People who visit this website have either of two concerns. Should my next array be RAID 10 or RAID 5? Or my RAID 10 has just dropped a drive: how worried should I be? Both can be addressed by answering one question: how does RAID 10 protect each copy, and what will happen if you lose another disk?
RAID 10 is a pairing of drive mirrors (twin sets) that then stripes data across those pairs to increase speed. It’s quick because each pair shares the load; resilient because every pair is a backup of itself.
RAID 5 places your data next to a calculated safeguard (parity) distributed among every disk: purchase N number of disks, retain the capacity of N − 1, and losing any one disk will still allow you to survive. RAID 10 ignores all of the maths and simply stores two copies of everything. The drawback of using RAID 10 is that you sacrifice half of your capacity; however, writes occur quickly since there is no calculation required. When you replace a failed disk, it is an exact duplicate from its identical twin, which allows for a rebuild to take place in hours, utilising only one operational disk instead of requiring all disks in the RAID to work together until the rebuild process completes.
There are a lot of articles showing “RAID 10 survives multiple failures.” However, there is always some sort of caveat as to where. Losing two different disks in two separate pairs: no problem. Losing a second disk within a pair which had already lost a disk will take that entire pair’s portion of every file down with it since they were the only copies. Reality also loads the dice against you. Disk pairs typically come from the same batch, were purchased at the same time, installed together at the same time, worked side-by-side for the exact same amount of time, etc., so if one fails, the other is often right behind and the heavy I/O usage during the rebuild causes the second one to fail.
The principle that allows you to save an injured RAID can be summed up on a Post-it note: shut it down, photograph and label every drive with its bay, and change nothing. If done correctly, we will image each member of the RAID read-only and then rebuild the RAID using those images. This way, the rebuild occurs using copies rather than your last original members. That basic principle is the backbone of how we perform RAID recovery regardless of the size or type of ‘array’ being recovered, whether it is a large rack server system or a small two-bay NAS sitting on a shelf.
No, and the mix-ups in there keep the recovery labs busy. RAID 10 protects you from just one thing: the loss of a drive. All other types of disasters (accidental deletions, ransomware attacks, corruptions to data on your drives, failures by controllers, fires, theft) copy the errors instantaneously to each of the mirrored disks. A RAID provides redundancy while allowing you to continue working through a single drive failure; a complete, separate, disconnected backup will protect you from all other disasters. Both are necessary for different uses.
The safety argument has merit; however, the performance advantage has diminished. The speeds at which SSDs operate are so much faster than the spinning hard disk drives used years ago that the benefits associated with striping are no longer as pronounced and many systems can be protected at a lower cost using a RAID 1 configuration. There is an area of use for RAID 10 with SSD storage: when doing large amounts of continuous write operations (e.g. databases and virtualisation) where high-speed throughput and low-impact (fast and gentle) rebuild processes provide value and therefore warrant a higher price.
It can wait, but that is how you get into two-drive disasters without realising it at first. You can be utilising a degraded array, having lost all protection for the ‘mirror’ while relying on the other drive (same model, same age, same mileage) to face the single drive workload alone. If you really have important data to recover, then take degraded as an event and not a state of operation. Get your backups performed now, replace the failed drive immediately and if at all you think the rebuild is being executed ‘wrong’, go ahead and image it before trying to execute the recovery.
Free diagnostic within 48 hours on multi-disk arrays: read-only image of all members, every rebuild run against copies, quote written before any work.