Geans: A GPU-accelerated Framework for Efficient End-to-End Anonymized Network Sensing

Anonymized network sensing enables privacy-preserving analysis of large-scale traffic data, supporting critical applications in security auditing, anomaly detection, and infrastructure monitoring. However, existing CPU-based solutions suffer from limited performance due to inefficient IP anonymization and fragmented block-wise matrix construction, hindering their ability to process high-volume traffic in real time. In this paper, we present Geans, a GPU-accelerated framework designed for efficient end-to-end anonymized network sensing. Geans introduces three key optimizations. First, it adopts a lookahead IP anonymization mechanism to eliminate redundant computations inherent in the prefix-preserving encryption algorithm CryptoPAN. Second, Geans employs the global matrix aggregation scheme to accelerate the construction of traffic matrices in coordinate format. Third, it integrates a heuristic hybrid encoding strategy to compress traffic matrices, thereby reducing the overhead of writing them to disk. Experimental results on both real-world and synthetic datasets demonstrate that Geans outperforms the CPU-based reference implementation of the Anonymized Network Sensing Graph Challenge by an average of 4.50× in terms of end-to-end running time improvement and up to 359.01× in terms of traffic matrix construction overhead reduction.