Section: New Results

SEP2P and DISPERS (Axis 2)

Participants : Luc Bouganim [correspondent] , Julien Loudet, Iulian Sandu Popa.

Personal Data Management Systems (PDMS) arrive at a rapid pace allowing us to integrate all our personal data in a single place and use it for our benefit and for the benefit of the community. This leads to a significant paradigm shift since personal data become massively distributed and opens an important question: how can users/applications execute queries and computations over this massively distributed data in a secure and efficient way, relying exclusively on peer-to-peer (P2P) interactions despite covert adversaries which could be executing the query? We first proposed a Secure and Efficient Peer-to-Peer protocol (SEP2P) to randomly select the nodes that will execute the query. This protocol leverages properties of distributed hash tables (DHT) to select nodes in a way that is, at the same time, secure, random and efficient. The security and randomness stem from the fact that we know, with a very high probability, that at least one honest node contributed to the creation and attestation of this list of nodes; while the efficiency stems from the fact that very few nodes are involved in this process. Building on top of SEP2P, we designed DISPERS, a protocol that applies three design rules: (D1) imposed randomness, enforced by SEP2P, (D2) knowledge dispersion, and (D3) task compartmentalization: Each user provides profile information to indexing nodes, chosen randomly thanks to the DHT (D1). Shamir secret-sharing techniques are used to avoid that any indexing node has a full knowledge of indexed nodes (D2). Then, for each query, a set of random nodes is selected (SEP2P) to coordinate the research for query targets using the indexing nodes. Each of these random nodes learns a part of the query targets IP address but does not know the query (D2, D3). Another set of random nodes is chosen to compute of the final answer based on partial local results from targets. These nodes learn part of the results but do not know the targets, thanks to proxies, nor the meaning of these results (D2, D3). These results are the core of Julien Loudet's thesis [1]. SEP2P was published at EDBT’19 [9] while a demonstration of DISPERS was published at VLDB'19 [8].Both works were also exposed/demonstrated at BDA'19 [13] [12] and APVP'19 [14] for the French research community in databases and security and privacy.