What is RAID 10?

Why RAID 10 matters

It is the standard choice for virtualisation storage because it tolerates a drive failure without the write-amplification penalty of parity RAID. RAID 5 and RAID 6 store parity instead of a full mirror, which is cheaper per usable terabyte but turns every small write into a read-modify-write cycle — ruinous for a host running dozens of database-backed guests. RAID 10 also rebuilds far faster after a failure, because it copies from one surviving mirror rather than recomputing parity across the whole array, and a long rebuild is exactly when a second drive is most likely to fail. Every OnionVPS NVMe pool is RAID 10; the Hold line uses RAID 6 on its bulk HDD arrays, where the access pattern is large and sequential and parity costs nothing noticeable.

RAID 10 in practice

When you're running a server on RAID 10, the first thing you check is the controller or software RAID status. You want to see all mirrors healthy and no degraded arrays. If a drive fails, replace it promptly; the rebuild reads from the surviving mirror, which is fast, but during that window you've lost redundancy. A second failure in the same mirror pair means total data loss. You also trade half your raw capacity for that safety, so capacity planning needs to account for it from day one.

What people get wrong about RAID 10

A common misconception is that RAID 10 can survive any two drive failures. In fact, it can only survive one failure per mirror pair. If two drives in the same mirrored pair fail, the array is lost. The correction: a RAID 10 array's fault tolerance is one drive per pair, not two overall.