Efficient HTTP-based streaming using Scalable Video Coding

Yago Sánchez de la Fuente, Thomas Schierl, Cornelius Hellge, Thomas Wiegand, Dohy Hong, Danny De Vleeschauwer, Werner Van Leekwijck, Yannick Le LouédecSignal Process. Image Commun 2012 · DOI

TL;DR

This paper explores how Scalable Video Coding (SVC) can significantly improve the efficiency of HTTP-based live video streaming. By leveraging SVC's ability to adapt video quality within a single stream, the authors demonstrate enhanced caching efficiency, faster adaptation to network changes, and reduced playout interruptions compared to traditional multi-rate AVC approaches.

What Problem Does This Paper Address?

The paper tackles the growing challenge of delivering high-quality live video streams over HTTP, a protocol increasingly favored for its ability to leverage existing network infrastructure (like caches) and bypass firewall/NAT issues common with UDP-based streaming. While HTTP streaming offers advantages, achieving low-latency, smooth playback, especially in dynamic network conditions, remains difficult. Traditional methods often rely on multiple, independently encoded versions of a video (Multi-Rate or MR), which can be inefficient for caching and slow to adapt. The authors investigate how Scalable Video Coding (SVC), which encodes video into a base layer and enhancement layers, can overcome these limitations for live streaming scenarios.

Methods

The core of the paper lies in comparing HTTP-based streaming using Scalable Video Coding (SVC) against traditional Multi-Rate (MR) streaming using H.264/AVC. The authors utilize the MPEG Dynamic Adaptive Streaming over HTTP (DASH) standard as the framework for their analysis.

Key methodological aspects include:

  • DASH Architecture: The paper describes the DASH architecture, including the Media Presentation Description (MPD) for describing available media representations and segments for chunked delivery.
  • SVC Integration: They detail how SVC's layered structure can be mapped to DASH representations, allowing clients to request different combinations of layers (base + enhancement) to match network conditions.
  • Receiver Behavior Analysis: The paper outlines the behavior of a DASH client, focusing on the scheduler/downloading controller responsible for selecting and requesting segments based on estimated throughput.
  • Priority-based Media Delivery (PMD): A key contribution is the application of a PMD algorithm, which prioritizes the download of essential SVC layers. This algorithm is designed to adapt more effectively to network variations.
  • Caching Efficiency Evaluation: The authors analyze the impact of SVC on HTTP cache performance, measuring cache-hit ratios under various conditions.
  • Rate Adaptation and Live Streaming Simulation: Extensive simulations are conducted to evaluate rate adaptation strategies for live streaming, considering factors like playout buffer size, network throughput variations, and latency. They compare SVC-PMD with MR-Streaming under different network scenarios.
  • Playout Interruption Probability Calculation: A Markov model is used to quantify the probability of playout interruptions for both MR-Streaming and SVC-PMD under varying network conditions.

Datasets

Not discussed in the paper. The authors mention using simulations based on real data statistics from a VoD service for caching performance evaluation and simulated network conditions (e.g., packet loss, throughput variations) for live streaming scenarios.

Key Insights and Ideas

  • SVC for Caching Efficiency: SVC significantly boosts HTTP cache efficiency. By removing redundancy between layers and allowing clients to request common base layers, caches can store more unique content or serve more requests from a single cached item, reducing server load and bandwidth.
  • Faster Adaptation with SVC: SVC, particularly when combined with a priority-based delivery mechanism like PMD, allows for much faster and more granular adaptation to network fluctuations compared to MR-Streaming. This is crucial for live streaming where buffer sizes are limited.
  • Reduced Playout Interruptions: SVC-PMD demonstrates a substantially lower probability of playout interruptions (e.g., freezing) compared to MR-Streaming, especially under challenging network conditions. This is achieved by prioritizing essential layers and adapting quality more gracefully.
  • Balancing Quality and Latency: For live streaming, there's a critical trade-off between playout delay and adaptation responsiveness. SVC offers a better balance, allowing for higher average video rates while managing quality shifts more smoothly than MR-AVC, which might force users to significantly lower their quality settings to avoid interruptions.
  • SVC's Advantage in Low-Buffer Scenarios: The benefits of SVC are particularly pronounced in scenarios with small playout buffers, which are typical for low-latency live streaming. SVC's ability to adapt quality by selectively downloading enhancement layers allows for better utilization of available bandwidth.

Key Results

  • Caching: SVC-based VoD services show a significantly higher average cache-hit-ratio (around 20% higher) compared to MR-VoD services.
  • Live Streaming Performance: SVC-PMD consistently achieves higher average download rates than MR-Streaming across various playout buffer sizes and network conditions. For example, in one scenario, SVC-PMD achieved an average download rate of 1.95 Mbps compared to MRS-80% at 1.8 Mbps for a 2-second pre-buffer.
  • Playout Interruption Probability: SVC-PMD significantly reduces the probability of playout interruptions. In a simulated scenario, SVC-PMD had a playout interruption probability of 0.19%, while MR-Streaming had 0.5%.
  • Adaptation Efficiency: SVC allows for adaptation to higher qualities more frequently, even for already buffered data, whereas MR-Streaming often requires switching to lower qualities for extended periods.

Limitations and Future Work

  • SVC Overhead: The paper acknowledges that SVC introduces a slight overhead (around 10%) compared to AVC for the same quality level. However, they argue this is offset by the benefits in adaptation and caching.
  • Complexity of Adaptation: While SVC improves adaptation, the paper notes that precisely estimating future network performance and selecting the optimal representation remains a difficult task, especially for short segments in live streaming.
  • Future Work: The authors suggest further research into more sophisticated scheduling mechanisms and rate estimation techniques to further optimize SVC-based streaming. They also highlight the need for a better understanding of the trade-offs between playout-buffer length, startup delay, and receiver capabilities.

Summary updated

BibTeX
@article{fuente2012efficient,
  author = {Yago Sánchez de la Fuente and Thomas Schierl and Cornelius Hellge and Thomas Wiegand and Dohy Hong and Danny De Vleeschauwer and Werner Van Leekwijck and Yannick Le Louédec},
  title = {{Efficient HTTP-based streaming using Scalable Video Coding}},
  journal = {Signal Process. Image Commun.},
  year = {2012},
  doi = {10.1016/j.image.2011.10.002},
}
Figure