Call Me By My Name: Simple, Practical Private Information Retrieval for Keyword Queries

We introduce ChalametPIR: a single-server Private Information Retrieval (PIR) scheme supporting fast, low-bandwidth keyword queries, with a conceptually very simple design. In particular, we develop a generic framework for converting PIR schemes for index queries over flat arrays (based on the Learning With Errors problem) into keyword PIR. This involves representing a key-value map using any probabilistic filter that permits reconstruction of elements from inclusion queries (e.g. Cuckoo filters). In particular, we make use of recently developed Binary Fuse filters to construct ChalametPIR, with minimal efficiency blow-up compared with state-of-the-art index-based schemes (all costs bounded by a factor of (≤ 1.08)). Furthermore, we show that ChalametPIR achieves runtimes and financial costs that are factors of between (6x)-(11x) and (3.75x)-(11.4x) more efficient, respectively, than state-of-the-art keyword PIR approaches, for varying database configurations. Bandwidth costs are additionally reduced or remain competitive, depending on the configuration. Finally, we believe that our application of Binary Fuse filters can have independent value towards developing efficient variants of related cryptographic primitives (e.g. private set intersection), that already benefit from using less efficient filter constructions.