Managing Microsoft Windows Server 2003 Disk Storage : Understanding Disk Storage Options

Physical Disks

Physical disks are the conglomeration of plastic, metal, and silicon that enable users to store enormous quantities of useless data and MP3s, and the occasional business document. Of course I’m being sarcastic here, but it is important to understand the difference between the physical disk, and its logical volume(s), which are discussed in the next paragraph. It is also helpful to remember that an advanced disk subsystem, such as hardware-based redundant array of independent disks (RAID) system, may consist of several physical disks, but its dedicated hardware controllers abstract the physical composition of the disk set so that Windows Server 2003 perceives and represents the disk system as a single physical disk.

Logical Volumes

A logical volume is the basic unit of disk storage that you configure and manage. A logical volume may include space on more than one physical disk. Logical volumes (also called logical disks in the context of performance monitoring) are physically distinct storage units, allowing the separation of different types of information, such as the operating system, applications, and user data. Logical volumes have traditionally been represented by a single drive letter.

As you dig into disk-related terminology, you will learn about partitions, logical drives, and volumes. Many resources will use all these terms interchangeably, which is possible because the technical distinctions between the terms are minuscule, and the user interface and command-line tools guide you clearly by exposing only the appropriate type of logical volume based on the task you are performing. Don’t get too hung up on the distinctions between the terms; they will become clear through experience if not through analysis.

Mounted Volumes

You noticed that we said, “Logical volumes have traditionally been represented by a single drive letter.” That structure severely limited (to 26, says my kindergarten teacher) the number of volumes you could create on a system, and the flexibility with which those volumes could be used. Windows Server 2003’s NTFS file system allows you to assign one or no drive letter to a volume. In addition, you can mount a volume to one or more empty folders on existing NTFS volumes. For example, you might create an empty folder Docs, on an existing volume with the drive letter X:, and mount a new 120 GB logical volume to that folder. When users navigate to X:\Docs, the disk subsystem redirects the input/output (I/O) requests to the new volume. All of this is transparent to the user.

The possibilities using this powerful feature are, as they say, “limitless.” By mounting a volume to a folder path, you can extend the available drive space on an existing volume. If the existing volume is not fault-tolerant, but the new volume is fault-tolerant, the folder to which the volume is mounted, X:\Docs, represents a fault-tolerant portion of the existing volume’s namespace. You could, theoretically, mount all logical volumes on a server to folders on the server’s C or D drive and thereby unify enormous storage capacity under the namespace of a single drive letter.

Fault Tolerance

Fault tolerance refers to a system’s ability to continue functioning when a component—in this case, a disk drive—has failed. Windows Server 2003 allows you to create two types of fault-tolerant logical volumes: mirrored (RAID-1) and striped with parity (RAID-5).

• In fault-tolerant disk configurations, two or more disks are used, and space is allocated to store data that will enable the system to recover in the event of a single drive failure.

• The fault tolerance options supported by Windows Server 2003 do not provide a means for a disk volume to continue functioning if two or more disks fail.

• The operating system allows you to use any two or more disk drives to create fault-tolerant volumes. You do not have to purchase any additional hardware or software to benefit immediately from fault-tolerant server configurations. However, if you use Windows Server 2003 mirrored or RAID-5 volumes, it is best practice to use similar or identical disk drives on the same bus. Combining a variety of disk hardware, or using drives connected to a variety of small computer systems interface (SCSI) or Integrated Device Electronics (IDE) buses can affect performance significantly.

• Speaking of performance, Windows Server 2003 fault tolerance is using processor cycles and other server resources to manage the volumes. RAID-5 can be particularly detrimental to server performance. It is possible, and affordable these days, to purchase hardware-based fault-tolerant disk arrays, known as hardware RAID. Hardware RAID uses dedicated controllers to manage fault tolerance, and such systems are generally faster and more flexible in both management and recovery than is Windows Server 2003 RAID.

• Because hardware RAID controllers offload the management duties from the operating system, a hardware RAID array appears to Windows Server 2003 as a single disk.

Separation of Data

It is a good idea to analyze storage requirements carefully before configuring the disk subsystem of a server. Administrators typically elect to install the operating system on a logical volume separate from applications and data. By isolating the operating system, it is easier to secure the operating system volume and to manage disk space so that the volume does not run out of space. It is also usual to configure some kind of fault tolerance for the operating system.

Applications are generally stored in a separate volume, and user data and files in a third. Again, isolation of data types allows you to manage security, performance, and fault tolerance separately for each data type. If an application uses a transaction log to prepare entries into a database, as do Microsoft Active Directory directory service and Microsoft Exchange Server, it is typical to store those logs in volumes that reside on physical disks separate from the database itself, allowing the application to rebuild the database from the logs if the database fails.

Once you have thoroughly analyzed your storage requirements as they relate to the data type, security, performance, and fault tolerance, you can begin to determine how many disks you require and how those disks should be configured.

Basic and Dynamic Disks

An operating system must have a way to make sense of the physical space on a disk drive. There are two structures that Windows Server 2003 can apply to help it apportion and allocate drive space: basic and dynamic storage, also called basic and dynamic disks.

Basic Disks, Partitions, and Logical Drives

Basic disks maintain the structure with which you are probably most familiar. Each basic disk is partitioned, and each partition functions as a physically separate unit of storage. The information about the location and size of each partition is stored in the partition table of the Master Boot Record (MBR) on the drive. A basic disk can contain as many as four partitions, consisting of either four primary partitions or three primary partitions and one extended partition.

The logical volumes on a basic disk are primary partitions and logical drives. The logical volume, as mentioned, can be represented by zero or more drive letters and can be mounted to folders on an existing NTFS volume.

• Primary partition Each primary partition maintains one logical volume on a basic disk. If a basic disk is used to start the operating system, one and only one primary partition on the disk must also be marked as active.

  Tip

  The computer’s basic input/output system (BIOS) looks to the active partition to locate the hardware-specific files required to load the operating system. That partition is technically referred to as the system partition and is usually assigned drive letter “C”. Once the boot process has begun, the operating system is loaded. Most servers are configured with the operating system on the C drive as well. The partition on which the operating system is stored is called the boot partition. Yes, it can get confusing, particularly because the same volume is referred to by the variable %Sysvol%. Fortunately, it’s not a distinction you’re likely to need to know, since most installations are completely on drive C, making the C drive the system partition, the boot partition, and %Sysvol%.

  
  

• Extended partition A basic disk may also contain an extended partition. Unlike primary partitions, extended partitions are not formatted or assigned drive letters. Instead, extended partitions are further divided into logical drives. Logical drives are logical volumes on a basic disk.

In earlier versions of Microsoft operating systems, including Windows 95, Windows 98, and MS-DOS, the operating system could only “see” the primary partition on which it was installed, plus the extended partition on the drive, if one existed. If you wanted additional storage segments on the drive, you had to configure an extended partition and apportion it into one or more logical drives. Because Windows NT, Windows 2000, Windows XP, and Windows Server 2003 can access all partitions on a disk, you only need an extended partition if you want more than four logical drives on a single disk.

Dynamic Disks and Volumes

Microsoft Windows 2000, Windows XP, and the Windows Server 2003 family also support dynamic storage. The storage units on dynamic disks are called volumes, and the first distinctions between basic and dynamic storage are that dynamic disks support an unlimited number of volumes, and that the configuration information about the volumes is stored in a database controlled by the Logical Disk Manager (LDM) service.

The logical volume of dynamic disks is the volume. Dynamic disks support simple volumes on a single disk. When a computer has more than one dynamic disk, you are provided more storage options from which to choose. Spanned, mirrored (RAID-1), striped (RAID-0), and striped with parity (RAID-5) volumes are logical volumes that utilize space on more than one physical disk. Each volume type uses disk space differently, and is characterized by a different level of fault tolerance.

• Simple volume The equivalent to a basic disk partition is a dynamic disk simple volume. Simple volumes utilize space on a single physical disk, and correspond to a single logical volume. Simple volumes can be extended by appending unallocated space on other regions of the same disk, allowing you to adjust a volume’s capacity with the growth of data stored in that volume. Because simple volumes exist on only one physical disk, they are not fault-tolerant.

• Spanned volume A spanned volume includes space on more than one physical disk. Up to 32 physical disks can participate in a spanned volume, and the amount of space used on each disk can be different. Data is written to the volume beginning with the space on the first disk in the volume. When the space on the first disk fills, the second disk is written to, and so on. Spanned volumes provide an option for increasing drive capacity. If a simple or spanned volume is filling up, you can extend the volume onto additional new storage capacity.

  But spanned volumes are not fault-tolerant, and cannot participate in any fault-tolerant configurations. Because their size tends to be greater, and because multiple physical disks are involved, the risk for failure increases. If any one disk in a spanned volume is corrupted or lost, data on the entire volume is lost as well. For these reasons, Windows Server 2003 will not allow the installation of the operating system on a spanned volume, nor can you extend or span the system volume. Spanned volumes are recommended only as a stop-gap measure when an existing volume fills to capacity, or else in situations where tolerance for failure is high—for example, a large library of read-only data that can easily be restored from tape backup in the event of failure.

• Striped volume A striped volume (RAID-0) combines areas of free space from multiple hard disks into one logical volume. Unlike a spanned volume, however, data is written to all physical disks in the volume at the same rate. Because multiple spindles are in use, read and write performance is increased almost geometrically as additional physical disks are added to the stripe. But like extended simple volumes and spanned volumes, if a disk in a striped volume fails, the data in the entire volume is lost.

• Mirrored volume A mirrored volume (also known as RAID Level 1, or RAID-1) consists of two identical copies of a simple volume, each on a separate hard disk. Mirrored volumes provide fault tolerance in the event that one physical disk fails.

• RAID-5 volume A RAID-5 volume is a fault-tolerant striped volume. Space on three or more physical disks is unified as a single volume. Data is written to all physical disks at the same rate, but unlike a striped volume, the data is interlaced with checksum information, called parity. Should a single disk in the volume fail, the data on that disk can be regenerated through calculations involving the remaining data and the checksum information. It is an interesting technical note that parity is distributed among all volumes in the RAID-5 set.

Basic vs. Dynamic Disks

So now that you know about basic and dynamic storage, and the types of partitions, logical drives, and volumes they support, which is better? The answer, as is frequently the case, is: “It depends.”

Dynamic disks that store data are easily transferred between servers, allowing you to move a disk from a failed server to a functioning server with little downtime. Dynamic disks flex their muscle when there is more than one dynamic disk in a computer. Each Windows 2000, Windows XP, and Windows Server 2003 computer can support one disk group, which itself can contain multiple dynamic disks. The LDM database is replicated among all disks in the disk group, which increases the resiliency of disk configuration information for all the group’s disks. In addition, disks can be configured to work together to create a variety of flexible and powerful volume types including spanned volumes, striped volumes (RAID-0), mirrored volumes (RAID-1), and striped-with-parity volumes (RAID-5).

Basic disks will continue to be used, however, for several reasons:

• Basic storage is the default in Windows Server 2003, so all new disks are basic disks until you convert them to dynamic.

• Dynamic disks do not offer advantages over basic disks in a computer that will have only one disk drive.

• The behavior of the LDM database also makes it difficult to transfer a dynamic disk used for starting the operating system to another computer when the original computer fails.

• Dynamic disks are not supported for removable media, and are not supported on laptops.

• Basic storage is the industry standard, so basic drives are accessible from many operating systems, including MS-DOS, all versions of Microsoft Windows, and most non-Microsoft operating systems (there are a few). Therefore, dynamic disks cannot be used if you need to dual-boot an earlier operating system that requires access to the disks. Keep in mind that we are talking about local access only. When a client of any platform accesses files over the network, the underlying storage and volume type are transparent to the client.

Tip