eBandit: Kernel-Driven Reinforcement Learning for Adaptive Video Streaming
Summary: arXiv:2604.08791v1 Announce Type: cross
Abstract: User-space Adaptive Bitrate (ABR) algorithms cannot see the transport layer signals that matter most, such as minimum RTT and instantaneous delivery rate, and they respond to network changes only after damage has already propagated to the playout buffer. We present eBandit, a framework that relocates both network monitoring and ABR algorithm selection into the Linux kernel using eBPF. A lightweight epsilon-greedy Multi-Armed Bandit (MAB) runs inside a sockops program, evaluating three ABR heuristics against a reward derived from live TCP metrics. On an adversarial synthetic trace eBandit achieves $416.3 \pm 4.9$ cumulative QoE, outperforming the best static heuristic by $7.2\%$. On 42 real-world sessions eBandit achieves a mean QoE per chunk of $1.241$, the highest across all policies, demonstrating that kernel-resident bandit learning transfers to heterogeneous mobile conditions.
Introduction
With the exponential growth of online video streaming, ensuring a smooth and high-quality user experience has become increasingly challenging. Traditional Adaptive Bitrate (ABR) algorithms operate in user space, limiting their ability to respond dynamically to network conditions. This often leads to buffering delays and a subpar viewing experience.
Overview of eBandit
eBandit is a novel framework designed to enhance the responsiveness of ABR algorithms by leveraging the Linux kernel. It introduces the following key components:
- eBPF Integration: eBandit utilizes eBPF (Extended Berkeley Packet Filter) to enable direct monitoring of network conditions from within the kernel.
- Multi-Armed Bandit Approach: A lightweight epsilon-greedy Multi-Armed Bandit (MAB) algorithm is employed to evaluate multiple ABR strategies, optimizing for the best performance based on real-time metrics.
- Real-Time Metrics: By accessing live TCP metrics, eBandit can make informed decisions to adapt the bitrate in response to current network conditions.
Performance Metrics
The performance of eBandit has been rigorously tested in both synthetic and real-world environments:
- In adversarial synthetic scenarios, eBandit achieved a cumulative Quality of Experience (QoE) score of 416.3 ± 4.9, surpassing the best static heuristic by 7.2%.
- In real-world testing across 42 sessions, eBandit recorded a mean QoE per chunk of 1.241, the highest observed among all tested policies.
Conclusion
eBandit represents a significant advancement in adaptive video streaming technology. By integrating ABR algorithm selection and network monitoring directly into the Linux kernel, it allows for a more responsive and efficient streaming experience. As streaming demands continue to evolve, frameworks like eBandit could pave the way for improved user experiences across diverse network conditions.