A Random Server Model for Private Information Retrieval

Abstract

Private information retrieval (PIR) schemes enable users to obtain information from databases while keeping their queries secret from the database managers.We propose a new model for PIR, utilizing auxiliary random servers to provide privacy services for database access. In this model, prior to any on-line communicationwhere users request queries, the database engages in an initial preprocessing setup stage with the random servers. Using this model we achieve the first PIR information theoretic solution inwhich the database does not need to give away its data to be replicated, and with minimal on-line computation cost for the database.This solves privacy and efficiency problems inherent to all previous solutions.

In particular, all previous information theoretic PIR schemes required multiple replications of the database into separate entitieswhich are not allowed to communicate with each other; and in all previous schemes (including ones which do not achieve information theoretic security), the amount of computation performed by the database on-line for every query is at least linear in the size of the database. In contrast, in our solutions the database does not give away its contents to any other entity; and after the initial setup stage which costs at most O(n log n) in computation, the database needs to perform only O(1) amount of computation to answer questions of users on-line. All the extra on-line computation is done by the auxiliary random servers.

An earlier version of this work appears asMIT technical reportMIT-LCS-TR-715. This work was done with the support of DARPA grant DABT63-96-C-0018.