Managing RAID on Linux - Derek Vadala [6]
Figure 1-2. Parity redundancy is accomplished by storing recovery data on specified drives.
Degraded
Degraded describes an array that supports redundancy, but has one or more failed disks. The array is still operational, but its reliability and, in some cases, its performance, is diminished. When an array is in degraded mode, an additional disk failure usually indicates data loss, although certain types of arrays can withstand multiple disk failures.
Reconstruction, resynchronization, and recovery
When a failed disk from a degraded array is replaced, a recovery process begins. The terms reconstruction, resynchronization, recovery, and rebuild are often used interchangeably to describe this recovery process. During recovery, data is either copied verbatim to the new disk (if mirroring was used) or reconstructed using the parity information provided by the remaining disks (if parity was used). The recovery process usually puts an additional strain on system resources. Recovery can be automated by both hardware and software, provided that enough hardware (disks) is available to repair an array without user intervention.
Whenever a new redundant array is created, an initial recovery process is performed. This process ensures that all disks are synchronized. It is part of normal RAID operations and does not indicate any hardware or software errors.
Striping
Striping is a method by which data is spread across multiple disks (see Figure 1-3). A fixed amount of data is written to each disk. The first disk in the array is not reused until an equal amount of data is written to each of the other disks in the array. This results in improved read and write performance, because data is written to more than one drive at a time. Some arrays that store data in stripes also support redundancy through disk parity. RAID-0 defines a striped array without redundancy, resulting in extremely fast read and write performance, but no method for surviving a disk failure. Not all types of arrays support striping.
Figure 1-3. Striping improves performance by spreading data across all available disks.
Stripe-size versus chunk-size
The stripe-size of an array defines the amount of data written to a group of parallel disk blocks. Assume you have an array of four disks with a stripe size of 64 KB (a common default). In this case, 16 KB worth of data is written to each disk (see Figure 1-4), for a total of 64 KB per stripe. An array's chunk-size defines the smallest amount of data per write operation that should be written to each individual disk. That means a striping array made up of four disks, with a chunk-size of 64 KB, has a stripe-size of 256 KB, because a minimum of 64 KB is written to each component disk. Depending on the specific RAID implementation, users may be asked to set a stripe-size or a chunk-size. For example, most hardware RAID controllers use a stripe-size, while the Linux kernel uses a chunk-size.
Figure 1-4. Stripe-size defines the size of write operations.
The RAID Levels: An Overview
Patterson, Gibson, and Katz realized that different types of systems would inevitably have different performance and redundancy requirements. The Berkeley Papers provided specifications for five levels of RAID, offering various compromises between performance and data redundancy. After the publication of the Berkeley Papers, however, the computer industry quickly realized that some of the original levels failed to provide a good balance between cost and performance, and therefore weren't really worth using.
RAID-2 and RAID-3, for example, quickly became useless. RAID-2 implemented a read/write level error correction code (ECC) that later became a standard firmware feature on hard drives. This development left RAID-2 without any advantage in redundancy over other RAID levels. The ECC implementation now required unnecessary overhead that hurt performance.