International Association
for Cryptologic Research

Private information retrieval (PIR) enables a client to retrieve data from a server while preserving the confidentiality of the client's query. When PIR is instantiated with fully homomorphic encryption (FHE), the protocol becomes non-interactive, requiring only a query-answer exchange, and it achieves asymptotically optimal communication and computation complexity. Although several FHE-based PIR protocols have been practically implemented with the desired properties, there has been little detailed comparison among them. As a result, it remains unclear which protocol is most efficient in practice with respect to various aspects such as performance and scalability.

In this paper, we revisit existing protocols by categorizing them into two different structures in order to analyze the advantages and disadvantages of each class in detail, with a focus on practical implementations. Furthermore, we introduce and compare various homomorphic algorithms that can be utilized for query optimization, discussing the strengths and limitations of each. Finally, with the goal of identifying the most efficient protocol in terms of computational cost and memory usage, based on database size. Additionally, we address common misconceptions that may lead to inefficient choices in real-world deployment scenarios and offer the best solutions. Consequently, our analysis and experimental results demonstrate that the less-explored design achieves a 90% reduction in communication cost and an 8× decrease in computational overhead compared to the other one, challenging the common misconception.