Performance Evaluation of Privacy Models for Data Streams on the Edge

Abstract

Recent advances in edge computing enable data stream privacy enforcement directly on resource-constrained devices, reducing latency and the exposure of sensitive information. In this paper, we extend and validate our previously proposed privacy-enforcing framework, which allows high-level privacy policies to be expressed as chains of triggers and transformations, executed at the edge. To assess its practical viability, we conduct a comprehensive performance profiling of multiple privacy models across heterogeneous edge hardware platforms. Six privacy-model chains, ranging from basic face detection to combined face-and-person anonymization, are evaluated across three representative edge devices. Key performance metrics (i.e., execution time, CPU uti- lization, memory usage, and power consumption) are measured to inform optimal placement of privacy transformations. Our evaluation offers critical insights into the effectiveness of the privacy-enforcing framework on resource-constrained devices, thereby guiding practitioners in selecting suitable deployment targets for privacy-preserving data stream analytics on the edge.