**Session Date/Time:** 17 Mar 2025 02:30

# iccrg

## Summary

The Internet Congestion Control Research Group (ICCRG) meeting covered several topics related to congestion control, including updates on active research group documents, pacing in transport protocols, congestion control in LEO satellite networks, improving cloud gaming traffic quality of service, and congestion control for AI/ML data centers. The meeting included presentations and discussions on each topic, with a focus on technical details and potential solutions to specific challenges.

## Key Discussion Points

*   **Ledbet++:**  An update on the Ledbet++ draft was presented, highlighting its maturity and readiness for a research group last call.
*   **Pacing in Transport Protocols:** A draft on pacing and transport protocols was discussed, including its evolution, new text on QDynamics and AQM interactions, and plans for further expansion.  The group expressed interest in adopting the document.
*   **Congestion Control in LEO Satellite Networks:** A presentation explored the challenges of congestion control in LEO satellite networks, emphasizing the impact of Leo-induced non-congestion variations and potential mitigation strategies based on satellite reconfiguration information. Questions and comments were raised about bandwidth and delay measurements, BBR performance, the use of machine learning, and the definition of "non-congestion losses."
*   **Improving Cloud Gaming Traffic Quality of Service:**  A class-based queuing policy and the L4S architecture were evaluated for improving cloud gaming traffic. The discussion highlighted the bufferbloat phenomenon, the unfair cohabitation between different CCAs, and the benefits of dual-queue solutions like HTB and L4S. Operational advantages and constraints of both approaches were also presented.
*   **Confucius - Gradual Bandwidth Share:** A proposal was made for a more gentle way of allocating bandwidth across different flows to avoid sudden changes and better control traffic congestion.
*   **Congestion Control for AI/ML Data Centers:** Challenges were described related to a relatively new application of Congestion Control algorithms. In particular for high performance data centers for training AI/ML models. The group discussed approaches for packet spraying, packet trimming and receiver driven flow control to mitigate congestion and improve flow completion time.

## Decisions and Action Items

*   **Ledbet++:** Proceed to a research group last call.
*   **Pacing in Transport Protocols:**  Adopt the draft as a research group document.
*   **Congestion Control in LEO Satellite Networks:**  The presenters will consider feedback and suggestions from the group for future work, including addressing the distinction between loss due to congestion and link errors.

## Next Steps

*   **Ledbet++:** The chairs will initiate the research group last call.
*   **Pacing in Transport Protocols:** The authors will incorporate feedback and prepare the document for formal adoption.
*   **Congestion Control for AI/ML Data Centers:**  Further discussion of this topic might be held in a side meeting during the week, since it was cut short due to time constraints.