We should be provisioning spot instances and handling failure of spot instances instead of more expensive ec2 vms