This is the level people prefer if they need redundancy and don’t want to pay the write penalty of parity, and the one that will behave most like expected should a failure occur, since there is nothing to do (i.e., “reverse”) from an arithmetic perspective.
Four disks can be paired up into two pairs where each member of the pair has identical data and data is striped across these pairs. Since there are two members per pair you lose fifty per cent of your raw storage capability and obtain fast write performance as well as simple predictable failure behaviour.
In order to create a RAID 10 configuration you first take the individual disks and pair them together, next you mirror each pair, and lastly you stripe the data across the mirrors. For example, with four total disks, you would have two pairs. With eight total disks, you would have four pairs. Each block of data will exist twice (once on each member of whatever pair contains it) and consecutive blocks will fall onto different pairs thereby allowing reads and writes to span across the entire set.
What makes this different from RAID 01 (which stripes first and then mirrors) is that RAID 10 disks are mirrored before they are striped. This results in greater fault tolerance since losing just one of the disks will affect only one mirror, as opposed to a whole stripe set being lost if one disk failed with RAID 01; therefore, many people deploy RAID 10.
For RAID 5 using parity, you can utilise the full capacity of all disks except one. In other words, four 4TB disks = 12TB compared to RAID 10’s 8TB. This is an advantage for RAID 5, and this advantage does exist. Disadvantages include: each time you write something to a RAID 5 volume, you will have to read out what is already written on your disks and the current parity information, then rewrite them both back. As such, RAID 5 performs poorly when under heavy write loads.
The major difference is how the arrays handle rebuilds. When rebuilding a RAID 5 array after losing one of its members, the system will read through every sector of all remaining member disks in order to recalculate lost sectors. During this process, there is maximum usage of these older disks. A RAID 10 rebuild involves copying data from the good mirror drive (the one with the matching sectors). Thus, none of the remaining drives in the array will be utilised during the rebuild, thereby creating a much shorter window of vulnerability. Since the days of smaller storage devices are behind us and we are now dealing with large storage devices (i.e., many terabytes per device), this difference becomes very important.
When you have a RAID 10 system using eight (8) hard drives, if you experience failures of up to four (4) hard drives and none of those failed hard drives were paired together as part of a mirror set, then your RAID will continue to function without interruption. If there is a failure on both parts of a mirror pair, then your volume will fail immediately. Therefore, once you experience a failure(s), the number of hard drives that fail does not matter; what matters is whether or not any two (2) of the failed disks are mirrored together.
The ability to recover from a RAID 10 failure is much easier compared to recovering from a RAID 5 or 6 because RAID 10 does not use parity, therefore when a drive fails, you do not need to calculate new parity information. What needs to be done in order to restore access to the data stored on the failed RAID 10 array, is determine which pairs of disks correspond to which mirrors, along with how they are striped across the disks. Once this is accomplished, provided one member of each pair is readable, you should be able to use its image to rebuild the full disk image of the missing disk. If both disks in a mirror pair fail, you merge the two images so that if a particular sector was clean on either image of the mirror pair, you would get the clean version of that sector.
All disks are imaged as read only and the configuration of the disks (i.e., which disks are paired together, the stripe width/size and order) determined by looking at the data itself, rather than trusting the configuration information provided by the RAID controller. Where both members of a pair are damaged, the two images are merged, which frequently produces a complete pair.
The main things to avoid are letting it sync back to a hard drive that you’ve reinserted, or creating the RAID again just to check if it works. We charge for all types of RAID/NAS/server recovery from £500 +VAT once we give you a free 48 hour diagnostic report and written quote.
A RAID 10 creates mirror sets, then uses these to create striped sets, so each block will be stored twice on its respective pair. Consecutive blocks will land on different pairs, so performance will distribute evenly throughout the RAID. Since this type of RAID doesn’t require any parity calculation when writing files, you’ll get much faster write speeds than the RAID 5 which does. You will lose approximately half of your raw capacity.
In terms of how many drives can fail at the same time before your RAID becomes completely unavailable, the answer depends upon how many mirrored pairs there are in total. For example, an 8 disk RAID has 4 mirrored pairs, therefore an 8 disk RAID can afford to lose up to 4 disks at the same time; but if both members of any of those 4 mirrors were to fail then the RAID would become unavailable regardless of how many other drives remained intact.
Depends on what you want to optimise. RAID 5 offers higher usable storage space. RAID 10 provides faster write operations as well as much more reliable rebuild processes. The reason rebuilds using RAID 10 are better is that they do not have to use each individual surviving sector when copying data. This means that rebuild time is generally going to be less than rebuild times experienced by RAID 5 systems. On larger modern disks this difference in rebuild time is typically greater than the potential loss due to reduced usable storage (capacity) afforded by the RAID 10 configuration.
Generally yes. Often easier or quicker than finding parity information on your disks. Without needing parity info to find out which pair a particular disk belongs to and how the pairs were ordered on the array, we can use the disks themselves to determine these items. If two members of a pair fail we can merge an image of one disk into another and then recover missing sectors from each disk.
For RAID, NAS and server recovery our costs begin at £500 + VAT depending upon the complexity of the recovery process. A fixed written quotation will always be provided after a free 48-hour diagnostic, prior to commencement of any work. Should it be necessary to physically repair any disk within a failed RAID assembly a deposit of 50% of total costs will need to be made prior to commencing. In such cases payment of the remaining amount will not be requested until successful completion of the recovery operation.