Abstract
HTTP-based Adaptive Streaming (HAS) has become the standard for media transmission, yet existing Adaptive Bitrate (ABR) algorithms often struggle in multi-client scenarios where users have varying priorities and playback speeds. This paper introduces PREPARE (Playback RatE and Priority Adaptive BitRate Selection), a novel algorithm that incorporates client priority and real-time playback rates into the system model. By utilizing server-side feedback and dynamic buffer management, PREPARE significantly improves fairness and stability. Experimental results using the NS-3 simulator demonstrate that PREPARE reduces unfairness by over 42% compared to state-of-the-art algorithms like ESTC, PANDA, and FDASH, while maintaining high efficiency.
Introduction
The explosion of online video consumption has made HTTP-based Adaptive Streaming (HAS) the backbone of modern digital media. At its core, HAS relies on Adaptive Bitrate (ABR) algorithms to select the optimal video quality based on network conditions. However, traditional ABR methods often fail to account for two critical real-world factors: varying user priorities (e.g., premium vs. free tiers) and non-standard playback rates (e.g., 1.5x or 2.0x speed). These omissions lead to network unfairness, quality instability, and frequent buffer underflows.
The Challenge of Diverse Playback
Standard ABR algorithms typically assume a 1x playback speed. When a user switches to 2x speed, their data consumption rate doubles, yet the algorithm may not adjust quickly enough, leading to a depleted buffer. Furthermore, in shared network environments, "selfish" clients often compete aggressively for bandwidth, causing instability and unfairness. Existing solutions often lack a global view of the network, making it difficult to balance the needs of multiple users with different service level agreements (SLAs).
The PREPARE Algorithm
To address these issues, the PREPARE algorithm introduces a collaborative framework between the client and the server. Unlike purely client-side solutions, PREPARE uses server feedback to help clients estimate their fair share of bandwidth more accurately.
As shown in the system architecture, the PREPARE client consists of a request controller, an ABR switcher, and a player, all interacting with a centralized buffer. The server plays a crucial role by tracking the bitrates and priorities of all active clients, providing a "correction factor" that prevents aggressive clients from destabilizing the link.
Key innovations in PREPARE include:
- Priority-Aware Fairness: The algorithm utilizes a modified Jain Fairness Index that incorporates both priority weights and playback rates.
- Dynamic Buffer Management: Buffer thresholds are dynamically adjusted based on the current playback speed to ensure that high-speed viewers maintain a safe margin against stalls.
- State-Based Control Logic: The bitrate selection process is governed by a robust state machine that ensures smooth transitions between quality levels, preventing jarring jumps in video quality.
Experimental Evaluation
The performance of PREPARE was evaluated using the NS-3 simulator across various network topologies, including wired and Wi-Fi environments. The algorithm was benchmarked against three prominent ABR methods: ESTC, PANDA, and FDASH.
Key Results
- Fairness Improvement: PREPARE achieved a reduction in unfairness ranging from 42% to 67% across different scenarios. By accounting for priority and playback speed, it ensures that bandwidth is distributed according to actual user needs and service tiers.
- Enhanced Stability: The algorithm significantly reduced the frequency of bitrate switches, providing a more consistent viewing experience.
- Resource Efficiency: Unlike some aggressive algorithms that cause "buffer bloat," PREPARE maintains a lean yet safe buffer, optimizing memory usage on the client device.
Conclusion
PREPARE represents a significant step forward in making HAS systems more equitable and robust. By treating playback rate and user priority as first-class variables, it provides a more accurate model of modern streaming behavior. Future work will focus on extending this model to highly dynamic 5G cellular networks and optimizing startup latency for short-form video content.
Summary updated