This project, supported by the APNIC Foundation, focuses on implementing the IETF-defined L4S (Low Latency, Low Loss, Scalable Throughput) architecture within the FreeBSD kernel to support real-time, congestion-aware internet applications. The work involves modifying queue management systems and using reinforcement learning to dynamically optimize performance for modern digital services including cloud gaming, real-time video, and industrial IoT.
Project Overview
This project explores the integration of Deep Reinforcement Learning (DRL) and Large Language Models (LLMs) with the Low Latency, Low Loss, and Scalable Throughput (L4S) architecture to improve Internet congestion control. Hosted by the lab and funded by the APNIC Foundation, the project implements modular congestion control algorithms within FreeBSD, evaluates them on custom-built testbeds, and disseminates experimental results, software and datasets to the wider community.
Key Outcomes
- DRL-based Multipath TCP (MPTCP) and AQM modules implemented in FreeBSD
- Public testbed infrastructure available for reproducible experimentation
- Peer-reviewed publications and ongoing contributions in academic/industry conferences
- Student internships, mentoring, and curriculum integration for skill development
- Ongoing industry and research partnerships
Open Source Software and Datasets
All source code and supporting data from this project have been made openly available to support reproducibility and collaboration:
- Experimental L4S testbed setup: MPTCP-FreeBSD/FreeBSD-L4S-Experiments
- DRL-enhanced congestion control stack for FreeBSD: MPTCP-FreeBSD/FreeBSD-DRL-L4S
- Predictive congestion marking using LLMs: MPTCP-FreeBSD/L4S-LLM
- Adaptive TCP fairness via fine-tuned LLMs: MPTCP-FreeBSD/LLM-TCP
- AQM decision models powered by LLM-based inference: MPTCP-FreeBSD/AQM-LLM
This project has also initiated and delivered new Agentic AI-driven L4S research for the community:
- LLM-guided AQM with adaptive invocation: MPTCP-FreeBSD/llm-invocation
- Reason-Action for L4S AQM: MPTCP-FreeBSD/Reason-Action-LLM
- ReWOO-AQM as a reasoning framework for L4S AQM: MPTCP-FreeBSD/rewoo-aqm
Publications
- S. R. Pokhrel, D. Satish, J. Kua, and A. Walid, “Distilling large language models for network active queue management,” IEEE Transactions on Networking, vol. 34, pp. 5501-5513, 4 May 2026. [Online]. Available: https://doi.org/10.1109/TON.2026.3690076
- S. Shrestha, J. Kua, and S. R. Pokhrel, “Visualizing TCP BBRv3 performance in AQM-enabled wireless networks,” in 50th IEEE Conference on Local Computer Networks (LCN), 14-16 October 2025, Sydney, Australia. [Online]. Available: https://doi.org/10.1109/LCN65610.2025.11146369
- S. R. Pokhrel, J. Kua, D. Satish, S. Ozer, J. Howe, and A. Walid, “DDPG-MPCC: An experience driven multipath performance oriented congestion control,” Future Internet, vol. 16, no. 2, p. 37, Feb. 2024. [Online]. Available: https://doi.org/10.3390/fi16020037
- S. K. Shrestha, S. R. Pokhrel, and J. Kua, “On the fairness of internet congestion control over WiFi with deep reinforcement learning,” Future Internet, vol. 16, no. 9, p. 330, Sept. 2024. [Online]. Available: https://doi.org/10.3390/fi16090330
- D. Satish, J. Kua, and S. R. Pokhrel, “Active Queue Management in L4S with Asynchronous Advantage Actor-Critic: A FreeBSD Networking Stack Perspective,” Future Internet, vol. 16, no. 8, p. 265, Aug. 2024. [Online]. Available: https://doi.org/10.3390/fi16080265
- S. R. Pokhrel, J. Kua, B. Fleming, S. Ozer, J. Howe, and A. Walid, “Multipath TCP implementation under FreeBSD-13 for pluggable machine learning models,” Computer Networks, vol. 252, p. 110671, 2024. [Online]. Available: https://doi.org/10.1016/j.comnet.2024.110671
- S. K. Shrestha, S. R. Pokhrel, and J. Kua, “Adapting large language models for improving TCP fairness over WiFi,” IEEE Transactions on Network and Service Management, August 2026 (revision submitted). [Online]. Available: https://arxiv.org/abs/2412.18200
- S. K. Shrestha, S. R. Pokhrel, and J. Kua, “TCP BBR performance over Wi-Fi 6: AQM impacts and cross-layer insights,” IEEE Transactions on Network and Service Management, August 2026 (under review).
Sponsors
Get Involved
Interested in collaborating or learning more? Our project continues to grow through academic and industry partnerships, conference engagement, and educational outreach.
- Contact: jonathan.kua@deakin.edu.au
- Contact: shiva.pokhrel@deakin.edu.au
