RAID levels · explained

RAID 10, explained properly.

A mirror of stripes is how RAID 10 works: speed through striping, safety through mirroring. This guide explains how this actually works, what type of failure it would survive and which combination of failures would quietly kill it. Also, we will show you one button that when pressed can turn an otherwise recoverable array into an irretrievable one.

Free diagnostic · arrays
Imaged before rebuilds
100% in-house · Belfast
// in one line

Two copies of half each.

Take four drives. Create two mirrors with them and then stripe your data across those pairs. Each block lives on exactly two drives, which is both the strength and fine print.

Min drives
4
Usable space
50%
Survives
1 per mirror
Kills it
Both of one pair
// why people choose it

What RAID 10 buys you — and what it costs.

Because the stripe half of this design allows both reads & writes to occur simultaneously over all drives within the array at once, the entire array will be running very quickly (fast) as there will be no parity calculations slowing down the write processes associated with RAID 5 & RAID 6. As well, the mirror half allows for the rebuilding of a failed drive using the exact identical data contained on the mirrored twin, thus allowing for a much faster rebuild process (hours, rather than the day-plus of a parity rebuild), and an immense amount of stress relief upon the remaining drives during said rebuild. It is exactly this combination (fast under load, fast recovery) that makes databases, virtual machine hosts & busy NAS systems so commonly found sitting on a RAID 10 system.

There are really only two costs associated with a RAID 10 system, the first being storage space / capacity. Everything is duplicated across each drive, resulting in half of your purchased space being utilised solely for duplicate copies. Also, there exists another somewhat subtle cost which warrants mentioning: positional safety. A RAID 10 does not provide protection against “any two failures”, but instead provides protection against “two failure(s) occurring in such a manner that they do not destroy a single mirror pair”.

// the fine print

The wrong two drives, and the rebuild trap.

Most of the RAID 10 systems that end up on our test bench fall into either one or both of the following categories:

Fact one: mirrors have partners. If you lose drive 1 & drive 3 (they are in different mirror pairs), the loss of these drives would allow the array to continue operation with a complete copy of all data. However, if you were to lose both drive 1 and drive 2 (they are the same mirror pair), then there would exist no surviving copy of those blocks that resided on that mirror. This seems like a reasonably acceptable scenario unless you take into consideration that most times when you lose a drive, the most probable candidate to also lose will be its mirrored partner. Therefore, since most people purchase paired drives from the same manufacturer, in the same lot, performing the same function, they are likely to be identical in age, and are more likely to experience failure at the same time due to their similarities.

Fact two: rebuilds are not neutral. In all cases where a controller offers to rebuild (or a degraded array receives a replacement drive), it begins a long copy process using a survivor that itself could develop problems. A rebuild against a failing twin can cause the twin to fail further in progress of rebuilding, causing a single drive failure to become the pair failure discussed above. If the wrong disk is removed during this period, the controller can write to the very disk that contained the last valid copy. When an array is degraded and you have data which must be recovered: power down, label every drive with its slot number, change nothing. Each member is then imaged read-only on the bench before the array is rebuilt, to ensure that no rebuild takes place on original disks. This method of recovering RAID arrays is what we use daily for RAID 10 recovery (QNAP/Synology) and other types of RAID recovery in rack servers. See NAS recovery for more information.

// sizing it

What RAID 10 costs in real bays.

Depending on your budget, and the number of drives you purchase, the 50% rule works differently. Before purchasing a new storage enclosure it’s well worth doing some maths to figure out what size enclosure will provide you the most bang for your buck. For example: four 8TB drives = 16TB of usable space (two mirrored pairs). Six drives will give you 24TB of usable space from 48TB purchased (three striped sets, so performance climbs with the bay count). With eight bays you get 32TB of usable space from 64TB purchased. Now compare the RAID 6 configuration at eight bays (48TB usable space with two-drive failure protection). It becomes apparent that there’s a trade-off here: RAID 10 provides faster access times and faster rebuild times (at the expense of total capacity); RAID 6 provides greater capacity and protects against any two failures occurring within the RAID set, rather than just the right two. Neither is incorrect, but spending money on something you don’t need is.

// questions

Asked before you ask, answered.

Half. Each block is duplicated (one to each side of a mirror) therefore 4 x 4 TB gives you 8 TB of usable space, not 16. That 50% penalty is what you pay for having a live second copy of everything; this is why RAID 10 is used for its ability to be fast as well as resilient, although not for storage capacity.

Both types of RAID optimise differently. RAID 5 uses less disk space for each additional level of redundancy compared to RAID 10, and provides protection against any single failure. RAID 10 uses more disk space per level of redundancy compared to RAID 5 and is also able to provide better performance during write operations, significantly shorter rebuild times, and protect against simultaneous failures provided those failures are confined to different mirrored pairs. When dealing with high volume database applications and/or large-scale VM environments, RAID 10 typically makes sense. When trying to maximise storage capacity on a budget for general-purpose storage RAID 5/6 may be more suitable.

This depends entirely upon which 2 you lose. Losing one drive from each mirror allows you to access the entire dataset because there is an additional complete copy of the data present in the remaining half of the mirrored set. In most cases this will allow us to recover what was lost, with many systems continuing to run. Losing both drives from the same mirror results in that portion of the data having no surviving copy within the array. Recovery of this portion of the data would require us to directly repair/extract data from the two original failed drives. A lab is designed specifically for this type of work. Regardless of the outcome: shut off the system immediately and do not attempt to perform a rebuild.

It shortens the dangerous window, as long as you’re careful about it. There’s nothing like having a “hot” (or available) spare drive. Once one member fails, rebuilding starts almost right away. This reduces how many hours your two drives (the mirror set) will be left without being protected by a third drive. On RAID 10 the rebuild process itself is simply copying data from one drive to another. That means no big headache in terms of rebuilding all the drives simultaneously when one drive fails. However, there is some risk involved. Automatic rebuilds occur based on what the controller thinks has happened. Therefore, if a drive appears to have failed but still has life in it, you could potentially do a rebuild against its equally tired twin at the absolute worst time. Spare drives work best for arrays which are monitored. No amount of spare drives replaces good monitoring.

// degraded array?

Stop before the rebuild — then call.

Our Belfast lab offers a complimentary diagnosis for multi-disk arrays; each member is imaged as read-only, the original array is built back using these images and a written quote provided prior to undertaking any repairs.

Call us — 028 9002 0144
Mon–Fri · 9am–5:30pm · No fix, no fee
Start a free diagnostic →
028 9002 0144