CCWG

Summary

The Congestion Control Working Group (CCWG) met at IETF 125 to discuss progress on core drafts and several new proposals. Key topics included updates to draft-ietf-ccwg-bbr, refinements to the SCReAM v2 algorithm for media, a "Rapid Start" proposal to improve slow start performance, Christian Huitema’s "C4" congestion control algorithm, and an analysis of LEO satellite mobility effects on congestion control. The chairs also issued a call to action for the working group to review recent changes to draft-ietf-ccwg-ratelimited-increase.

Key Discussion Points

Working Group Drafts

draft-ietf-ccwg-ratelimited-increase: The chairs noted that non-trivial changes were made after the last call. The working group was asked to review the latest version to ensure continued consensus.
BBR Updates (regarding draft-ietf-ccwg-bbr):
- Ian Swett presented recent logic refinements, including improved pacing text, handling loss recovery when SACK is unavailable, and an exit condition for the DRAIN phase (after 3 RTTs) to prevent flows from getting stuck.
- Editorial work continues to remove TCP-specific language to better support QUIC and other transports.
- A Pull Request (PR) for standardized test cases is in progress to help implementers avoid common pitfalls.
- Yusuf Guven and Ian Swett discussed a paper regarding BBRv3 vs. Cubic coexistence, noting that while BBRv3 fixes bugs from v2, work continues to optimize fairness.

Media Congestion Control

SCReAM V2:
- Ingemar Johansson detailed updates to the SCReAM algorithm, focusing on adaptive reference window overhead and L4S support.
- Stuart Cheshire questioned the prescriptive nature of discarding late frames, noting that dropping a single late P-frame can destroy the decoding pipeline. Ingemar Johansson agreed that the draft needs to clarify when discarding is appropriate.
- Harald Alvestrand confirmed that Google has a WebRTC implementation in Chrome (Canary) and intends to push SCReAM as a standard for WebRTC if experiments succeed.
- Polls: The room showed strong interest in continuing work on media congestion control and using SCReAM v2 as a starting point.

Startup Phase Improvements

Rapid Start:
- Kazuho Oku proposed three changes to slow start: full RTT pacing in the first RTT, a 3x (rather than 2x) growth rate until queue buildup is detected, and a smoother reduction of the congestion window upon entering recovery.
- Production data from a CDN showed significant Time to Last Byte (TTLB) reductions (10–20%).
- Martin Duke requested simulations on how Rapid Start interacts with other flows at a bottleneck.
- Neal Cardwell noted that the algorithm's requirements (bytes acked/lost per ACK) are available in Linux TCP and suggested looking at how these ideas could apply to BBR.
- Polls: There was significant interest in the working group adopting work on slow start improvements based on this proposal.

New Algorithms and Research

Presenting C4 (Christian’s CC Code):
- Christian Huitema presented C4, a "simple" algorithm using ECN, Max RTT, and pacing. It uses sensitivity curves to enable Multiplicative Increase Multiplicative Decrease (MIMD) to converge toward fairness.
- Lars Eggert suggested that a new general-purpose congestion control algorithm should undergo academic peer review (e.g., at a conference like SIGCOMM or IMC) before being adopted by the IETF.
- Suhas Nandakumar mentioned Cisco is experimenting with C4 for Media-over-QUIC (MoQ) due to its performance in high-jitter environments like Wi-Fi.
Analysis for the Adverse Effects of LEO Mobility on Internet Congestion Control:
- Zonglun Li discussed measurements from Starlink, showing that LEO satellite mobility causes "Path Phase Boundaries" (PPB)—sharp shifts in RTT and bandwidth every ~15 seconds due to satellite handovers.
- Existing CCAs often misinterpret these as congestion. The author suggested the transport layer may need explicit signaling for path-phase changes to prevent stale estimates in algorithms like BBR.

Decisions and Action Items

Review Request: Working group members are tasked with reviewing draft-ietf-ccwg-ratelimited-increase following recent updates.
BBR Development: The editors of draft-ietf-ccwg-bbr will prioritize finishing the PR for test cases.
SCReAM Clarification: Ingemar Johansson will clarify the logic regarding frame discarding in the next version of the SCReAM draft.

Next Steps

The chairs will take the adoption/interest polls for SCReAM v2 and Rapid Start to the mailing list to confirm consensus.
The C4 algorithm and LEO mobility analysis will likely continue as discussion topics within ICCRG and CCWG to determine future standardization potential.
Lars Eggert encouraged proponents of new slow start or CC algorithms to share specific metrics for telemetry so other implementers (e.g., Mozilla) can perform apples-to-apples comparisons.

Automatic IETF Minutes