In this Daily Drill Down, I’ll discuss disk management strategies for the Windows NT Server. There are several different strategies from which you can choose. Depending on your particular situation, you can opt for speed over redundancy (or vice versa), or you can opt for maximum storage capacity over speed and redundancy. I’ll give some examples so that you can decide what’s best for you.

Different types of configurations
I’ll be discussing RAID (redundant array of independent disks) throughout this Daily Drill Down. There are many different types of RAID configurations, but I’ll focus primarily on RAID 0, RAID 1, and RAID 5.

Disk striping without parity (RAID level 0)
Disk striping without parity (RAID level 0) is a disk management strategy that provides no data redundancy. It isn’t fault-tolerant, either. It requires a minimum of two disks, and it can be formatted with FAT or NTFS partitions. Unlike RAID 5, which stripes the disks with parity, there’s no parity stripe with RAID 0, and data can be written across all disks evenly. RAID 0 is the fastest disk strategy available. By allowing concurrent requests to be processed on all drives simultaneously, it offers the highest level of read-and-write performance of any available disk management strategy. The downside to this strategy is that, if one disk in the array goes bad, all data will be lost.

This particular disk strategy is best suited for scientific analysis or imaging, where you can tolerate compromised system reliability. The system and boot partitions can’t become part of a stripe set without parity. A good backup routine will help minimize the loss of data and build in some indirect redundancy for RAID 0.

Disk mirroring (RAID level 1)
Disk mirroring (RAID level 1) supports only two hard drives. Mirror sets are the only form of fault tolerance that can include system and boot partitions. Using this method, you run both drives off of the same controller. With disk mirroring, data will be written to both hard drives simultaneously, which degrades the overall performance. However, if one disk fails, the other continues to provide data redundancy. If the disk controller fails, then data isn’t written to either drive.

Disk duplexing is similar to mirroring, except that it uses two disk controllers instead of one (a hardware enhancement vs. a software enhancement). If you use disk duplexing, you can lose a drive or controller and continue working.

Disk mirroring is the least cost-effective disk management strategy because you lose half of your disk capacity. Of course, if both disks fail, all data would be lost, so it’s important that you back up the data. Disk mirroring and disk duplexing are strategies that should be applied in situations when redundancy is needed, but speed and cost are not the main focus. If you have only two disks, then mirroring is the only disk strategy with redundancy.

Disk striping with parity (RAID level 5)
Disk striping with parity (RAID level 5) provides fault tolerance. It requires a minimum of three physical disks, and it can use as many as 32 disks. All partitions in a stripe set are the same size. Before you create a fault-tolerant set, however, perform a full backup of all data. Back up anything that can’t be reinstalled and back up any data that you can’t regenerate on physical media.

If you have a stripe set that contains five disks, the parity information is written across all disks. Parity information starts at the first stripe set on Disk 0, continues at the second stripe on Disk 1, the third stripe on Disk 2, and so on until the fifth stripe on disk 4. After that, it starts over again on Disk 0.

With software RAID, information about fault-tolerant sets is kept in the registry. It’s preferable that you back up the disk key in the system hive separately. You can do so by using the Save option of the Configuration entry under the Partition menu in Disk Administrator. This option will copy the disk key in an uncompressed format to a floppy disk. The disk key keeps the disk signature and any fault-tolerant set information off of the drives in your system.

If you select free disk areas of different sizes when you create a stripe set, no stripe will be larger than the smallest free disk area. For example, if you have 200 MB, 400 MB, 600 MB, and 800 MB free on four drives, respectively, only 200 MB will be used on each drive. The entire stripe set will be 800 MB in size. The space that’s equivalent to one partition will be used for parity information. So, in this case, 25 percent of the total available disk space is used for storing parity information (200 MB), and only 600 MB of data can be stored on the stripe set.

Regardless of how many disks are used in a stripe set with parity, data will be recoverable only if no more than one disk is lost. If two or more disks are lost, the data will be unrecoverable. Therefore, even with a redundant disk strategy, a good backup routine is very important. This strategy is not as fast as RAID 0 because the parity information must be written—making everything slower. Currently, it’s the most widely used method of providing hard disk redundancy. It works well for Microsoft Exchange Servers, for Microsoft Systems Management Servers, and for any situation where you must have fault tolerance but can’t sacrifice speed.

The pagefile shouldn’t be placed on a stripe set with parity because redundant data will be written and performance will degrade. After a stripe set is created, you can’t enlarge or extend it unless you back up the data and reformat the set. You can’t incorporate the Windows NT boot files into a stripe set.

Windows NT Server support for RAID is software-based; information about the configuration is kept in the registry. Hardware-based RAID has many advantages over software-based RAID, including support for hot swapping hard drives. I’ll discuss hardware-based RAID in more detail below.

Volume sets
A volume set allows you to combine the free space of as many as 32 disks and to create a single volume with a single drive letter that’s transparent to the user. Volume sets provide no fault tolerance; if even one area of disk space in the set is lost, all data on the disks will be lost. Backing up data on a volume set is essential. Volume sets comprise the only Windows NT disk-partition management option that allows more than one area of disk space in the set to reside on the same physical hard disk. Furthermore, volume sets comprise the only Windows NT disk-partition management option that allows the individual areas of disk space, which are making up the volume, to be of different sizes.

The main benefit of volume sets is that they offer the most efficient use of hard disk space. System and boot partitions can’t become part of a volume set, but all other partitions can. Since only one drive performs disk access at a time, a volume set is the slowest access method. Mainly, this disk strategy is used to maximize hard disk utilization. Instead of having hard disks run at less than full capacity, you can implement a volume set to take advantage of stray sections of available space. This disk strategy is best when fault tolerance isn’t necessary but maximum use of hard disk space is. The pagefile shouldn’t be placed on a volume set because performance degrades.

Other RAID configurations
There are several other RAID configurations. These configurations aren’t widely used, but RAID 10 is growing in popularity. It’s implemented as a striped array (RAID 0) with segments that are mirrored arrays (RAID 1), and it has the same fault tolerance that RAID 1 does. It’s an excellent solution for sites that have to use mirrored sets but must have an additional performance boost.

High I/O rates can be achieved with this method. All drives must move parallel to the proper track—a circumstance that lowers sustained performance. This configuration has a very limited scalability and a very high inherent cost. It takes a minimum of four disks to implement this configuration.

A word about hard disks
Each time that you double the number of disks, you double the probability that you’ll have a disk failure. Implementing a RAID configuration doesn’t prevent disk failures; RAID is intended to prevent system failures in the event of a disk failure.

All RAID levels have a theoretical mean time between failures (MTBF) that is worse than the MTBF of a single drive system. The advantage is that a drive failure won’t cause the system to quit working immediately. Generally, all drives perform well in early life and in the absence of error conditions. Within both SCSI and IDE, significant variation can be found in performance and reliability among manufacturers. While drive selection is crucial to reliability, this important decision is based too often on cost. Make sure that you consider all factors when you’re selecting a drive.

A look at Disk Administrator
Disk Administrator is the graphical tool that Windows NT gives you in order to manage your hard disks. It allows you to create partitions, extend partitions, format drives, create stripe sets, create stripe sets with parity, create mirror sets, create and extend volume sets, and assign drive letters. To open to Disk Administrator, choose Start | Programs | Administrative Tools | Disk Administrator. Your version of Disk Administrator should look similar to the one that appears in Figure A. It will show all of your physical hard drives and the partitions on each. You can set up Disk Administrator to use colors or patterns in order to identify different disk strategies. Your CD-ROM and Zip or Jazz drive will appear in Disk Administrator. Windows NT Disk Administrator doesn’t support the disk strategy RAID 10.

Figure A
Your Disk Administrator should show all of your physical hard drives and partitions.

Hardware-based RAID
Hardware-based RAID provides better performance because it runs independently of the operating system and doesn’t need to scan a series of drivers. The operating system can be placed on a hardware-based RAID because the array is built in the controller BIOS, not by a driver that the operating system supplies. Many hardware-based RAID configurations allow for hot swapping drives. If a drive goes bad, you can replace it without shutting the server down. There is a degree of risk when you hot swap drives. If you lose more than one drive in a hardware-based stripe set, all data will be lost. Backing up your data is as important with a hardware-based RAID as it is with a software-based RAID.

Many vendors, such as Dell and Compaq, provide software utilities that manage the hardware-based RAID. I’ll show you Compaq’s Array Configuration Utility and let you contrast it with Windows NT Disk Administrator. Figure B shows the logical drive view in the Configuration Utility that Compaq provides. Figure C shows the physical drive view.

Figure B
The logical drive view will appear in the Configuration Utility.

Figure C
The physical drive view will also appear.

Controller Settings
The Controller Settings screen (see Figure D) lets you select the operating system that you’re using, the Rebuild Priority and Expand Priority options, and the memory percentages for the Array Accelerator’s read and write cache.

Figure D
The Controller Settings screen allows you to select several options.

  • Operating System (see Figure E): Optimizes the number of heads, sectors per track, and cylinders for each logical drive. Select the operating system that you’re currently using or the one that you plan to use.
  • Rebuild Priority: Reflects the importance of rebuilding data after a failed drive has been replaced.
  • Expand Priority: Affects how the controller moves data after you’ve chosen to expand the capacity of an array.
  • Accelerator Ratio: Determines the amount of memory that’s allocated to the read and write caches. The options that are listed for Maximum Read and Write are dependent upon the type of controller that you’re configuring and the amount of memory that’s installed on the Array Accelerator.

Figure E
The Operating System selection optimizes several parts of each logical drive.

Modify Drive Array
This is a graphical way of expanding the array and of assigning drives to the current stripe set, as shown in Figure F. It’s similar to Windows NT Disk Administrator. Click on the drive that you want to assign and click Assign Drive(s) To Array A. Not all hardware-based RAID configuration utilities are graphical. Some must be accessed during the boot process before Windows is launched.

Figure F
You can expand the array and assign drives to the current stripe set graphically.

There are many different types of disk management strategies to consider. Each has advantages and disadvantages. If you need data redundancy, choose RAID 1 (disk mirroring) or RAID 5 (disk striping with parity). If speed is your only concern, RAID 0 (disk striping without parity) may be your best choice. To utilize all of the disk space that may be scattered on several disks or partitions, you may want to implement a Volume Set. There are several other RAID configurations—such as RAID 10—that you could consider. The more built-in redundancy and speed, the higher the cost will go. Regardless of which strategy you implement, make sure that you have a good backup routine in place, too.

Troy Thompson, MCSE+I, has worked in the automation field for 15 years, and he has dealt with a variety of systems, including Wang OIS, Unisys BTOS, UNIX, Windows 3.11, Novell NetWare, Windows NT 3.51, and Windows NT 4.0. He’s worked as an administrator of a Novell and an NT network and as a systems analyst for an IBM mainframe. Currently, Troy is the Information System Security Officer at the Information Management shop at Fort Knox. If you’d like to contact Troy, send him an e-mail.

The authors and editors have taken care in preparation of the content contained herein, but make no expressed or implied warranty of any kind and assume no responsibility for errors or omissions. No liability is assumed for any damages. Always have a verified backup before making any changes.