Case Studies in Differential Privacy for Computer Networking Research

Abstract

We conduct two case studies on the use of differential privacy in computer networking research: private analysis of 1) Internet performance measurements from the Measuring Broadband America dataset and 2) flow-based network traces from the NF-UNSW-NB15 Netflow dataset. We survey two open-source tools for this analysis, Ektelo and Tumult Analytics, and evaluate the experience for a data practitioner at each step of designing a differentially private statistical release with each of these tools. In Ektelo, we asses the privacy versus utility trade-off for 5 algorithms (Identity, H2, HB, GreedyH, and DAWA) and provide examples of context-specific utility functions and post-processing techniques for the Internet measurement data.