Multi-server private information retrieval over unsynchronized databases

Giulia Fanti, Kannan Ramchandran

Abstract: Search histories contain detailed and sensitive information about people. Private information retrieval (PIR) aims to hide search histories from service providers by allowing a user to retrieve the wth record of a database without revealing w to the server. However, most known PIR schemes are either very inefficient (and therefore unlikely to gain traction in a practical sense) or reliant on some restrictive assumptions. In this paper, we consider an efficient class of schemes called multi-server PIR. Multi-server PIR assumes that the client communicates with multiple, non-colluding servers, each possessing an identical copy of the database. The current literature does not address the assumption that servers store perfectly-synchronized databases. This seems implausible, especially if servers are not meant to collude. We propose the first multiserver PIR scheme to return the desired record even when servers' databases are not perfectly synchronized. Our scheme asymptotically has the same computational and communication complexity as state-of-the-art PIR schemes for synchronized databases; this comes at the expense of probabilistic success guarantees and two rounds of communication. As a secondary result, our approach can efficiently process multiple concurrent queries in one round of PIR.