DiagSys: a dataset collection framework for root cause analysis from end-devices

For Internet operators, on-line service providers and end-users, representative operational measurements are crucial to monitor and diagnose the performance of networks and on-line services. While numerous approaches have been proposed to measure performance, only a few works fully adopt an end-user perspective by taking measurements from within web browsers. In this paper we propose and describe DiagSys, a novel crowd-sourced data collection system designed to monitor the performance of network- and web-services from a range of diverse viewpoints. DiagSys leverages the web browsers running on end-user devices to probe dedicated remote measurement points and third-party web services. It uses a JavaScript snippet embedded within webpages and/or a dedicated browser extension to this end, while staying compatible with recent browser capabilities and security restrictions. We also present interesting case studies based on the data already collected in our DiagSys deployment.