DuxkDB's query engine is inspired by the same paper! You'll be surprised what you can process on a single node with DuckDB - it takes 33 Spark nodes to match performance! That still puts it ahead of Photon, and it is open source.