BurstRTC: Harnessing Variable Bit-Rate of RTC Through Frame-Bursting Congestion Control
Zhidong Jia,Yihang Zhang,Qingyang Li,Xinggong Zhang
DOI: https://doi.org/10.1145/3663408.3665821
2024-01-01
Abstract:The rapid growth of online interactive video applications reflects the increasing popularity of real-time communication (RTC). Despite advancements in network and video technologies, the worse quality of experience (QoE) such as large delay, rebuffering and low image quality, etc. remains to be complained generally. We argue that this is mainly due to the legacy network-oriented congestion control (CC), which assumes continuous stream of packets are sent. But it is not satisfied for RTC since bit-rate variation is inherent to RTC’s video encoder. In response, we propose BurstRTC, a novel paradigm for RTC transport protocol. Packets are emitted by frame-bursting, not longer by continuous packets stream. A novel frame-level congestion control mechanism is also proposed. Network congestion is detected by side-by-side packets of frame, and a simple queue-based congestion model is used to predict congestion. For better fairness and QoE of RTC, a QoE objective function and video frame model are also proposed. By adjusting video coding parameters, frame size is optimally decided to avoid network congestion while achieve good QoE and fairness. Preliminary experimental results indicate that BurstRTC has achieved the expected outcomes, reducing frame delay by up to 66.03%, improving image quality by SSIM up to 30.1%, while preserving good fairness.