7.10. RAID 5-Distributed Data Guarding

7.10. RAID 5 Distributed Data Guarding

RAID 5 distributed data guarding, also referred to as stripe sets with parity, breaks data up in chunks, calculates parity, and then writes the data chunks in "stripes" to the disk drives, saving one stripe on each drive for the parity data. The total amount of disk space used for redundancy (parity data) is equivalent to the capacity of a single drive.

RAID 5 distributes the parity information among all disk members. By doing so, it eliminates the performance bottleneck at the parity disk, and if one drive fails, the failed disk can be re-created after it is replaced. This is illustrated in Figure 7-10.

Figure 7-10. RAID 5 distributed data guarding.

RAID 5 is based on the RAID 4 principle of generating new checksums when write operations occur. Unlike RAID 4, parity data is not stored on a dedicated drive but is distributed evenly across all drives.

RAID 5 requires a minimum of three drives. Available disk space is the sum of the size of all disks minus the size of one disk. Five 9GB disks produce a logical drive size of (5 1) times 9GB, which is 36GB.

A disadvantage of RAID 5 is that only a single drive can fail at one time without loss of data. If a second drive fails, before the failed drive has been replaced and recovered, the data will be lost.

7.10.1 RAID 5 Performance

RAID 5 is the most cost-effective of the fault-tolerant RAID solutions, but is slower in performance than RAID 1+0 and loses additional performance when a member disk is missing.

RAID 5 performs well in environments where the I/O profile consists predominately of read requests. Read operations can occur in parallel and because of distributed parity, all drives are available to service read operations. RAID 5 is often used in database transaction processing and multiuser file services environments.

Because the array controller needs to perform an average of four I/O operations for each write request, RAID 5 has a write overhead penalty that slows the overall performance of the system. For multiple write operations, parity updates can occur in parallel because of the interleaved parity.

These four I/O operations, illustrated in Figure 7-11, are as follows:

1. The controller must read the old data block (the data block that is about to be overwritten).

2. The controller must read the old parity block. (This parity block must be updated to reflect the new data.)

3. The controller must recalculate and write the new parity information.

4. The controller must write the new data block.

Figure 7-11. RAID 5 I/O operations.

Where RAID 1 mirroring requires a 100% increase in capacity to protect the data, RAID 5 requires as little as 50% (2 + 1), and commonly only 20% (5 + 1). The larger the RAID 5 set, the more burdensome it is to manage and the more readily performance could suffer.