Management of redundant systems can be complicated. The HP AutoRAID attempts to provide a solution to this problem with a two-level storage hierarchy that is abstracted behind a single disk-array controller. Write-active data are mirrored for better performance while write-inactive data are stored in RAID 5 for better cost. The system begins by allocating mirrored space until no more can be stored this way and automatically reallocates some of the storage space to RAID 5. The HP AutoRAID adapts to the workload changes: newly active data are promoted to mirrored storage while data the have become less active are demoted to RAID 5. The HP AutoRAID also allows for easy disk upgrades of disks of different capacity, which allows for easy expansion and upgrade. The system can even have a second controller that it can automatically change over to in the case of failure of the primary controller. The system is simple to administer and setup. The HP AutoRAID system seems to do a good job of providing performance close to that of nonredundant systems for many workloads. At the same time, it provides full data redundancy and can tolerate failures of any single array component.
This system is a classic example of a system that does achieves its function while abstracting away how the function is carried out. Has this system become obsolete with the trend towards clusters of computers networked together? The cost of storage has gotten so cheap that RAID is not needed in such systems. Redundancy can be achieved by replicating data across multiple computers.