I believe the interviewer was expecting a distributed computing based solution, esp when "100 billion records" are involved. With the limited knowledge of Distributed Computing I have, I would suggest you to look into Distributed Hash Table and map-reduce (for parallel query processing)