FRCC: Towards Provably Fair and Robust Congestion Control
summary
Claims that FRCC provably bounds unfairness under jitter by encoding only link fraction (flow count) into congestion signals and separately probing capacity. Implements capacity probes that scale with estimated flow count, uses sustained-delay measurements rather than instantaneous deltas, and proves convergence with dynamical-system analysis plus exhaustive worst-case checks via Z3 SMT solver. Delivers a Linux kernel implementation and experiments showing FRCC prevents BBR-like starvation across synthetic jitter, heterogeneous RTTs, and multi-bottleneck setups.
AI-written summary · about
chapters (9)
- 0:00FRCC and fairness problem
- 1:22BBR starvation on asymmetric delays
- 3:10Congestion control via encodings
- 6:38Noise causes starvation
- 8:44Filtering noise insufficient
- 9:49Decoupling capacity and flows
- 11:57Link capacity under noise
- 14:04FRCC formal guarantees
- 15:27Summary and future
related talks
17mTowards provably performant congestion control
18mRevisiting Congestion Control for Lossless Ethernet
16mFLB: Fine-grained Load Balancing for Lossless Datacenter Networks
17mBarre: Empowering Simplified and Versatile Programmable Congestion Control in...
17mEnabling Users to Control their Internet
17mPyrrha: Congestion-Root-Based Flow Control to Eliminate Head-of-Line Blocking in...
19mSwCC: Software-Programmable and Per-Packet Congestion Control in RDMA Engine
25mPCC Vivace: Online-Learning Congestion Control
