Section: New Results

Performance of large scale data-oriented operations under TEE constraints (Axis 3)

Participants : Robin Carpentier [correspondent] , Nicolas Anciaux, Iulian Sandu Popa, Guillaume Scerri.

The rise of Trusted Execution Environments like Intel SGX, and their more and more widespread use for data processing raises the question of their impact on performance, specifically for data oriented operations. While some works aim at embedding either the entirety of part of a database engine within a TEE, the direct impact of processing data with TEEs as opposed to more classical environment has not been studied yet. In particular, the cryptographic overhead of accessing persistent data outside the TEE enclave, the limited RAM amount of each TEE enclave, the cost of external function calls and memory access overheads, may slow the computing by orders of magnitude compared to a regular environment, and have to be taken into account. Preliminary results presenting both a benchmark of data operations within Intel SGX, together with optimisation of search algorithm dealing with the specific way of accessing external memory from inside SGX have been presented at BDA'18 [16].