Input/Output Systems and Peripheral Devices 1

6. THE SCSI INTERFACE

This laboratory work presents various types of SCSI electrical interfaces, the SCSI
standards that have been developed, the SCSI bus, the structure of SCSI commands, the SCSI
device configuration, and the main SCSI programming interfaces.

6.1. Overview of the SCSI Interface

SCSI (Small Computer System Interface) originates from the SASI (Shugart Associ-
ates System Interface) interface, which has been developed in 1981 by the hard disk drive
manufacturer company Shugart Associates and has been designed for connecting disk drives
to the computer. This interface used logical addresses instead of physical addresses and com-
mands of 6 bytes each. In the same year, the X3T9 working committee of the American stand-
ards institute ANSI used the SASI specifications as basis for developing a standard of a paral-
lel interface, standard that has later been named SCSI-1 and has been published in 1986. The
same X3T9 committee developed the SCSI-2 standard, which has been published in its final
form in 1994. In 1993, another working committee, T10, started the activity for developing
the SCSI-3 version of the standard. The documents of this version have been published sepa-
rately, during several years, starting from 1996. These documents continue to be updated to-
day with new versions.
SCSI is not a disk interface, that is, a certain type of controller, but rather a system
interface composed of a bus to which several devices can be attached. One of these devices,
the host adapter, operates as a bridge between the SCSI bus and the system bus. The SCSI bus
does not communicate directly with the peripheral devices, such as disk drives, but with the
controller that is included into these drives.
A single SCSI bus can accept up to 8 or 16 physical units, called SCSI units, out of
which one is the SCSI adapter. The physical units can be magnetic disk drives, magnetic tape
drives, optical drives, scanners, printers. Most systems can support up to four SCSI adapters.
One of the reasons that delayed acceptance of the SCSI interface in the personal com-
puter market was the lack of a standard for the host adapter, interface drivers, and BIOS pro-
gram. Due to the lack of an interface standard, several problems occurred, such as the impos-
sibility to use the disk drives outside the SCSI bus, the impossibility to boot the operating
system from these drives, and to use some operating systems. These problems were solved by
developing the SCSI standards.
SCSI is an interface designed especially for workstations and high-performance serv-
ers. In the first version of the SCSI interface, the bus clock frequency was 5 MHz, and the
maximum transfer rate was 5 MB/s. In today’s versions of this interface, the bus clock fre-
quency is 80 MHz or 160 MHz, and the maximum transfer rates are of 320 MB/s and 640
MB/s, respectively.
Besides these parallel versions of the SCSI interface, a serial version of the interface
has also been developed. This serial SCSI interface, called Serial Attached SCSI, will gradu-
ally replace the parallel SCSI interface. Both the parallel and the serial SCSI technologies are
promoted by the SCSI Trade Association (www.scsita.org).

6.2. Types of SCSI Electrical Interfaces

There are two main types of SCSI electrical interfaces: normal and differential. In
case of a normal SCSI interface (also called Single-Ended – SE), for each signal transmitted
2 6. The SCSI Interface

on the bus there is a single wire. The receiver circuits at the other end of the cable detect the
electrical voltages on the bus lines. The cost of such an interface is low, but the signals trans-
mitted are affected by noise and electromagnetic interference.
With differential SCSI interfaces, for each signal there is a pair of wires. One of the
wires carries the same type of signal that is carried by normal SCSI interfaces. The second
wire carries a signal that is obtained by the logical inversion of the original signal. The re-
ceiver circuits do not have to detect the absolute value of the voltage, but only the difference
between the signals received on the two wires. Differential interfaces have several ad-
vantages: increased immunity to noise, the possibility of transfers with higher speeds, and
greater cable length. With normal SCSI interfaces, the cable length can be up to 6 m for low
frequencies of the SCSI bus, or up to 3 m for higher frequencies. With differential SCSI inter-
faces, the cable length can be up to 25 m or 12 m, depending on the bus frequency.
In the early differential SCSI interfaces, called High Voltage Differential (HVD), rela-
tively high voltages were used for signal transmission. For this reason, low-power and small-
sized interface chips could not be developed. For implementing the interface, circuits using
several chips were required, which increased the cost of the interface. Another problem of this
differential interface was that devices with differential interface could not be attached to the
same bus to which devices with normal interface were attached, because the higher voltages
of the differential interface could destroy the receiver circuits of the devices with normal in-
terface. Because of the high cost and the incompatibility with devices having normal inter-
face, the HVD differential interface was very rarely used, and its specifications were removed
from the latest version of the SCSI standard.
Instead of the HVD electrical interface, a new differential interface has been devel-
oped, called LVD (Low Voltage Differential), which uses low voltages. This interface can be
implemented with low-power and inexpensive circuits. Another advantage of this interface is
that it allows connecting LVD devices to a normal SE SCSI bus, without the risk of damaging
the devices’ interface circuits. Optionally, devices with LVD interface can be designed so that
they can be connected to both LVD and SE buses. These multimode devices detect if they are
connected to an SE bus and will operate in the SE mode that is compatible with this bus.
When there is even a single SE device connected to a SCSI bus, all LVD devices connected to
the same bus will run in SE mode, at a reduced frequency of the bus (up to 20 MHz).
Because the SCSI connectors are the same for various types of the electrical interface,
and connecting a device with HVD interface to a bus with SE or LVD devices may cause dam-
ages to these devices, it is necessary to differentiate between various types of interfaces. De-
vices with normal interface can be distinguished from those with HVD or LVD differential
interfaces by the symbols labeled on them. Different symbols have been adopted in the indus-
try for the normal and differential SCSI interfaces (Figure 6.1).

Figure 6.1. Symbols for the SCSI electrical interfaces: (a) SE normal interface; (b) HVD differential interface;
(c) LVD differential interface; (d) LVD/SE multimode interface.
 Input/Output Systems and Peripheral Devices 3

6.3. SCSI Standards

The SCSI interface standards have been developed by several working groups within
the ANSI institute, which functioned or function as accredited committees for standardization.
The SCSI standards define the physical and electrical parameters of an I/O bus, which is used
for daisy-chaining the peripheral devices.

6.3.1. SCSI-1
The first standard of the SCSI interface, SCSI-1 (ANSI X3.131-1986), has been ap-
proved by the ANSI institute in 1986. Even before the approval of this standard, the hard disk
manufacturers were worried that many of the commands and features specified by the stand-
ard were optional. For that reason, no guarantee existed that a particular peripheral will sup-
port all these commands. To solve this problem, the industry requested the X3T9 committee
that developed the SCSI-1 specifications to extend the mandatory commands and features of
these specifications. Because the specifications were nearly finished, the ANSI institute has
approved the standard in the existing form and established a new working group to develop a
set of 18 basic SCSI commands. This set has been called CCS (Common Command Set) and
will become the minimum set of commands supported by all peripherals. This command set
became the basis of the SCSI-2 standard.

6.3.2. SCSI-2
The SCSI-2 standard is an improved version of the SCSI-1 standard, with new fea-
tures and options added. Normally, SCSI-1 and SCSI-2 devices are compatible, but devices
conforming to the SCSI-1 standard do not support the additional features introduced in the
SCSI-2 standard.
The X3T9 working committee finalized the SCSI-2 standard in 1990, but the docu-
ment has been withdrawn at the end of the same year for some changes, to be made before the
final publishing by the ANSI institute. The final version has been approved only at the begin-
ning of 1994, although this document (ANSI X3.131-1994) contains very few changes com-
pared to the initial version from 1990. Basically, all the specifications of the SCSI-1 standard
can also be found in the SCSI-2 standard.
In addition to supporting the set of 18 basic SCSI commands, the SCSI-2 standard
also contains new specifications, such as new commands for CD-ROM drives (including the
possibility to use CD audio discs), tape drives, and other peripherals. Likewise, a faster ver-
sion of the interface has been defined, called Fast SCSI-2, and a 16-bit version, called Wide
SCSI-2. Another improvement defined by the SCSI-2 standard is the possibility to write the
commands into a command queue, which allows a peripheral to accept several commands and
to execute them in an order that is considered the most efficient. This possibility is important
for multitasking operating systems.
Some of the changes specified by the SCSI-2 standard are minor. For instance, in the
SCSI-1 standard the parity on the SCSI bus was optional, while in the SCSI-2 standard it is
mandatory to use a parity bit. Another requirement is that initiator devices, such as host
adapters, have to provide power to the terminator devices of the interface. However, most
devices already fulfilled this requirement.
The SCSI-2 standard defines the following optional features:
 Fast SCSI;
 Wide SCSI;
 Command queuing;
 New commands;
 Improved terminators.
Fast SCSI refers to the capability to perform synchronous transfers at higher speeds.
With this version, transfer rates of 10 MB/s can be achieved on the standard 8-bit SCSI bus.
4 6. The SCSI Interface

When combined with a 16-bit Wide SCSI interface, this configuration allows transfer rates of
20 MB/s.
Wide SCSI enables data transfers on 16-bit or 32-bit buses. For these variants other
cables are required than for the 8-bit variant. The standard 50-conductor cable used for 8-bit
transfers is called the A cable. For the 16-bit Wide SCSI variant a 68-conductor P cable is
required. For the 32-bit Wide SCSI variant, which was never actually implemented, two ca-
bles are required: the 68-conductor P cable and the 68-conductor Q cable.
According to the SCSI-1 standard, an initiator device, such as a host adapter, can send
a single command per device. According to the SCSI-2 standard, the host adapter can send up
to 256 commands to a single device, which will store the commands in a command queue and
will process them before sending a response on the SCSI bus. The device can even change the
order of commands to enable the most efficient execution possible.
The commands from the Common Command Set (CCS), which were already used in
the industry, have been included officially into the SCSI-2 standard. The common command
set was defined especially for disk drives and did not include specific commands for other
types of devices. Many of the old commands have been changed and new commands have
been added. For instance, new command sets have been added for CD-ROM drives, other
optical drives, scanners, communication devices etc.
For correct operation of the single-ended SCSI bus, termination resistors with tight
tolerances are needed. The 132- passive terminators, defined in the SCSI-1 standard, are not
adequate for the higher transfer speeds allowed by the SCSI-2 standard. These passive termi-
nators can cause signal reflections, and errors can occur when transfer rates increase or when
more devices are attached to the bus. According to the SCSI-2 standard, active components
should be used as terminators, which ensure an impedance of 110  and improve system in-
tegrity.

6.3.3. SCSI-3
SCSI-3 represents a collection of standards that were developed and published sepa-
rately. These standards have been divided into categories that include: standards for primary
commands, standards for commands that are specific to various device classes, standards for
the communication protocols, and standards for the physical interconnects. In addition, a
SCSI Architectural Model (SAM) exists for the physical and electrical interfaces. The SCSI-3
standards have been developed and are updated by the T10 technical committee within the
InterNational Committee on Information Technology Standards (INCITS) that is accredited by
ANSI. The working documents of the T10 committee are available at address www.t10.org.
The main improvements introduced by the SCSI-3 standard include the following:
 Ultra2 (Fast-40) SCSI;
 Ultra3 (Fast-80DT) SCSI;
 Ultra4 (Fast-160DT) SCSI;
 Ultra5 (Fast-320DT) SCSI;
 New Low Voltage Differential signaling;
 Elimination of High Voltage Differential signaling.
Figure 6.2 presents the main components of the SCSI-3 collection of standards. Most
of the individual standards have several versions, which are indicated in the figure.
The main components of the SCSI-3 family of standards are described next.
SCSI Architecture Model (SAM) defines the model of SCSI systems, the functional
partitioning of the SCSI-3 set of standards, and the requirements applicable to all SCSI-3 im-
plementations.
 Input/Output Systems and Peripheral Devices 5

Figure 6.2. The main components of the SCSI-3 collection of standards.

Commands represent specifications that define the classes of devices and a device
model for each class. These specifications define the commands that must be implemented by
all devices or those that are specific to various classes of devices, and describe the rules that
must be followed by an initiator device when it sends commands to another device. The main
standards related to the commands are the following:
 Primary Commands (SPC): basic commands for all SCSI devices;
 Block Commands (SBC): commands for direct-access devices such as magnetic disk
drives;
 Stream Commands (SSC): commands for sequential-access devices such as magnetic
tape drives;
 Medium Changer Commands (SMC): commands for medium changer devices such as
jukeboxes for audio discs;
 Multimedia Commands (MMC): commands for optical disc drives such as CD-ROM,
CD-R/E (Recordable/Erasable), DVD;
 Controller Commands (SCC): commands for I/O controllers, e.g., for RAID (Redun-
dant Array of Independent Disks) disk drive sets.
Protocols represent specifications that define the communication rules between vari-
ous SCSI devices.
Interconnects contain specifications that define various physical interfaces. The
SCSI-3 collection of standards defines several types of interfaces. For a long time, the most
used was the SCSI Parallel Interface (SPI), until it begun to be replaced with the Serial At-
tached SCSI (SAS) serial interface. Another serial interface is IEEE 1394, especially used for
video applications. Fibre Channel is a high-performance serial interface, which allows com-
munication over optical fiber. The Serial Storage Architecture (SSA) serial interface is de-
signed to connect disk drives or RAID disk arrays to servers. InfiniBand is another high-
performance serial interface, intended for the connection of processors with high-speed pe-
ripherals such as storage devices. iSCSI (Internet SCSI) allows server computers to access
remote disk volumes using the existing network infrastructure and the TCP/IP protocol.
We consider the parallel SCSI interface. Several versions of this interface have been
developed, as the electrical protocol has been improved. The various types of the SCSI parallel
interface are presented in Table 6.1.
6 6. The SCSI Interface

Table 6.1. Types of parallel SCSI interfaces specified in the SCSI-3 standards.

Clock Maximum
Standard Technology Alternate Name Frequency Width Transfer Rate
(MHz) (MB/s)
SPI Fast-20 Ultra 20 8 20
SPI Fast-20/Wide Ultra/Wide 20 16 40
SPI-2 Fast-40 Ultra2 40 8 40
SPI-2 Fast-40/Wide Ultra2/Wide 40 16 80
SPI-3 Fast-80DT Ultra3 (Ultra160) 80 16 160
SPI-4 Fast-160DT Ultra4 (Ultra320) 80 16 320
SPI-5 Fast-320DT Ultra5 (Ultra640) 160 16 640

In Table 6.1, DT (Double Transition) indicates two transfers performed in each clock
cycle, one at each edge of the clock signal. For the SPI-3, SPI-4, and SPI-5 interfaces, the SCSI
bus width is 16 bits. Starting with the SPI-2 interface, only the LVD differential signaling is used.

6.4. The SCSI Bus

6.4.1. Communication on the SCSI Bus

Communication on the SCSI bus takes place between a device that initiates the trans-
fer and a destination device. At any time, communication is performed only between two
devices, one initiator that selects and controls the target device that performs the requested
operation. Usually, a SCSI device has a fixed role of initiator or target, but some devices can
fulfill both roles.
An initiator device can address up to eight logical units attached to a target device.
For each data block logical addresses rather than physical addresses are used. For devices
with direct addressing, each logical unit can be interrogated to determine the number of data
blocks it contains. A logical unit may coincide with a peripheral device or it may be part of
that device.
The SCSI standards define the signal levels on the bus, their logical function, the
communication protocol, and the command sequences. All devices must allow to use the
asynchronous protocol defined by the standards for data transfers. In addition, an optional
protocol is defined for synchronous transfers. Similarly, a protocol is specified for sending
messages in order to control the interface.
The SCSI bus uses a distributed arbitration system to support multiple initiators and
the concurrent execution of I/O operations. A priority system allows granting the bus to the
SCSI device with the highest priority out of those that request the bus. The time required to
perform the arbitration is independent of the number of devices that request the bus and is
lower than 10 s.
The initiator may request the SCSI bus and may select a certain target device. The
target may request the transfer of data, command, or status information on the data bus, and in
some cases it may request the bus and it may reselect the initiator in order to continue an op-
eration.

6.4.2. Asynchronous and Synchronous SCSI Protocol

The SCSI interface allows to use either the asynchronous or synchronous protocol for
data transfers. By using the synchronous protocol faster data transfers can be achieved. Im-
plementation of the asynchronous protocol is mandatory for all devices, while implementation
of the synchronous protocol is optional.
In the asynchronous SCSI protocol, a device first issues a request (by asserting the
REQ signal) and then waits for an acknowledgement from the initiator. The initiator places a
byte or word on the data bus and asserts the ACK signal. The device reads the byte or word,
de-asserts the REQ signal, and waits for the initiator to de-assert the ACK signal. These opera-
tions are then repeated for the next byte or word.
 Input/Output Systems and Peripheral Devices 7

This REQ/ACK handshake protocol requires that the signals propagate through the
SCSI cable twice for each data transfer. This requires a certain time; typically, the propaga-
tion delay is about 5.25 ns for each meter. The propagation delay is the primary speed limita-
tion of asynchronous SCSI transfers when long cables are used.
In the synchronous SCSI protocol, an acknowledgement is also required for each
request sent, but the acknowledgement can be delayed. Consequently, a device can send data
packets one after another without the propagation delays required in the asynchronous proto-
col. The speed is determined by the bus cycle time without regard to propagation delay, and is
therefore independent of cable length.
The first clock frequency specified for synchronous transfers was 5 MHz. This fre-
quency has been increased gradually in the newer versions of the interface, with each version
doubling the frequency of its predecessor version. Starting with the SPI-3 interface, the DT
(Double Transition) technology is used, so that data transfers take place on both the rising and
falling edge of each clock cycle.

6.4.3. SCSI Bus Signals

The 8-bit SCSI bus that uses the A cable contains 18 signals, out of which 9 data sig-
nals and 9 control signals. For the 16-bit and 32-bit versions there are bus extensions. The bus
signals are described next.
 BSY (Busy): A wired-OR signal that indicates the busy state of the bus.
 SEL (Select): A wired-OR signal used by an initiator device to select a target device
or by a target device to reselect an initiator device. The identifier of the selected de-
vice will appear on the data lines.
 C/D (Control/Data): Used by the target device to specify whether control or data in-
formation is sent on the data bus. The asserted state of this signal specifies that con-
trol information is sent.
 I/O (Input/Output): The target device controls with this signal the direction of data
transfer. The direction is considered from the initiator device’s viewpoint. The assert-
ed value indicates an input operation for the initiator. This signal is also used to dis-
tinguish between the Selection and Reselection phases.
 MSG (Message): The target indicates with this signal that a message is sent on the bus
(in the Message phase).
 REQ (Request): Generated by the target device to specify a transfer request using the
asynchronous protocol.
 ACK (Acknowledge): Generated by the initiator device to acknowledge an
asynchronous transfer request made by a target device by asserting the REQ signal.
 ATN (Attention): Used by an initiator device to indicate an attention condition for the
target device.
 RST (Reset): A wired-OR signal that initialized the SCSI bus and resets all the devic-
es attached to the bus.
 DB (7..0, P) (Data Bus). Represent the bidirectional data signals and the parity bit
signal, which form a data bus. DB (7) is the most significant bit and it has the highest
priority during the arbitration phase. DB (P) is the odd-parity bit. The parity is defined
during the arbitration phase.
 DB (31..8, P1, P2, P3) (Data Bus). Represent the extension of the data bus. DB (P1,
P2, P3) are odd-parity bits for DB (15..8), DB (23..16), and DB (31..24), respectively.
 TERMPWR (Terminator Power). Represents the power signal for the bus terminators.
8 6. The SCSI Interface

6.4.4. SCSI Bus Phases

During operation, the SCSI bus passes through several distinct states, called phases.
A phase describes the direction of transfer and the contents of information transferred. The
following phases can be distinguished:
 Bus Free;
 Arbitration;
 Selection;
 Reselection;
 Command;
 Data;
 Status;
 Message.
After reset or power up, the bus enters the Bus Free phase. This is followed by the
Arbitration phase, in which one of the devices gains control of the bus. If the arbitration fails,
the bus returns to the Bus Free phase. If the arbitration succeeds, the bus enters the Selection
or Reselection phase, in which a target and an initiator device that will execute a command
are selected. After determining the two devices, one or more information transfer phases
(Command, Data, Status, Message) follow. The last information transfer phase is normally
the Message In phase, in which a DISCONNECT or COMMAND COMPLETE message is sent
to the initiator, followed by the Bus Free phase.
Some examples of messages are described next.
 COMMAND COMPLETE: Sent by the target device to the initiator to indicate that a
command has been completed and a valid status has been sent to the initiator.
 DISCONNECT: Sent by the target device to inform an initiator that the existing con-
nection is going to be interrupted and a later reconnect will be required to complete
the current operation.
 INITIATOR DETECTED ERROR: Sent by the initiator to inform a target device that
an error (e. g., parity error) has occurred that does not preclude the target from retry-
ing the operation.
 ABORT: Sent by the initiator to the target device to abandon the current operation.

 SYNCHRONOUS DATA TRANSFER: Sent by the initiator to a target device to select

the synchronous transfer protocol.

6.4.5. SCSI Command Example

A READ command will be used as example to explain the various bus phases and
signals. This command transfers data from a target device to the initiator.
The command begins in the Bus Free phase. Next, an Arbitration phase follows, in
which one or more devices compete to take control of the bus. Each of the devices asserts the
BSY signal and one of the data lines. Each device has a unique ID from 0 to 7 (or from 0 to
15), and each device asserts one of the data lines corresponding to its own ID. Each ID is
assigned a priority, with 7 (or 15) having the highest priority and 0 the lowest priority. If more
than one device asserts its ID during the Arbitration phase, then the device with the highest
priority takes control of the bus.
The device that has won the arbitration becomes the initiator. This device enters the
Selection phase by asserting the SEL signal. During this phase, the initiator asserts both data
lines corresponding to its own ID and the target device’s ID. After a delay, the initiator de-
asserts the BSY signal. When the target device detects that the SEL signal is asserted, BSY and
 Input/Output Systems and Peripheral Devices 9

I/O are de-asserted, and recognizes its ID, it will assert the BSY signal. When the initiator
detects that the BSY signal is asserted, it releases the data bus and de-asserts the SEL signal.
Next, the target device indicates that it has entered the Command phase by asserting
the C/D signal. This signal will remain asserted during this phase. The target device then as-
serts the REQ signal to request the first byte of the command from the initiator. The initiator
places the first byte of the command on the data bus and asserts the ACK signal. The target
device reads the byte and it de-asserts the REQ signal; the initiator then de-asserts the ACK
signal. The first byte of the command contains the operation code, which indicates the number
of bytes remained to be transferred. These additional bytes are transferred with the same
REQ/ACK protocol.
After the target device has received and interpreted the command, it places the bus in
the Data In phase by de-asserting the C/D signal (indicating data information) and it asserts
the I/O signal (indicating the direction from the target to the initiator). The target device plac-
es the first byte or word of the requested data on the data bus and asserts the REQ signal. The
initiator asserts the ACK signal after it has read the data. Additional data bytes or words are
transferred with the same REQ/ACK protocol.
After transferring all the requested data, the target device places the bus in the Status
phase and transfers a status byte to the initiator, indicating that it has successfully completed
the transfer. The C/D signal is again asserted and the I/O signal remains asserted. The
REQ/ACK protocol is used to transfer the status byte.
Finally, the target device places the bus in the Message In phase by asserting the MSG
signal and transferring the COMMAND COMPLETE message byte. Once this message is re-
ceived by the initiator, the target device releases all bus signals to place the bus in the Bus
Free phase.

6.5. SCSI Commands

The SCSI standards specify a high-level command set that must be supported by
SCSI devices. Mandatory and optional commands are defined, out of which some are com-
mon for all device types, while others are specific for various types of devices.

6.5.1. Structure of a Command Descriptor Block

A command is specified as a Command Descriptor Block, which is sent to the target
device. For some commands, the command descriptor block is followed by a list of parame-
ters that are sent during the Data Out phase. The command descriptor block begins with an
operation code in the first byte and ends with a control byte.
There are typical structures of the descriptor block for 6-byte, 10-byte, and 12-byte
commands. Figure 6.3 presents the typical structure of a command descriptor block for 10-byte
commands.

Bit 7 6 5 4 3 2 1 0
Byte
0 Operation Code
1 Logical Unit Number Reserved
2 MSB
3 Logical Block Address
4
5 LSB
6 Reserved
7 MSB Transfer Length, Parameter List Length,
8 Allocation Length LSB
9 Control
Figure 6.3. Structure of a descriptor block for 10-byte SCSI commands.

The operation code contains two fields: the group code (bits 7..5) and command code
(bits 4..0). The three bits of the group code allow 8 groups of codes. The five bits of the
10 6. The SCSI Interface

command code allow 32 command codes in each group. Therefore, there are a total number of
256 possible operation codes. The group code defines separate groups for 6-byte, 10-byte, or
12-byte commands, as well as for vendor-specific codes.
The logical unit number is defined in the IDENTIFY message. The target will ignore
the logical unit number specified in the command descriptor block if an IDENTIFY message
has been received. It is recommended to set to zero the logical unit number in the descriptor
block. This field was included into the command descriptor block for compatibility with some
SCSI-1 devices.
The logical block address within a logical unit or a volume partition starts with block
zero and must be contiguous up to the last logical block of the logical unit or partition. A de-
scriptor block for 6-byte commands contains a logical block address of 21 bits. Descriptor
blocks for 10-byte and 12-byte commands contain logical block addresses of 32 bits.
The transfer length specifies the length of data that must be transferred, usually, in
number of blocks. For some commands, the transfer length specifies the number of bytes that
are to be transferred. For commands that use a byte for the transfer length, a transfer length
between 1 and 255 indicates the number of blocks that must be transferred by a single com-
mand. A value of 0 indicates 256 blocks. For commands that use several bytes for the transfer
length, a length of 0 indicates that no data are to be transferred.
The parameter list length is used to specify the number of bytes that are transferred
during the Data Out phase, bytes that represent the parameters sent to the target device.
The allocation length specifies the maximum number of bytes allocated by the initia-
tor device for the data sent from the target device. The target must finish the Data In phase
when a number of bytes indicated by the allocation length have been transferred. This length
is used to limit the number of bytes returned to the initiator device.
The control field has the following structure:

7 6 5 4 3 2 1 0
Vendor-specific Reserved Flag Link

The Link bit is used to continue the I/O process after the current command completes
successfully. If the Link bit is 1, the target must enter the Command phase after the current
command completes. Implementation of this bit is optional.
The Flag bit specifies the message that must be returned by the target device to the
initiator if the Link bit is 1 and the command completes without errors. Usually, this bit is
used to generate an interrupt of the initiator device between linked commands. Implementa-
tion of this bit is optional.
If the Link bit is 0 and the Flag bit is 1, the target device must return the Check Con-
dition status. This status indicates that an unpredicted event occurred during the operation.
The initiator device must send an additional command (Request Sense) to determine the un-
predicted event.
If the Link bit is 1 and the Flag bit is 0, and the command completes successfully, the
target must send the LINKED COMMAND COMPLETE message. If the Link bit is 1 and the
Flag bit is 1, and the command completes successfully, the target must send the LINKED
COMMAND COMPLETE (WITH FLAG) message. These messages indicate the end of a
linked command.

6.5.2. Command Examples

The SCSI-3 standard defines commands that can be used for all types of devices and
commands that are specific for various types of devices. The main device types for which
commands are defined are the following:
 Direct-access devices (magnetic disks);
 Sequential-access devices (magnetic tapes);
 Printers;
 Processors (intelligent devices);
 Input/Output Systems and Peripheral Devices 11

 WORM (Write Once, Read Multiple) discs;

 CD-ROM discs (including audio discs);
 DVD discs;
 Scanners;
 Optical memories (several optical discs, e. g., CD-R);
 Communication devices (network nodes).
Table 6.2 presents example SCSI commands for direct-access devices.

Table 6.2. SCSI commands for direct-access devices.

Command Name Code Command Name Code

COPY 0x18 REQUEST SENSE 0x03
COPY AND VERIFY 0x3A RESERVE (6) 0x16
FORMAT UNIT 0x04 RESERVE (10) 0x56
INQUIRY 0x12 SEEK (10) 0x2B
READ (6) 0x08 SEND DIAGNOSTIC 0x1D
READ (10) 0x28 TEST UNIT READY 0x00
READ (12) 0xA8 VERIFY 0x2F
READ BUFFER 0x3C WRITE (6) 0x0A
READ CAPACITY 0x25 WRITE (10) 0x2A
READ DEFECT DATA (10) 0x37 WRITE (12) 0xAA
READ DEFECT DATA (12) 0xB7 WRITE AND VERIFY 0x2E
RELEASE (6) 0x17 WRITE BUFFER 0x3B
RELEASE (10) 0x57 WRITE SAME 0x41

6.6. SCSI Device Configuration

To configure a SCSI device, two operations must be performed: setting the SCSI
identifier and installing the terminators.

6.6.1. Setting the SCSI Identifier

Up to 8 or 16 SCSI devices can be attached to a SCSI bus and each of them must
have a unique SCSI identifier (ID) in order to avoid conflicts. One of these identifiers, usually
7 or 15, which has the highest priority, is assigned to the host adapter. There are adapters that
allow booting the operating system only from a disk drive with a specific ID. For example,
older Adaptec host adapters required the boot disk drive to have the ID 0. Newer adapters
allow booting the operating system from any drive, regardless its ID.
Usually, setting the ID requires positioning of three or four jumpers on the drive. The
configuration of jumpers results from the binary representation of the ID values from 0 to 7 or
from 0 to 15. In other cases, the ID can be set with a rotary switch.
Current SCSI systems are Plug-and-Play and the ID assignment is performed
automatically by the operating system and the SCSI adapter. A protocol called SCAM (SCSI
Configured Automagically) is used, which interrogates the SCSI devices and assigns unique
identifiers for each of them, so that no conflicts will exist.

6.6.2. Installing the Terminators

The SCSI bus, like other buses, needs terminators at both ends of the bus. The incor-
rect termination of the bus lines represents one of the problems that may occur when using
SCSI devices. If the host adapter is at one end of the bus, it must have its terminator enabled.
If the adapter is in the middle of the bus and both internal and external bus links are present,
the terminator in the adapter must be disabled and the devices at the two ends of the bus must
have their terminators installed.
There are several types of terminators for the SCSI bus:
 Passive terminators;
 Active terminators;
12 6. The SCSI Interface

 FPT (Forced Perfect Termination).

Passive terminators, composed of resistors, allow fluctuations of the bus signals. The
signal levels depend on the voltage drop on these resistors. Usually, passive terminators are
adequate for short distances, of up to 1 m, but for longer distances active terminators are
needed.
Active terminators use, instead of voltage dividers composed of resistors, one or more
voltage regulators that ensure a constant voltage. The voltage regulators ensure termination of
signals on the SCSI bus at a correct voltage level. The SCSI-2 and SCSI-3 standards recom-
mend using active terminators at both ends of the bus.
FPT terminators represent a type of active terminators that use stable voltage levels
obtained with diodes. They eliminate fluctuations of signal levels, especially at high transfer
speeds or great cable lengths.
Usually, external SCSI devices have an input and an output SCSI connector, so that
several devices can be connected in a chain. When the device is at one end of the SCSI bus, a
terminator must be installed in the output connector.
Some devices have embedded terminators that may be enabled or disabled through a
switch or by their removal. Other devices do not have embedded terminators and for these
external terminator modules must be used. Terminator modules are available in various con-
figurations of the connectors, which also include pass-through terminators. These are neces-
sary for devices installed at the end of the bus and which only have a single SCSI connector.
Pass-through terminators are also commonly used in internal installations in which the devic-
es do not have embedded terminating resistors. Many disk drives use such terminators for
internal installations to save space on the logic board.

6.6.3. Other Configurations

There are additional options that can be configured for the SCSI drives. Following are
the most common additional settings:
 Start on command;
 SCSI parity;
 Terminator power;
 Synchronous transfer negotiation.

6.6.3.1. Start on Command

If there are multiple disk drives connected to a system, they can be configured so that not
all the drives start simultaneously, immediately when the system is powered on. In case of
simultaneous start, the power consumption would increase significantly, the power supply can be
overloaded, which can cause the system to hang or have other startup problems.
Usually, SCSI drives allow delaying the start of the spindle motor. When the SCSI
adapters initialize the bus, most of them send out a command called Start Unit to each of the
disk drives in succession. By positioning a jumper on the disk drive, the spindle motor will
only start when the drive receives the Start Unit command form the host adapter. Because the
adapter sends this command to all the disk drives, in succession, from the highest priority
drive to the lowest priority drive, the higher priority drives can be configured to start first, and
then lower priority drives to start sequentially. There are some adapters that do not send the
Start Unit command. In this case, disk drives can be configured so that they delay the start
with a certain number of seconds, rather than wait for the Start Unit command.
There is no need to enable the delayed start function for drives with separate power
supply. This function is needed for internal drives that are powered from the same power sup-
ply that runs the system.
 Input/Output Systems and Peripheral Devices 13

6.6.3.2. SCSI Parity

Using a parity bit allows limited error detection. Most of the host adapters support
parity checking, so this option should be enabled on every device. This function is optional,
because there are older adapters that do not use the SCSI parity bit, so for these adapters the
parity checking must be disabled.

6.6.3.3. Terminator Power

Active terminators positioned at each end of the SCSI bus must be powered from at
least one device attached to the bus. Usually, the terminator power is supplied by the host
adapter, but there are some exceptions.
It is not a problem that the terminator power is supplied by multiple devices, because
each source is protected with diodes. For simplicity, often all devices are configured to supply
terminator power. If none of the devices supplies terminator power, the bus will not be
terminated correctly and will not function properly.

6.6.3.4. Synchronous Transfer Negotiation

By default, the SCSI bus uses the asynchronous transfer protocol. It is possible to
select the synchronous transfer protocol, which is faster, through a procedure called synchro-
nous transfer negotiation. Before information is transferred across the bus, the initiator device
and the target device negotiate how the transfer will take place. If both devices support the
synchronous protocol, the transfer will be performed with this protocol.
Some older devices do not respond to a request for synchronous transfer and can even be
disabled when they receive such requests. For this reason, host adapters and devices that support
synchronous transfer negotiation often have a jumper that can be used to disable this negotiation,
so they can work with older devices.

6.7. SCSI Adapters

The most important manufacturers of SCSI adapters are Adaptec and Future Domain.
SCSI adapters can be easily installed, because all of their functions can be configured by
software. The configuration information is written to a memory on the logic board of the
adapter. The most important features of these adapters are the following:
 The ROM memory of the adapters contains software tools that allow to fully config-
ure the adapters;
 The interrupt levels, ROM addresses, DMA and I/O port addresses, SCSI bus parity,
and physical addresses of devices can be configured by the software;
 Terminators can be enabled and disabled by the software;
 No additional drivers are needed for more than two disk drives;
 Start on command of the drives is supported;
 The operating system can be loaded from any SCSI device.

6.8. SCSI Drivers

Each of the SCSI devices attached to the bus, except for disk drives, needs an external
SCSI driver. Usually, disk drives have the drivers included in the BIOS program of the SCSI
adapter. The external drivers are designed for a specific type of device and are specific to a
certain SCSI adapter.
Two types of standard drivers for interfacing with the host adapter became popular. If a
standard driver is available for the host adapter, manufacturers can create more easily new drivers
for peripheral devices, which can communicate with the universal host adapter driver. This
14 6. The SCSI Interface

eliminates dependence on a particular type of host adapter. Universal drivers perform the link
between the host adapter and operating system.
One of these drivers is the Advanced SCSI Programming Interface (ASPI), initiated
by the Adaptec company. Later on, other SCSI device vendors have used the ASPI driver for
their products. Many operating systems include an ASPI driver for several SCSI host adapt-
ers.
The second driver is called CAM (Common Access Method) and has been created by
Future Domain and NCR companies. The CAM driver is a protocol approved by ANSI and
enables to control several SCSI host adapters. Future Domain also provides a CAM-ASPI
converter for its host adapters.
Current Windows operating systems use the SCSI Pass Through Interface (SPTI) for
accessing SCSI devices. This programming interface has been developed by Microsoft Corpo-
ration and is accessible to software using the DeviceIoControl() function, which allows to
send a control code directly to a device driver.

6.9. The Serial Attached SCSI Interface

Serial Attached SCSI (SAS) represents the serial version of the SCSI interface. It uses
a point-to-point serial protocol and the standard SCSI command set. The SAS serial interface
offers compatibility with second-generation SATA (Serial ATA – Serial Advanced Technology
Attachment) disk drives, which may be connected to SAS backplanes.
The SAS standards have been developed by the T10 Technical Committee of INCITS
(International Committee for Information Technology Standards). The first version of the
SAS interface standard has been published in 2003, and an improved version (SAS-1.1) has
been published in 2005. Both versions specify a serial interface with a maximum speed of 3
Gbits/s. The SAS-2 standard, which has been published by the INCITS committee in 2009,
defines the second generation of the SAS interface. This version of the standard introduces a
serial link with a maximum speed of 6 Gbits/s, a physical layer that is compatible with the
SATA interface, and protocols for transferring SCSI commands to SAS devices and ATA
commands to SATA devices. The SAS-2.1 version of the standard, which has been approved
in 2010, defines a number of enhancements to the SAS-2 standard, including additional con-
nectors and power management features. The SAS-3 version of the standard, which has been
published in 2013, defines the third generation of the SAS interface, with a maximum speed
of 12 Gbits/s. The SAS-4 version of the standard, which is under development and expected
to be released in 2017, will increase the maximum speed of the interface to 24 Gbits/s; pro-
ducts based on this version are expected in 2018 and 2019.
Figure 6.4 illustrates the symbol of the SAS, SAS-2, and SAS-3 interfaces.

Figure 6.4. Symbols of the SAS, SAS-2, and SAS-3 interfaces.

The SAS interface is a point-to-point architecture, with each device connecting direct-
ly to a SCSI port rather than connecting to a shared bus. Since the bandwidth is not shared by
several devices, as in the case of the parallel SCSI interface, transfers with higher speeds are
possible. By using a point-to-point connection, data reliability and the ability to locate failures
are improved compared to a shared-bus architecture.
The SAS protocol specifies a full-duplex communication between the SAS controller
and a disk drive. Therefore, read and write operations can be performed at the same time,
which increases performance. Comparatively, SATA drives use half-duplex communication,
so that when data are being sent to the drive for writing, any data that needs to be read from
the drive must wait for the previous communication to complete.
 Input/Output Systems and Peripheral Devices 15

A large number of SAS or SATA disk drives can be connected to a SAS controller
port by using SAS expanders. An expander allows a single initiator to communicate with a
number of SAS/SATA target devices. A SAS expander is similar to a switch in a network,
which allows multiple systems to be connected using a single switch port. The cost of a sys-
tem containing an expander is much lower compared to the cost of a system containing a large
port-count SAS controller or multiple smaller port-count controllers. By using expanders, up
to 16,384 physical links are possible.
A SAS domain consists of a set of SAS devices that communicate with one another
by means of cables and backplanes, with or without expanders. Each SAS port and expander
device in a SAS domain is assigned a globally unique identifier by the device manufacturer.
This 64-bit identifier that represents the SAS address is called World Wide Name (WWN) and
it uniquely identifies the device in the SAS domain just as a SCSI ID identifies a device at-
tached to a parallel SCSI bus. Out of the 64 bits, 24 bits represent the vendor company identi-
fier and 40 bits represent the vendor-specific identifier. Unlike for the parallel SCSI interface,
there is no need to set manually the addresses in a serial SCSI system, since all the configura-
tion is performed automatically. The SAS interface does not require installing terminators like
the parallel SCSI interface.
The SAS interface uses differential signaling and data scrambling to reduce electro-
magnetic interference. It is possible to combine up to four ports with the same address into a
wide port, which enable to increase the data rate.
SAS disk drives are dual-ported, which means that they can be directly connected to
and controlled by two SAS controllers at the same time. This capability allows a redundant
system to be built. When one of the SAS controllers fails, the other is still able to access the
SAS disk drives and the data stored on those drives.
A SAS system can use either SAS or SATA disk drives. A SATA drive can be con-
nected to a SAS system via an expander or a chip that implements the SAS protocol. This
compatibility is possible because the SAS and SATA drive connectors are similar. These con-
nectors are shown in Figure 6.5. Both connectors have the same number of pins and the pins
have the same size and shape. However, the SATA drive connector has a notch, which is miss-
ing in the SAS drive connector. This notch prevents a SAS drive from being plugged into a
SATA system. A SAS system uses a connector that allows a disk drive with a notch ( SATA) or
a disk drive without the notch (SAS) to be installed.

Figure 6.5. SATA and SAS disk drive connectors.

SAS disk drives have higher performance compared to SATA disk drives. The spindle
speed of SAS drives is between 10,000 and 15,000 revolutions per minute (RPM), while the
spindle speed of SATA drives is between 5,400 and 7,200 RPM. The higher spindle speed
reduces the access time. The full-duplex communication supported by SAS drives also con-
tributes to the higher performance of these drives. SAS drives are dual-ported, and this allows
them to communicate with two host adapters or controllers simultaneously, which improves
16 6. The SCSI Interface

data availability. SAS drives are also much more reliable than SATA drives and are designed
for much more intensive use.

6.10. Applications
6.10.1. Answer the following questions:
a. What are the improvements introduced by the SCSI-3 standards for the parallel
SCSI interface?
b. What is the difference between the asynchronous and synchronous SCSI protocols?
c. What are the operations required for configuring a SCSI system?
d. What are the advantages of the serial SCSI interface compared to the parallel
SCSI interface?
6.10.2. Present examples of possible configurations for a SCSI bus composed of de-
vices A, B, C, and D (including the host adapter). Show the placement of the adapter and ter-
minators.
6.10.3. Create a Windows application for checking whether the ASPI manager is ini-
tialized correctly and for displaying the number of SCSI adapters in the system. As model for
the Windows application, use the AppScroll-e application available on the laboratory web page in
the AppScroll-e.zip archive. Perform the following operations to create the application project:
1. In the Microsoft Visual Studio programming environment, create a new project by se-
lecting General  Empty Project in the New Project dialog window.
2. Copy to the project folder the files contained in the AppScroll-e.zip archive and add all
the files to the project.
3. Copy to the project folder the files from the WNASPI32.zip archive, available on the la-
boratory web page. Add to the project the wnaspi32.h and scsidefs.h header files.
4. Open the AppScroll-e.cpp source file, delete the #include "Hw.h" directive and add
#include directives to include the wnaspi32.h and scsidefs.h header files.
5. In the AppScroll() function, delete the sequences for initializing the Hw library
with the HwOpen() function and for closing the Hw library with the HwClose()
function.
6. Select Build  Build Solution and make sure that the application builds without er-
rors.
Use the specifications of the ASPI programming interface from the ASPI32.pdf docu-
ment (available on the laboratory web page or at reference [1]) and the wnaspi32.dll library to
write a function with no input parameters that performs the following operations:
 Load the wnaspi32.dll library with the LoadLibrary() function;
 Determine the address of the GetASPI32SupportInfo() and SendASPI32-
Command() functions with the GetProcAddress() function.

 Call the GetASPI32SupportInfo() function and display whether the ASPI manag-
er is initialized correctly.
 If the ASPI manager is initialized correctly, display the number of SCSI adapters in
the system.
After writing the function, include the call to this function in the AppScroll() func-
tion and verify the operation of the function.
6.10.4. Extend Application 6.10.3 by writing a function that performs the following
operations for each SCSI adapter, starting with adapter number 0:
 Input/Output Systems and Peripheral Devices 17

 Call the SendASPI32Command() function with command code SC_HA_INQUIRY

and display the following information from the SRB_HAInquiry structure: the string
describing the ASPI manager (the HA_ManagerId member of the structure); the
SCSI identifier of host adapter (the HA_SCSI_ID member of the structure); the string
describing the host adapter (the HA_Identifier member of the structure).
 Determine the maximum number of SCSI devices (targets); this number is contained
in the HA_Unique[3] byte of the SRB_HAInquiry structure.
 For every SCSI device and for every logical unit (from 0 to 7) of a device, call the
SendASPI32Command() function with command code SC_GET_DEV_TYPE, and
display the SCSI device number, the logical unit number, and the device type (the
SRB_DeviceType member of the SRB_GDEVBlock structure).
After writing the function, include the call to this function in the AppScroll() func-
tion and verify the operation of the function.
Note
 For obtaining information about some SCSI devices with the wnaspi32.dll library,
administrator privileges are required.

Bibliography
[1] Adaptec, Inc., “ASPI for Win32 Technical Reference”, 2001,
http://www.zianet.com/jgray/dat/files/ASPI32.pdf.
[2] Baruch, Z., Sisteme de intrare/ieşire, Îndrumător de lucrări de laborator, U.T.PRES,
Cluj-Napoca, 1998.
[3] Czekalski, M., “Serial Attached SCSI Architecture”, SCSI Trade Association, 2004,
http://www.scsita.org/library/presentations/SAS_Architecture_Overview.pdf.
[4] Mueller, S., Upgrading and Repairing PCs, 16th Edition, Que Publishing, 2005.
[5] Rosch, W. L., Hardware Bible, Sixth Edition, Que Publishing, 2003.
[6] Stallings, W., Computer Organization and Architecture. Designing for Performance,
Fifth Edition, Pearson Education, 2001.
[7] T10 Committee, “Information Technology – SCSI-3 Block Commands (SBC)”, T10
Project 0996-M, Revision 08c, 1997, http://www.t10.org/cgi-bin/ac.pl?t=f&f=sbc-
r08c.pdf.
[8] The PC Guide, “Small Computer Systems Interface (SCSI)”, 2001,
http://www.pcguide.com/ref/hdd/if/scsi/index.htm.
[9] Wikimedia Foundation, Inc., “Serial Attached SCSI”, 2015,
https://en.wikipedia.org/wiki/Serial_attached_SCSI.