Improving the Robustness of Private Information Retrieval

Since 1995, much work has been done creating protocols for private information retrieval (PIR). Many variants of the basic PIR model have been proposed, including such modifications as computational vs. information-theoretic privacy protection, correctness in the face of servers that fail to respond or that respond incorrectly, and protection of sensitive data against the database servers themselves. In this paper, we improve on the robustness of PIR in a number of ways. First, we present a Byzantine-robust PIR protocol which provides information-theoretic privacy protection against coalitions of up to all but one of the responding servers, improving the previous result by a factor of 3. In addition, our protocol allows for more of the responding servers to return incorrect information while still enabling the user to compute the correct result. We then extend our protocol so that queries have information-theoretic protection if a limited number of servers collude, as before, but still retain computational protection if they all collude. We also extend the protocol to provide information-theoretic protection to the contents of the database against collusions of limited numbers of the database servers, at no additional communication cost or increase in the number of servers. All of our protocols retrieve a block of data with communication cost only O(lscr) times the size of the block, where lscr is the number of servers. Finally, we discuss our implementation of these protocols, and measure their performance in order to determine their practicality.