18.4 About Software RAID

The Redundant Array of Independent Disks (RAID) feature allows you to spread data across the drives to increase capacity, implement data redundancy, and increase performance. RAID is usually implemented either in hardware on intelligent disk storage that exports the RAID volumes as LUNs, or in software by the operating system. Oracle Linux kernel uses the multidisk (MD) driver to support software RAID by creating virtual devices from two or more physical storage devices. You can use MD to organize disk drives into RAID devices and implement different RAID levels.

The following software RAID levels are commonly used with Oracle Linux:

Linear RAID (spanning)

Combines drives as a larger virtual drive. There is no data redundancy or performance benefit. Resilience decreases because the failure of a single drive renders the array unusable.

RAID-0 (striping)

Increases performance but does not provide data redundancy. Data is broken down into units (stripes) and written to all the drives in the array. Resilience decreases because the failure of a single drive renders the array unusable.

RAID-1 (mirroring)

Provides data redundancy and resilience by writing identical data to each drive in the array. If one drive fails, a mirror can satisfy I/O requests. Mirroring is an expensive solution because the same information is written to all of the disks in the array.

RAID-5 (striping with distributed parity)

Increases read performance by using striping and provides data redundancy. The parity is distributed across all the drives in an array but it does not take up as much space as a complete mirror. Write performance is reduced to some extent from RAID-0 by having to calculate parity information and write this information in addition to the data. If one disk in the array fails, the parity information is used to reconstruct data to satisfy I/O requests. In this mode, read performance and resilience are degraded until you replace the failed drive and it is repopulated with data and parity information. RAID-5 is intermediate in expense between RAID-0 and RAID-1.

RAID-6 (striping with double distributed parity)

A more resilient variant of RAID-5 that can recover from the loss of two drives in an array. RAID-6 is used when data redundancy and resilience are important, but performance is not. RAID-6 is intermediate in expense between RAID-5 and RAID-1.

RAID 0+1 (mirroring of striped disks)

Combines RAID-0 and RAID-1 by mirroring a striped array to provide both increased performance and data redundancy. Failure of a single disk causes one of the mirrors to be unusable until you replace the disk and repopulate it with data. Resilience is degraded while only a single mirror remains available. RAID 0+1 is usually as expensive as or slightly more expensive than RAID-1.

RAID 1+0 (striping of mirrored disks or RAID-10)

Combines RAID-0 and RAID-1 by striping a mirrored array to provide both increased performance and data redundancy. Failure of a single disk causes part of one mirror to be unusable until you replace the disk and repopulate it with data. Resilience is degraded while only a single mirror retains a complete copy of the data. RAID 1+0 is usually as expensive as or slightly more expensive than RAID-1.