How do you design an OLTP system that handles terrabytes of data and needs to scale 200% per year ? Or 200% per month ?
By the way, all queries still need to respond in subseconds. And no, your budget is not unlimited.
Enter Database sharding – the art of splitting data into manageable chunks, each running on independent commodity hardware.
Come to this presentation to learn the concepts behind data sharding, its strenghts and weaknesses and why (or why not) you might want to shard your data.