Stochastic Forecasts Achieve High Throughput and Low Delay over Cellular Networks
summary
Sprout is an end-to-end transport that has the receiver infer a probabilistic link-speed “cloud,” forecast 100 ms into the future at 95% confidence, and pace the sender to bound the risk of queueing delays (5% chance of exceeding 100 ms). Evaluation uses a trace-based emulator fed with traces from four commercial LTE/3G networks and an EC2 image; Sprout cuts self-inflicted delay by ~7–9× versus Skype/Hangouts/FaceTime while increasing throughput (2.2×,4.4×,1.9×), and matches TCP Cubic+CoDel without modifying carrier equipment.
AI-written summary · about
chapters (22)
- 0:00Cellular Networks
- 0:59Network Delivery
- 1:41Interactive Applications
- 2:45Skype
- 4:14Why is Skype Performance so bad
- 4:46Sprout
- 5:11Bounded Risk
- 5:55Inference
- 8:10Stochastic Model
- 9:05Control Strategy
- 12:18Results
- 13:35Buffer Bloat
- 14:41Competing Traffic
- 15:08Stanford Networking Class
- 15:39MIT Networking Class
- 16:34Preliminary Results
- 17:10Student Results
- 17:35Optimal Tradeoffs
- 18:01Limitations
- 18:54Bottom Line
- 20:05Summary
- 21:17Questions
related talks
28mSalsify: Low-Latency Network Video through Tighter Integration
17mAUGUR: Practical Mobile Multipath Transport Service for Low Tail Latency in Real-Time...
45mReducing Network Latency: Innovations for a Faster Internet
29mDealing with Unreliable Networks
15mPOLYCORN: Data-driven Cross-layer Multipath Networking for High-speed Railway
20mSTORM: a Multipath QUIC Scheduler for Quick Streaming Media Transport under...
16mBolt: Sub-RTT Congestion Control for Ultra-Low Latency
28mFrom Feast to Famine: Managing Mobile Network Resources Across Environments