RAID 10 is built as mirrored pairs with striping across them, so whether it survives a second failure depends entirely on which disk fails. If you lose one disk from each mirror set you will be fine. However if you lose both disks from one mirror set then your entire RAID 10 volume will go down. The single-parity cousin has its own failure mode, described in RAID 5 data recovery.
You make mirrored pairs out of your disks. When you add new data to the RAID 10 system, the data is written across all of the pairs. It gives you the ability to quickly write data to the RAID 10 system. It also provides very predictable behaviour when dealing with failures. In addition, since you use mirrored pairs, you reduce your usable space by half compared to using individual disks.
The order of construction is the most important thing about how it will behave.
A RAID 10 with four disks will be two mirrored sets of disks — one set consisting of disks one and two, which contain the exact same data, and another set consisting of disks three and four, also containing the same data — and the volume will be striped over these two pairs. When a write occurs for any given stripe, it writes to both members of that stripe's respective pair.
This approach provides the write performance associated with striping, plus the redundancy of mirroring, making it the default method for databases and busy virtual machine hosts. However, this approach also represents that you have paid for the redundancy of half of your raw storage space. For example, using eight 4TB disks results in 16TB of useable space and not 28TB that would be provided by RAID 5.
A key structural aspect: Redundancy exists within individual pairs and not throughout the entire array. Therefore, there is no parity that spans across all devices in the array. This means if one pair fails on an array resulting in two failed disks, neither disk will have anything to rebuild or recover from.
Same number of disk failures yields completely different results.
Survivable. Each mirrored pair now retains at least one active member; the volume remains active; and the loss in performance is minor. In theory, if you have an eight disk array, you could lose up to four disks like this.
Complete failure of an entire volume. That pair held the only copies of its stripes, and nothing else in the array can reconstruct them.
The "dangerous" window. When rebuilding a pair after losing a member, a full read is made of the surviving member of the pair. If that member was also degraded, it can fail under the load of the rebuild — taking the pair with it.
Deletions. Corruption. Ransomware. Duplicated to both members of every pair simultaneously. No protection against these types of failures exists using RAID 10.
The primary threat you will face is a direct result of the design's success.
When a member fails in a RAID 5, the system slows down significantly, because each time you access the failed disk's data, you have to compute it by using the parities of all remaining members. People see it.
If an active member fails in a RAID 10, then the system will just read off the other half of the failing pair. There is no calculation, no noticeable degradation and nothing a user would feel. The array runs at nearly full speed with a pair reduced to one disk and no margin left in it.
Therefore, the RAID 10 arrays we receive have been running in a degraded mode for months because nobody ever looked. When you are running a RAID 10 array, alerting matters more than it does on any other level — there is no performance indication to fall back on.
The rebuild process occurs in reverse of how the build occurred.
Although the controller has removed the disk from service for failure, a majority of that disk's data is often still present on the disk; in a mirrored set, it may be the only source for whatever its partner is unable to retrieve.
Mirrored members can be identified through direct comparison of their image data. Each member will contain identical data with identical offset, making the association obvious after a few megabytes.
Stripe size, stripe order and stripe offset will be determined by trial-and-error testing of possible candidates until a coherent structure to the file systems is achieved.
A virtual representation of the volume will be constructed using the external images, while the file system will be restored from its internal backup structures.
It all depends on which ones. One disk out of each mirrored pair can be lost; on an eight-disk array that is theoretically four. If both disks from the same pair are lost then you lose everything, as nothing else in the array has those stripes.
For rebuild risk, considerably. In rebuilding a RAID 10 pair you only read the other side of the pair, and thus do not read through the entire array. This greatly lowers the load. However you pay for that by sacrificing space in your RAID — half of the raw storage in the RAID is used for redundancy.
RAID 10 provides no performance indication at time of failure; it simply reads off the other half of the pair. There is no calculation and therefore no measurable slow down. Alerting matters more here than on any other level.
No. The errors (deletion, corruption, ransomware) occur above the RAID layer, and all data written to each member of the mirror pairs is immediately copied to the peer member of that pair. Those can only be protected against by making either snapshots or backing up to a non-array device.
Only by recovering the physical disks in a failed pair will you have enough information to recreate the missing portions of the stripes for both disks in that pair. Each disk in the failed pair is then imaged, including any that were ejected by the RAID controller; those images are then merged to restore continuity in each other's stripe.
From £500 plus VAT after a free 48hr diagnostic. We also ask for a 50% deposit if any of your drives need physical drive-level work. Please send us all disks, labelled by bay where you know the order.