Why Buying Provisioned IOPS on RDS Might Be a Mistake

general purpose SSD offers a good bang for your buck. The "best/most consistent" disk performance on amazon is still (unfortunately) ephemeral disks in a raid set. But if you turn that instance off (or Amazon decides to turn it off for you...) bye-bye to your local data. Not an acceptable risk for a database.

And still, after all these years, AWS won't replace a failed ephemeral disk in your instance. Is there a technical reason for this??

If one of the large cloud providers wants to become competitive in this space, all they need to do is offer instances with local storage that are in a hardware raid set with truly large volumes. Heavy disk-io applications are still much better suited on-prem where you have the flexibility to build these types of monsters. 5TB spinning disks are going for under $200 these days.

Or AWS could just start replacing disks on their m1 series machines...