Private Information Retrieval from Transversal Designs

International audiencePrivate information retrieval (PIR) protocols allow a user to retrieve entries of a database without revealing the index of the desired item. Information-theoretical privacy can be achieved by the use of several servers and specific retrieval algorithms. Most of known PIR protocols focus on decreasing the number of bits exchanged between the client and the server(s) during the retrieval process. On another side, Fazeli et. al. introduced so-called PIR codes in order to reduce the storage overhead on the servers. However, only a few works address the issue of the computation complexity of the servers.In this paper, we show that a specific encoding of the database provides PIR protocols with reasonable communication complexity, low storage overhead and optimal computational complexity for the servers. This encoding is based on incidence matrices of transversal designs, from which a natural and efficient recovering algorithm is derived. We also present instances of our construction, making use of finite geometries and orthogonal arrays, and we finally give a generalisation of our main construction for resisting collusions of servers