Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Prink: $k_s$-Anonymization for Streaming Data in Apache Flink

Philip Groneberg, Saskia Nuñez von Voigt, Thomas Janke, Louis Loechel, Karl Wolf, Elias Grünewald, Frank Pallas

TL;DR

This paper addresses privacy in streaming data by introducing Prink, a semantics-aware $k_s$-anonymization framework that extends CASTLE to non-numerical attributes and integrates with Apache Flink. It develops domain-specific generalization via Domain Generalization Hierarchies and introduces information-loss metrics (GLM, NCP, PRL) along with per-attribute weights and multi-attribute $\ell$-diversity for robust utility preservation. The architecture leverages dynamic rule broadcasting and a CastleFunction to enable scalable, distributed anonymization with real-time output, while maintaining privacy guarantees. Empirical evaluation on the ASHRAE dataset demonstrates feasible latency and acceptable information loss, validating Prink’s practicality for enterprise-grade streaming pipelines and highlighting trade-offs among anonymity, utility, and performance. The work advances privacy-preserving streaming analytics by enabling non-numerical data handling, semantic generalization, and runtime configurability in a Flink-based deployment, with open-source potential for broad adoption.

Abstract

In this paper, we present Prink, a novel and practically applicable concept and fully implemented prototype for ks-anonymizing data streams in real-world application architectures. Building upon the pre-existing, yet rudimentary CASTLE scheme, Prink for the first time introduces semantics-aware ks-anonymization of non-numerical (such as categorical or hierarchically generalizable) streaming data in a information loss-optimized manner. In addition, it provides native integration into Apache Flink, one of the prevailing frameworks for enterprise-grade stream data processing in numerous application domains. Our contributions excel the previously established state of the art for the privacy guarantee-providing anonymization of streaming data in that they 1) allow to include non-numerical data in the anonymization process, 2) provide discrete datapoints instead of aggregates, thereby facilitating flexible data use, 3) are applicable in real-world system contexts with minimal integration efforts, and 4) are experimentally proven to raise acceptable performance overheads and information loss in realistic settings. With these characteristics, Prink provides an anonymization approach which is practically feasible for a broad variety of real-world, enterprise-grade stream processing applications and environments.

Prink: $k_s$-Anonymization for Streaming Data in Apache Flink

TL;DR

This paper addresses privacy in streaming data by introducing Prink, a semantics-aware -anonymization framework that extends CASTLE to non-numerical attributes and integrates with Apache Flink. It develops domain-specific generalization via Domain Generalization Hierarchies and introduces information-loss metrics (GLM, NCP, PRL) along with per-attribute weights and multi-attribute -diversity for robust utility preservation. The architecture leverages dynamic rule broadcasting and a CastleFunction to enable scalable, distributed anonymization with real-time output, while maintaining privacy guarantees. Empirical evaluation on the ASHRAE dataset demonstrates feasible latency and acceptable information loss, validating Prink’s practicality for enterprise-grade streaming pipelines and highlighting trade-offs among anonymity, utility, and performance. The work advances privacy-preserving streaming analytics by enabling non-numerical data handling, semantic generalization, and runtime configurability in a Flink-based deployment, with open-source potential for broad adoption.

Abstract

In this paper, we present Prink, a novel and practically applicable concept and fully implemented prototype for ks-anonymizing data streams in real-world application architectures. Building upon the pre-existing, yet rudimentary CASTLE scheme, Prink for the first time introduces semantics-aware ks-anonymization of non-numerical (such as categorical or hierarchically generalizable) streaming data in a information loss-optimized manner. In addition, it provides native integration into Apache Flink, one of the prevailing frameworks for enterprise-grade stream data processing in numerous application domains. Our contributions excel the previously established state of the art for the privacy guarantee-providing anonymization of streaming data in that they 1) allow to include non-numerical data in the anonymization process, 2) provide discrete datapoints instead of aggregates, thereby facilitating flexible data use, 3) are applicable in real-world system contexts with minimal integration efforts, and 4) are experimentally proven to raise acceptable performance overheads and information loss in realistic settings. With these characteristics, Prink provides an anonymization approach which is practically feasible for a broad variety of real-world, enterprise-grade stream processing applications and environments.
Paper Structure (29 sections, 3 equations, 5 figures, 1 table)

This paper contains 29 sections, 3 equations, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Anonymization Techniques and Anonymity Guarantees
  4. Anonymity Guarantees for Dynamic and Streaming Data
  5. Apache Flink
  6. Data Cluster Anonymization -- CASTLE
  7. Proposal: Prink
  8. Rule Broadcasting and CastleFunction
  9. Cluster Logic
  10. Semantic Generalization Metrics
  11. Attribute Weights in Information Loss Calculation
  12. Design Choices and Extensibility
  13. Evaluation
  14. Influencing Factors
  15. Parameters
  16. ...and 14 more sections

Figures (5)

  • Figure 1: Prink Architecture in the context of Apache flink data streaming infrastructure.
  • Figure 2: Benchmarking experiment architecture diagram
  • Figure 3: Evaluation of accuracy through the information loss of attributes.
  • Figure 4: Evaluation of end-to-end latency, i.e., total time a data tuple spends within Prink
  • Figure 5: Evaluation of internal processing times, focusing on key functions: bestSelection, delayConstraint, and waitTime (the interval between these operations).