RAID Data Recovery

RAID 5 Example

Raid 5 arrays are built for performance and redundancy. Instead of reading/writing 1 file to 1 disk, RAID 5 arrays read/write 1 file to multiple disks in parallel.

They also calculate a parity block when writing so that data can be recovered if a disk fails. The following is a simple example illustrating how RAID 5 works.

Say we have a text file that consists of three characters "ABC" and a RAID 5 array of three disks with a stripe/block size of four bits (equal to one half of a character).

Here's a data representation of our text file:

Here's what happens when we save the file:

The first block of four bits (first half of letter A) is written to Disk 1 and the second block of four bits (second half of letter A) is written to Disk 2. Disk 3 is parity for Block 0. Next, the third block (first half of letter B) is written to Disk 1 and the fourth block (second half of letter B) is written to Disk 3. Disk 2 is parity for Block 1. Finally, the fifth block (first half of letter C) is written to Disk 2 and the sixth block (second half of letter C) is written to Disk 3. Disk 1 is parity for Block 2. Notice how the parity is on a different disk for every block. The order of the parity is called the parity order, parity map, or parity rotation. In this example, the parity order is backwards (Disk 3, Disk 2, Disk 1).

*Pink data represents the parity calculation for the block

Now consider what happens when Disk 3 fails and all we have left is Disk 1 and Disk 2. Continue reading about RAID Arrays at one of our informational pages listed below.

As you can see, the raw data from Disk 1 and Disk 2 represents the string "Dqd" which is nonsense. The data should be the string "ABC". To recover the data from the two remaining disks, we must properly reassemble the stripes and use the XOR operator to recalculate the lost data from Disk 3. The following chart illustrates how to properly reassemble the RAID 5 array:

As you can see, the RAID 5 reconstruction process is quite complex. The reconstruction must be done exactly right or the resulting data will be useless.