Grouping Disks: Stripe Sets and RAID

---

While a PC normally just has a single disk, in enterprise applications separate physical disks are frequently grouped logically for added performance, size, and reliability. Disks can be grouped in several ways to achieve this:

• Stripe sets — This approach involves using several disk drives and splitting the data between these drives to improve performance. Because the hard drives are slower than other system elements, using multiple drives speeds up the data access by allowing simultaneous access to parts of the file from more than one drive.

• Logical drives — Drives can also be grouped logically so that the operating system treats them as a single drive. For example, three 40-GB drives could be combined logically so that the operating system considers them as a single 120-GB drive.

• Disk duplexing — Two disks are operated by separate controllers and the data to both disks is simultaneously recorded for reliability and performance. When one disk goes down, the other can still be used for failover. A performance boost can be gained by setting it up so that a seek operation is sent to whichever disk offers the quickest turnaround at that time.

In addition, the various Redundant Arrays of Independent Disks (RAID) layers can be used. RAID is basically a disk subsystem that increases performance or provides some means of fault tolerance. The subsystem is composed of two or more disk drives as well as a controller. RAID is used mainly on servers but is now starting to find applications in PCs, as well.

Some of the various RAID levels are:

• RAID level 0 — Uses striping for added performance but has no data redundancy. RAID-0 writes data to more than one drive to speed the process. Information is striped to one disk and then to others, back and forth, so the data is distributed among multiple disks. If one drive fails, the data is lost, as no redundancy is built into RAID-0

• RAID level 1 — Uses full disk mirroring; that is, the exact same data is put on two or more disks. Read speeds are higher than for a single disk, as more than one disk can be read at the same time. Write speeds, however, are the same as for a single disk, as all the data is written to each of the disks. RAID-1 is expensive in that twice the storage space must be used to achieve redundancy. In comparison, RAID-3 and RAID-5 are less expensive.

• RAID level 2 — Uses striping for performance and stores error checking codes (ECCs), which allow data errors to be detected and corrected; this is a relatively rarely used method.

• RAID level 3 — Combines the speed advantage of striping, as for RAID-0, but with the addition of redundancy. Rather than doing full mirroring as in RAID-1, however, it takes the data from two disks, XORs it, and places the result on a third disk. In this way, it requires half as many disks as RAID-1 to store the redundant data. The acronym XOR stands for "exclusive-OR": A RAID system takes one bit from each of the two data disks. If both bits are the same (i.e., both 0 or both 1), it will record a 0 on the redundant disk; if the two bits are different (i.e., one is 0 and the other 1), it records a 1. This is also known as parity.

• RAID level 4 — Does not use striping; stores parity bits on a separate disk. Like RAID-2, this method is rarely used.

• RAID level 5 — Although similar to RAID-3 in terms of striping and replication, RAID-5 differs in that the redundant data is interspersed with the user data on the same disks rather than being stored on separate disks. The read speed is similar to that for RAID-0 and RAID-3, but the write speed is slower because both the user data and the redundant data must be written to the same disks. This is the most widely used RAID level

• RAID level 10 — Actually RAID-1,0, in that it utilizes both striping (RAID-0) and mirroring (RAID-1), RAID-10 combines the improved performance of striping with the fault tolerance of mirroring.

The levels of RAID most commonly used for reliability are levels 1, 3, and 5. Several other RAID levels exist in addition to the ones listed above. Full specifications on all the RAID levels are available from the RAID Advisory Board.

---