On 05/12/14 03:35 PM, Pascal Hambourg wrote:
You can think of the RAID algorithms as parity checks. A mirror is even parity. While the disks are not physically assigned to be data or parity, you can recreate a failed RAID 5 disk by recalculating the parity based on the surviving disks.Hello, Some mistakes in what you wrote. Gary Dale a écrit :RAID 1 and RAID 5 are both immune to single disk failures in their most common configurations (1 or more data disks with 1 parity disk). RAID 10 is also immune to single disk failure but uses half the disks for parity.RAID 1 and 10 are just mirrors, they have no parity. I guess you mean "redundancy". RAID 5 does not use data disks and parity disks. Data and parity are distributed among all disks in the array. RAID 1 with N disks can survive N-1 disk failures.
If you have different sized disks, yes. The more usual case is to use similar disks. If one disk is not striped, you lose some of the performance improvement. RAID 10 with two disks makes little sense.If you are concerned about availability, with 4 disks (the simplest RAID 10 configuration)Linux can use a special RAID 10 mode (mirror+stripe) with two or three disks.
You misread the sentence. You can run RAID with any number of parity disks by tweaking the algorithms. Most people don't bother using more than 2 parity disks but there is no theoretical reason why you couldn't, to get as much safety as desired. Prior to being able to boot to an mdadm RAID 5 array, I regularly had 3 to 5 disk RAID 1 /boot partitions - why not use the disks since they are already there and it keeps the partitioning the same across the drives.with 6 disks, RAID 6 will give you double the capacity of 4 disks or get you immunity to 3 disks failing.RAID 6 can survive 2 disk failures regarless of the number of disks in the array.
RAID 6 can be considered a tweaked RAID 5 and RAID 5 can be considered a tweaked RAID 1.