Private Database Synthesis for Outsourced System Evaluation

The goal of this paper is to permit secure outsourced system evaluation. We propose a method for generating synthetic databases and obfuscating a workload of queries in order to protect the sensitive information present in the database. The synthetic database and workload can be used by a third party to accurately carry out performance tuning, index selection, or other system evaluation tasks. As a result, an untrusted third party can evaluate whether a new technology would benefit the data owner without the risk of a privacy breach. Our approach is to employ state-of-the-art privacy mechanisms to compute the sufficient statistics of a statistical model of the true database. These statistics are safe to release, so a third party can then use them to generate one or more synthetic databases to be used as a surrogate for the true database.