This paper presents an industrial report on the implementation of a system that supports execution of queries over encrypted data. While this idea is not new, e.g. [HILM02, AKRX04, PRZB11], the implementation in a real world large scale in-memory database is still challenging. We will provide an overview of our architecture and detail two use cases to give the reader an insight into how we technically realized the implementation. We then provide three main contributions, reporting that: a) We significantly improve functionality by intelligently splitting query execution, i.e. which parts of a query can be performed in the cloud and which on the client. b) We share some initial performance measurements with the community on basis of the TPCH benchmark. c) We present a domain-specific analysis of three data sets that shows the effects of executing queries over encrypted data and what adjustments are required with respect to the encryption of individual columns. The three made observations on query execution, execution time measurements and domain-specific query analysis will lead us to the conclusion that although searching over outsourced encrypted data is always a trade off between functionality, performance and security, it is realistic to assume that working solutions can be provided in the not too distant future to the market.

There are a few reliable privacy mechanisms for cloud applications. Data usually needs to be decrypted in order to be processed by the cloud service provider. In this paper we explore how an encrypted database can (technically) ensure privacy. We study the use case of a mobile personalized healthcare app. We show that an encrypted database can ensure data protection against a cloud service provider. Furthermore we show that if privacy is considered in application design, higher protection levels can be achieved, although encrypted database are a transparent privacy and security mechanism.

Recent databases are implemented as in-memory column-stores. Adjustable encryption offers a solution to encrypted database processing in the cloud. We show that the two technologies play well together by providing an analysis and prototype results that demonstrate the impact of mechanisms at the database side (dictionaries and their compression) and cryptographic mechanisms at the adjustable encryption side (order-preserving, homomorphic, deterministic and probabilistic encryption).

In order to perform a join in a deterministically, adjustably encrypted database one has to re-encrypt at least one column. The problem is to select that column that will result in the minimum number of re-encryptions even under an unknown schedule of joins. Naive strategies may perform too many or even infinitely many re-encryptions. We provide two strategies that allow for a much better performance. In particular the asymptotic behavior is O(n 3/2) resp. O(n logn) re-encryptions for n columns. We show that there can be no algorithm better than O(n logn). We further extend our result to element-wise re-encryptions and show experimentally that our algorithm results in the optimal cost in 41% of the cases.