Enhancing security for columnar storage and data lakes

In this PhD project, we will investigate enhancements for columnar storage file formats, which play a crucial role in database work- loads, but also increasingly in machine-learning workloads, with the overall goal of improving security in next-gen Data Lakes. This is investigated in three lines of proposed work to develop (i) a vector-friendly encryption file layout that allows efficient process- ing on both CPUs and GPUs in our new FastLanes format, (ii) hybrid encrypted query processing, in which decryption happens client-side, introducing new techniques that allow cloud servers to skip encrypted data in predicate pushdown; and (iii) oblivious data access techniques that exploit the compaction processes already necessary for data life-cyle management in Data Lakes.