Choose how big your data is and how much throughput you need. Each point is a benchmarked instance × disk; everything right of the target line clears your requirement, and the cheapest of those is the pick.
Want the raw numbers instead? See the full benchmark data →