Exactly what I was thinking. If you have a system of input data snapshots that's also distributed and highly available, it essentially becomes the database (as far as complexity is concerned).