13.1 IDE
Integrated Drive Electronics
(IDE), formally called the AT
Attachment (ATA) interface, is the
means used by most computers to interface hard disks and other drives
to the computer. Early hard disk interfaces used a separate hard disk
and controller card, which limited throughput. In 1986, Compaq and
Western Digital combined the hard disk and controller card into one
unit, thereby inventing IDE. Early IDE drives used an expansion card
called a paddle card, but this card was not really a disk controller.
It simply provided an interface between the bus and the embedded disk
controller on the hard drive itself. All modern systems have IDE
interfaces embedded on the
motherboard.
You may or may not need
to understand much about IDE to install and configure a hard drive,
depending on how old your hardware is. The most recent drives,
motherboards, and BIOSs handle the hard parts for you, automatically
detecting and configuring the drive, interface, and BIOS for optimum
performance. So, if all you're doing is installing a
new IDE hard drive in a recent PC, you can safely skip this entire
section.
If, however, you're upgrading an older system with a
new drive, installing an older drive in a new system, or installing
an additional drive in a system more than a couple of years old, keep
reading. If any of the components is more than a few years old, you
need to understand quite a bit about IDE, not just to configure the
PC for optimum performance, but also to ensure that your data is not
corrupted by running data transfer rates higher than your hardware
can safely support. Understanding the fundamentals of IDE also helps
you make good decisions when you purchase drives, system boards, and
add-on IDE interface cards.
13.1.1 IDE/ATA Standards and Implementations
Although IDE was
originally proprietary, the cost and performance advantages of IDE
quickly made it the standard hard disk interface. By 1990, most
computer systems came with IDE hard disks. A slew of acronyms and
standards has arisen as IDE proliferates, some meaningful and many
that are just marketing hype. The formal ATA standards are maintained
by Technical Committee T13 of the
National Committee on Information Technology
Standards (NCITS), and may be viewed
at http://www.t13.org. These
standards include:
- ATA-1
-
The original IDE specification, ATA defines a standard 40-pin
interface that supports two hard disk devices on one cable. ATA-1,
adopted as T13 0791M and ANSI
X3.221-1994, was withdrawn as a standard in 1999.
- ATAPI (ATA Packet Interface)
-
The first ATA standard supported only hard disks. Manufacturers soon
realized that the ubiquity, high performance, and low cost of the IDE
interface also made it ideal for nondisk devices such as CD-ROM and
tape drives. The ATAPI standard was developed to allow these nondisk
devices to be connected to a standard ATA port. ATAPI hardware
connects to and works with any standard IDE or EIDE port (explained
later in this section). Note that, although ATAPI devices connect to
ATA ports, they are not ATA devices, and differ significantly from an
ATA hard drive. This is not a problem with most motherboard IDE ports
and IDE interface cards, but caching controllers and other
intelligent interfaces must be explicitly ATAPI-compliant to support
ATAPI devices.
- ATA-2
-
Advances in hard disk technology soon made it clear that the original
ATA standard was too confining. Since the original standard was
completed, several developments made it desirable to produce an
updated version of the specification. That updated specification,
ATA-2, adds faster PIO and DMA modes, improves Plug-and-Play support,
and adds Logical Block Addressing (LBA). ATA-2, adopted as
T13 0948D and ANSI
X3.279-1996, was withdrawn as a standard in 2001.
- ATA-3
-
Provided several minor improvements to the ATA-2 standard, including
enhanced reliability, especially for PIO-4, better power management,
and the incorporation of Self Monitoring Analysis and
Reporting Technology
(SMART), which allows the drive to
warn the operating system of impending problems. ATA-3 did not add
any PIO or DMA modes faster than those defined in ATA-2. ATA-3 is
officially known as T13 2008D AT Attachment-3
Interface and ANSI X3.298-1997 AT Attachment-3
Interface.
- ATA/ATAPI-4
-
Merges ATA-3 and ATAPI into a single integrated book-length standard
and formalizes Ultra DMA 33 as a part of that standard. ATA-4 is
officially known as T13 1153D AT Attachment-4 with Packet
Interface Extension and ANSI NCITS 317-1998 AT
Attachment-4 with Packet Interface Extension.
- ATA/ATAPI-5
-
The current standard. ATA/ATAPI-5 formalizes Ultra DMA 66, but
otherwise makes only relatively minor enhancements to ATA/ATAPI-4.
ATA-5 is officially known as T13 1321D AT Attachment-5 with
Packet Interface and ANSI NCITS 340-2000 AT
Attachment-5 with Packet Interface.
- ATA/ATAPI-6
-
The forthcoming ATA/ATAPI standard, which is already implemented in
practice by drive makers. ATA/ATAPI-6 formalizes Ultra DMA 100 and
48-bit LBA, which expands the maximum disk size possible under ATA
from 128 GB to 128 PB (i.e., a binary million times larger), but
otherwise makes only relatively minor enhancements to ATA/ATAPI-5. As
of July 2003, ATA-6 is still in development and is officially known
as T13 1410D AT Attachment-6 with Packet
Interface.
- ATA/ATAPI-7
-
The next and likely final ATA/ATAPI standard, before SATA supersedes
ATA/ATAPI. ATA/ATAPI-7 is expected to formalize Ultra DMA 133, but
otherwise make only relatively minor enhancements to ATA/ATAPI-6. As
with ATA/ATAPI-6, drive makers will incorporate the important parts
of ATA/ATAPI-7 before it is formally standardized. As of July 2003,
ATA/ATAPI-7 is still in development and is officially known as
T13 1532D AT Attachment-7 with Packet Interface.
Although these formal standards define the ATA interface, the hard
drives and interfaces you can actually buy are marketed using the
following ad hoc standards:
- Enhanced IDE (EIDE)
-
A Western Digital IDE implementation that incorporates and extends
the ATA-2 and ATAPI standards. EIDE supports: two devices each on
primary and secondary ATA interfaces, for a total of four devices;
fast transfer modes (PIO-3 or better and multiword DMA-1 or better);
Logical Block Addressing (LBA) mode; and connecting ATAPI devices
such as CD-ROM and tape drives to the ATA interface. EIDE includes
the Western Digital Enhanced BIOS, which eliminates the 504/528 MB
limitation under DOS. Recent Western Digital hard drives and EIDE
interfaces support some or all of the ATA/ATAPI-4 and proposed
ATA/ATAPI-5 standards, including Ultra-DMA/16 and /33 (UDMA modes 0
and 1) and Ultra-DMA/66 modes 2, 3, and 4.
- Fast ATA
-
A Seagate IDE implementation, later endorsed by Quantum, that
counters the Western Digital EIDE initiative. Fast ATA and Fast ATA-2
are based on ATA and ATA-2, but not on ATAPI. Fast ATA supports: fast
transfer modes (PIO-3 and multiword DMA-1); LBA mode; and Read/Write
Multiple commands, also called Block Mode. Fast ATA-2 adds support
for PIO-4 and multiword DMA-2. Recent Fast-ATA hard drives support
Ultra ATA-33, -66, and -100.
- Ultra ATA
-
An extension to ATA-2, first proposed by Quantum and Intel, and now a
part of the ATA/ATAPI-4 standard (as Ultra ATA-33 for UDMA modes 0
and 1), the ATA/ATAPI-5 standard (as Ultra ATA-66 for UDMA modes 2,
3, and 4), the proposed ATA/ATAPI-6 standard (as Ultra ATA-100 for
UDMA mode 5), and the proposed ATA/ATAPI-7 standard (as Ultra ATA-133
for UDMA mode 6). Ultra-ATA enhances the earlier SDMA and MDMA modes
by adding CRC error detection to prevent data corruption during fast
DMA transfers.
To benefit from the increased data transfer rates provided by
Ultra-DMA, the drive, the BIOS, the ATA interface, and the operating
system must all support it. By late 1997 some PCs and motherboards
included embedded Ultra ATA-33 support. Ultra ATA-66 drives and
interfaces began shipping in early 1999. ATA-100 drives and
interfaces first appeared in early 2001, and soon became standard on
new systems and motherboards. ATA-133, championed by Maxtor, is
unlikely to achieve critical mass. Only Maxtor makes ATA-133 drives,
and Intel seems likely to transition its chipsets and motherboards
directly from ATA-100 to SATA. It's easy to upgrade
older systems to Ultra ATA simply by adding an inexpensive IDE
interface card, such as those made by Promise, SIIG, and
others.
13.1.2 IDE/ATA Data Transfer Modes
To understand ATA data transfer modes,
it's necessary to understand something about how
data is read from and written to the hard drive. Real-mode operating
systems such as 16-bit Windows and DOS make read and write requests
to the BIOS, which passes the command to the drive. Protected-mode
operating systems such as Windows NT/2000/XP and Linux bypass the
real-mode BIOS and use their own protected-mode I/O subsystems to
accomplish the same purpose.
Data transfer commands are controlled by the BIOS or I/O subsystem,
but execution speed—and therefore data transfer rate—is
determined by the strobe frequency of the ATA interface hardware. The
time needed to complete one full cycle, measured in nanoseconds (ns),
is called the cycle time for the interface. A
shorter cycle time allows more cycles to be completed in a given
period, and therefore provides a higher data transfer rate. For
example, a 600 ns cycle time yields 1.66 million cycles/second.
Because each cycle transfers one word (16 bits or two bytes), a 600
ns cycle time translates to a data transfer rate of 3.33 MB/s.
ATA supports two data transfer modes, called Programmed
Input/Output (PIO) mode and Direct Memory Access
(DMA) mode. Both of these have
several submodes that use different cycle times and have different
data transfer rates. When an ATA interface interrogates a modern
drive with the Identify Drive command, the drive
returns its model, geometry, and a list of the PIO and DMA modes it
supports, allowing the interface, given proper BIOS support, to
automatically configure the best settings for optimum drive
performance. With an older drive, an older BIOS, or both,
it's up to you to configure these settings yourself,
so it's important to understand what they mean to
avoid either crippling drive performance by choosing too slow a mode
or risking your data by choosing one that's too fast
for your hardware.
 |
Your operating system also determines whether an ATA hard disk is
configured automatically for optimal performance. Windows 98/2000/XP
and Linux (usually) detect DMA-capable interfaces and drives and
configure them automatically to operate in the more efficient DMA
mode. Windows 95 and Windows NT 4 use the less-efficient PIO mode by
default, and must be configured manually to use DMA mode on hardware
that supports it. Configuring DMA mode on ATA drives is described in
Chapter 14.
|
|
13.1.2.1 Programmed Input/Output (PIO) Modes
PIO
is a means of data transfer that
requires the CPU to intermediate data exchanges between the drive and
memory. This dependence on the CPU places unnecessary demands on it
and slows CPU performance under multitasking operating systems.
Accordingly, modern drives and interfaces substitute DMA modes,
described later, to provide high-speed data transfer. PIO modes
remain important, however, when connecting older drives to newer
interfaces or vice versa. Table 13-1 lists PIO
modes, not all of which are supported by all drives and all
interfaces.
Table 13-1. ATA PIO modes|
|
PIO-0
|
600 ns
|
3.3 MB/s
|
ATA
| |
PIO-1
|
383 ns
|
5.2 MB/s
|
ATA
| |
PIO-2
|
330 ns
|
8.3 MB/s
|
ATA
| |
PIO-3
|
180 ns
|
11.1 MB/s
|
ATA-2
| |
PIO-4
|
120 ns
|
13.3 MB/s
|
ATA-2
| |
PIO-5
|
90 ns
|
22.2 MB/s
|
Never implemented
|
PIO Modes 0, 1, and 2 are not used on recent systems except to
support older drives. Using PIO Mode 3 or 4 provides reasonably fast
transfer rates, but requires that the drive support the IORDY line
for hardware flow control. If the interface delivers data faster than
a PIO Mode 3 or 4 drive can accept it, the drive de-asserts IORDY to
notify the interface to stop delivering data. Running a drive that
does not support IORDY on an interface configured for PIO Mode 3 or 4
may appear to work, but risks corrupting data. If you are not certain
that your drive supports PIO Mode 3 or 4, configure the interface to
use PIO Mode 2.
13.1.2.2 Direct Memory Access (DMA) Modes
DMA
is a data transfer mode that allows
bidirectional transfer of data between drives and memory without
intervention from the processor. If you use a multitasking operating
system such as 32-bit Windows or Linux, using DMA mode increases
performance by freeing the CPU to do other things while data is being
transferred. DMA doesn't improve performance with
single-tasking operating systems such as DOS or 16-bit Windows
because the processor must wait until the transfer is complete before
doing anything else. Table 13-2 lists DMA modes,
not all of which are supported by all drives and all interfaces. Note
that ATA-3 introduced no new modes.
Table 13-2. ATA DMA modes|
|
Single Word Mode 0 (SDMA-0)
|
960 ns
|
2.1 MB/s
|
ATA
| |
Single Word Mode 1 (SDMA-1)
|
480 ns
|
4.2 MB/s
|
ATA
| |
Single Word Mode 2 (SDMA-2)
|
240 ns
|
8.3 MB/s
|
ATA
| |
Multiword Mode 0 (MDMA-0)
|
480 ns
|
4.2 MB/s
|
ATA
| |
Multiword Mode 1 (MDMA-1)
|
150 ns
|
13.3 MB/s
|
ATA-2
| |
Multiword Mode 2 (MDMA-2)
|
120 ns
|
16.6 MB/s
|
ATA-2
| |
Ultra-DMA Mode 0 (UDMA-0)
|
240 ns
|
16.6 MB/s
|
ATA-4
| |
Ultra-DMA Mode 1 (UDMA-1)
|
160 ns
|
25.0 MB/s
|
ATA-4
| |
Ultra-DMA Mode 2 (UDMA-2)
|
120 ns
|
33.3 MB/s
|
ATA-4
| |
Ultra-DMA Mode 3 (UDMA-3)
|
90 ns
|
44.4 MB/s
|
ATA-5
| |
Ultra-DMA Mode 4 (UDMA-4)
|
60 ns
|
66.7 MB/s
|
ATA-5
| |
Ultra-DMA Mode 5 (UDMA-5)
|
40 ns
|
100.0 MB/s
|
ATA-6
| |
Ultra-DMA Mode 6 (UDMA-6)
|
30 ns
|
133.3 MB/s
|
ATA-6/7
|
The SDMA modes were obsoleted by ATA-3, and are useful only for older
drives and interfaces. MDMA modes provide reasonable transfer rates,
and are useful for drive/interface configurations that are ATA-2
compliant, but where the drive and/or interface is not Ultra-ATA
compliant. Using UDMA modes requires that the drive, interface, BIOS,
and operating system be Ultra-ATA compliant. MDMA-2 and UDMA-0
provide identical transfer rates, but MDMA-2 does not use hardware
flow control, while UDMA-0 uses CRC error detection to prevent data
corruption.
Hardware that supports UDMA Modes 3 and 4, also called UDMA/66, began
shipping in early 1999, and had become standard for new systems,
motherboards, and hard drives by mid-2000. UDMA/100, also called
Ultra ATA-100 or simply ATA-100, is an extension of UDMA/66 developed
by Maxtor Corporation and since licensed by other drive makers.
ATA-100 components began shipping in volume in early 2001, and are
now standard.
ATA-100 provides 100 MB/s throughput for reads, but only 88 MB/s
throughput for writes. ATA-100 uses the same 80-wire, 40-pin ATA
cable as UDMA/66, which is sufficient to reduce reflections, noise,
and inductive coupling to the level required by ATA-100, but reduces
signaling voltages to 3.3V from the 5.0V used by all previous ATA
standards. Installing a UDMA/66 or /100 drive on an older system
requires adding a UDMA/66 or /100 interface card and using a special
80-wire, 40-pin ATA cable.
Ultra ATA-133, also called FastDrive/133, is another Maxtor
initiative that has not been widely adopted by other manufacturers.
It uses the same 80-wire, 40-pin cable as ATA-66 and ATA-100, and
ATA-133 drives and interfaces are backward-compatible with earlier
standards.
|
ATA-133 is a solution in search of a problem. As of July 2003, no ATA
hard drive can saturate an ATA-100 interface. Even if that were
possible, the ATA data would occupy the entire bandwidth of the PCI
bus. Only one device on an ATA channel can be active at a time, which
means that PATA will be replaced entirely by SATA before hard drives
can saturate an ATA-100 interface. Contrast this with SCSI, which
does allow multiple devices to use the channel simultaneously. The
fastest SCSI hard drives exceed 60 MB/s throughput, which means that
three drives on a channel can saturate even an Ultra160 SCSI
interface. Our opinion is that ATA-100 will suffice for desktop
systems until SATA replaces PATA, so we consider ATA-133 to be a
marketing gimmick rather than a useful technology.
|
|
There are
two ways to implement DMA. First-party DMA, also
called bus-mastering DMA, uses a DMA controller
embedded in the device itself to arbitrate possession of the bus and
data transfer. Third-party DMA instead depends
on the DMA controller that resides on the motherboard. If the
motherboard is of recent vintage, either first-party or third-party
DMA can be used for high-speed DMA transfer modes.
13.1.3 ATA Multichannel Support
The original ATA specification defined
one dual-channel ATA interface that supported one or two ATA devices.
ATA-2 defines a second interface, with the primary
interface using standard ATA IRQ and base address, and the
secondary interface using the alternate IRQ and
base address that had always been set aside for that purpose. Any
modern motherboard or interface card provides dual-ported ATA
support.
Windows 9X/NT/2000/XP and Linux recognize and use the secondary
interface automatically. DOS and 16-bit Windows are not dual-port
ATA-aware, so installing more than two ATA drives with these
operating systems requires either BIOS support for the secondary port
or supplemental drivers. If BIOS Setup allows you to enter drive
parameters for four drives, you already have the BIOS support you
need. If BIOS Setup lists only two drives, you must (in order of
desirability) install an updated system BIOS or
use a dual-ported bus-mastering ATA adapter with an onboard dual-port
BIOS or install dual-port device drivers.
In addition to the well-standardized primary and secondary ATA
interfaces, some devices support semistandard tertiary and quaternary
ATA interfaces. In theory, at least, you can support up to eight ATA
devices on one computer by adding a dual-ported interface card that
can be configured as tertiary/quaternary to a system that includes a
standard dual-ported ATA interface, assuming that you have the four
IRQs needed to support all four ports. We have not seen a BIOS that
includes direct support for ATA Ports 3 and 4, although the current
Phoenix BIOS specification allows ATA ports beyond 1 and 2 to be
defined arbitrarily. Table 13-3 lists the IRQs and
base addresses for ATA Ports 1 through 4.
Table 13-3. ATA port IRQ and base address assignments|
|
Primary
|
14
|
0x1F0 - 0x1F7 and 0x3F6 - 0x3F7
| |
Secondary
|
15 or 10
|
0x170 - 0x177 and 0x376 - 0x377
| |
Tertiary
|
12 or 11
|
0x1E8 - 0x1EF and 0x3EE - 0x3EF
| |
Quaternary
|
10 or 9
|
0x168 - 0x16F and 0x36E - 0x36F
|
13.1.4 The ATA Interface
Table 13-4 lists
the pins and signal names for the ATA interface. Signal names that
end with # are active-low. Signal names contained in square brackets
are specific to the secondary ATA interface.
Table 13-4. ATA interface pinout and signal definitions|
|
1
|
Reset IDE
| |
2
|
Ground
| |
3
|
Data 7
| |
4
|
Data 8
| |
5
|
Data 6
| |
6
|
Data 9
| |
7
|
Data 5
| |
8
|
Data 10
| |
9
|
Data 4
| |
10
|
Data 11
| |
11
|
Data 3
| |
12
|
Data 12
| |
13
|
Data 2
| |
14
|
Data 13
| |
15
|
Data 1
| |
16
|
Data 14
| |
17
|
Data 0
| |
18
|
Data 15
| |
19
|
Ground
| |
20
|
Key (no pin)
| |
21
|
DDRQ0 [DDRQ1]
| |
22
|
Ground
| |
23
|
I/O Write#
| |
24
|
Ground
| |
25
|
I/O Read#
| |
26
|
Ground
| |
27
|
IOCHRDY
| |
28
|
P_ALE (CSEL pull-up)
| |
29
|
DDACK0# [DDACK1#]
| |
30
|
Ground
| |
31
|
IRQ 14 [IRQ 15]
| |
32
|
(reserved)
| |
33
|
DAG1 (Address 1)
| |
34
|
GPIO_DMA66_Detect_Pri [GPIO_DMA66_Detect_Sec]
| |
35
|
DAG0 (Address 0)
| |
36
|
DAG2 (Address 2)
| |
37
|
Chip Select 1P# [Chip Select 1S#]
| |
38
|
Chip Select 3P# [Chip Select 3S#]
| |
39
|
Activity#
| |
40
|
Ground
| |
41
|
+5VDC (logic)
| |
42
|
+5VDC (motor)
| |
43
|
Ground
| |
44
|
-TYPE (0=ATA)
|
The standard ATA interface used in desktop systems uses only pins 1
through 40. Pins 41 through 44 are implemented only in notebook ATA
interfaces, for use with 2.5-inch and smaller drives. The additional
four pins provide power and ground to the drives, which are not large
enough to accept a standard power connector.
|
ATA interfaces are enabled or disabled in BIOS Setup. Disabling an
unused interface frees the interrupt that would otherwise be used by
that interface. If there are no devices connected to the secondary
ATA interface, disable it in BIOS. If you are running an all-SCSI
system, disable both the primary and secondary ATA interfaces.
Conversely, if you install an ATA/ATAPI device as the first device on
the secondary interface and find that it is not recognized, check
BIOS Setup to verify that the interface is enabled. Although most
recent systems automatically detect and configure PIO mode versus DMA
mode, some systems require you to specify PIO mode versus DMA mode
for each interface in Setup.
|
|
13.1.5 Configuring ATA Devices
Each ATA interface (often called a
channel) can have zero, one, or two ATA and/or ATAPI devices
connected to it. Every ATA and ATAPI device has an embedded
controller, but ATA permits only one active controller per interface.
Therefore, if only one device is attached to an interface, that
device must have its embedded controller enabled. If two devices are
attached to an ATA interface, one device must have its controller
enabled and the other must have its controller disabled.
In ATA terminology, a device whose controller is enabled is called a
Master; one whose controller is disabled is
called a Slave. In a standard PC with two ATA
interfaces, a device may therefore be configured in any one of four
ways: Primary Master, Primary
Slave, Secondary Master, or
Secondary Slave.
13.1.5.1 Assigning masters and slaves
ATA/ATAPI devices are assigned as
Master or Slave by setting jumpers on the device. ATA devices have
the following jumper selections:
- Master
-
Connecting a jumper in the Master position enables the onboard
controller. All ATA and ATAPI devices have this option. Select this
jumper position if this is the only device connected to the
interface, or if it is the first of two devices connected to the
interface.
- Slave
-
Connecting a jumper in the Slave position disables the onboard
controller. (One of our technical reviewers notes that he has taken
advantage of this to retrieve data from a hard drive whose controller
had failed, a very useful thing to keep in mind.) All ATA and ATAPI
devices can be set as Slave. Select this jumper position if this is
the second device connected to an interface that already has a Master
device connected.
|
When you jumper a device as Master or Slave, it assumes that role
regardless of which position it connects to on the ATA cable. For
example, if you jumper a device as Master, it functions as Master
regardless of whether you connect it to the drive connector at the
end of the ATA cable or the drive connector in the middle of the ATA
cable.
|
|
- Cable Select
-
All recent ATA/ATAPI devices have a third jumper position labeled
Cable Select, CS, or
CSEL. Connecting a jumper in the CSEL position
instructs the device to configure itself as Master or Slave based on
its position on the ATA cable. When you use CSEL, the device
connected nearest the ATA interface is always configured as Master
and the device on the end of the cable is always configured as Slave.
CSEL has long been a part of the ATA Plug-and-Play standard and was
first formalized in the Intel/Microsoft PC97 standard, but was little
used until recently. If the CSEL jumper is connected, no other
jumpers may be connected. For more information about CSEL, see the
following section.
- Sole/Only
-
When functioning as Master, a few older ATA/ATAPI devices need to
know whether they are the only device on the channel, or if a Slave
device is also connected. Such devices may have an additional jumper
position labeled Sole or
Only. For such devices, jumper as Master if it
is the first of two devices on the interface, Slave if it is the
second of two devices on the interface, and Sole/Only if it is the
only device connected to the interface.
- Slave Present
-
A few older drives have a jumper designated Slave
Present, or SP. This jumper performs
the inverse function of the Sole/Only jumper, by notifying a device
jumpered as Master that there is also a Slave device on the channel.
For such devices, jumper as Master if it is the only device on the
interface, or Slave if it is the second of two devices on the
interface. If it is the Master on a channel that also has a Slave
installed, connect both the Master and Slave Present jumpers.
13.1.5.2 Using Cable Select
All recent ATA/ATAPI drives provide a
Cable Select (CS or CSEL) jumper in addition to the standard
Master/Slave jumpers. If you jumper a drive as Master (or Slave),
that drive functions as Master (or Slave) regardless of which
connector it is attached to on the ATA cable. If you jumper a drive
as CSEL, the position of the drive on the cable determines whether
the drive functions as a Master or a Slave.
CSEL was introduced in the ATA-2 and ATA-3 specifications as a means
to simplify ATA configuration. The goal was that eventually all
ATA/ATAPI devices would be configured to use CSEL. That would mean
that drives could simply be installed and removed without changing
jumpers, with no possibility of conflict due to improper jumper
settings. Although CSEL has been around for years, it has only
recently become popular with system makers. If you work on a PC built
in the last year or two, you may encounter CSEL, so
it's good to be aware of it.
Using CSEL requires the following:
If one drive is installed on the interface, that drive must support
and be configured to use CSEL. If two drives are installed, both must
support and be configured to use CSEL. The ATA interface must support CSEL. For this to be true, the ATA
interface must ground Pin 28. On many older ATA interfaces, Pin 28 is
open or high, and so cannot be used for CSEL. If Pin 28 is not
grounded on the interface, any drive configured as CSEL that connects
to that interface is configured as Slave. The ATA cable must be a special CSEL cable. Unfortunately, there are
three types of CSEL cable: A 40-wire CSEL cable differs from a standard 40-wire ATA cable in
that Pin 28 is connected only between the ATA interface and the first
drive position on the cable (the middle connector). Pin 28 is not
connected between the interface and the second drive position (the
end connector on the cable). With such a cable, the drive attached to
the middle connector (with Pin 28 connected) is Master; the drive
attached to the connector furthest from the interface (with Pin 28
not connected) is Slave. All 80-wire (Ultra DMA) ATA cables support CSEL, but with exactly the
opposite orientation of the 40-wire standard CSEL cable. With such a
cable, the drive attached to the middle connector (with Pin 28 not
connected) is Slave; the drive attached to the connector furthest
from the interface (with Pin 28 connected) is Master. This is
actually a better arrangement, if a bit nonintuitive—how can a
wire be connected to the end connector but not to the one in the
middle?—because the standard 40-wire CSEL cable puts the Master
drive on the middle connector. If only one drive is installed on that
cable, that leaves a long "stub" of
cable hanging free with nothing connected to it. Electrically,
that's a very poor idea because an unterminated
cable allows standing waves to form, increasing noise on the line and
impairing data integrity. A 40-wire CSEL Y-cable puts the interface connector in the middle
with a drive connector on each end, one labeled Master and one Slave.
Although this is a good idea in theory, in practice it seldom works.
The problem is that IDE cable length limits still apply, which means
that the drive connectors don't have enough cable to
get to the drives in all but the smallest cases. If you have a tower,
you can forget it.
40-wire CSEL cables are supposed to be clearly labeled, but we have
found that this is often not the case. It is not possible to identify
such cables visually, although you can verify the type using a DVM or
continuity tester between the two end connectors on Pin 28. If there
is continuity, you have a standard ATA cable. If not, you have a CSEL
cable.
 |
Keep unlabeled 40-wire CSEL cables segregated from standard cables.
If you substitute a CSEL cable for a standard cable, drives that are
jumpered as Master or Slave function properly. If you substitute a
standard cable for a CSEL cable and connect one drive jumpered CSEL
to that cable, it will function properly as Master. But if you
connect two CSEL drives to a standard cable, both function as Master,
which may result in anything from subtle problems to (more likely)
the system being unable to access either drive.
|
|
13.1.5.3 Master/Slave guidelines
When deciding how to allocate
devices between two interfaces and choose Master versus Slave status
for each, use the following guidelines:
Always assign the main hard drive as Primary Master. Particularly if
your hard drive supports fast DMA modes (ATA-66 or ATA-100), do not
connect another device to the primary ATA interface unless both
positions on the secondary interface are occupied. ATA forbids simultaneous I/O on an interface, which means that only
one device can be active at a time. If one device is reading or
writing, the other device cannot read or write until the active
device yields the channel. The implication of this is that if you
have two devices that need to perform simultaneous I/O—e.g., a
CD writer that you use to duplicate CDs from a CD-ROM drive—you
should place those two devices on separate interfaces. If you are connecting an ATA device (a hard drive) and an ATAPI
device (e.g., an optical drive or a tape drive) to the same
interface, set the hard drive as Master and the ATAPI device as
Slave. If you are connecting two similar devices (ATA or ATAPI) to an
interface, it generally doesn't matter which device
is Master and which Slave. There are exceptions to this, however,
particularly with ATAPI devices, some of which really want to be
Master (or Slave) depending on which other ATAPI device is connected
to the channel. If you are connecting an older device and a newer device to the same
ATA interface, it's generally better to configure
the newer device as Master because it is likely to have a more
capable controller than the older device. Try to avoid sharing one interface between a DMA-capable device and a
PIO-only device. If both devices on an interface are DMA-capable,
both use DMA. If only one device is DMA-capable, both devices are
forced to use PIO, which reduces performance and increases CPU
utilization dramatically. Similarly, if both devices are DMA-capable,
but at different levels, the more capable device is forced to use the
slower DMA mode.
13.1.6 ATA Cables
All desktop ATA cables have three
40-pin connectors, one that connects to the ATA interface and two
that connect to ATA/ATAPI drives. All standards-compliant ATA
cables—there are many that do not comply—share the
following characteristics:
An ATA cable must be at least 10 inches (0.254m) long, and no longer
than 18 inches (0.457m). The center-to-center distance between one end connector and the
middle connector must be 5 inches (0.127m) minimum to 12 inches
(0.304m) maximum. The distance between the center connector and the
other end connector must be 5 inches minimum to 6 inches (0.152m)
maximum. Pin 1 is indicated by a colored stripe on the cable (usually a red
stripe on a gray cable). ATA cables are keyed, either by blocking Pin 20, by using a keyed
shroud, or both.
ATA cables come in three varieties:
- Standard
-
A standard ATA cable uses a 40-wire ribbon cable and 40-pin
connectors in all three positions. All 40 conductors connect to all
three connectors. The only real variation, other than cable quality,
is the positioning of the three connectors. A standard ATA cable
requires using Master/Slave jumpering for connected devices. All
connectors on a standard ATA cable are interchangeable, which is to
say that any drive may be connected to any position, and the
interface itself may be connected to any position, including the
center connector. A standard ATA cable may be used with any ATA/ATAPI
device through ATA-33. If a standard ATA cable is used to connect
ATA-66/100 devices, those devices will function properly, but will
fall back to operating at ATA-33.
- Standard/CSEL
-
A standard/CSEL ATA cable is identical to a standard ATA cable except
that Pin 28 is not connected between the middle drive connector and
the end drive connector. A standard/CSEL ATA cable supports either
Master/Slave jumpering or CSEL jumpering for connected devices.
Connector position is significant on a standard/CSEL cable. The
interface connector on a CSEL cable is either labeled or a different
color than the drive connectors. The center connector is for the
Master device, and the end connector opposite the interface connector
is for the Slave device.
- Ultra DMA (80-wire)
-
An Ultra DMA cable uses an 80-wire ribbon cable and 40-pin connectors
in all three positions. The additional 40 wires are dedicated ground
wires, each assigned to one of the standard 40 ATA pins. An Ultra DMA
cable may be used with any ATA/ATAPI device—and should be for
more reliable functioning—but is required for full performance
with ATA-66 and ATA-100 devices. All Ultra DMA cables are CSEL
cables, and may be used either with drives jumpered CSEL or those
jumpered Master/Slave.
The following points about Ultra DMA cables are worth being aware of:
The Ultra DMA cable was first defined for UDMA Modes 0 through 2, for
transfer speeds up to 33 MB/s, but is optional for those modes. For
UDMA Mode 3 and higher, the Ultra DMA cable is required. All Ultra DMA cables have Pin 28 wired to support CSEL mode, and so
can be used to connect drives that use either CSEL or Master/Slave
jumpering. Color-coded connectors were not specified for earlier ATA cables. The
Ultra DMA cable specification requires the following connector
colors: - Blue
-
One end connector is blue, which indicates that it attaches to the
ATA interface.
- Black
-
The opposite end connector is black, and is used to attach the Master
drive (Device 0), or a single drive if only one is attached to the
cable. If CSEL is used, the black connector configures the drive as
Master. If standard Master/Slave jumpering is used, the Master drive
must still be attached to the black connector. This is true because
ATA-66, ATA-100, and ATA-133 do not allow a single drive to be
connected to the middle connector, which results in standing waves
that interfere with data communication.
- Gray
-
The middle connector is gray, and is used to attach the Slave drive
(Device 1), if present.
|
Note the difference between using a 40-wire CSEL cable and an 80-wire
cable for CSEL operation. Although all Ultra DMA cables support
drives jumpered either Master/Slave or CSEL, that does not mean you
can freely substitute an 80-wire cable for a 40-wire cable. If the
drives are jumpered Master/Slave, substituting an 80-wire cable works
fine. However, if the drives are jumpered CSEL, replacing a 40-wire
CSEL cable with an 80-wire cable causes the drives to exchange
settings. That is, the drive that was Master on the 40-wire cable
becomes Slave on the 80-wire cable, and vice
versa. |
|
Because an Ultra DMA
cable is required for ATA-66, ATA-100, or ATA-133 operation, the
system must have a way to detect if such a cable is installed. This
is done by grounding Pin 34 in the Blue connector, which attaches to
the interface. Pin 34 carries the
GPIO_DMA66_Detect_Pri signal (primary interface)
or the GPIO_DMA66_Detect_Sec signal (secondary
interface). These signals are also referred to as
/PDIAG:/CBLID. Because 40-wire ATA cables do not
ground Pin 34, the system can detect at boot whether a 40-wire or
80-wire cable is installed. In some Ultra ATA cables, the grounding
is done internally and is therefore invisible. In others, the signal
conductor for Pin 34 is cut in the cable, as shown in Figure 13-1. Figure 13-2 shows a 40-wire
standard ATA cable and an 80-wire UltraATA cable side by side for
comparison. Note that the 80-wire cable uses a gray end connector,
and so is not standards-compliant.
13.1.7 Compatibility Between Old and New IDE Devices
With only minor exceptions, there are
no compatibility issues when using new ATA devices with old ATA
interfaces or vice versa. Newer drives cannot
yield their highest performance when connected to an ATA interface
that does not support the fastest transfer modes the drive is capable
of, just as a new interface can't improve the
performance of an older drive. But you can connect any ATA drive to
any ATA interface with reasonable assurance that it will function,
although perhaps not optimally.
Nearly all modern ATA interfaces—including the ubiquitous Intel
PIIX3 and PIIX4—support independent timing for Master and Slave
devices. This means that you can safely put an old, slow ATA or ATAPI
device on the same cable as a new, fast hard drive without concern
that the older device will cripple the throughput of the newer one.
|
Note that independent timing does not mean that you should mix DMA
and PIO devices on the same interface. If an interface has one
PIO-mode device connected, the other device must also run PIO. Also,
independent timing does not mean that an interface supports multiple
DMA modes simultaneously. For example, if you connect an ATA-66 drive
and an ATA-33 drive to the same interface, both run as ATA-33
devices.
|
|
13.1.8 IDE Capacity Limits
Various IDE capacity limits exist,
which depend on the BIOS, interface hardware, operating system, and
other factors. There's a lot of information and
misinformation about these limits, so it's worth
getting the facts straight before you accept unnecessary limitations.
Most of these limits are a result of interactions between the methods
that the BIOS and the ATA interface use to address sectors on a hard
disk. Even the oldest BIOS or ATA interface can address large hard
disks. In combination, however, interactions between the BIOS and the
interface may limit the number of addressable sectors to a fraction
of the number either could address alone. Table 13-5 summarizes these limits.
Table 13-5. ATA addressing and BIOS addressing limitations|
|
Cylinders
|
65,536 (216)
|
1,024 (210)
|
1,024
| |
Heads
|
16 (24)
|
256 (28)
|
16
| |
Sectors per track
|
255 (28 - 1)
|
63 (26 - 1)
|
63
| |
Disk size (bytes)
|
136,902,082,560
|
8,455,716,864
|
528,482,304
|
ATA addressing uses four registers that total 28 bits. ATA numbers
cylinders starting at 0. The cylinder address is
a 16-bit value, divided as a least-significant 8-bit
Cylinder Low register and a most-significant
8-bit Cylinder High register, allowing up to
65,536 cylinders to be addressed. Each cylinder has a number of
heads, also numbered starting at 0, which are addressed using a 4-bit
value stored in the lowest four bits of the
Device/Head register, allowing up to 16 heads to
be addressed. Sectors are numbered starting at 1, and are addressed
using an 8-bit value stored in the Sector Number
register, allowing up to 255 sectors to be addressed. These
capacities mean that ATA can address up to (65,536 x 16
x 255) 512-byte sectors, which equals 136,902,082,560
bytes.
BIOS Int13 API addressing uses three registers that total 24 bits.
Cylinder Low is an 8-bit register that stores
the least-significant eight bits of the 10-bit cylinder address.
Cylinder High/Sector Number is an 8-bit register
whose least-significant two bits store the most-significant two bits
of the cylinder address and whose most-significant six bits store the
sector number. In conjunction, these addresses allow up to 1,024
cylinders and 63 sectors/track to be addressed. Head
Number is an 8-bit register, which allows up to 256 heads
to be addressed. These capacities mean that Int13 can address up to
(1,024 x 256 x 63) 512-byte sectors, which
equals 8,455,716,864 bytes.
The impact of these limits depends entirely on the addressing method
used by the BIOS. From the oldest to the newest, here are the methods
used by various BIOSs:
- Cylinder-Head-Sectors (CHS)
-
Cylinder-Head-Sectors (CHS)
is the access method used by early BIOSs to address individual
sectors on a hard disk by identifying them directly by the cylinder,
head, and sector where they reside. CHS in an ATA/Int13 environment
is subject to the maximum common address register size supported for
each element by ATA and Int13 addressing. This means that CHS can
address at most 1,024 cylinders, 16 heads, and 63 sectors per track,
which results in the 504/528 MB limit.
- Extended Cylinder-Head-Sectors (ECHS)
-
Extended Cylinder-Head-Sectors
(ECHS) is the most popular of several access
methods (others being Large and Big
IDE) used by so-called enhanced or translating
BIOSs to support hard disks larger than those supported by
CHS. ECHS or bit shift translation works by
lying about disk geometry, translating an unsupported physical
geometry to a supported logical geometry whose address will fit the
available registers. For example, ECHS translation may represent a 1
GB drive whose physical geometry is 2,048 cylinders, 16 heads, and 63
sectors per track as having a logical geometry of 1,024 cylinders, 32
heads, and 63 sectors per track, allowing the full 1 GB to be
recognized and used. The problem with ECHS translation is that it is
nonstandard, and different BIOSs handle it differently.
- Logical Block Addressing (LBA)
-
Logical Block Addressing
(LBA) dispenses with CHS addressing entirely,
and addresses each sector sequentially. With LBA addressing, all 28
address bits (16+4+8) are used as one LBA address, allowing up to
228 (268,435,456) sectors or
137,438,953,472 bytes to be addressed. Note that, with LBA
addressing, sector numbers start at 0. LBA assist translation can be
used only with BIOSs and drives that support LBA. With LBA
translation, the drive reports its actual geometry to the BIOS, which
then multiplies the cylinders, heads, and sectors reported by the
drive to determine the total number of sectors to be used for LBA
translation. All modern BIOSs support LBA.
Various environments may enforce the following limits:
- 504/528 MB
-
Results from the interaction between ATA addressing and CHS Int13
addressing, which allows 1,024 cylinders, 16 heads, 63 sectors/track,
limiting drive capacity to 528,482,304 bytes, or 504 MB. To get
around this limit, you may upgrade the PC BIOS to one with LBA
support, disable the embedded interface and replace it with one that
has LBA support, or install a BIOS extender card that provides LBA
support with the existing interface. If a hardware solution is not
feasible, installing a third-party driver such as Ontrack Disk
Manager allows using large disks, but with some limitations relative
to the hardware solutions.
- 2.0/2.1 GB (hardware)
-
Various manufacturers attempted to solve the 504 MB limit by various
methods, including modifications to the way registers were
manipulated. One such method was to
"steal" the two previously unused
high-order bits assigned to the head register and assign them to the
cylinder register, expanding it from 10 to 12 bits. This increased
the number of addressable cylinders to 4,096, but in doing so limited
the number of heads that could be addressed to 64. Because this
translation method was not commonly used, the two high-order head
bits cannot safely be assumed to be assigned to either the head count
or the cylinder count, which allows 1,024 cylinders, 64 heads, and 63
sectors/track. Multiplying these numbers yields 2,113,929,216 bytes,
usually stated as 2.0 or 2.1 GB. You've encountered
this limit if the system hangs during POST when the BIOS attempts to
translate the cylinders and heads. The solutions are the same as for
the 504 MB limit.
- 2.0/2.1 GB (software)
-
This limit derives from how the FAT filesystem tracks hard disk
space. FAT16 cannot use clusters larger than 32,768 bytes, or 64
512-byte sectors per cluster, and can track at most 65,536 clusters.
Multiplying the numbers yields 2,147,483,648 bytes, or 2 GB. The only
solution is use multiple partitions, each no larger than 2 GB. The
FAT32 filesystem used by Windows 95B/98/2000 and the filesystems used
by Windows NT/2000/XP and Linux are not subject to this limit.
- 3.27 GB
-
Some BIOSs do not properly handle cylinder counts higher than 6,322,
enforcing a limit of about 3.27 GB. A BIOS has this limit if the PC
hangs when you enter a value of 6,322 cylinders or more. The
solutions are the same as those listed for the 504 MB limit.
- 4.2 GB
-
Some interfaces store the number of heads reported by the BIOS as an
8-bit number, calculated modulo-256. If the BIOS reports 256 heads,
these interfaces fill the 8-bit register with zeros (the least
significant eight bits), assume that the drive has zero heads, and
refuse to allow it to be configured. This problem occurs if the drive
reports 16 heads and 8,192 or more cylinders to the bit shift
translation, effectively limiting drive size to 1,024 cylinders
x 128 heads x 63 sectors/track x 512
bytes/sector = 4,227,858,432 bytes. Because LBA translation never
reports more than 255 heads, this problem does not occur with BIOSs
that use LBA translation. A BIOS has this limit if the computer hangs
at boot after you create partitions on the drive. The solutions are
the same as those listed for the 504 MB limit.
- 8.4 GB
-
LBA translation uses ID words 1, 3, and 6 of the Identify
Drive command. These three 8-bit words allow maximum LBA
values of 16,384 cylinders, 16 heads, and 63 sectors/track, for a
total capacity of 8,455,716,864 bytes. Exceeding this limit requires
BIOS support for Extended Int13 functions, which is available with
recent BIOSs. Extended Int13 functions no longer transfer disk
addresses via host registers. Instead, they pass the
"address of an address" by placing
a 64-bit LBA Device Address packet in host
memory. The least-significant 28 bits of this packet contain a
standard 28-bit LBA address. If the drive uses LBA, the
least-significant 28 bits of this packet may be passed directly to
the ATA registers. If the drive does not support LBA, the host
translates the full LBA address to a CHS address, allowing the full
capacity of the disk drive to be used, within overall ATA and LBA
limitations.
The BIOS may have this limit if: (1) the PC hangs during POST when it
tries to translate cylinders and heads; (2) the total disk capacity
reported to the operating system is 8.4 GB or less; or (3) strange
things happen when you partition or format the disk. For example,
fdisk may refuse to partition the drive,
claiming that it is read-only, or format may
report a huge number of bad sectors. The solutions are the same as
those listed for the 504 MB limit.
- 32 GB
-
An Award BIOS dated earlier than June 1999 does not recognize drives
larger than 32 GB. Current versions of the Award BIOS no longer have
this limit. Many motherboard manufacturers have posted updates for
their products that use an Award BIOS.
- 128 GB / 137 GB
-
This is an absolute limit that results from the 28-bit LBA used by
ATA/ATAPI interfaces prior to ATA/ATAPI-6, which limits the total
number of sectors to 268,435,456. Using standard 512-byte sectors,
137,438,953,472 bytes (128 GB) is the largest drive that can be
supported by ATA/ATAPI-5.
- 128 PB / 144 PB
-
ATA/ATAPI-6 supports 48-bit LBA, which increases the maximum possible
drive size using 512-byte sectors to 144,115,188,075,855,872 bytes
(128 PB). It would require about 500,000 of the largest hard drives
currently available to total 128 PB, so we don't
expect this limit to become an issue anytime soon.
13.1.8.1 The Maxtor Big Drive Initiative
The 28-bit LBA used by ATA/ATAPI-5 and earlier interfaces limits hard
drive sizes to 128 GB (binary, although drive makers use decimal, and
so refer to 128 GB as 137 GB). Before 2001 this was not an issue, but
during 2001 drive makers began shipping SCSI hard drives larger than
128 GB. They wanted to produce larger ATA drives as well, but 28-bit
LBA made that impossible unless those drives used sector sizes larger
than the standard 512 bytes. Using nonstandard sector sizes
introduces severe compatibility problems, so there was no easy
solution.
Maxtor Corporation solved the problem with its Big Drive initiative,
which was subsequently adopted in ATA/ATAPI-6. Big Drive specifies a
new ATA interface that uses 48-bit LBA. Those additional 20 bits
allow the new interface to address a binary million
(220) times more sectors than a 28-bit ATA
interface, or (248 x 512
bytes/sector) = 144,115,188,075,855,872 bytes = 128 PB. Drive makers,
of course, call that "144 PB."
|
Don't bother telling us that binary values are
actually kibibytes, mebibytes, and gibibytes, which use
"bi" to indicate binary rather than
decimal values. We're aware of this ill-considered
initiative, and we don't know anyone who uses those
words in real life. As an alternative, we propose everyone use the
original definition of a kilobyte as 1,024 bytes, and substitute
kidebyte, medebyte, or gidebyte when referring to decimal values. You
heard it here first.
|
|
Compatibility between old and new drives
and interfaces was a key issue, and Maxtor did an excellent job of
maintaining that. 28-bit ATA/ATAPI drives can be used on a 48-bit
interface, although doing so limits that interface to 28-bit LBA for
all drives connected to it. Conversely, 48-bit hard drives can be
installed on a 28-bit ATA interface, although doing so limits that
drive to 128 GB usable capacity. SATA uses 48-bit LBA natively, and
so has no realistic limit on drive capacity.
13.1.8.2 Working around ATA capacity limits
The cheapest and easiest workaround for ATA capacity limits is to
update the system BIOS to a version that supports extended Int13. A
BIOS update, if available, allows an older motherboard to support
drives up to the 128 GB limit. The best (and possibly only) place to
get an updated BIOS is from the computer or motherboard maker. This
is true because, although many makers use BIOSs based on Phoenix,
Award, or AMI core BIOS code, they may have made changes to it that
render an updated generic BIOS unusable and may have also impacted
the extended Int13 support present in the core BIOS code. That said,
here are the BIOS levels necessary to eliminate the 8.4 and 32 GB
limits:
- American Megatrends, Inc. (AMI)
-
AMI BIOSs dated January 1, 1998 or later include extended Int13
support. Contact AMI directly for update information (http://www.ami.com or 800-828-9264).
- Award
-
Award BIOSs dated after November 1997 include extended Int13 support
and can be used with disks larger than 8.4 GB. Award recommends
contacting Unicore for BIOS updates (http://www.unicore.com or 800-800-2467).
- Phoenix
-
All Phoenix BIOSs are Version 4. It's the revision
level that counts. Phoenix BIOSs at Revision 6 or higher have
extended Int13 support. Phoenix recommends contacting Micro Firmware
for BIOS updates (http://www.firmware.com or 800-767-5465).
A BIOS update cannot eliminate the 128 GB ATA/ATAPI-5 limit because
older ATA interface hardware supports only 28-bit LBA. If you require
48-bit LBA to support drives larger than 128 GB, install a PCI ATA
interface card from Promise, SIIG, or another manufacturer, or
replace the motherboard.
13.1.9 ATA RAID
RAID
(Redundant Array of Inexpensive Disks) is a
means by which data is distributed over two or more physical hard
drives to improve performance and increase data safety. A RAID can
survive the loss of any one drive without losing data. RAID was
formerly a SCSI technology and was generally limited by cost to use
on servers and professional workstations. That changed a few years
ago when Promise Technology (http://www.promise.com) introduced a line of
ATA-based RAID controllers that combine the benefits of RAID with the
low cost of ATA, making RAID a realistic alternative for small
servers and individual PCs.
|
ATA drives are so inexpensive that RAID is increasingly popular for
protecting data on desktop systems. Do note, though, that using RAID
isn't a substitute for backing up. RAID protects
against drive failures, but does nothing to protect you against
corrupted or accidentally deleted files. Never count on RAID to
protect you against anything but drive failures.
|
|
For several
years, we've used Promise ATA RAID cards on a couple
of our servers and desktop systems. If you're
building a small server, or even if you just want to protect the data
on your PC, look into ATA RAID products. Many manufacturers now
produce ATA RAID cards, and some premium motherboards include
embedded ATA RAID support.
There are five defined
levels of RAID, numbered RAID 1 through RAID 5, although not all of
those levels are appropriate for PC environments. Some or all of the
following RAID levels are supported by current ATA RAID adapters:
- JBOD
-
JBOD
, or Just a Bunch Of
Drives, is a non-RAID operating mode that most RAID
adapters support. With JBOD, two or more physical drives can be
logically melded to appear to the operating system as one larger
drive. Data is written to the first drive until it is full, then to
the second drive until it is full, and so on. In the past, when drive
capacities were smaller, JBOD arrays were used to create single
volumes large enough to store huge databases. With 100+ GB drives now
common, there is seldom a good reason to use JBOD. The downside of
JBOD is that failure of any drive renders the entire array
inaccessible. Because the likelihood of a drive failure is
proportionate to the number of drives in the array, a JBOD is less
reliable than one large drive. The performance of a JBOD is the same
as that of the drives that make up the array.
- RAID 0
-
RAID 0
, also called disk
striping, is not really RAID at all because it provides no
redundancy. With RAID 0, data is written interleaved to two or more
physical drives. Because writes and reads are split across two or
more drives, RAID 0 provides the fastest reads and writes of any RAID
level, with both write and read performance noticeably faster than
that provided by a single drive. The downside of RAID 0 is that the
failure of any drive in the array causes the loss of all data stored
on all drives in the array, which means data stored on a RAID 0 array
is actually more at risk than data stored on a single drive. RAID 0
is most appropriate for use as
"scratchpad" storage—e.g.,
for temporary working files, or as part of a stacked RAID (covered
later).
- RAID 1
-
RAID 1
, also called disk
mirroring, duplicates all writes to two or more physical
disk drives. Accordingly, RAID 1 offers the highest level of data
redundancy at the expense of halving the amount of disk space visible
to the operating system. The overhead required to write the same data
to two drives means RAID 1 writes are typically a bit slower than
writes to a single drive. Conversely, because the same data can be
read from either drive, an intelligent RAID 1 adapter may improve
read performance slightly relative to a single drive by queuing read
requests for each drive separately, allowing it to read the data from
whichever drive happens to have its heads nearest the requested data.
It is also possible for a RAID 1 array to use two physical host
adapters to eliminate the disk adapter as a single point of failure.
In such an arrangement, called disk duplexing,
the array can continue operating after the failure of one drive, one
host adapter, or both (if they are on the same channel).
- RAID 5
-
RAID 5
,
also called disk striping with parity, requires
at least three physical disk drives. Data is written blockwise to
alternating drives, with parity blocks interleaved. For example, in a
RAID 5 array that comprises three physical drives, the first 64 KB
data block may be written to the first drive, the second data block
to the second drive, and a parity block to the third drive.
Subsequent data blocks and parity blocks are written to the three
drives in such a way that data blocks and parity blocks are
distributed equally across all three drives. Parity blocks are
calculated such that if either of their two data blocks is lost, it
may be reconstructed using the parity block and the remaining data
block. A failure of any one drive in the RAID 5 array causes no data
loss because the lost data blocks can be reconstructed from the data
and parity blocks on the remaining two drives.
A RAID 5 array with a failed drive (called a degraded
array) can be read from and written to, although until the
failed drive is replaced redundancy is lost and performance is
snail-like because much of the data must be reconstructed on-the-fly
using parity data. After the failed drive is replaced, the array must
be rebuilt. This process takes a long time because each data and
parity block from the failed drive must be reconstructed from the
data and parity blocks on the good drives. During a rebuild, read and
write performance of the RAID 5 is typically so slow that the array
is for all practical purposes unusable until the rebuild completes.
In normal operation, a RAID 5 provides somewhat better read
performance than a single drive. RAID 5 write performance is
typically a bit slower than that of a single drive because of the
overhead involved in segmenting the data and calculating parity
blocks. Because most PCs and small servers do more reads than writes,
RAID 5 is often the best compromise between performance and data
redundancy.
-
Stacked RAID
-
By using four or more drives, it is possible to combine the benefits
of different RAID levels by using an array of arrays. For example,
you can combine the speed of RAID 0 with the security of RAID 1 by
creating two mirror sets (each of which has complete data redundancy)
and then stripe those mirror sets for increased performance. Such a
stacked RAID arrangement is informally referred
to as RAID 10, RAID 0+1, or
RAID 1+0. In a four-drive RAID 0+1 array, any
one drive can fail without loss of data. Any two drives can fail
without loss of data if the two drives are members of different
mirror sets. Similarly, a workstation or server may use more complex
stacked RAID arrangement, such as stacked RAID 5 and RAID 0, called
RAID 50, RAID 0+5, or
RAID 5+0.
RAID Levels 2 and 4 are seldom used in the PC environment. A few ATA
RAID adapters, such as the Promise UltraTrack, support RAID 3, which
is optimized for sequential reads of large files in applications such
as streaming video.
 |
