ETC: An Elastic Transmission Control Using End-to-End Available Bandwidth...
summary
ETC claims a new congestion signal: the pulling rate, measured from the instantaneous receipt rate of N-packet microbursts, and uses a convex elastic step function to update send rate. Implementation is C++ user-space over UDP, shipped as an SDK in Tencent X P2P and tested between servers in Beijing, Shenzhen, Mumbai and Virginia. Results show ~15% higher throughput than Copa, ~20% lower latency than BBR, near-zero rebuffer at 60 Mbps (2% at 15 Mbps), and a 10× improvement in convergence time.
AI-written summary · about
chapters (16)
- 0:00ETC overview
- 0:42Congestion control targets
- 2:05Limitations of mainstream signals
- 3:48Current design limitations
- 5:14ETC design principles
- 6:40Pulling rate as signal
- 8:04Challenges in pulling rate data
- 9:53Fine-grained clocking with ACKs
- 10:56Rate acceleration in ETC
- 12:22Rate deceleration in ETC
- 13:48ETC implementation
- 14:33ETC vs state-of-the-art
- 16:01ETC fairness and speed
- 17:06Compatibility with Cubic
- 17:49ETC in video apps
- 18:53Summary of ETC
related talks
14mTECC: Towards Efficient QUIC Tunneling via Collaborative Transmission Control
19mTackling DC Congestion and Bursts
16mPowerTCP: Pushing the Performance Limits of Datacenter Networks
16mSöze: One Network Telemetry Is All You Need for Per-flow Weighted Bandwidth Allocation...
23mACK to the Future: TCP in 2025
16mR-TCP: A Framework to Optimize TCP Performance Over Rate-Limiting Networks
22mAdapting TCP for Reconfigurable Datacenter Networks
12mP4EC: Enabling Terabit Edge Computing in Enterprise 4G LTE
