No abstract

Database outsourcing is an emerging data management paradigm which has the potential to transform the IT operations of corporations. In this paper we address privacy threats in database outsourcing scenarios where trust in the service provider is limited. Specifically, we analyze the data partitioning (bucketization) technique and algorithmically develop this technique to build privacy-preserving indices on sensitive attributes of a relational table. Such indices enable an untrusted server to evaluate obfuscated range queries with minimal information leakage. We analyze the worst-case scenario of inference attacks that can potentially lead to breach of privacy (e.g., estimating the value of a data element within a small error margin) and identify statistical measures of data privacy in the context of these attacks. We also investigate precise privacy guarantees of data partitioning which form the basic building blocks of our index. We then develop a model for the fundamental privacy-utility tradeoff and design a novel algorithm for achieving the desired balance between privacy and utility (accuracy of range query evaluation) of the index.

Database outsourcing is an emerging data management paradigm which has the potential to transform the IT operations of corporations. In this paper we address privacy threats in database outsourcing scenarios where trust in the service provider is limited. Specifically, we analyze the data partitioning (bucketization) technique and algorithmically develop this technique to build privacy-preserving indices on sensitive attributes of a relational table. Such indices enable an untrusted server to evaluate obfuscated range queries with minimal information leakage. We analyze the worst-case scenario of inference attacks that can potentially lead to breach of privacy (e.g., estimating the value of a data element within a small error margin) and identify statistical measures of data privacy in the context of these attacks. We also investigate precise privacy guarantees of data partitioning which form the basic building blocks of our index. We then develop a model for the fundamental privacy-utility tradeoff and design a novel algorithm for achieving the desired balance between privacy and utility (accuracy of range query evaluation) of the index.

In this paper, we study the problem of supporting multidimensional range queries on encrypted data. The problem is motivated by secure data outsourcing applications where a client may store his/her data on a remote server in encrypted form and want to execute queries using server's computational capabilities. The solution approach is to compute a secure indexing tag of the data by applying bucketization (a generic form of data partitioning) which prevents the server from learning exact values but still allows it to check if a record satisfies the query predicate. Queries are evaluated in an approximate manner where the returned set of records may contain some false positives. These records then need to be weeded out by the client which comprises the computational overhead of our scheme. We develop a bucketization procedure for answering multidimensional range queries on multidimensional data. For a given bucketization scheme, we derive cost and disclosure-risk metrics that estimate client's computational overhead and disclosure risk respectively. Given a multidimensional dataset, its bucketization is posed as an optimization problem where the goal is to minimize the risk of disclosure while keeping query cost (client's computational overhead) below a certain user-specified threshold value. We provide a tunable data bucketization algorithm that allows the data owner to control the trade-off between disclosure risk and cost. We also study the trade-off characteristics through an extensive set of experiments on real and synthetic data.

In