Highly-Scalable Searchable Symmetric Encryption with Support for Boolean Queries

Abstract

Abstract. This work presents the design and analysis of the first searchable symmetric encryption (SSE) protocol that supports conjunctive search and general Boolean queries on outsourced symmetricallyencrypted data and that scales to very large databases and arbitrarilystructured data including free text search. To date, work in this area has focused mainly on single-keyword search. For the case of conjunctive search, prior SSE constructions required work linear in the total number of documents in the database and provided good privacy only for structured attribute-value data, rendering these solutions too slow and inflexible for large practical databases.In contrast, our solution provides a realistic and practical trade-off between performance and privacy by efficiently supporting very large databases at the cost of moderate and well-defined leakage to the outsourced server (leakage is in the form of data access patterns, never as direct exposure of plaintext data or searched values). We present a detailed formal cryptographic analysis of the privacy and security of our protocols and establish precise upper bounds on the allowed leakage. To demonstrate the real-world practicality of our approach, we provide performance results of a prototype applied to several large representative data sets, including encrypted search over the whole English Wikipedia (and beyond).

SHIELD: Scalable Homomorphic Implementation of Encrypted Data-Classifiers

“…Secure computations on binary decision trees were introduced in [8], yet they have long execution time due to the use of the IBM HElib library. Searching an encrypted database was previeously addressed by [4], [6], [7], [14]. One drawback in [4], [6] was the need for a special key to aid the server in performing the search request.…”

“…Our data-classifier takes the idea of PIR one step forward, it homomorphically computes on the encrypted data retrieved by the PIR to obtain useful pieces of information. Other PIR constructions [2], [10], [19], [21], [31], [44] cannot implement data-classifiers the way we do because they either: (a) cannot compute with the PIR responses, or, (b) their plaintext field is only mod 2 (or modulo a small prime, for efficiency purposes) and thus they cannot do integer addition as required by (14). Our HE on the other hand has the advantage of being able to use the full modulo-q domain for plaintext additions.…”

Private Database Access with HE-over-ORAM Architecture

“…Some notable examples of work along this direction is the CryptDB system of Popa et al [23], and recent works on searchable symmetric encryption due to Pappas et al [22] Cash et al [4], and Jarecki et al [17].…”

“…When using the two-ORAM solution, we can often use plain ORAM for the data without the secure-computation layer above it. 4 The reason is that for many ORAM protocols (including path-ORAM), the entire view of the client during the access protocol is uniquely determined by the results of all queries to date, so nothing new is leaked to the client during access to the data ORAM.…”

A practical privacy-preserving targeted advertising scheme for IPTV users

“…Several works on multi-keyword search that enable conjunctive and disjunctive search options exist [11][12][13][14], but these schemes incur large overhead in computation and communication costs. A recent work by Cash et al [15] introduces a searchable encryption scheme that supports multi-keyword search operation over encrypted database. While due to scalability the scheme is suitable for big data applications, it does not allow to rank the responses based on their relevancy to a given query.…”