cid 760d5985e4ac6476c3acad6a2a7ab8f673bb0bc20a83ad3199cf8145235c10c7 · original · via openalex
Title: Prometheus Authors: Vaneet Aggarwal; Emir Halepovic; Jeffrey Pang; Shobha Venkataraman; He Yan Venue: IEEE Workshop on Mobile Computing Systems and Applications (WMCSA) Year: 2014 Canonical: doi:10.1145/2565585.2565600 URL: https://doi.org/10.1145/2565585.2565600 Collector: openalex OA: false ABSTRACT Cellular network operators are now expected to maintain a good Quality of Experience (QoE) for many services beyond circuit-switched voice and messaging. However, new smart-phone "app" services, such as Over The Top (OTT) video delivery, are not under an operator's control. Furthermore, complex interactions between network protocol layers make it challenging for operators to understand how network-level parameters (e.g., inactivity timers, handover thresholds, middle boxes) will influence a specific app's QoE. This paper takes a first step to address these challenges by presenting a novel approach to estimate app QoE using passive network measurements. Our approach uses machine learning to obtain a function that relates passive measurements to an app's QoE. In contrast to previous approaches, our approach does not require any control over app services or domain knowledge about how an app's network traffic relates to QoE. We implemented our approach in Prometheus, a prototype system in a large U.S. cellular operator. We show with anonymous data that Prometheus can measure the QoE of real video-on-demand and VoIP apps with over 80% accuracy, which is close to or exceeds the accuracy of approaches suggested by domain experts.