Google Research

SQP: Congestion Control for Low-Latency Interactive Video Streaming

  • Connor Smith
  • David Chu
  • Devdeep Ray
  • Srinivasan Seshan
  • Teng Wei
arXiv, arXiv (2022)

Abstract

This paper presents the design and evaluation of SQP, a congestion control algorithm (CCA) for low-latency interactive video streaming applications like AR streaming and cloud gaming. SQP couples network measurements with frame transmissions, and responds to congestion primarily via modulation of the video bitrate. SQP’s tight integration with the traffic pattern of interactive video streaming also enables a unique rate- and delay-based approach for measuring the net-work bandwidth. This combination of features enables SQP to perform better than prior video-agnostic CCA designs by minimizing end-to-end frame delay due to sender-side queuing, and achieving low network queuing delay, while remaining competitive in the presence of queue-building cross traffic.In real-world A/B testing against Copa in X’s AR stream-ing platform, SQP improves the number of sessions that have high bandwidth and low frame delay by 27% points on LTE, and 15% points on Wi-Fi. On emulated Wi-Fi and LTE links,SQP achieves≈2×higher throughput compared to WebRTC, where as the frame delays of Copa (with mode switching),Sprout and BBR are 140-290% higher. When competing with queue-building traffic (e.g. Cubic, BBR), SQP achieves2−3×higher bandwidth compared to WebRTC, Sprout and Vivace, and comparable performance to Copa (with mode switching). SQP rapidly adapts to dynamic link conditions, quickly converges to fairness, and has good RTT fairness.

Learn more about how we do research

We maintain a portfolio of research projects, providing individuals and teams the freedom to emphasize specific types of work