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

Abstract

Web browsing is one of the key applications of the Internet. In this paper, we address the problem of mobile Web and App QoE monitoring from the Internet Service Provider (ISP) perspective, relying on in-network, passive measurements. Our study targets the analysis of Web and App QoE in mobile devices, including mobile browsing in smartphones and tablets, as well as mobile apps. As a proxy to Web QoE, we focus on the analysis of the well-known Speed Index (SI) metric. Given the wide adoption of end-to-end encryption, we resort to machinelearning models to infer the SI of individual web page and app loading sessions, using as input only packet level data. Empirical evaluations on a large, multi mobile-device corpus of Web and App QoE measurements for top popular websites and selected apps demonstrate that the proposed solution can properly infer the SI from in-network, encrypted-traffic measurements, relying on learning-based models. Our study also reveals relevant network and web page content characteristics impacting Web QoE in mobile devices, providing a complete overview on the mobile Web and App QoE assessment problem.

Fingerprinting Web Pages and Smartphone Apps from Encrypted Network Traffic with WebScanner

“…To construct labeled datasets for both web browsing and apps, we use a measurement testbed developed in [2], based on WebPageTest (WPT) [14] and Appium libraries [15]. This testbed offers network emulation capabilities to simulate different access technologies in terms of network performance (e.g., latency, throughput, packet-loss), which adds variability and heterogeneity to the datasets.…”

Marina: Realizing ML-Driven Real-Time Network Traffic Monitoring at Terabit Scale

“…This has given rise to a wider adoption of artificial intelligence and machine learning (ML) technology to improve traffic monitoring at scale. These approaches have been successfully applied to flow-level data [4], [5], [6] and packet-level (time series) data [7], [8], [9], even for the case of encrypted network traffic [10], [11], [12], [13]. Still, running ML-driven monitoring applications in real-time and at line rate is challenging.…”

Not all Web Pages are Born the Same Content Tailored Learning for Web QoE Inference

“…In this paper we build on our previous work [4], [5] to address this personalization approach, firstly by discovering classes of web pages sharing similar content-related properties through clustering techniques, and then by training per-class supervised learning models to infer the SI of individual web page loading sessions. We take the well-known SI metric as a proxy to Web QoE, based on the rich literature on Web QoE analysis [4]- [9].…”