MARLIN: Soft Actor-Critic based Reinforcement Learning for Congestion Control in Real Networks

TL;DR

MARLIN employs a Soft Actor-Critic reinforcement learning approach to develop a congestion control algorithm that adapts to diverse network scenarios, achieving performance comparable to TCP Cubic with minimal tuning.

Contribution

This work introduces MARLIN, a novel RL-based congestion control algorithm using maximum entropy RL, trained on real networks to improve adaptability across heterogeneous environments.

Findings

MARLIN performs comparably to TCP Cubic in file transfer tasks.

Training on real networks reduces sim-to-real transfer issues.

Minimal hyperparameter tuning is needed for effective performance.

Abstract

Fast and efficient transport protocols are the foundation of an increasingly distributed world. The burden of continuously delivering improved communication performance to support next-generation applications and services, combined with the increasing heterogeneity of systems and network technologies, has promoted the design of Congestion Control (CC) algorithms that perform well under specific environments. The challenge of designing a generic CC algorithm that can adapt to a broad range of scenarios is still an open research question. To tackle this challenge, we propose to apply a novel Reinforcement Learning (RL) approach. Our solution, MARLIN, uses the Soft Actor-Critic algorithm to maximize both entropy and return and models the learning process as an infinite-horizon task. We trained MARLIN on a real network with varying background traffic patterns to overcome the sim-to-real mismatch that researchers have encountered when applying RL to CC. We evaluated our solution on the task of file transfer and compared it to TCP Cubic. While further research is required, results have shown that MARLIN can achieve comparable results to TCP with little hyperparameter tuning, in a task significantly different from its training setting. Therefore, we believe that our work represents a promising first step toward building CC algorithms based on the maximum entropy RL framework.