PCC Vivace: Online-Learning Congestion Control
summary
Vivace leverages online convex optimization to replace fixed TCP rules with a strictly concave latency-aware utility that penalizes RTT inflation and loss, and a gradient-ascent probing controller that probes slightly above and below the current rate to estimate utility gradients. Implementation uses a UDT user-space implementation and emulators (Emulab, mahimahi) plus EC2 tests. Results show up to ~90% lower RTT inflation versus PCC-BBR in some setups, a 5% random-loss tolerance parameter, faster convergence and a formal no-regret guarantee.
AI-written summary · about
chapters (16)
- 0:00Intro
- 0:28Internet Congestion Control
- 4:12Strong Assumptions Cause Problem
- 4:57PCC Utility Framework
- 5:27PCC Rate Control
- 6:08PCC Design Challenges
- 6:52PCC Allegro
- 8:29PCC Vivace
- 12:41Evaluation
- 13:15Vivace Utility Performance
- 14:08Vivace Rate Control Performance
- 15:07Vivace upon Convergence
- 15:42TCP Friendliness
- 17:06Limitation in Extremely Dynamic Networks
- 19:16Conclusion
- 20:08THE ADVANCED COMPUTING SYSTEMS ASSOCIATION
related talks
17mTowards provably performant congestion control
15mMortise: Auto-tuning Congestion Control to Optimize QoE via Network-Aware Parameter
18mCCC: Re-architecting Delay-based Congestion Control in Datacenter Networks
23mACK to the Future: TCP in 2025
20mETC: An Elastic Transmission Control Using End-to-End Available Bandwidth...
17mBarre: Empowering Simplified and Versatile Programmable Congestion Control in...
19mPudica: Toward Near-Zero Queuing Delay in Congestion Control for Cloud Gaming
13mAUTO: Adaptive Congestion Control Based on Multi-Objective Reinforcement Learning...