| Title | : | AI-Driven Traffic Management Mechanisms for Enhanced Quality of Experience in Smartphones |
| Speaker | : | Madhan Raj Kanagarathinam (IITM) |
| Details | : | Thu, 27 Jun, 2024 11:00 AM @ By Google Meet |
| Abstract: | : | This thesis presents a suite of solutions designed to manage network traffic intelligently, enhance application performance, and facilitate seamless connectivity transitions in the evolving context of real-time (RT) traffic on smartphones. We implement and evaluate the proposed solutions in live-air using COTS smartphones. When RT video calling and gaming apps race with non-real-time (NRT) traffic, severe degradation in the Quality of Experience (QoE) of RT apps. Hence, an Application Prioritization Engine (APE) framework was proposed and implemented to improve user experience by dynamically managing bandwidth to the different apps. APE helps improve the end-user experience by detecting and prioritizing RT traffic over concurrent best-effort traffic. APE uses the eBPF (extended Berkeley Packet Filter) to control NRT traffic without additional packet processing overhead. Building on APE, an AI-driven framework, Game Stabilizer (GS), was developed to enhance the online mobile gaming experience. Utilizing machine learning to manage non-real-time (NRT) traffic, the GS significantly reduces latency by up to 60%, ensuring smooth and responsive gaming experiences under diverse network conditions. Furthermore, many applications lack proper QoS configurations, leading to suboptimal user experiences. To address this issue, we propose QBOX (Quality Box), an app-agnostic approach that leverages Wi-Fi Access Category to enhance QoS in smartphone applications. QBOX dynamically detects the RT traffic and upgrades its Wi-Fi priority access category from Best-effort (BE) to Voice (VO) or Video (VI), ensuring efficient and reliable data delivery. It improves application latency by up to 24%, resulting in smoother and more responsive experiences.
Poor Wi-Fi conditions and network handovers can lead to sub-par gaming and calling experience. Hence, a framework called ODIN (On-Device Intelligence) was developed. This leverages the QUIC protocol alongside AI to facilitate seamless handovers between cellular and Wi-Fi networks. This AI-driven approach enables predictive analytics to determine the most opportune moments for network transitions, significantly minimizing disruptions and latency in RT applications. Furthermore, network traffic classification is essential to managing and optimizing modern network environments. An eBPF-assisted AI method was developed to provide efficient and accurate smartphone traffic classification. This is based on a power-efficient approach that selectively processes specific app traffic, optimizing resource utilization and enhancing classification efficiency. Integrating eBPF and BPF maps enabled real-time, low-latency traffic classification with improved accuracy and scalability.
Web Conference Link :https://meet.google.com/swh-trof-gvy |
