Database Reference
In-Depth Information
61218
24 …
DISK 1
DISK 5
DISK 6
1713
19 …
51117
23 …
2814
20 …
41016
22 …
DISK 2
DISK 4
3915
21 …
DISK 3
Figure 12-10
Striping units of data across a disk array.
10110101
The first parity bit written on the check disk depends on whether the total
number of 1s written as the first bit on the data disks is odd or even. In our case,
this number is 5. Because the number 5 is odd, the first parity bit is written as 1. If
the number were even, the first parity bit would be written as 0.
When a disk failure occurs, you must be able to recover from the failure. You
must be able to recreate the contents of the failed disk. Let us assume that the fourth
disk failed. After the failure, the recovery mechanism counts the number of 1s in
the first bit positions of all the disks other than the failed disk. This number of 1s is
4—an even number. Now examine the first parity bit on the check disk. It is 1. There-
fore, because of the way parity bits are written and because the number of 1s in the
first bit position of the remaining disk is even, the first bit of the failed disk must
have been 1. In this manner, by looking at the data in the remaining disks and
reviewing the parity data, the data in the failed disk may be completely recovered.
RAID systems may be implemented at different levels. The number of check disks
needed to write parity data depends on the RAID level.
RAID Levels
Let us discuss the implementation at different RAID levels for data
on four data disks. The disk array will then consist of these four data disks and a
certain number of check disks for parity data. As far as storage of data is concerned,
the check disks do not contribute to space utilization for data storage; the check
disks contain redundant data.
Level 0—Nonredundant
•
Just data striping
•
Best write performance
•
Reliability still a problem
•
RAID system consists of 4 data disks and no check disks
•
Effective space utilization 100%
Search WWH ::
Custom Search