RAID10

RAID 10, also known as RAID 1+0, is a nested RAID level that combines disk mirroring (RAID 1) and striping (RAID 0) to provide redundancy and improved performance. It requires at least four drives and is implemented by grouping drives into mirrored pairs and then striping data across the pairs. With N drives (N even), the array consists of N/2 mirrored pairs; each pair stores a complete copy of the data while stripes are distributed across pairs. Usable capacity equals half of total raw capacity, assuming equal-sized disks (e.g., four 2 TB disks yield 4 TB usable).

Fault tolerance and rebuild: RAID 10 can tolerate the failure of one disk in each mirrored pair, so up to floor(N/2) disks can fail provided no two failed disks are in the same pair. If both disks in a single mirror fail, the array loses data. Rebuilds occur by reconstructing a failed member from its mirror, and performance may degrade during rebuild.

Performance: Read operations can be served from any disk in a pair, and multiple pairs allow parallel reads, yielding high read throughput. Writes must be performed to both disks in each mirror, so write operations are effectively mirrored; the array can achieve high write throughput through parallelism across pairs, but each write is duplicated within its pair.

Implementation notes: RAID 10 is supported by many hardware RAID controllers and by software RAID solutions such as Linux mdadm and Windows Storage Spaces; it requires an even number of disks, typically a minimum of four. It is often favored for databases and other I/O-intensive workloads where fast rebuilds and robust fault tolerance are beneficial. It offers strong fault tolerance characteristics compared with some other RAID levels, depending on failure distribution.