A disk subsystem that includes a RAID configuration
enables the disks in the system to work in concert to improve
performance, fault tolerance, or both.
Implementing Disk Fault Tolerance
Fault tolerance is the ability of a computer or operating system to
respond to a catastrophic event, such as a power outage or hardware
failure, so that no data is lost and that work in progress is not
corrupted. Fully fault-tolerant systems using fault-tolerant disk arrays
prevent the loss of data. You can implement RAID fault tolerance as
either a hardware or software solution.
Hardware Implementations of RAID
In a hardware
solution, the disk controller interface handles the creation and
regeneration of redundant information. Some hardware vendors implement
RAID data protection directly in their hardware, as with disk array
controller cards. Because these methods are vendor specific and bypass
the fault tolerance software drivers of the operating system, they offer
performance improvements over software implementations of RAID.
Consider the following points when deciding whether to use a software or hardware implementation of RAID:
Hardware fault tolerance is more expensive than software fault tolerance and might limit equipment options to a single vendor.
Hardware fault tolerance generally provides faster disk I/O than software fault tolerance.
Hardware
fault tolerance solutions might implement hot swapping of hard disks to
allow for replacement of a failed hard disk without shutting down the
computer and hot sparing so that a failed disk is automatically replaced
by an online spare.
Software Implementations of RAID
Windows Server 2003
supports one RAID implementation (striped, RAID-0) that is not
fault-tolerant and two implementations that provide fault tolerance:
mirrored volumes (RAID-1) and striped volumes with parity (RAID-5). You
can create fault-tolerant RAID volumes only on dynamic disks formatted
with NTFS.
With Windows
Server 2003 implementations of RAID, there is no fault tolerance
following a failure until the fault is repaired. If a second fault
occurs before the data lost from the first fault is regenerated, you can
recover the data only by restoring it from a backup.
Striped Volumes
A striped volume, which
implements RAID Level 0, uses two or more disks and writes data to all
disks at the same rate. By doing so, I/O requests are handled by
multiple spindles, and read/write performance is the beneficiary.
Striped volumes are popular for configurations in which performance and
large storage area are critical, such as computer-aided design (CAD) and
digital media applications.
Note
You
might not experience a performance improvement on IDE unless you use
separate controllers. Separate controllers—ideally, one for each
drive—will improve performance by distributing I/O requests among
controllers as well as among drives. |
Creating a Striped Volume
To create a striped
volume, you must have unallocated space on at least two dynamic disks.
Right-click one of the spaces and choose Create Volume. The New Volume
Wizard will step you through the process of selecting a striped volume
and choosing other disk space to include in the volume. Striped volumes
can be assigned a drive letter and folder paths. They can be formatted
only with NTFS.
Up
to 32 disks can participate in a striped volume. The amount of space
used on each disk in the volume will be equal to the smallest amount of
space on any one disk. For example, if Disk 1 has 200 GB of unallocated
space, and Disk 2 has 120 GB of space, the striped volume can contain,
at most, 240 GB as the size of the stripe on Disk 1 can be no greater
than the size of the stripe on Disk 2. All disk space in the volume is
used for data; there is no space used for fault tolerance.
Recovering a Striped Volume
Because data is
striped over more than one physical disk, performance is enhanced, but
fault tolerance is decreased—there is more risk because if any one drive
in the volume fails, all data on the volume is lost. It is important to
have a backup of striped data. If one or more disks in a striped volume
fails, you must delete the volume, replace the failed disk(s) and
recreate the volume. Then you must restore data from the backup.
Tip
Striped
volumes provide maximum storage and performance but support no fault
tolerance. The only recovery potion is that of your regular backup
routine. |
Mirrored Volumes
A mirrored volume
provides good performance along with excellent fault tolerance. Two
disks participate in a mirrored volume, and all data is written to both
volumes. As with all RAID configurations, use separate controllers (by
adding a controller, you create a configuration called “duplexing”) for
maximum performance. Mirrored volumes relate to RAID-1 hardware
configurations.
Create Mirrored Volumes
To create a mirrored
volume, you must have unallocated space on two dynamic disks.
Right-click one of the spaces and choose Create Volume. The New Volume
Wizard will step you through the process of selecting a mirrored volume
and choosing space on another disk to include in the volume. Mirrored
volumes can be assigned a drive letter and folder paths. Both copies of
the mirror share the same assignment.
You can also mirror an
existing simple volume by right-clicking the volume and choosing Add
Mirror and selecting a drive with sufficient unallocated space.
Once you have
established the mirror, the system begins copying data, sector by
sector. During that time, the volume status is reported as Resynching.
Recovering from Mirrored Disk Failures
The recovery process for
a failed disk within a mirrored volume depends on the type of failure
that occurs. If a disk has experienced transient I/O errors, both
portions of the mirror will show a status of Failed Redundancy. The disk
with the errors will report a status of Offline or Missing, as seen in Figure 1.
After correcting
the cause of the I/O error—perhaps a bad cable connection or power
supply—right-click the volume on the problematic disk and choose
Reactivate Volume or right-click the disk and choose Reactivate Disk.
Reactivating brings the disk or volume back online. The mirror will then
resynchronize automatically.
If you want to stop mirroring, you have three choices, depending on what you want the outcome to be:
Delete the volume
If you delete the volume, the volume and all the information it
contains is removed. The resulting unallocated space is then available
for new volumes.
Remove the mirror
If you remove the mirror, the mirror is broken and the space on one of
the disks becomes unallocated. The other disk maintains a copy of the
data that had been mirrored, but that data is of course no longer
fault-tolerant.
Break the mirror
If you break the mirror, the mirror is broken but both disks maintain
copies of the data. The portion of the mirror that you select when you
choose Break Mirror maintains the original mirrored volume’s drive
letter, shared folders, paging file, and reparse points. The secondary
drive is given the next available drive letter.
Knowing
that information, how do you suppose you would replace a failed disk—a
member of the mirrored volume that simply died? Well, after physically
replacing the disk, you will need to open Disk Management to rescan,
initialize the disk and convert it to dynamic. After all that work you
will find that you can’t remirror a mirrored volume, even though half of
it doesn’t exist. So far as the remaining disk is concerned, the
mirrored volume still exists—its partner in redundancy is just out to
lunch. You must remove the mirror to break the mirror. Right-click the
mirror and choose Remove Mirror. In the Remove Mirror dialog box, it is
important to select the half of the volume that is missing; the volume
you select will be deleted when you click Remove Mirror. The volume you
did not select will become a simple volume. Once the operation is
complete, right-click the healthy, simple volume and choose Add Mirror.
Select the new disk and the mirror will be created again.
Tip
Mirrored
volumes provide fault tolerance and better write performance than
RAID-5 volumes. However, because each disk in the mirror contains a full
copy of the data in the volume, it is the least efficient type of
volume in terms of disk utilization. |
RAID-5 Volumes
A RAID-5 volume uses
three or more physical disks to provide fault tolerance and excellent
read performance while reducing the cost of fault tolerance in terms of
disk capacity. Data is written to all but one disk in a RAID-5. That
volume receives a chunk of data, called parity, which acts as a checksum
and provides fault tolerance for the stripe. The calculation of parity
during a write operation means that RAID-5 is quite intensive on the
server’s processor for a volume that is not read-only. RAID-5 provides
improved read performance, however, as data is retrieved from multiple
spindles simultaneously.
As data in a file is
written to the volume, the parity is distributed among each disk in the
set. But from a storage capacity perspective, the amount of space used
for fault tolerance is the equivalent of the space used by one disk in
the volume.
From a storage capacity
perspective, that makes RAID-5 more economical than mirroring. In a
minimal, three disk RAID-5 volume, one-third of the capacity is used for
parity, as opposed to one-half of a mirrored volume being used for
fault tolerance. Because as many as 32 disks can participate in a RAID-5
volume, you can theoretically configure a fault-tolerant volume which
uses only 1/32 of its capacity to provide fault tolerance for the entire
volume.
Configure RAID-5 Volumes
You need to have
space on at least three dynamic disks to be able to create a RAID-5
volume. Right-click one disk’s unallocated space and choose New Volume.
The New Volume Wizard will step you through selecting a RAID-5 volume
type, and then selecting the disks that will participate in the volume.
The
capacity of the volume is limited to the smallest section of
unallocated space on any one of the volume’s disks. If Disk 2 has 50 GB
of unallocated space, but Disks 3 and 4 have 100 GB of unallocated
space, the stripe can only use 50 GB of space on Disks 3 and 4—the space
used on each disk in the volume is identical. The capacity, or Volume
Size reported by the New Volume Wizard will represent the amount of
space available for data after accounting for parity. To continue our
example, the RAID-5 volume size would be 100 GB—the total capacity minus
the equivalent of one disk’s space for parity.
RAID-5 volumes can be assigned a drive letter or folder paths. They can be formatted only with NTFS.
Because RAID-5 volumes
are created as native dynamic volumes from unallocated space, you cannot
turn any other type of volume into a RAID-5 volume without backing up
that volume’s data and restoring into the new RAID-5 volume.
Recovering a Failed RAID-5 Volume
If a single disk fails
in a RAID-5 volume, data can continue to be accessed. During read
operations, any missing data is regenerated on the fly through a
calculation involving remaining data and parity information. Performance
will be degraded and, of course, if a second drive fails it’s time to
pull out the backup tapes. RAID-5 and mirrored volumes can only sustain a
single drive failure.
If the drive is returned
to service, you may need to rescan, and then you will need to
right-click the volume and choose Reactivate Volume. The system will
then rebuild missing data and the volume will be fully functional again.
If the drive does not
offer a Reactivate option, or if you have had to replace the disk, you
may need to rescan, initialize the disk, convert it to dynamic, then
right-click the volume and choose Repair Volume. You will be asked to
select the disk where the missing volume member should be recreated.
Select the new disk and the system will regenerate the missing data.
Mirrored Volumes versus RAID-5 Volumes
Mirrored volumes
(RAID-1) and RAID-5 volumes provide different levels of fault tolerance.
Deciding which option to implement depends on the level of protection
you require and the cost of hardware. The major differences between
mirrored volumes and RAID-5 volumes are performance and cost. Table 1 describes some differences between software-level RAID-1 and RAID-5.
Table 1. RAID Performance and Costs
| Mirrored Volumes (RAID-1) | Striped Volumes with Parity (RAID-5) |
|---|
| Can protect system or boot partition | Cannot protect system or boot partition |
| Requires two hard disks | Requires a minimum of three hard disks and allows a maximum of 32 hard disks |
| Has a higher cost per MB | Has a lower cost per MB |
| 50 percent redundancy | 33 percent maximum redundancy |
| Has good read and write performance | Has excellent read and moderate write performance |
| Uses less system memory | Requires more system memory |
Creating Fault Tolerance for the System Volume
Because RAID-5 is a
native dynamic volume, it is not possible to install or start the
Windows Server 2003 operating system on a RAID-5 volume created by the
Windows Server 2003 fault-tolerant disk technologies.
Tip
Hardware RAID,
however, is invisible to Windows Server 2003, so the operating system
can (and should, where available) be installed on hardware RAID arrays. |
The only option for
creating fault tolerance for the system, without buying hardware RAID,
is thus to mirror the system volume. You can mirror the system volume by
following the procedures described for creating a mirrored volume:
right-click the system volume and choose Add Mirror. Unlike Windows
2000, you do not need to restart, and the BOOT.INI file is updated
automatically so that you can start to the secondary drive if the
primary drive fails.
If the drives are attached
to IDE controllers, and the primary drive fails, you may have to remove
that drive, change the secondary drive to the primary controller and set
its jumpers or cable position so that it is the master. Otherwise, the
system may not boot to the secondary drive.
Tip
If
you are going to mirror the system volume, do so on one or two SCSI
controllers. If you use two controllers, make sure they are of the same
type. This configuration will be the most easily supported and
recovered. |
There
are two potential “gotchas” when you upgrade disks from previous
versions of Windows, or attempt to move disks to a Windows Server 2003
computer from a computer running a previous version of Windows.
First, if a disk
was configured in a Windows 2000 computer as a basic disk, then was
converted to dynamic, you cannot extend that disk’s simple volumes onto
other disks using Windows Server 2003. In other words, if you move that
disk to a Windows Server 2003 computer, or upgrade the operating system
to Windows Server 2003, you cannot create spanned volumes out of the
disk’s simple volumes.
Second,
Windows Server 2003 no longer supports multidisk arrays created in
Windows NT 4. Windows NT 4 created mirrored, striped, and
striped-with-parity (RAID-5) sets using basic disks. Windows 2000
permitted the use of those disk sets, although it was important to
convert the sets to dynamic quickly in order to facilitate
troubleshooting and recovery. Windows Server 2003 does not recognize the
volumes. On the off chance that you upgrade a server from Windows NT 4
to Windows Server 2003, any RAID sets will no longer be visible. You
must first back up all data prior to upgrading or moving those disks,
and then, after recreating the fault-tolerant sets in Windows Server
2003, restore the data.
