Comment by lobster_johnson

That criticism is specific to S3, not EC2 or Hadoop. It's perfectly feasible and probably preferable to have Hadoop work on local files in instance store volumes (or EBS if you're mad).