What storage type stripes data and performs a parity check of data over multiple disks that can recover from a hard disk failure?

Redundant Array of Independent Disks (RAID)

A RAID is a method of combining two or more hard drives in a format that enhances reliability or performance. There are many types of RAID, but RAID 0 (zero) and RAID 5 arrays (see Figure 9.13) are most popular because they can “stripe” information across multiple drives for maximum performance. When you’re upgrading, consider purchasing a computer with built-in RAID functionality (usually listed as a feature of the motherboard), which allows you to use RAID formatting from at least a pair of identical hard drives. Although many systems use RAID 0 arrays, they are very risky. If one of the drives in a RAID 0 array fails, they all fail. See Table 9.3 for a list of available RAID levels.

Table 9.3. RAID Formatting Variations

RAID

RAID uses your disk subsystem to provide enhanced read/write performance, protection against data lost due to disk failures, or both. It can be implemented using hardware specific RAID controllers (Hardware RAID) or functionality embedded within the operating system (Software RAID).

Did You Know?

RAID subsystems require two or more disks to form one virtual disk. The differences between software and hardware RAID solutions are as follows:

Hardware-based RAID performs faster than the software-based RAID implementation because software-based RAID requires more CPU time and has additional memory requirements.

Software-based RAID is operating system dependent and hardware-based RAID is vendor independent.

Hardware-based RAID may be more expensive than software-based RAID as it may require additional hardware components (for example, RAID controllers), unless incorporated on your motherboard.

RAID technology requires an understanding of three basic concepts: striping, mirroring, and parity. Striping joins the hard disk drives together to form one large disk drive. For example, three 300 MB drives joined together in a striping array will form a single 900 MB drive. Striping evenly writes data across all of the disks contained in the array and will evenly read data from all disks contained in the array, increasing overall disk subsystem performance. The downside to striping is that it does not provide any fault tolerance.

Mirroring forms one disk drive whose size is determined by the size of the smallest drive, and the same data is written to both drives, providing a level of redundancy in the event one drive crashes. The disadvantage of mirroring is the impact of having to record the same data twice across two different drives which reduces the disk subsystem performance. Parity stores information in the disk array subsystem that can be used to rebuild files or lost data in the event one of the disks in the disk subsystem array fails. Unlike striping and mirroring, parity requires a minimum of three disks inside the disk array subsystem. Each of these techniques is assigned a different RAID level.

RAID-6

RAID-6 (block-level striping with double distributed parity) provides fault tolerance up to two failed drives. This makes larger RAID groups more practical, especially for high-availability systems. This becomes increasingly important as large-capacity drives lengthen the time needed to recover from the failure of a single drive. Like RAID-5, a single drive failure results in reduced performance of the entire array until the failed drive has been replaced and the associated data rebuilt. This RAID is also used in some CCTV recorders.

RAID 5

RAID 5 is disk striping with parity. With this level of RAID, data is striped across three or more disks, with parity information stored across multiple disks. Parity is a calculated value that's used to restore data from the other drives if one of the drives in the set fails. It determines the number of odd and even bits in a number, and this information is used to reconstruct data if a sequence of numbers is lost, which is the case if one of the disks fail.

Because RAID 5 requires parity information to be calculated, the performance of your server can decrease when data is written using this level of RAID. However, performance can increase when data is read, because the data is read from multiple disks at once. Since there can be up to 32 disks making up RAID 5 array, this can be a significant advantage. However, when a disk fails, the performance decreases because the server must reconstruct the data from parity information on the other disks.

RAID 5 also offers better disk utilization than RAID 1.When you mirror a disk, you must purchase twice the amount of disk space that you'll actually use for data. Fifty percent of the total disk space is used for redundancy. With RAID 5, the amount of space used for parity is equal to one disk in the array, so the more disks you use, the higher your percentage of disk utilization. For example, if you have 10 disks in the array, only one-tenth of the total disk space is used for redundancy and nine-tenths of the space is available for data.

17.5.1.3.3 RAID

RAID (Redundant Array of Independent Devices) arrays use software to combine numerous drives together to make them appear as a single device. There are several levels of RAID that emphasize increasing performance, enhancing redundancy, or both.

Implementing a RAID array for your application isn’t without certain drawbacks. Some potential downsides of implementing a RAID array are:

It’s another layer of complexity that needs to be managed. If neither you nor your organization has experience with RAID arrays, then this is an additional area of expertise that needs to be acquired.

It isn’t free. There is a cost to the additional disk drives and the software that manages a RAID subsystem. Be certain that the benefits outweigh the costs for your installation.

3.2.6 Hierarchical RAID

RAID structures can be organized hierarchically. Before explaining this organization, we introduce one level of RAID that does not include any redundancy: RAID level 0.

Level 0 is an arrangement, where the data are interleaved between two disks. There is no parity bit or other form of redundancy here, just parallelism. (In that sense the term RAID is misapplied here: the array of disks contains no redundancy.) The purpose is to increase data throughput to and from the disks by allowing parallel accesses.

Let us now consider hierarchical organization. The nomenclature for the level is Level ij, Level i/j or Level i+j, denoting that the disks consist of a RAID Level j organization of units, where each unit is organized as a RAID Level i entity.

Example: RAID level 50 (see Fig. 3.26) consists of a RAID level 0 organization of units, each unit consisting of a RAID level 5 structure. So, given a file consisting of segments 0,1,2,3,⋯, we would assign segments 0,2,4,⋯ to be stored in one group of disks, whereas 1,3,5,⋯ would be stored in the other. Each of these two groups is organized as a RAID Level 5 structure.

Similarly, RAID31 consists of a RAID level 1 organization of units, each unit consisting of a Level 3 structure. The Level 1 organization is mirroring, so Level 31 consists of two mirrored RAID level 3 entities, i.e., two Level 3 entities, which are copies of each other. Each of these organizations has its own distinctive characteristics.

Example: Note that the striping is not directly “visible” from outside the RAID group. Thus for example, if a file is striped inside a RAID group, the individual stripes are not separately accessible from outside. This makes a difference in the speed with which recovery can be effected.

To see why, let us compare RAID10 and RAID01. Suppose each of these arrangements contains six disks in all. In RAID10 (Fig. 3.27), these six disks would be organized into three groups of mirrored disks (RAID 0 at the lower level; RAID 1 above). In RAID01 (Fig. 3.28), we would have two groups at the higher level mirroring each other; each of these groups consisting of three disks.

Suppose Disk 0 fails in RAID01. The lower-level controller cannot, by itself, fix the problem, since it does not have any redundant data within its reach. It has, instead, to obtain the data from the controller of the other low-level group.

By contrast, suppose Disk 0 fails in the RAID10 configuration. The lower-level controller in charge of that disk can resolve the problem by itself: when a new disk is inducted into the system, this controller simply copies its (good) disk, i.e., Disk 1, into the new one. The volume of data copied is therefore much less in RAID10 than for RAID01. This translates into a quicker recovery from failure; note that the system is vulnerable to a second failure while it is still recovering from the first. As disk sizes increase and the recovery time increases, this difference may be significant.

Redundant Array of Independent (or Inexpensive) Disks

Acquiring an entire redundant array of independent (or inexpensive) disks (RAID) set disk by disk and then reassembling them in EnCase is probably the easiest way to deal with a RAID and may be the only way to capture a software RAID. Hardware RAIDs can be most efficiently captured using a boot disk. This allows the capture of a single volume that contains all unique data in an array. It can be trickier to configure, and as with everything, practice makes perfect. Be sure you understand how it works. The worst thing that can happen is that the wrong disk or an unreadable disk gets imaged and the job has to be redone at your expense.

Creating a RAID 5 volume

The process of creating a RAID-5 volume is very similar to creating a mirrored volume. This time you will need to ensure that you have at least three disk drives to be included in the RAID array. Once the drives are installed, you are ready to create the RAID-5 volume.

Log on to the server, open Server Manager, and expand Disk Management.

You may need to bring the newly installed drives online by right clicking on each drive and choosing the online option.

Once the disk drives are online, they can be initialized by right clicking on each drive and choosing Initialize.

You can now establish a new RAID-5 volume by right clicking on the first drive to be used in the array and choosing New RAID-5 volume as seen in Figure 2.19.

The New RAID-5 Volume Wizard will launch. Click Next to begin.

Add the drives you want to include in the RAID-5 volume, using the Select Disks page (see Figure 2.20). Remember that you will need to include three or more disk drives to create a RAID-5 volume. After you have selected the disk drives to use, click on the Next button to continue.

Select a drive letter to be used and click Next.

Choose the option to Perform a Quick Format and optionally give the new volume a meaningful label and then click Next.

At the summary page of the wizard, verify the settings you have selected, and then click on the Finish button to create the RAID-5 volume.

You will be prompted to convert the drives to dynamic. Click Yes to convert the disks to dynamic.

You have now established a newly formatted RAID-5 volume as seen in Figure 2.21.

Notes from the field

Resilience using RAID

Probably the most common type of resilience, especially on a server, is Redundant Array of Independent Discs (RAIDs).26 There are several forms of RAID and we consider two (levels 1 and 5) here. RAID level 1 is a simple disc image, as per replication above – only in real-time. The replication is handled by the RAID controller (a disc controller with additional functionality) which writes any information to both discs simultaneously. If one disc fails, then the other can be used to keep the system operational. The failed disc may then be replaced (often without halting the system – known as “hot swapping”) and the RAID controller builds the new disc into a copy of the current primary over a period of time, depending on the amount of data held and the processing load.

Other forms of RAID do not replicate the data directly, but spread it across several discs instead, adding in some error correction as well. The form of spreading (known as “striping”) and the type of error correction are different for each level of RAID.

RAID 5 uses block-level striping and parity data, spread across all discs in the array. In all disc storage, the disc is divided up into a set of blocks, a block being the smallest unit of addressable storage. A file will therefore occupy at least one block, even if the file itself is only one byte in size. A read or write operation on a disc will thus read or write a set of blocks. In RAID 5 these blocks are spread across several discs, with parity data stored on another one (see Fig. 8.2). Thus RAID 5 always requires a minimum of three discs to implement.

Parity, which we met in “encryption”, is a computer science technique for reducing data corruption. It was originally designed for data transmission and consisted of adding an extra bit to the data.27 This extra bit forced the sum of the bits to be odd (odd parity) or even (even parity). It was added at transmission and checked at reception (although this would only detect one error).28 More complex forms used more bits which enabled the data not just to be better checked, but also to be corrected. Such codes are known as Hamming codes.29 The parity used within RAID 5 not only checks that the data is correct, but enables it to be re-built, should a disc fail and have to be replaced.30 A RAID controller is also able to recreate the missing data in real time, resulting in minimal degradation of service.

In the figure, the distribution of the blocks and the parity can clearly be seen. Distributing the parity blocks distributes the load across all the discs, as this is where bottlenecks may appear (to read blocks B3 to C1, for example, also requires two parity blocks to be read – in this case each disc only has one read operation to perform). RAID 5 has found favour as it is viewed as the best cost-effective option providing both good performance and good redundancy. As write operations can be slow, RAID 5 is a good choice for a database that is heavily read-oriented, such as image review or clinical look-up.

As has been mentioned, a RAID controller can keep a system operational even when a disc has failed, so much so that users may not notice. It is therefore imperative to monitor such clusters as one failure may be easily fixed but two may be catastrophic.

Traditionally computer systems and servers have stored operating systems and data on their own dedicated disc drives. With the data requirements expanding it has now become more common to have separate large data stores using NAS or SAN technology. These differ in their network connectivity but both rely on RAID for resilience. NAS uses TCP/IP connections and SANs use Fibre Channel connections.