Performance Comparison of Transcoding and Bitrate-Aware Caching in Adaptive Video Streaming

Abstract

Video traffic is growing in dominance in today's Internet, prompting new challenges in timely delivery of video content. As Dynamic Adaptive Streaming (DAS) is becoming the de facto paradigm for video delivery, there is growing evidence on how caching improves users' Quality of Experience (QoE) in DAS. However, there is no consensus on how to maximize the utilization of in-network caching resources. Specifically, there are conflicting proposals on the impact of caching based on its distance (in hops) from the network edge. That is, contrasting ubiquitous network-wide caching to edge-caching. Supporters of the ubiquitous caching paradigm propose bitrate-aware caching schemes for optimizing video streaming, while counter-proposals suggest that edge-caching only the highest bitrate with online transcoding, may offer superior performance to ubiquitous caching. In this paper, we answer a contentious question: Can transcoding at the edge outperform bitrate-aware ubiquitous caching for DAS? We devise an extensive simulation environment using NS-3 to contrast both paradigms, experimenting with different bandwidth fluctuation patterns, under the FESTIVE user-based bitrate adaptation protocol. Caching performance was evaluated under five established QoE metrics, gauging delivered video quality, playback freezing and bitrate oscillation. We further assume zero processing delay for online transcoding at the network edge, to contrast to an upper bound performance from the edge-caching paradigm. Our experiments demonstrate that neither transcoding nor bitrate-aware caching offer a silver bullet for all cases. We present our insights on networking scenarios where each model would dominate in performance, and present our concluding remarks on their development.