you can use redundancy where the recovery time doesn't scale with size