EMZETT.RAID
In short: Redundant Array of Independent Disks — a method for combining several physical hard drives so they appear as a single logical drive, with advantages in speed and/or fault tolerance.
In more detail: Different RAID levels offer different trade-offs: RAID 0 spreads data across several drives for higher speed (but with no redundancy — if one drive fails, all data is gone), RAID 1 mirrors data onto two drives (redundancy, but half the usable capacity), RAID 5/6 combine speed with redundancy via parity data.
In Depth
| Level | Principle | Fault tolerance | Usable capacity |
|---|---|---|---|
| RAID 0 | Data spread across all drives (striping) | None | 100% |
| RAID 1 | Full mirroring onto 2 drives | 1 drive failure | 50% |
| RAID 5 | Data + parity spread across ≥3 drives | 1 drive failure | (n-1)/n |
| RAID 6 | Like RAID 5, double parity | 2 drive failures | (n-2)/n |
With RAID 5/6, data isn’t simply copied — a checksum (“parity”) is stored spread across several drives in addition to the actual data — if one drive fails, its content can be reconstructed from the remaining data and the parity, without having to keep all data twice as with RAID 1. RAID 6 increases safety with a second, independent parity calculation, so even two simultaneous drive failures can be survived — relevant, because the rebuild process after a failure itself puts heavy strain on the remaining drives, making a second failure during this phase more likely than during normal operation.
Important: RAID is not a replacement for a backup — it protects against the failure of individual hard drives, not against accidental deletion, ransomware, or a complete system failure (fire, theft), since all drives usually sit in the same device/room.
Hardware vs. software RAID
RAID can be implemented at two levels: hardware RAID uses a dedicated RAID controller (its own board with its own processor), which computes the distribution/parity completely independently of the main system — fast and transparent to the operating system (it only sees one logical drive), but if the controller itself fails, access to the data can be difficult if no identical replacement device is available. Software RAID (e.g. Linux mdadm, Windows Storage Spaces) handles the same task in the operating system itself, without special hardware — more flexible and cheaper, but with somewhat higher CPU load and potentially lower speed for very data-intensive applications.
Rebuild time as a practical risk
After a drive failure, the RAID system has to reconstruct the missing data from the remaining drives (and parity, if applicable) — the “rebuild”. For very large, modern hard drives (several terabytes), this process can take many hours to days, during which the remaining drives are under heavy strain. This is a real practical risk factor: drives of the same model/batch, installed at the same time, often age at a similar rate — if one fails, the probability of a second failure during the strenuous rebuild process is higher than during normal operation, which is a well-known risk especially for RAID 5 (only one tolerated failure).
RAID in the context of modern storage solutions
In cloud environments and modern NAS systems, classic RAID is increasingly supplemented or replaced by more flexible approaches like ZFS (with its own, integrated redundancy concept and additional protection against silent data corruption) or distributed storage systems spread across several physical servers — but the basic idea (redundancy through distribution across several physical storage units) remains the same as with classic RAID.
See also: Hard Drives, SSDs, HDDs