On the performance of certain Private Set Intersection protocols. (And some remarks on the recent paper by Huang et al. in NDSS'12)

Private Set Intersection (PSI) is a useful cryptographic primitive that allows two parties (client and server) to interact based on their respective (private) input sets, in such a way that client obtains nothing other than the set intersection, while server learns nothing beyond client set size. This paper considers one PSI construct from [DT10] and reports on its optimized implementation and performance evaluation. Several key implementation choices that significantly impact real-life performance are identified and a comprehensive experimental analysis (including micro-benchmarking, with various input sizes) is presented. Finally, it is shown that our optimized implementation of this RSA-OPRF-based PSI protocol markedly outperforms the one presented in [HEK12].