MARC: Motion-Aware Rate Control for Mobile E-commerce Cloud Rendering
summary
MARC is a server-side, frame-level rate controller that predicts short-term motion with a lightweight Markov model and uses frame-level stochastic optimization to assign different bitrates for motion vs non-motion frames. Deployed to 1M+ sessions, it claims 71% session freeze reduction and 20% increased interaction time while adding ~1.3% server CPU overhead; simulations and A/B tests show MARC traces out a better bitrate vs 95th-percentile latency frontier than WebRTC and several baselines.
AI-written summary · about
chapters (13)
- 0:00Introduction
- 1:45Latency versus visual quality
- 3:34Latency sensitivity during motion
- 5:42Motion patterns impact
- 7:28QE mismatch in systems
- 8:12MARC design principles
- 9:17Motion and network predictor
- 10:23QE optimizer with penalties
- 11:08Frame delay cascade strategy
- 12:14Simulation results
- 12:58Real-world A/B test results
- 13:41Computational overhead summary
- 14:47Q&A
related talks
22mProviding better video experiences for the next billion users.
15mEnabling High-Quality Real-Time Communications with Adaptive Frame-Rate
15mProactive Energy-Aware Adaptive Video Streaming on Mobile Devices
10mRTC @Scale, Future RTC Experiences Session - Live Q&A
19mUsing application informed pacing to be a friendly internet neighbor
17mContent Aware Encoding for low latency live streaming encoders using deep learning
26mBandwidth Ballet: Optimizing Video Performance for Web Maestros
19mBeyond the Screen: Streaming the Immersive Internet with Mawari