Peeking without Spying: Collecting End-Host Measurements to Improve User Experience

Obtaining data directly from end hosts could benefit a broad set of applications, including application performance diagnosis, network troubleshooting, security profiling and even energy-saving policies. However, the data central to the study and development of such tools is fundamentally hard to collect. This situation has arisen both for technical reasons and for psychological ones. We believe that a lack of such end-host data has been a limiting factor in research on these topics. In this paper, we explore the design space of a measurement tool to run on end-hosts. Many tradeoffs arise such as the utility or benefit of collecting a given piece of data, the user’s comfort with that data being monitored, and the overhead incurred via the technique used to collect the data. We combine both qualitative and quantitative elements to drive our own decisions for designing such a tool: we conduct a user survey to understand user comfort issues, and we carry out evaluations of the overhead and utility of different techniques for collecting various data. We build an instance of such a data collection tool which exposes these tradeoffs in an explicit way and describe its design.