Homomorphic Encryption (HE) has emerged as a promising approach for data processing without exposing sensitive information. Despite significant advances, the practical strategies for the integration of HE into widely used database management systems (DBMSs) remain limited due to performance constraints and architectural challenges. This paper explores HE integration strategies within DBMS, focusing on SQL Server, PostgreSQL, and M̃ariaDB. A methodology is proposed to assess the feasibility and performance of multiple HE schemes, including BFV, CKKS, BGV, TFHE, Paillier, and RSA (without padding). The evaluation considers different integration strategies, namely Python-based execution and native C++ extensions, across Windows and Debian environments. Experimental results obtained from four configurations demonstrate that the choice of HE scheme and integration strategy significantly impacts performance. Lattice-based schemes (BFV, CKKS, BGV) provide a balanced trade-off between functionality and efficiency, while TFHE incurs high computational costs due to its bit-level design. Native C++ integrations consistently outperform Python-based approaches, although the latter offer greater flexibility and ease of development. The findings highlight the feasibility of integrating HE into DBMS while emphasizing the importance of selecting appropriate schemes and integration mechanisms to meet application-specific requirements. The proposed evaluation framework provides preliminary insights into the relative behavior of different HE schemes and integration strategies under controlled experimental conditions, supporting future work on privacy-preserving DBMS design.
Jorge et al. (Sat,) studied this question.