Network-Aware Optimization of Video Chunking for MEC-Assisted Video-on-Demand Streaming

While existing video chunking (or segmentation) methods are at best content-aware, they often fail to consider the diverse characteristics of wireless networks. Variations in bandwidth, latency, and packet loss rates experienced by clients in different wireless networks, however, render network-agnostic video chunking at the content server suboptimal for wireless video-on-demand streaming. In this work, we present QAOCS for network-aware chunking optimization that leverages the role of the multi-access edge computing (MEC) server to learn the best chunking strategy for the target network. With a dualclip proximal policy optimization (Dual-PPO) algorithm, QAOCS simultaneously adjusts the video quality and segment duration to adapt to varying network conditions and video content for chunking optimization. We evaluate QAOCS using diverse video datasets and real-world network traces and we find that QAOCS consistently out-performs baseline chunking methods in terms of video quality, stall probability, and quality switch ratio, thus substantiating the benefits of network-aware optimization of video chunking for wireless video-on-demand streaming.