Video Streaming: Impact of Rate Adaptation, Multipath, and Link Preference
Video streaming today accounts for up to 55% of mobile traffic and it is expected to account for 75% by 2023. This increase has forced service providers to enhance their infrastructures to support high-quality video streaming. Despite these efforts, users frequently experience low Quality-of-Experience (QoE) metrics such as choppy videos and playback stalls. In this work, we explore streaming algorithms for videos encoded using Advanced Video Coding (AVC) and Scalable Video Coding (SVC) schemes. We study video streaming problems for the single path and multi-path with and without considering MPTCP. In particular, we formulate the video streaming as an optimization problem whose objective is to optimizes a novel QoE metric that minimizes the video’s stall/skip duration as the first priority and maintains a trade-off between maximizing the playback rate of every chunk and ensuring fairness among all the video chunks. The objective function is constrained by a set of constraints such as the chunks’ deadlines, the available bandwidth, and the playback buffer size. We include both real and non-real time video streaming. In real time streaming, if a chunk can’t meet its deadline it will be skipped while in the non-real time scenario, the video freezes until the chunk is fully downloaded. We extend our formulation to the multipath streaming in which multiple interfaces (e.g WiFi and LTE) can be used to download the video. However, since some links could be less preferable than others (e.g LTE is more expensive and less energy efficient), we impose link preference constraint in our formulation. Moreover, we consider multipath streaming with and without MPTCP. For all variants of the problem, we develop a set of novel and polynomial complexity algorithms to solve the proposed problems. We show that the proposed algorithms are optimal for single path problem. Moreover, we show the proposed algorithms of link preference aware streaming over multipath are also optimal for some special cases. The results give a class of discrete non-convex problems that are optimally solvable in polynomial time. Further, we propose online versions of the proposed algorithms where several challenges including handling bandwidth prediction errors, variable bit-rates, and short bandwidth prediction duration are addressed. Extensive simulations, emulations, and real implementations with real AVC/SVC encoded videos, real bandwidth traces of public datasets reveal the robustness of our scheme and demonstrate its significant performance improvement as compared to the state-of-the-art AVC/SVC video streaming algorithms.
Bell, Purdue University.
Electrical engineering|Computer science|Web Studies
Off-Campus Purdue Users:
To access this dissertation, please log in to our