Groping On Trains And Busses

Berkely The paper defines five types of RAID implementation with the addition of a technique called non-redundant striping RAID 0.


Striping: disks are logically combined in one unit.

RAID 0. This is an array but not redundant, then the name was incorrect RAID but universally accepted. This is actually concatenated disk striping: instead of in sequence, the data capacity of each disk is partitioned into small functional units called stripes and joined together in an overall logical unit that includes all disks stripig. The process of interleaving (pass me the term taken from the world of memories) is possible through a technique called round-robin. The size of the stripes is variable and configurable. RAID 0 offers no fault tolerance but also significantly increases the performance since the read and write operations can occur simultaneously on all disks. The minimum number of disks required to implement RAID 0 is two.



Raid 0: drives combined round-robin improve the transfer since more I / O occur simultaneously satisfied by all disks in the array.


RAID 1: disk mirroring. Two disks are configured as a copy of the other and then record both the same data. Redundancy is assicuratadal fact that in the event of failure of one, the other disk can continue to operate by providing access to data. The performance improved in reading since the applications are fulfilled simultaneously by all the disks, while in writing remain the same, having to write to both disks. the minimum number of disks required to implement RAID 1 is two.
Raid 1: both discs recorded the same data: it is given a writing at a time but multiple reads simultaneously satisfied by all disks in the array.


RAID 2. In this old configuration of at least three disks, two are striped, and one contains information ECC fault tolerance. Because now each disk contains its own error correction code this implementation has been abandoned. Furthermore, the data are arranged (ie, distributed by a method of spanning) among all the disks and reading and writing should therefore continue to access data arranged in different positions, moving from one disk to another and slowing the flow of throughput.



Raid 2: each operation (reading and writing) is distributed among all disks. The record ECC records information of fault tolerance.


RAID 3. It 'similar to RAID 2, except that instead of having more units that handle information ECC, there is a drive that records only the parity bit. If a drive fails the missing stripe can be restored by calculating the XOR (parity information) of stripes positioned equally on other disks and restoring the data perduto.Per able to achieve this feature requires that the recorded data is smistino among all disks in the array in order to have a transfer of shares of each record in parallel, maximizing speed. However since the size of the stripes should be small, smaller than a typical record, should be strengthened demands of long sequential data except that the continuous parallel access to disks slow down operations, also allows only RAID 3 operation I / O time wing, limiting Workgroups environment with many users.


example equal

Disk 1 Disk 2 Disk 3 equal

1 + 3 + 6 = 10
2 + 1 + 7 = 10 + 2 + 3
x = 10

parity - disk1 - Disc 2
x = X = 5 (since restored)



Raid 3: each operation is distributed among all disks.


RAID 4. In this implementation, a minimum of 3 drives are the same features as RAID 3, except that the stripe is larger than a typical record, allowing him to reside entirely in a single drive array. This division of the records allows more read operations simultaneously, thus a greater flow of data. However, every write must update the parity drive specific so it can not be more than one time.



Raid 4: each write to update the parity. More I / O simultaneously.


RAID 5. With a technique called distributed parity, RAID 5 overcomes the limitations imposed by the single drive to RAID 4 for the equal, in dedicating each disk sectors. The total space distributed between disks dedicated to parity is equal to that of a single whole drive, allowing multiple write operations occur simultaneously, as well as readings. A slight delay is dall'update equality. In addition, the stripes can be defined in size, adapting the array to your working environment.



RAID 5 distributed parity. More I / O simultaneously.