Smart phones and tablets are fast becoming the medium of choice for Internet media access, with streaming video being one of the most popular applications. The content to be streamed could be sourced from a number of different servers or access points, some of which could themselves be peer devices. The unreliability of the wireless medium, coupled with heterogeneity of servers implies that the arrival rate of packets would follow some stochastic process. In order to account for this unreliability, it is customary for streaming applications to buffer some packets before playback begins. If the playback buffer is empty at any point, users experience a loss in quality due to interruption in playback. The main goal in this paper is to define and understand the fundamental trade-offs among Quality of user Experience (QoE) metrics in the context of video streaming on mobile devices.
The user experience metrics the authors consider in this work are the initial buffering (delay) before the media playback, and the probability of experiencing an interruption throughout media playback. Interruption probability captures the reliability of media playback. Such metrics best capture the user experience for most of media streaming applications e.g. Internet video, TV, Video on Demand (VoD), where the user may tolerate some initial delay, but expects a smooth sequential playback.
In this paper, the authors focus on the scenario of streaming pre-prepared content over unreliable wireless channels and study the trade-off among user experience metrics. In particular, the authors analyze the transient effect, i.e., the first time that video playback is interrupted as a function of the initial amount of buffering. For infinite file sizes, the techniques used to compute the optimal trade-off curves are similar to those used in the literature of Ruin Theory, which studies insurer’s vulnerability to insolvency. For the finite file size case, the authors characterize probabilities of crossing a time-varying threshold for which such methods are not effective.
The basic model of the receiver’s buffer is a queue with Poisson arrivals and deterministic departures. The authors provide an analytical characterization of the QoE metric trade-offs for finite and infinite file sizes.
Their analysis shows that for arrival rates slightly larger than the playback rate (positive drift case), the minimum initial buffering required to achieve a certain level of interruption probability remains bounded as the file size grows and grows logarithmically in a given interruption probability. This result is analogous to information theoretic error exponent results relating the error probability of a code to the block length of that code. Furthermore, for finite file sizes, the authors show that the minimum initial buffering grows as the square root of the file size when the arrival rate and the play rate match (zero drift case). They also present an analytical characterization of the QoE trade-offs even in the negative drift case, which is applicable in low bandwidth or high-resolution scenarios.
In order to address more general scenarios, the authors consider channels with memory, which can be modeled using Markovian arrival processes. They again establish the above relation and characterize the optimal trade-off curves for the infinite file size case in such Markovian environments.