Comprehensive Multi-step Prefetching Strategy for Short Video Streaming

Currently, the short video market grows rapidly. As a typical bandwidth-intensive application, short video streaming can easily lead to bandwidth bottlenecks on servers. The prefetching strategy is proposed to improve Quality of Experience (QoE) and reduce bandwidth resource wastage, spending computility for performance optimization. In this work, we reveal that the state-of-the-art prefetching strategy Dashlet fails to maximize the weighted sum of QoE and bandwidth usage, i.e., the utility. To address this, we propose the Comprehensive Multi-step Prefetching Strategy (CMPS) for Short Video Streaming. CMPS computed the prefetching urgency of chunks, and incorporated it with expected utility to comprehensively evaluate multi-step decision sequences, achieving higher QoE and bandwidth utilization. Extensive evaluations show that CMPS improves average utility in all kinds of scenarios by 16.1% compared with the state-of-the-art strategy Dashlet.