RAID Levels

RAID Levels

RAID levels 0, 1, and 5 are the most commonly available hardware RAID solutions for Windows NT systems, and NT includes them. Asterisks denote RAID levels available for NT only through hardware controllers or combinations of hardware and software.

RAID 0 is ideal for environments in which performance (read and write) is more important than fault tolerance or you need the maximum amount of available drive capacity in one volume. Drive parallelism increases throughput because all disks in the stripe set work together on every I/O operation. For greatest efficiency, all drives in the stripe set must be the same capacity. Because all drives are used in every operation, RAID 0 allows for single-threaded I/O only (i.e., one I/O operation at a time). Environments with many small simultaneous transactions (e.g., order entry systems) will not get the best possible throughput.

RAID 1 is useful for building a fault-tolerant system or data volume, providing excellent availability without sacrificing performance. (NT does not support a RAID set as the boot volume--this capability requires a hardware controller). However, you lose 50 percent of assigned disk capacity. Read performance is somewhat higher than write performance because NT reads data off the drive whose head is closest to the desired sector (called locality of reference); all write operations are made to both disks simultaneously.

RAID 3 offers the performance of RAID 0 with the fault tolerance of RAID 5, but with cautions. As with RAID 0, RAID 3 uses single-threaded I/O but with a standard 256KB data transfer block. Streaming data applications such as video-editing systems benefit simultaneously from disk parallelism, fault tolerance, and the large-block transfers. But database applications or others with many small transactions issued simultaneously might have problems.

RAID 2 is similar in concept to RAID 3 but writes in single bits to the drives in the stripe set rather than in blocks, thus requiring many drives for a volume (however, few--if any--implementations are available for NT).

RAID 4 is similar to RAID 3 but uses a round-robin algorithm for writing data to the stripe set. Because RAID 4 has to generate parity data on one drive, write operations are single-threaded, but reads are multithreaded (i.e., are performed via multiple simultaneous I/O requests) because not all drives are involved in every transaction. RAID 4 is better for an environment requiring more discrete transactions per second. Both RAID 3 and 4 use the minimum number of drives to achieve fault tolerance.

RAID 5 is multithreaded for both reads and writes because both normal data and parity data are distributed round-robin. This is one reason why RAID 5 offers better overall performance in server applications than either RAID 3 or 4. Random I/O benefits more from RAID 5 than does sequential I/O, and writes take a performance hit because of the parity calculations. RAID 5 is ideal for database applications.

RAID 6 is essentially RAID 5 enhanced, such that two drives in the stripe set can fail because the distributed parity information has its own parity and is thus redundant.

You can build RAID 10 either directly through the RAID controller (depending on the controller) or by combining software mirroring and controller striping, or vice versa (called RAID 01).