Towards provably performant congestion control
Talkat 21st USENIX Symposium on Networked Systems Design and Implementation (NSDI '24) · 2024-04-16..04-18·
summary
Framework creates a belief set that exhaustively maps observed send/ACK time series to possible bottleneck bandwidth, buffer, and delay states, and uses SMT-based synthesis to generate short, implementable rate-decision programs that come with formal guarantees and proven impossibility results. Results show a concrete loss vs convergence trade-off: tolerating more loss lets algorithms converge in time logarithmic in bandwidth (as BBR-like probes do) versus linear for conservative AIMD, and synthesized CCAs are implemented and validate controlled probing and draining behavior.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:07Traditional congestion control limits
- 4:41Belief set representation
- 9:20Impossibility results with belief sets
- 13:40Synthesizing congestion control algorithms
- 14:23Trade-offs in objectives
- 16:49Summary
related talks
16mFRCC: Towards Provably Fair and Robust Congestion Control
18mCCC: Re-architecting Delay-based Congestion Control in Datacenter Networks
24mCopa: Practical Delay-Based Congestion Control for the Internet
23mACK to the Future: TCP in 2025
25mPCC Vivace: Online-Learning Congestion Control
17mBarre: Empowering Simplified and Versatile Programmable Congestion Control in...
13mAUTO: Adaptive Congestion Control Based on Multi-Objective Reinforcement Learning...
16mR-TCP: A Framework to Optimize TCP Performance Over Rate-Limiting Networks
