From Network Traffic Measurements to QoE for Internet Video

Abstract

Video streaming is a dominant contributor to the global Internet traffic. Consequently, monitoring video streaming Quality of Experience (QoE) is of paramount importance to network providers. Monitoring QoE of video is a challenge as most of the video traffic of today is encrypted. In this paper, we consider this challenge and present an approach based on controlled experimentation and machine learning to estimate QoE from encrypted video traces using network level measurements only. We consider a case of YouTube and play out a wide range of videos under realistic network conditions to build ML models (classification and regression) that predict the subjective MOS (Mean Opinion Score) based on the ITU P.1203 model along with the QoE metrics of startup delay, quality (spatial resolution) of playout and quality variations, and this is using only the underlying network Quality of Service (QoS) features. We comprehensively evaluate our approach with different sets of input network features and output QoE metrics. Overall, our classification models predict the QoE metrics and the ITU MOS with an accuracy of 63-90% while the regression models show low error; the ITU MOS (1-5) and the startup delay (in seconds) are predicted with a root mean square error of 0.33 and 2.66 respectively.

Estimating Video Streaming QoE in the 5G Architecture Using Machine Learning

“…Making use of those relations to classify video QoE from encrypted network traffic is recently widely discussed and numerous works have been published in the past few years addressing this topic. The works presented in [7,11,13,16,22] study the capabilities of different ML-based algorithms to classify values for QoE influence factors (QoE-IFs), such as stallings or video quality. Compared to previous works, we do not classify objective QoE-IFs, but estimate actual QoE values.…”

“…We use the QoE on MOS scale as computed by the standardized ITU-T P.1203 model [20], which uses a set of objective quality assessment modules to measure the subjective application quality perceived by a user. It is used in the context of classifying QoE from encrypted network traffic in [17] and [11].…”

On Accounting for Screen Resolution in Adaptive Video Streaming: A QoE-Driven Bandwidth Sharing Framework

“…To build these QoE functions, we rely on two publicly available datasets that map the QoS to the QoE. The first dataset is built by controlled experiments and links the network throughput to the QoE modeled according to ITU P.1203 recommendation [10], while the second dataset is based on the work of the Video Quality Experts Group (VQEG) [2] and maps the video bit rate to the MOS. On these datasets, we apply curve fitting methods (e.g., non-linear least squares) with the canonical function given in Equation 1to build our target QoE function taking each time as input the network throughput and the video bit rate respectively.…”

“…1) Network throughput to QoE: The dataset for this model is built by controlled experiments in the lab [10]. It consists of 100k unique YouTube video playouts under different tracedriven emulated network conditions.…”

“…In this paper, we study the problem of resource allocation for several video flows sharing the same bottleneck, and develop an optimization framework to maximize the overall QoE taking into consideration terminal display capabilities. We use two data sets to fit QoE functions that link throughput [10] or video bit rate [2] to a QoE level. We formulate a nonlinear optimization problem and propose heuristics to solve it, then we conduct numerical and network simulations under different system settings to validate the interest of a QoEdriven bandwidth allocation approach.…”