCP CUBIC Congestion Control with Machine Learning - 2023 James Knott

Aug 3, 2026

Transmission Control Protocol (TCP) is widely used for reliable internet transfers, yet congestion-driven packet loss can degrade performance through increased delays and reduced throughput. This project proposes combining machine learning with existing congestion control rather than replacing TCP mechanisms, focusing specifically on TCP CUBIC, the default Linux Kernel congestion control variant. A model-free Q-learning approach optimizes CUBIC’s beta parameter to improve throughput and reduce loss across diverse simulated network conditions. Extensive testing demonstrates strong adaptability and measurable gains, and the report documents key development decisions and directions for future work.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages10

Reading Time25 min

AccessFree to download

Introduction

Related Work

Methodology

Q-Learning for TCP CUBIC Parameter Optimization

Evaluation

Simulated Network Conditions

Results and Discussion

Development Decisions and Future Work

Improving TCP CUBIC Congestion Control with Machine Learning (2023) James Knott Abstract—Transmission Control Protocol (TCP) is commonly used for reliable internet data transfers. However, TCP can experience packet loss due to network congestion. Packet loss happens when data doesn't reach its destination for various reasons. In recent years, there has been a growing inclination towards adopting novel, clean-slate learning-based designs as alternatives to traditional congestion control mechanisms for the Internet. However, we posit that integrating machine learning techniques with the current congestion control schemes can achieve comparable, if not superior outcomes. This project endeavoured to address this gap and implement a system that can utilised with TCP CUBIC. Our method looked to enhance the efficiency of the TCP CUBIC congestion control by incorporating machine learning techniques. TCP CUBIC, the default congestion control variant in the current Linux Kernel, modifies the congestion window size based on a loss-based algorithm, thereby influencing the rate of data transmission. TCP CUBIC uses a parameter, beta, to modify the rate at which the congestion window grows. Our approach involves employing a model-free reinforcement learning algorithm, specifically a Q-learning algorithm to optimize the TCP CUBIC beta parameter, targeting an increase in throughput for TCP CUBIC connections. Through extensive testing performed in various simulated network conditions we demonstrate the performance and adaptability of the Q-Learning algorithm. Furthermore, this report details the various development decisions undertaken and their driving influences. It also provides an insight into the project's results, expanding on the existing system design, and elaborates on the potential for future work in this area. Index Terms—TCP, CUBIC, Q-Learning, Reinforcement Learning, Machine Learning I. INTRODUCTION Transmission Control Protocol (TCP) is a protocol that is frequently used by internet users for reliable data transfers. However, TCP can suffer from packet loss through network congestion. Packet loss is the loss of data during network transmission. It occurs when one or more packets fail to reach their destination, which can happen due toa variety of reasons. When packets are lost the receiving node may not receive all the data it requires to properly reconstruct the data. The implications of packet loss are considerable, leading to network performance degradation in terms of packet delays, a decrease in throughput, and reduced application performance. As a response to these challenges, various TCP alternatives equipped with congestion control algorithms have been developed to curb such adverse effects. TCP CUBIC (CUBIC) is a notable example of these alternatives and has been This project was supervised by Winston Seah (primary), Alvin Valera. universally adopted across standard operating systems, including Windows, Linux, and Mac [1] . CUBIC controls congestion window growth using a cubic function [1] . This project looked to reduce packet loss and increase throughput of TCP connections using machine learning. The proposed solution looked to develop a machine learning algorithm that modifies CUBIC parameters to improve throughput and packet loss by 15% . We did not look to redesign CUBIC but to create an algorithm that can work alongside its current implementation. Furthermore, in the current literature, there is a noticeable gap in machine learning algorithms used with the already existing CUBIC. Most studies look to create their own protocol that works in conjunction with machine learning [3][4][5][6][7][8] . This project looked to address this gap and implement a system that can utilised with CUBIC. Using a Q-learning algorithm tooptimise CUBIC’s congestion control parameters. Specifically, CUBIC’s β (beta) parameter. CUBIC follows a cubic algorithm and has parameters that affect the growth of the congestio

FAQs

Why does TCP experience packet loss and what are the effects?

TCP can lose packets when congestion prevents data from reaching its destination. Packet loss increases delays, reduces throughput, and degrades overall application performance.

How does the project improve TCP CUBIC congestion control?

It integrates a Q-learning reinforcement learning algorithm to optimize CUBIC’s beta parameter, which influences how the congestion window decreases after losses and how it grows.

What improvements did the Q-learning approach achieve in testing?

Across simulated network conditions, the approach increased throughput by up to 13% and reduced packet loss by up to 8%, showing adaptability under loss scenarios.

Aug 3, 2026

Upload by Lucas Martin