[reiddraper]

Another thought. If availability is your goal, with the trend toward 'operations as code', I think a small development team can build a system on top of AWS that can automatically respond to arbitrary node/resource/data-center failure. Netflix seems to do this to an extent, with their Chaos Monkey.

That being said, there are situations where you may truly need single-node performance that isn't available on AWS.

[jedberg]

Chaos Monkey causes chaos, it does not fix it. :)

But yes, you are right. The goal is to have a system that scales itself. Not an easy task for sure.