International Association
for Cryptologic Research

Distributed ORAM (DORAM) is a multi-server variant of Oblivious RAM. Originally proposed to lower bandwidth, DORAM has recently been of great interest due to its applicability to secure computation in the RAM model, where the circuit complexity and rounds of communication are equally important metrics of efficiency. All prior DORAM constructions either involve linear work per server (e.g., Floram) or logarithmic rounds of communication between servers (e.g., square root ORAM). In this work, we construct the first DORAM schemes in the 2-server, semi-honest setting that simultaneously achieve sublinear server computation and constant rounds of communication. We provide two constant-round constructions, one based on square root ORAM that has O(sqrt{N} log N) local computation and another based on secure computation of a doubly efficient PIR that achieves local computation of O(N^e) for any e > 0 but that allows the servers to distinguish between reads and writes. As a building block in the latter construction, we provide secure computation protocols for evaluation and interpolation of multi- variate polynomials based on the Fast Fourier Transform, which may be of independent interest.

We consider a scenario where a server holds a huge database that it wants to make accessible to a large group of clients. After an initial setup phase, clients should be able to read arbitrary locations in the database while maintaining privacy (the server does not learn which locations are being read) and anonymity (the server does not learn which client is performing each read). This should hold even if the server colludes with a subset of the clients. Moreover, the run-time of both the server and the client during each read operation should be low, ideally only poly-logarithmic in the size of the database and the number of clients. We call this notion Private Anonymous Data Access (PANDA). PANDA simultaneously combines aspects of Private Information Retrieval (PIR) and Oblivious RAM (ORAM). PIR has no initial setup, and allows anybody to privately and anonymously access a public database, but the server’s run-time is linear in the data size. On the other hand, ORAM achieves poly-logarithmic server run-time, but requires an initial setup after which only a single client with a secret key can access the database. The goal of PANDA is to get the best of both worlds: allow many clients to privately and anonymously access the database as in PIR, while having an efficient server as in ORAM.In this work, we construct bounded-collusion PANDA schemes, where the efficiency scales linearly with a bound on the number of corrupted clients that can collude with the server, but is otherwise poly-logarithmic in the data size and the total number of clients. Our solution relies on standard assumptions, namely the existence of fully homomorphic encryption, and combines techniques from both PIR and ORAM. We also extend PANDA to settings where clients can write to the database.

We initiate the study of fully homomorphic encryption for RAMs (RAM-FHE). This is a public-key encryption scheme where, given an encryption of a large database D, anybody can efficiently compute an encryption of P(D) for an arbitrary RAM program P. The running time over the encrypted data should be as close as possible to the worst case running time of P, which may be sub-linear in the data size.A central difficulty in constructing a RAM-FHE scheme is hiding the sequence of memory addresses accessed by P. This is particularly problematic because an adversary may homomorphically evaluate many programs over the same ciphertext, therefore effectively “rewinding” any mechanism for making memory accesses oblivious.We identify a necessary prerequisite towards constructing RAM-FHE that we call rewindable oblivious RAM (rewindable ORAM), which provides security even in this strong adversarial setting. We show how to construct rewindable ORAM using symmetric-key doubly efficient PIR (SK-DEPIR) (Canetti-Holmgren-Richelson, Boyle-Ishai-Pass-Wootters: TCC ’17). We then show how to use rewindable ORAM, along with virtual black-box (VBB) obfuscation for specific circuits, to construct RAM-FHE. The latter primitive can be heuristically instantiated using existing indistinguishability obfuscation candidates. Overall, we obtain a RAM-FHE scheme where the multiplicative overhead in running time is polylogarithmic in the database size N. Our basic scheme is single-hop, but we also extend it to obtain multi-hop RAM-FHE with overhead $$N^\epsilon $$ for arbitrarily small $$\epsilon >0$$ .We view our work as the first evidence that RAM-FHE is likely to exist.

We consider a setting where users store their encrypted documents on a remote server and can selectively share documents with each other. A user should be able to perform keyword searches over all the documents she has access to, including the ones that others shared with her. The contents of the documents, and the search queries, should remain private from the server.This setting was considered by Popa et al. (NSDI ’14) who developed a new cryptographic primitive called Multi-Key Searchable Encryption (MKSE), together with an instantiation and an implementation within a system called Mylar, to address this goal. Unfortunately, Grubbs et al. (CCS ’16) showed that the proposed MKSE definition fails to provide basic security guarantees, and that the Mylar system is susceptible to simple attacks. Most notably, if a malicious Alice colludes with the server and shares a document with an honest Bob then the privacy of all of Bob’s search queries is lost.In this work we revisit the notion of MKSE and propose a new strengthened definition that rules out the above attacks. We then construct MKSE schemes meeting our definition. We first give a simple and efficient construction using only pseudorandom functions. This construction achieves our strong security definition at the cost of increasing the server storage overhead relative to Mylar, essentially replicating the document each time it is shared. We also show that high server storage overhead is not inherent, by giving an alternate (albeit impractical) construction that manages to avoid it using obfuscation.