Understanding Disk Terminology
Before going into the details of managing disks and storage, let's review some definitions:
- Physical drive The actual hard disk itself, including the case, electronics, platters, and all that stuff. This is not terribly important to the disk administrator.
- Partition A portion of the hard disk. In many cases, this is the entire hard disk space, but it needn't be.
- Allocation unit The smallest unit of managed disk space on a hard disk or logical volume. It's also called a cluster.
- Primary partition A portion of the hard disk that's been marked as a potentially bootable logical drive by an operating system. MS-DOS can support only a single primary partition, but Windows Server 2008 can support multiple ones. There can be only four primary partitions on any hard disk.
- Extended partition A nonbootable portion of the hard disk that can be subdivided into logical drives. There can be only a single extended partition per hard disk, but it can be divided into multiple logical drives.
- Extended volume Similar to, and sometimes synonymous with, a spanned volume. This is any dynamic volume that has been extended to make it larger than its original size. When an extended volume uses portions of more than one physical disk, it is more properly referred to as a spanned volume.
- Logical drive A section or partition of a hard disk that acts as a single unit. An extended partition can be divided, for example, into multiple logical drives.
- Logical volume Another name for a logical drive.
- Basic disk A traditional disk drive that is divided into one or more partitions, with a logical drive in the primary partition, if present, and one or more logical drives in any extended partitions. Basic disks do not support the more advanced functions of Disk Management, but they can be converted to dynamic disks in many cases.
- Dynamic disk A managed hard disk that can be used to create various volumes.
- Volume A unit of disk space composed of one or more sections of one or more disks. Prior versions of Windows Server used volume only when referring to dynamic disks, but Windows Server 2008 uses it to mean partitions as well.
- Simple volume Used interchangeably with partition in Windows Server 2008, earlier versions of Windows used simple volume only when referring to a dynamic disk. A portion of a single disk, a simple volume can be assigned either a single drive letter or no drive letter and can be attached (mounted) on zero or more mount points.
- RAID (redundant array of independent [formerly "inexpensive"] disks) The use of multiple hard disks in an array to provide for larger volume size, fault tolerance, and increased performance. RAID comes in different levels, such as RAID-0, RAID-1, RAID-5, and so forth. Higher numbers don't necessarily indicate greater performance or fault tolerance, just different methods of doing the job.
- Spanned volume A collection of portions of hard disks combined into a single addressable unit. A spanned volume is formatted like a single drive and can have a drive letter assigned to it, but it will span multiple physical drives. A spanned volume—occasionally referred to as an extended volume—provides no fault tolerance and increases your exposure to failure, but does permit you to make more efficient use of the available hard disk space.
- Striped volume Like a spanned volume, a striped volume combines multiple hard disk portions into a single entity. A striped volume uses special formatting to write to each of the portions equally in a stripe to increase performance. A striped volume provides no fault tolerance and actually increases your exposure to failure, but it is faster than either a spanned volume or a single drive. A stripe set is often referred to as RAID-0, although this is a misnomer because plain striping includes no redundancy.
- Mirror volume A pair of dynamic volumes that contain identical data and appear to the world as a single entity. Disk mirroring can use two drives on the same hard disk controller or use separate controllers, in which case it is sometimes referred to as duplexing. In case of failure on the part of either drive, the other hard disk can be split off so that it continues to provide complete access to the data stored on the drive, providing a high degree of fault tolerance. This technique is called RAID-1.
- RAID-5 volume Like a striped volume, a RAID-5 volume combines portions of multiple hard disks into a single entity with data written across all portions equally. However, it also writes parity information for each stripe onto a different portion, providing the ability to recover in the case of a single drive failure. A RAID-5 volume provides excellent throughput for read operations, but it is substantially slower than all other available options for write operations.
- SLED (single large expensive disk) Now rarely used, this strategy is the opposite of the RAID strategy. Rather than using several inexpensive hard disks and providing fault tolerance through redundancy, you buy the best hard disk you can and bet your entire network on it. If this doesn't sound like a good idea to you, you're right. It's not.
- JBOD Just a bunch of disks. The hardware equivalent of a spanned volume, this has all the failings of any spanning scheme. The failure of any one disk will result in catastrophic data failure.
Real World: Disk Technologies for the Server
The first time we wrote a chapter about disk management, basically three possible technologies were available: Modified Field Modification (MFM), Pulse Frequency Modulation (PFM), and Small Computer System (or Serial) Interface (SCSI). Unless you were a total geek (and had oodles of money), your systems used either MFM or PFM, and RAID wasn't even an option. Over time, SCSI became the only real choice for the vast majority of servers and even became mainstream on high-end workstations. Servers at the high end might use fiber, but SCSI had the vast majority of the server disk market. SCSI has changed over the years to support faster speeds, more disks, and greater ease of configuration and use, but is finally reaching its limits as a parallel interface.
Integrated Device Electronics (IDE), later called Advanced Technology Attachment (ATA), became the standard on the personal computer. However, IDE never made a serious inroad into the server market because, while fast for single tasks, it lacked the inherent multitasking support and bus mastering that a server disk interface technology required, and no real hardware RAID solutions supported it.
Recently, the introduction of Serial ATA (SATA) technology has made serious inroads into the lower end of the server marketplace. With SATA RAID controllers built into many motherboards, and stand-alone SATA RAID boards that support 8 or more SATA drives and have substantial battery-backed RAM cache onboard, many low- to mid-range servers are finding that SATA RAID solutions provide a cost-effective alternative to SCSI. While most SATA RAID controllers lack the ability to hot-swap a failed drive, and don't have the performance potential of SCSI or Serially Attached SCSI (SAS), they are still quite attractive alternatives where cost is a primary factor. SATA also makes sense as secondary or "near-line" storage for a server.
The new kid on the block, however, is SAS. This is the most interesting addition to the server storage equation in quite a while. Using the same thin cables and connectors as SATA, with none of the configuration nuisance of traditional SCSI, SAS is definitely the way to go. When combined with new 2.5-inch drives, the ability to put a really large amount of very fast storage in a small space has taken a significant step forward. SAS drives interoperate with SATA drives to combine the two technologies on the same controller. SAS disk controllers can control SATA drives as well, though the reverse is not true.
With the main bottleneck for servers continuing to be I/O in general, and especially disk I/O, there will continue to be pressure to find new and faster methods to access disk-based storage. SAS, combined with 2.5-inch drives, enables fast and flexible storage arrays in remarkably smaller spaces. Because 64-bit servers are the only real option, and because of the enormous datasets supported on 64-bit Windows Server 2008, the need for fast and easily expandable disk storage keeps increasing. Windows virtualization technology and the move to greater virtualization in the data center also drive the need for faster disk and I/O subsystems.
< Back Next >
© Microsoft. All Rights Reserved.