RAID HARDWARE On board SATA RAID controller SATA RAID controller card RAID drive caddy (hot swappable) Anne Watson 1
RAID The word redundant means an unnecessary repetition. The word array means a lineup. A Redundant Array of Inexpensive Disks (RAID) is essentially where a network file server uses extra disks to keep multiple copies of data. (Recently the word 'inexpensive' has been replaced by 'independent'.) In these situations, the data itself is more valuable than the cost of the disks. That s why the disks are termed inexpensive. Making repeated copies of the valuable data isn t actually unnecessary; still, that s what the computer people called it, using the word redundant. The concept of RAID data protection has been around long enough that there are now six different levels of protection (0 through 5). Each level indicates a higher degree of protection. Dynamic Disks/Volume Sets One of the important ways Windows 2000 and XP differs from Windows 98 in its use of disk space is that several partitions can be joined together in a single, logical area called a volume set. These partitions can be on a single physical disk, or located on several physical disks running off different controllers. Another interesting thing about a volume set is that it can include different types of disks. A volume set can have a maximum of 32 different partitions. When data is written to the volume set, the process takes place in a sequential fashion. When the first partition in the volume set fills up, the data is written to the next partition, and so on down the line. Neither the system partition nor the boot partition can be part of a volume set. Volume sets combine free space in partitions on 1 to 32 disks, writing data in sequential order. Striped partitions Striped partitions are partitions of equal size on one or more physical disks, joined into a logical area called a set. Striped sets provide no fault tolerance and can be set up using anywhere from 2 to 32 partitions on one or more physical disks. Striped sets are also known as RAID 0. Striped sets with parity combine 3 to 32 partitions, where one of the partitions is used as the parity stripe. Striped sets with parity do provide fault tolerance, and are also known as RAID 5. Striped Sets A striped set is somewhat like a volume set because data is written across different physical disks. The partitions on each disk are equal in size, giving rise to the term stripes. Unlike a volume set, data can be written to whatever disk is sitting idle, making for a more efficient (and faster) I/O process. Much like random access read/ writes take place on different platters of a hard drive, striped sets allow for something akin to random access read/writes to whole sets of disks. Striped sets can be set up on anywhere from 2 to 32 physical disks. One of the problems with striped sets is that if any of the disks in the set fails, all data on that particular physical disk will be lost. This kind of setup is called no fault tolerance because the system has no tolerance at all for any faulty machinery. Boot partitions can be mirrored or duplexed, but not striped. System partitions cannot be used in a striped set. Striped sets provide better read/write efficiency by writing to whichever disk is idle at the moment. Anne Watson 2
Striped Sets with Parity Striped sets can be configured either with or without parity. Striped sets with parity can be set up on anywhere from 3 to 32 partitions and work in a fashion vaguely similar to the parity circuits used to test main memory. In this configuration, one of the partition stripes in the disk array (group, or set) is set up as a parity check. If any of the disks in the set fails that data can be reconstructed using the parity information in the parity stripe. An error message will be generated, but it continues to function until the system administrator finds a convenient time to replace the failed disk. This requires shutting down the network, so a convenient time would be when network use is very low. Striped sets with parity can be set up on anywhere from 3 to 32 partitions. Disk Mirroring With volume sets and striped sets, many physical disks are set up to act as a virtual, single entity. Different data is written to any of the partitions within the two types of sets. Disk mirroring, on the other hand, is where two partitions are mirrors of each other. Each partition contains an exact copy (not actually reversed like a real mirror) of every piece of data on the other mirrored partition (also called a mirrored set). We call it disk mirroring because these two partitions are almost always on two separate disks. The advantage of disk mirroring is that if one physical disk fails, no reconstruction is necessary. The network continues operating using the data from the mirrored (constantly updated backup) copy of whatever information was on the failed disk. An error message, anwill be generated and, once again, the failed disk drive can be replaced at a more convenient time for the system administrator. Disk Duplexing Disk duplexing is almost the same as mirroring, except each disk is connected to a different drive controller. One of the problems with both mirroring and duplexing is that any amount of data requires twice the physical resources. In other words, rather than having room for two files on two disks, this data redundancy means having one file on two disks, with an exact copy on each disk. On the other hand, this means less chance of losing any data. Mirrored sets are set up on two physical drives using a single controller, as with an IDE or SCSI controller. If the two drives use separate controllers, this is called duplexing or disk duplexing. Anne Watson 3
RAID 0 RAID level 0 increases the logical drive capacity by treating two or more of the drives as a single logical drive, but includes only one copy of data. Therefore it does not let you recover from failure. Raid 0 is one method of disk striping in which more than one hard drive is treated as a single volume and data is written across more than one physical drive. For an array of two hard drives, some data is written to one hard drive and some to the other; both drives make up one logical drive, or volume and data is only written once. If one hard drive fails, data on the entire logical drive can be lost Windows NT, Windows 2000 and Windows XP support RAID 0, but not Windows 9X. However, you should know that RAID can be implemented with a hard drive controller without the involvement or knowledge of the operating system. This method is called hardware RAID; the method by which the operating system manages the RAID array is called software RAID. Anne Watson 4
RAID 1 RAID Level 1 is designed to protect data from a hard drive failure by writing the data twice, once to each of two drives. One type of RAID Level 1 is called disk mirroring in which the two hard drives both use the same controller. The advantages of RAID 1 disk mirroring are: 1. If either drive fails, the data is still safe on the other drive 2. Disk reads are speeded up because the system has two places to read from. The disadvantages of this type of RAID 1 are: 1. Having two drives in the same computer is more expensive. 2. Because the data is written twice, hard drive capacity is, in effect, cut in half. 3. Disk writes are slowed down because data must be written twice. One weakness in drive mirroring is that both drives are using the same controller. If the controller fails, neither hard drive is accessible. An improvement on disk mirroring, also considered RAID 1, is duplexing, in which each hard drive has its own controller on its own adapter card. The initial cost is higher, but if one adapter fails, you are protected from losing data on both drives. Anne Watson 5
RAID 2 RAID 2 is similar to striped sets, but the data is moved at a bit level, rather than in 64KB blocks. One of the disks in RAID 2 is set aside for data recovery. This method of data protection is currently unavailable through Windows NT. RAID 3, 4 RAID 3 is disk striping with parity, with data being moved at the bit level rather than the block level. RAID 4 is disk striping at the block level, with one disk reserved for parity recovery. RAID 4 uses four drives grouped as one virtual drive, the data divided into three segments. Each segment is written to one of the first three drives, and parity information is written to the fourth drive. Anne Watson 6
RAID 5 RAID 5, disk striping with distributed parity, is currently the most popular RAID implementation and provides optimum fault tolerance. RAID 5 is an interesting variation of RAID 0 combined with RAID 1 and RAID 4, and improves fault tolerance and drive capacity. RAID 5 requires at least three hard drives. RAID 4, disk striping with parity is not often used today, but to understand RAID 5 you must understand how RAID 4 works. RAID 5, rather than storing all the parity data on a single, dedicated parity drive, distributes parity information over all the drives, avoiding the performance bottleneck created by having a single parity drive. Anne Watson 7
OVERVIEW Anne Watson 8