Salsify: Low-Latency Network Video through Tighter Integration
summary
Salsify tightly couples a transport that predicts the next frame size with a per-frame optimization loop and a functional video codec that can save/restore encoder state and produce two quality variants or discard frames. The transport picks the variant that stays under its instantaneous capacity estimate, avoiding queueing and recovery delays. Evaluation on reproducible traces and real systems shows lower end-to-end video delay and about 2 dB higher visual quality than FaceTime, Skype, Hangouts and WebRTC (with/without SVC); encoding two variants doubles CPU work.
AI-written summary · about
chapters (4)
related talks
15meXpressSFU: Toward Super-Scalable Video Conferencing with SmartNICs
26mReliable Video Delivery with Minimal Resources
17mLessons Learned: Low Latency Ingest
43mDASH and Media-over-QUIC Transport Face-Off: Performance Showdown at Low Latency
20mTambur: Efficient loss recovery for videoconferencing via streaming codes
22mSMPTE ST 2110 - Highest Quality, Lowest Latency, Pro Video over IP
27mStochastic Forecasts Achieve High Throughput and Low Delay over Cellular Networks
19mLive Media Over QUIC