RAID Information

General Tech

MapInfo / GIS

Oracle / Database

Misc / Useless

RAID Information

Redundant Array of Inexpensive Disks - provide data redundancy in case of disk failure. Most common are RAID 0, 1 3, and 5. Except for RAID 1, writes tend to be slower and reads faster since the RAID controller spreads data over several physical drives. The striping of data, besides balancing I/O, also allows you to create larger logical files than the operating system will usually allow. However, it also will not allow a single datafile to be placed on a single physical drive - will make tuning more complex and recover time longer (all disks involved as part of logical RAID device will need to restored, not just single physical disk that went down). The storage of ECC's require up to 20% more disk space. The OS will handle all aspects of implementing RAID. Datafiles and archive logs can be placed on RAID devices, but redo logs shouldn't since they are accessed sequentially (however, they should be mirrored).

RAID Levels
Level	Description
0	stripe (spans) data cross multiple disks - get no data redundancy, but you get the best speeds be writing and reading from only a single logical disk at a time - requires at least 2 disks - great for adding disk capacity to a server without adding a new volume name
1	disk mirroring - provides the most redundant fault tolerance available - requires twice as much disk space, no parity information is involved so corrupt data will be mirrored also - writes are no faster or slower - makes sense only for mission critical information - requires at least 2 disks.
2	redundancy is provided through ECC's (error correction codes) a.k.a. parity - these are calculated for each chunk of data and can be used to recreate the data - requires less disk space than mirroring, but makes writes slower
3	parity checking or error-correcting (ECC's) are calculated - stripes data across multiple disks at the byte level and adds redundancy by using a dedicated parity disk that recreates the data after a disk failure - costly since need to dedicate an entire disk to parity information - good for applications like video that send and receive large files since it writes data in big chunks - requires at least 3 disks to implement
4	same as RAID 3, but with block level parity
5	most popular solution - ECC's provide redundancy, but parity information is stored with the data, thus it is striped across several physical disks - is the most redundant of the striping levels because if one disk fails the parity information on the other can work together to rebuild it - writes data at the block (not byte) level so it is better suited to muliple small transactions like email, word processing - requires at least 3 disks to implement
6	ECC's are calculated in a more complex way - redundancy is provided with an advanced 2-dimensioanl parity.

RAID systems and what they cost
Level	Description	Extra Cost	Performance Change Over RAID Level 0
0	Hardware striping only	Little	N/A
1	Disk mirroring, no striping	100%	lack of striping is costly but can speed up small reads
0+1	Combination	100%	none - some systems get a slight improvement by using both disks for reads and/or using the disk with the closest head to the block for reads
3	Bit-level redundancy	20%	writes are slowed significantly, reads somewhat - parity disk can become a bottleneck
5	Block-level redundancy	20%	writes are slowed significantly, reads somewhat - parity is distributed so the parity disk is not a bottleneck
6	P+Q Redundancy	40%	large writes are slower