At the disk level, automatic memory management using low-cost disk mirroring set up and use the concept of fault group to prevent any single node failure.