Paper 2022/691

QuORAM: A Quorum-Replicated Fault Tolerant ORAM Datastore

Abstract

Privacy and security challenges due to the outsourcing of data storage and processing to third-party cloud providers are well known. With regard to data privacy, Oblivious RAM (ORAM) schemes provide strong privacy guarantees by not only hiding the contents of the data (by encryption) but also obfuscating the access patterns of the outsourced data. But most existing ORAM datastores are not fault tolerant in that if the external storage server (which stores encrypted data) or the trusted proxy (which stores the encryption key and other meta- data) crashes, an application loses all of its data. To achieve fault-tolerance, we propose QuORAM, the first ORAM datastore to replicate data with a quorum-based replication protocol. QuORAM’s contributions are three-fold: (i) it obfuscates access patterns to provide obliviousness guarantees, (ii) it replicates data using a novel lock-free and decentralized replication protocol to achieve fault-tolerance, and (iii) it guarantees linearizable semantics. Experimentally evaluating QuORAM highlights counter-intuitive results: QuORAM in- curs negligible cost to achieve obliviousness when compared to an insecure fault-tolerant replicated system; QuORAM’s peak throughput is 2.4x of its non-replicated baseline; and QuORAM performs 33.2x better in terms of throughput than an ORAM datastore that relies on CockroachDB, an open- source geo-replicated database, for fault tolerance.