Secure Yannakakis: Join-Aggregate Queries over Private Data

In this paper, we describe a secure version of the classical Yannakakis algorithm for computing free-connex join-aggregate queries. This protocol can be used in the secure two-party computation model, where the parties would like to evaluate a query without revealing their own data. Our protocol presents a dramatic improvement over the state-of-the-art protocol based on Yao's garbled circuit. In theory, its cost (both running time and communication) is linear in data size and polynomial in query size, whereas that of the garbled circuit is polynomial in data size and exponential in query size. This translates to a reduction in running time in practice from years to minutes, as tested on a number of TPC-H queries of varying complexity.