Hi David101, unfortunately you are not correct there. Raid 0+1, you /can/ loose two drives and not loose data, but in reality, because of the way the extra load is put on the second disk in the pair, you will more often loose data after the second disk failure.
DiddyDave. There is nothing wrong with taking lots and lots of backups. You can never be too safe. At the moment, I am considering having my main data on a raid 5 system, as my main working system. Then I will take a backup to an external disk drive. Once a month, I am going to clone that external drive to an offsite area (well, in fact I am going to retrieve the offsite disk drive, and swap it with the current external one). This need not be expensive, consider a family member or a friend who doesn't mind you popping in once a month. The reasoning behind this is if you are burgled or the premises are burnt down. Also, each month, I can check the consistency of the offsite backup
On the raid side of things, a quick explanation. These explanations are mainly based on /hardware/ raid controllers. Software raid controllers may not offer all of the benefits (hardware raid is a separate device inside the computer, software raid the operating system performs the tasks). Also assumed that all disks are the same size. With hardware raid based systems, you can usually take disks out and replace them whilst the machine is still in use. (not with a raid 0 of course)
Raid 0. This offers no redundancy, but allows you to read and write data faster. Data is interlaced between disks, You can have multiple disks in the raid set (2+). The speed benefit is gained by having twice as much cache local to the disks. Cache being a buffer memory based on the disk controller itself. Often called stripe raid. Storage size is about the size of all the disks added together.
Raid 1. Mirror. Data is sent to both disks at the same time, in a mirror. If one disk fails, the other is used immediately. Often raid controllers will continually/schedule checks to ensure that the data on the disks match, and alert you if there are any errors likely to be encountered (not an awful lot it can do to fix them though). Mirrors have to be in pairs.
Raid 0+1. This is a stripe of mirrors. This gives you the resiliency of a mirror, you can loose a disk, but with the extra speed performance of a stripe. However, experience shows that should one disk fail, the other disk in its mirror, starts to get a lot more use. This can cause this disk to fail quicker than the other two disks left, meaning that you will more often than not, loose data on a second disk failure. For raid 0+1, you have half the disk space, minimum requirement of 4 disks, any even number above that is valid.
Raid 3. This raid system is no longer used (in preference to raid 5 if it exists), it is however easier to explain raid 3 then raid 5. With raid 3, the computer writes the data to all of the disks less 1. So if you have 4 disks in a raid 3, the data would be written to disk 1, 2, 3 . For disk 4, a checksum of the data is written. This checksum allows the computer to calculate whether any of the items, 1, 2, 3 have changed, and if they have, it can re-create the value that they had (conversely, of the data on disk 4 has changed, it can recreate that). With a raid 3, if one disk fails, then there is no loss of data. You have to have a minimum of 3 disks. You loose one disk space for the parity information.
The downside found with raid 3, was that when a disk fails, the amount of extra usage put on the disk for the parity calculations, caused it to fail quicker.
Raid 5. This is a similar theory to raid 3. The difference is that instead of there being 1 disk reserved for parity, the parity is interspersed with the standard data. Similar to the below (A,B.C being the data, P being the parity).
ABCP
BCPA
CPAB
PABC
Most (proper) hardware raid systems will also perform consistency checks of the data, and should there be a problem showing with the disk, they will copy the failing disk to a spare disk within the system, usually called a hot spare. It isn't infallible, but around 80% of the returns I see are from systems which have correctly determined an upcoming failure and automatically copied and replaced the failing disks..
