Performance Analysis of Homomorphically-Encrypted Heterogeneous Multi-Layer Graph Databases
John Long, Ram Dantu, Jacob White · 2023
Encrypted databases allow for information retrieval on encrypted data; however, the security and information visibility vary for these systems depending on the encryption scheme. Homomorphic encryption allows for computation over encrypted ciphertexts with minimal and static information disclosure, unlike searchable encryption which reveals additional information for everv querv as part of the query process, We saw a gap in the current literature regarding Homomorphically encrypted databases for heterogeneous multi-layer networks. In light of this, we mathematically defined our heterogeneous multi-layer networks and then implemented a HE relational database and a HE network database using TFHE. Our implementation can support arbitrary queries. We measured the performance of both systems on synthetic data and two realworld data sets. The results show the potential of these systems currently for low-volume queries. Future large-scale feasibility will require additional work in accelerating HE operations. Our measurements with select queries on the HE network database against increasing amount of data show the expected quadratic increase in runtime. We also show that our parallel system scales as expected by observing the runtime decreasing linearly with the number of parallel process workers. The above-described HE databases will allow for secure third-party computation without the worry of information disclosure which is useful for data cooperatives and computational outsourcing in the cloud.