AI in Congestion Control
Author
Term
4. semester
Education
Publication year
2025
Submitted on
2025-06-04
Pages
30
Abstract
Congestion control algorithms is a field that has been studied for many years. As a result many different protocols have been developed for the purpose of ensuring reliable end to end communication, that ensures that lost data is retransmitted, and packets arriving out of order will be reordered. However in recent years AI methods such as reinforcement learning have been proposed in the context of congestion control algorithm. This project investigates, how Q-learning can be implemented in TCP, and how the reward function will affect the behavior of the agent. Different versions of reward functions have been designed, implemented and tested in various network setups. The agents are even trained and tested against rule-based TCP protocols such as TCP Vegas and TCP New Reno. Hyper-parameter tuning has also been performed on the agent, and the results showed that hyper-parameter tuning was the deciding factor in regards to agents reaching their goal.
Documents