Learning fault-tolerance planning with EC2 instances