Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Camel: Frame-Level Bandwidth Estimation for Low-Latency Live Streaming under Video Bitrate Undershooting

Liming Liu, Zhidong Jia, Li Jiang, Wei Zhang, Lan Xie, Feng Qian, Leju Yan, Bing Yan, Qiang Ma, Zhou Sha, Wei Yang, Yixuan Ban, Xinggong Zhang

TL;DR

This paper tackles the problem of bandwidth undershooting in low-latency live streaming, where encoder frame-level bursts distort network feedback and lead to stalling. It introduces Camel, a frame-level congestion control framework comprising a Bandwidth and Delay Estimator, a Congestion Detector, and a Bursting Length Controller, and gates transmission with per-frame bursts to accurately probe capacity while avoiding buffer overflow. The key contributions include (i) a frame-level bandwidth-delay-product estimator, (ii) an inflight-delay gradient-based congestion detector, and (iii) a dynamic bursting controller that adjusts bursts to buffer conditions, all validated by large-scale real-world deployment and extensive simulations. Results show Camel improves bitrate and resolution, reduces stalling, and maintains latency comparable to baselines, demonstrating practical impact for production LLS systems under variable network conditions and undershooting scenarios.

Abstract

Low-latency live streaming (LLS) has emerged as a popular web application, with many platforms adopting real-time protocols such as WebRTC to minimize end-to-end latency. However, we observe a counter-intuitive phenomenon: even when the actual encoded bitrate does not fully utilize the available bandwidth, stalling events remain frequent. This insufficient bandwidth utilization arises from the intrinsic temporal variations of real-time video encoding, which cause conventional packet-level congestion control algorithms to misestimate available bandwidth. When a high-bitrate frame is suddenly produced, sending at the wrong rate can either trigger packet loss or increase queueing delay, resulting in playback stalls. To address these issues, we present Camel, a novel frame-level congestion control algorithm (CCA) tailored for LLS. Our insight is to use frame-level network feedback to capture the true network capacity, immune to the irregular sending pattern caused by encoding. Camel comprises three key modules: the Bandwidth and Delay Estimator and the Congestion Detector, which jointly determine the average sending rate, and the Bursting Length Controller, which governs the emission pattern to prevent packet loss. We evaluate Camel on both large-scale real-world deployments and controlled simulations. In the real-world platform with 250M users and 2B sessions across 150+ countries, Camel achieves up to a 70.8% increase in 1080P resolution ratio, a 14.4% increase in media bitrate, and up to a 14.1% reduction in stalling ratio. In simulations under undershooting, shallow buffers, and network jitter, Camel outperforms existing congestion control algorithms, with up to 19.8% higher bitrate, 93.0% lower stalling ratio, and 23.9% improvement in bandwidth estimation accuracy.

Camel: Frame-Level Bandwidth Estimation for Low-Latency Live Streaming under Video Bitrate Undershooting

TL;DR

This paper tackles the problem of bandwidth undershooting in low-latency live streaming, where encoder frame-level bursts distort network feedback and lead to stalling. It introduces Camel, a frame-level congestion control framework comprising a Bandwidth and Delay Estimator, a Congestion Detector, and a Bursting Length Controller, and gates transmission with per-frame bursts to accurately probe capacity while avoiding buffer overflow. The key contributions include (i) a frame-level bandwidth-delay-product estimator, (ii) an inflight-delay gradient-based congestion detector, and (iii) a dynamic bursting controller that adjusts bursts to buffer conditions, all validated by large-scale real-world deployment and extensive simulations. Results show Camel improves bitrate and resolution, reduces stalling, and maintains latency comparable to baselines, demonstrating practical impact for production LLS systems under variable network conditions and undershooting scenarios.

Abstract

Low-latency live streaming (LLS) has emerged as a popular web application, with many platforms adopting real-time protocols such as WebRTC to minimize end-to-end latency. However, we observe a counter-intuitive phenomenon: even when the actual encoded bitrate does not fully utilize the available bandwidth, stalling events remain frequent. This insufficient bandwidth utilization arises from the intrinsic temporal variations of real-time video encoding, which cause conventional packet-level congestion control algorithms to misestimate available bandwidth. When a high-bitrate frame is suddenly produced, sending at the wrong rate can either trigger packet loss or increase queueing delay, resulting in playback stalls. To address these issues, we present Camel, a novel frame-level congestion control algorithm (CCA) tailored for LLS. Our insight is to use frame-level network feedback to capture the true network capacity, immune to the irregular sending pattern caused by encoding. Camel comprises three key modules: the Bandwidth and Delay Estimator and the Congestion Detector, which jointly determine the average sending rate, and the Bursting Length Controller, which governs the emission pattern to prevent packet loss. We evaluate Camel on both large-scale real-world deployments and controlled simulations. In the real-world platform with 250M users and 2B sessions across 150+ countries, Camel achieves up to a 70.8% increase in 1080P resolution ratio, a 14.4% increase in media bitrate, and up to a 14.1% reduction in stalling ratio. In simulations under undershooting, shallow buffers, and network jitter, Camel outperforms existing congestion control algorithms, with up to 19.8% higher bitrate, 93.0% lower stalling ratio, and 23.9% improvement in bandwidth estimation accuracy.
Paper Structure (18 sections, 7 equations, 20 figures, 1 table)

This paper contains 18 sections, 7 equations, 20 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Motivation
  3. Common Undershooting Causes Low QoE
  4. Insight: Frame-Level Burst
  5. Challenges
  6. Overview of Camel
  7. Design of Camel
  8. Bandwidth and Delay Estimator
  9. Congestion Detector
  10. Bursting Length Controller
  11. Evaluations
  12. Large-Scale Real-World A/B Test
  13. Robustness under undershooting
  14. Performance in weak network
  15. Component Analysis
  16. ...and 3 more sections

Figures (20)

  • Figure 1: Without video bitrate undershooting, every packet can detect the network information; When video bitrate undershooting, feedback signals are also influenced by bitrate undershooting, causing estimation distortion; When using frame-level congestion control, some network information can be correctly measured within the individual frame burst.
  • Figure 2: Distribution of stalling ratio and uplink bandwidth utilization among live streams' uploading links. Most live streams only utilize a moderate portion of the estimated uplink bandwidth. But the stalling ratio remains high. Additionally, in traces where stalling occurs, the video bitrate is typically further below the estimated uplink bandwidth.
  • Figure 3: Distribution and impact of undershooting in RTC-based upstream video streams.
  • Figure 4: Lower ETR always leads to lower resolution.
  • Figure 5: Lower ETR always leads to higher stalling ratio.
  • ...and 15 more figures