Protecting Privacy in Key-Value Search Systems

Yinglian Xie
Carnegie Mellon University
USA

David O'Hallaron
Carnegie Mellon University
USA

Michael Reiter
Carnegie Mellon University
USA

This paper investigates the general problem of efficiently performing key-value search at untrusted servers without loss of user privacy. Given key-value pairs from multiple owners that are stored across untrusted servers, how can a client efficiently search these pairs such that no server, on its own, can reconstruct the key value pairs?

We propose a system, called Peekaboo, that is applicable and practical to any type of key-value search while protecting both data owner privacy and client privacy. The main idea is to separate the key-value pairs across different servers. Supported by access control and user authentication, Peekaboo allows search to be performed by only authorized clients without reducing the level of user privacy.

Keywords: privacy, key-value search, distributed systems

Read Paper Read Paper (in PDF)