Data-oblivious Algorithms Research Articles

Enhancing Privacy in Graph Algorithms: Data-Oblivious Approaches to DFS and Dijkstra's Algorithm

Data obliviousness is characterized by a consistent sequence of operations irrespective of input data and data-independent memory accesses, making it a suitable solution for users utilizing outsourced storage data services who aim to conceal their data access patterns. In ACM SIGSAC 2013, data-oblivious algorithms were introduced for breadth-first search, single-source single-destination (SSSD) algorithms, maximum flow, and minimum spanning tree. In this study, we present novel data-oblivious algorithms designed for depth-first search and single-source shortest path (Dijkstra’s algorithm). Our proposed data-oblivious algorithms demonstrate efficiency comparable to non-data-oblivious counterparts, particularly for graphs containing fewer than 1000 nodes. This research contributes to advancing data privacy in outsourced storage services by providing effective data-oblivious solutions for common graph algorithms.

Find Thy Neighbourhood: Privacy-Preserving Local Clustering

Identifying a cluster around a seed node in a graph, termed local clustering, finds use in several applications, including fraud detection, targeted advertising, community detection, etc. However, performing local clustering is challenging when the graph is distributed among multiple data owners, which is further aggravated by the privacy concerns that arise in disclosing their view of the graph. This necessitates designing solutions for privacy-preserving local clustering and is addressed for the first time in the literature. We propose using the technique of secure multiparty computation (MPC) to achieve the same. Our local clustering algorithm is based on the heat kernel PageRank (HKPR) metric, which produces the best-known cluster quality. En route to our final solution, we have two important steps: (i) designing data-oblivious equivalent of the state-of-the-art algorithms for computing local clustering and HKPR values, and (ii) compiling the data-oblivious algorithms into its secure realisation via an MPC framework that supports operations over fixed-point arithmetic representation such as multiplication and division. Keeping efficiency in mind for large graphs, we choose the best-known honest-majority 3-party framework of SWIFT (Koti et al., USENIX'21) and enhance it with some of the necessary yet missing primitives, before using it for our purpose. We benchmark the performance of our secure protocols, and the reported run time showcases the practicality of the same. Further, we perform extensive experiments to evaluate the accuracy loss of our protocols. Compared to their cleartext counterparts, we observe that the results are comparable and thus showcase the practicality of the designed protocols.

Differentially Oblivious Data Analysis With Intel SGX: Design, Optimization, and Evaluation

A privacy-preserving data analytics system enables a cloud user to perform the distributed job in a secure manner such that the data privacy can be guaranteed during the cloud-outsourced computation. However, many SGX-based solutions are vulnerable to some side-channel attacks, including the access pattern leakage from both network and memory. Several data-oblivious algorithms with full obliviousness have been proposed in the literature, but they are impractical to be used in the cloud due to the expensive computational overhead. In this article, we propose a DPSpark system with the security defined in a notion of <inline-formula><tex-math notation="LaTeX">$(\epsilon,\delta)$</tex-math></inline-formula> -differentially private obliviousness ( <inline-formula><tex-math notation="LaTeX">$(\epsilon,\delta)$</tex-math></inline-formula> -DPO), which relaxes full obliviousness to enable an efficiency improvement. Based on this definition, we present a perturbation-shuffle-analysis (PSA) computing architecture and design several typical differentially oblivious operators. In further, we optimize the system efficiency by reducing the number of oblivious shuffles and choosing an appropriate privacy budget. Finally, we benchmark the system in different parameters. The experimental results show that DPSpark significantly outperforms two state-of-the-art solutions, only with 10.1-85.4 percent additional overhead performing an SGX-based data analysis application.