Answering aggregation queries in a secure system model

TLDR: A comprehensive study on answering SUM and AVG aggregation queries in a read-optimized DBMS for data warehousing applications by using a secure homomorphic encryption scheme in a novel way, demonstrating that the performance of such a solution is comparable to a traditional symmetric encryption scheme.

Abstract: As more sensitive data is captured in electronic form, security becomes more and more important. Data encryption is the main technique for achieving security. While in the past enterprises were hesitant to implement database encryption because of the very high cost, complexity, and performance degradation, they now have to face the ever-growing risk of data theft as well as emerging legislative requirements. Data encryption can be done at multiple tiers within the enterprise. Different choices on where to encrypt the data offer different security features that protect against different attacks. One class of attack that needs to be taken seriously is the compromise of the database server, its software or administrator. A secure way to address this threat is for a DBMS to directly process queries on the ciphertext, without decryption. We conduct a comprehensive study on answering SUM and AVG aggregation queries in such a system model by using a secure homomorphic encryption scheme in a novel way. We demonstrate that the performance of such a solution is comparable to a traditional symmetric encryption scheme (e.g., DES) in which each value is decrypted and the computation is performed on the plaintext. Clearly this traditional encryption scheme is not a viable solution to the problem because the server must have access to the secret key and the plaintext, which violates our system model and security requirements. We study the problem in the setting of a read-optimized DBMS for data warehousing applications, in which SUM and AVG are frequent and crucial.