Are you on Mobile or Desktop? On the Impact of End-User Device on Web QoE Inference from Encrypted Traffic

Abstract

Web browsing is one of the key applications of the Internet, if not the most important one. We address the problem of Web Quality-of-Experience (QoE) monitoring from the ISP perspective, relying on in-network, passive measurements. As a proxy to Web QoE, we focus on the analysis of the wellknown SpeedIndex (SI) metric. Given the lack of applicationlevel-data visibility introduced by the wide adoption of end-toend encryption, we resort to machine-learning models to infer the SI and the QoE level of individual web-page loading sessions, using as input only packet-and flow-level data. In this paper, we study the impact of different end-user device types (e.g., smartphone, desktop, tablet) on the performance of such models. Empirical evaluations on a large, multi-device, heterogeneous corpus of Web-QoE measurements for the most popular websites demonstrate that the proposed solution can infer the SI as well as estimate QoE ranges with high accuracy, using either packetlevel or flow-level measurements. In addition, we show that the device type adds a strong bias to the feasibility of these Web-QoE models, putting into question the applicability of previously conceived approaches on single-device measurements. To improve the state of the art, we conceive cross-device generalizable models operating at both packet and flow levels, offering a feasible solution for Web-QoE monitoring in operational, multi-device networks. To the best of our knowledge, this is the first study tackling the analysis of Web QoE from encrypted network traffic in multi-device scenarios.

Mobile Web and App QoE Monitoring for ISPs - from Encrypted Traffic to Speed Index through Machine Learning

“…In [12], [13], the impact of network quality fluctuations and outages on user Web QoE was studied. Other components besides network quality influence Web QoE, linked to the specific web page content -usability [14], aesthetics [15], etc., as well as device type -desktop, smartphone, tablets [1], [16]. Important to our study, these papers show that smartphones and tablets have their own characteristics, not only regarding screen sizes but also in terms of content rendering and web designs.…”

“…Recent work [3], [4] takes a step further to directly infer the SI metric, using ML techniques mapping network (encrypted) traffic features to SI. In [1], we complemented [3], [4] and showed that the performance of previously proposed ML models built on desktop measurements does not generalize to mobile browsing, resulting in poor Web QoE inference performance when applied to web browsing in smartphones.…”

“…By using page load controlled experiments, where network data is simultaneously collected with ground-truth Web QoE metrics such as SI, AFT, etc., their study built a labeled dataset and trained supervised ML models to infer these QoE-related metrics from network traffic features, computed on the stream of collected bytes. In [1], we showed that the performance of these ML models built on desktop measurements does not generalize to mobile browsing, resulting in poor Web QoE inference performance when applied to web browsing in smartphones. Finally, in [2] we have recently introduced a novel measurement framework to measure loading times in apps, which can be used to extend the analysis of Web QoE to those mobile apps where perceived waiting times determine the experience of the end-user -e.g., different from video streaming, where re-buffering and video quality are the key to user experience.…”

“…In this paper we follow a similar approach to [1], [2], focusing the analysis on Web and App QoE in mobile devices exclusively. Web and App QoE is paramount for mobile ISPs, as the lion's share of Internet-access devices is today smartphones, with nearly three quarters of the world population using just their smartphones to access the Internet by 2025 [5].…”

Improving Web QoE Monitoring for Encrypted Network Traffic through Time Series Modeling

“…They evaluate their model on different browsers and under different conditions, and outline the problem of model generalization, introduced by the highly variable and dynamic nature of the Web. Finally, in [11] we have recently investigated the same problem, extending previous work to the mobile devices scenario. These papers are the closest to this work.…”