R-TCP: A Framework to Optimize TCP Performance Over Rate-Limiting Networks
summary
Implements in-kernel Linux changes that detect token-bucket rate limiting by estimating bucket size B and rate R, trigger only when loss exceeds ~20%, then classify and mitigate by capping cwnd for Cubic or pacing for BBR; pipeline of detection → estimation → classification → mitigation runs every ~20 RTTs. Evaluated on five mobile operators and real short‑video and DASH workloads, it claims up to 88% retransmission reduction, cuts retransmits by ~40–60% in apps, and targets the ~28% of flows that become rate‑limited.
AI-written summary · about
chapters (8)
- 0:00R-TCP framework intro
- 1:02Rate-limiting in mobile networks
- 3:33TCP performance impact
- 5:19R-TCP algorithm design
- 8:53Detecting rate-limiting
- 10:18Congestion control with R-TCP
- 12:08R-TCP performance evaluation
- 15:25Conclusion & Q&A
related talks
34mStrategies for Rate Limiting Network Packet Ingress
23mACK to the Future: TCP in 2025
17mEnabling Users to Control their Internet
19mTackling DC Congestion and Bursts
16mBolt: Sub-RTT Congestion Control for Ultra-Low Latency
22mAdapting TCP for Reconfigurable Datacenter Networks
17mTowards provably performant congestion control
14mA large-scale deployment of DCTCP