Quantifying the Privacy-Vs-performance Trade-Offs for Fine-Grained Wireless Network Measurement Data.

Real-world network measurements are critical to building performant and resilient networks at scale. However, access to such data exposes end-users to significant privacy risks; and this is particularly true for wireless network measurements. In this paper, we apply six state-of-the-art differentially private (DP) algorithms, that span data-independent/dependent and workload-aware/unaware classes, to privatize queries from real-world WiFi traces on a large-scale campus network. We analyze utility-vs-privacy trade-offs involved in constructing privatized queries for canonical network resource provisioning tasks. We present the following results: (1) for count and histogram queries, the utility of the Laplacian-algorithm shows comparable (or better) performance compared to more complex data-aware DP algorithms, (2) for a given query-type and DP algorithm, the utility-to-noise trade-off varies for each distinct network metric, and finally, (3) we implement a state-of-the-art DP algorithm for trajectory analysis that reveals that there exist significant challenges in accurately reconstructing privatized network mobility trajectories, for relatively small trajectory lengths, even with relaxed privacy budgets.