[Paper Review] Avoiding Interruptions - QoE Trade-offs in Block-coded Streaming Media Applications

Ali ParandehGheibi, Muriel Médard+2 more|arXiv (Cornell University)|Jan 12, 2010

Cooperative Communication and Network Coding25 references4 citations

TL;DR

This paper analyzes Quality of Experience (QoE) trade-offs in block-coded video streaming over unreliable wireless networks by modeling the receiver buffer as an M/D/1 queue with Poisson arrivals and deterministic playback. It derives asymptotically tight upper and lower bounds on the minimum initial buffering required to achieve a target interruption probability, showing that when the arrival rate slightly exceeds the playback rate, the required buffer remains bounded as file size grows—contrasting with the square-root growth observed when rates are equal.

ABSTRACT

We take an analytical approach to study Quality of user Experience (QoE) for video streaming applications. First, we show that random linear network coding applied to blocks of video frames can significantly simplify the packet requests at the network layer and save resources by avoiding duplicate packet reception. Network coding allows us to model the receiver's buffer as a queue with Poisson arrivals and deterministic departures. We consider the probability of interruption in video playback as well as the number of initially buffered packets (initial waiting time) as the QoE metrics. We characterize the optimal trade-off between these metrics by providing upper and lower bounds on the minimum initial buffer size, required to achieve certain level of interruption probability for different regimes of the system parameters. Our bounds are asymptotically tight as the file size goes to infinity.

Motivation & Objective

• To model the user experience in block-coded video streaming over unreliable wireless channels using queueing theory.
• To quantify the trade-off between initial buffering time and playback interruption probability (buffer underflow).
• To derive tight analytical bounds on the minimum initial buffer size required to achieve a target interruption probability.
• To characterize how this trade-off scales with file size and the relationship between packet arrival rate and playback rate.
• To show that when the arrival rate exceeds the playback rate by a small margin, the required buffer remains bounded as file size increases.

Proposed method

• Models the receiver buffer as an M/D/1 queue with Poisson packet arrivals and deterministic playback departures.
• Applies random linear network coding to simplify block recovery and reduce redundant requests.
• Uses martingale-based bounds and large-deviation techniques to analyze the first passage time of the buffer level to zero.
• Derives upper and lower bounds on the minimum initial buffer size using exponential martingale inequalities and the intermediate value theorem.
• Employs the intermediate value theorem to prove existence of critical buffer thresholds for different parameter regimes.
• Applies Glynn’s and other Poisson tail bounds to quantify the probability of buffer underflow under varying arrival and playback rates.

Experimental results

Research questions

• RQ1 What is the minimum initial buffering required to achieve a target playback interruption probability in block-coded streaming?
• RQ2 How does the required initial buffer scale with file size when the packet arrival rate is slightly above the playback rate?
• RQ3 What happens to the required buffer size when the arrival rate exactly matches the playback rate?
• RQ4 Can tight analytical bounds be derived for the optimal trade-off between initial buffering and interruption probability?
• RQ5 Under what conditions does the required buffer remain bounded as file size grows?

Key findings

• When the packet arrival rate is slightly larger than the playback rate, the minimum initial buffer size required to achieve a given interruption probability remains bounded as file size increases.
• When the arrival rate exactly matches the playback rate, the minimum buffer size grows as the square root of the file size.
• The derived upper and lower bounds on the required initial buffer are asymptotically tight as file size tends to infinity.
• For all parameter regimes, the bounds are derived using martingale methods and exponential inequalities, with the intermediate value theorem used to establish existence of critical buffer thresholds.
• The analysis shows that random linear network coding simplifies block recovery and enables tighter analytical characterization of QoE trade-offs.
• The results are robust across different regimes: when arrival rate is less than, equal to, or slightly greater than playback rate.