Almost Optimal Private Information Retrieval

A private information retrieval (PIR) protocol allows a user to retrieve one of N records from a database while hiding the identity of the record from the database server. With the initially proposed PIR protocols to process a query, the server has to process the entire database, resulting in an unacceptable response time for large databases. Later solutions make use of some preprocessing and oo-line communication, such that only O(1) on-line computation and communication are performed to execute a query. The major drawback of these solutions is oo-line communication, comparable to the size of the entire database. Using a secure coprocessor we construct a PIR scheme with O(1) on-line computation and communication, periodical oo-line preprocessing, and zero oo-line communication. The protocol is almost optimal. The only parameter left to improve is the server's oo-line preprocessing complexity-the least important one.