i would say RAID5 going by size, co$t, speed, data availability(redundancy) type of use etc...
to Repeat: RAID is not a backup; please always have at least 1 VERIFIED backup..
OS Array:
Non-Parity RAID (0,1,10) favored for OS so that overhead of parity calculations on WRITEs for CONSTANT winRegistry and virtual mem/paging file don't bog the system (like they would in RAID5,6,50,60 for any writes or degraded array reads).
RAID0 Array:
RAID is Redundant Array etc.; so RAID 0 is kinda an oxymoron, in that it is the nonRedundant-Redundant Array of Indep. Disks; It is the only RAID level that doesn't give higher data availability (just speed and space increased).
non-OS Array:
For non-OS Arrays (apps, data, database) we can have many more reads, than writes (so not calculating parity unless array is degraded / missing HD). So, changes things a bit. Also, some databases specifically are more setup to read from a stripe across 0,5,6,10,50,60 would be reading stripe across (not RAID1 or 01). If feeding a data base, that has it's own software caching on, it is best to turn off the hardware RAID caching thru the RAID controller.
co$t of Array:
RAID 5 can be cheapest redundant array to deploy space wise (RAID0 cheaper, but not redundant).
maintenance of Array:
backup, verified. Updates Reset to pristine redundancy/mirror/parity monthly to comb out any bad blocks before having a HD fall offline. In single fault tolerant array scheme, would want rest of array pristine; when a HD fails/falls offline. A RAID controller set to narrower tolerances, could be a better controller, demanding more; but seem to have more failed or just HD's fallen offline..
amount of HD's in stripe across of array:
Reading across a stripe will get faster with wider stripe(more hd's) slowing down in RAID5 about HD8 (9th drive), as the overhead form the parity calculation becomes so enormous; assume this paradigm is hit earlier in double parity RAID6 types. The more HD's in array stripe, the greater chance 1 will fall offline and have to be rebuilt back into the sequence of the array. BUT Also: the more HD's in array, the greater chance that when a hd fails and there is a bad block of data, that it will not be on the failed HD, and thus PUNCTURE single fault tolerant array.
Double Redundancy in array:
RAID6 can be more in vogue on HUGE array, or more critical; but not speed is double fault tolerant. When in rebuild, take more chance of double fault on fault tolerant array (increasing risk/exposure) in a larger array rebuild. Larger multi level arrays(10,50,60etc.), can have even more fault tolerance to face running day to day risks, as well as rebuild faults.
maintenance X amount of HD's in array:
More HD's give greater chance of 1 falling offline X larger chance of puncturing array = more risk/care in handling of larger array.
RAID1 does not provide n/2 but 1/n storage space. The data is the same on every disk, there is no "half of the space for mirroring". If you have 5 disk of 2TB, the space you'll have will still be 2TB (but you'll have a high fault tolerance). – m4573r – 2012-10-29T12:49:50.930
@m4573r Yes, using RAID 1 with more than 2 disks and no striping you would effectively mirror the same data over all disks. I modified my answer accordingly. – speakr – 2012-10-29T13:20:05.683