FAULT TOLERANCE

Fault tolerance involves storing data on multiple hard drives to insure against the failure of any individual drive. There are several levels of fault tolerance, referred to as RAID (Random Array of Inexpensive Disks) levels. However, both A+ and Microsoft tests deal with the three that are most commonly implemented.

RAID LEVEL 0 – DISK STRIPING

It is actually a misnomer to refer to this type of disk striping as fault tolerant. It consists of at least two hard disks. Beginning on the first disk, data is written in stripes that span all disks in the set, starting over again on the first disk as needed until each disk is full. No error correction is used; therefore, if one disk in a stripe set crashes, the data on that disk is unavailable. While this method will increase read/write performance, it will not provide fault tolerance.

RAID LEVEL 1 – MIRROR SET

This level uses exactly two hard drives and one disk controller. The second drive(X’) is an exact duplicate, or mirror, of the first drive(X). If one drive crashes, the data is still available on the second drive.

A variation of RAID 1 called disk duplexing uses two hard drives and two disk controllers. With a mirror set, neither disk is available if the controller fails. Two controllers adds an extra measure of hardware-based fault tolerance.

Mirror sets and disk duplexing are both used to protect boot or system partitions, which cannot be part of a stripe set. The drawback to both is hardware redundancy: the expense involved in installing a second hard drive, and, in the case of disk duplexing, an additional controller as well.

RAID LEVEL 5 – STRIPE SET WITH PARITY

This method uses three or more hard disks to create a stripe set like that used in RAID 0. The difference is that some disk space is used for parity information. This results in a loss of storage space, but provides for fault tolerance: in the event one disk fails, the data it contained can be reconstructed from the parity information on the other disks.

RAID 5 can only handle the loss of one disk. However, the odds of two disks failing at the same time are highly unlikely. A stripe set offers very effective data protection for less overhead than RAID 1 requires. The space required for parity information is 1/X, where X is the number of hard drives in the stripe set. In a stripe set using three disks, for example, 1/3 of the total disk size will be reserved for parity information. In a five-disk stripe set, 1/5 of the total disk space is used for parity information.