Wiley Encyclopedia of Computer Science and Engineering
Tham khảo tài liệu 'wiley encyclopedia of computer science and engineering', công nghệ thông tin, quản trị web phục vụ nhu cầu học tập, nghiên cứu và làm việc hiệu quả
Wiley Encyclopedia of Computer
Science and Engineering
FullTitle of Book: Wiley Encyclopedia Of Computer Science And Engineering
Editor(s): Wah
Publisher: Wiley-interscience
YearPublished: Nov., 2008
ISBN-10: 0471383937
ISBN-13: 978-0471383932
Size& Format: 2362 pages
• Applications
• Computer Vision
• Computing Milieux
• Data
• Foundation and Theory
• Hardware and Architecture
• Image Processing and Visualization
• Intelligent Systems
• IS
• Parallel and Distributed Systems
• Software
A
ASYNCHRONOUS TRANSFER MODE peak rate, statistical multiplexing allows a large number of
NETWORKS bursty sources to share the network’s bandwidth.
Since its birth in the mid-1980s, ATM has been fortified
Asynchronous transfer mode, or ATM, is a network transfer by a number of robust standards and realized by a signifi-
technique capable of supporting a wide variety of multi- cant number of network equipment manufacturers. Inter-
media applications with diverse service and performance national standards-making bodies such as the ITU and
requirements. It supports traffic bandwidths ranging from independent consortia like the ATM Forum have developed
a few kilobits per second (e.g., a text terminal) to several a significant body of standards and implementation agree-
hundred megabits per second (e.g., high-definition video) ments for ATM (1,4). As networks and network services
and traffic types ranging from continuous, fixed-rate traffic continue to evolve toward greater speeds and diversities,
(e.g., traditional telephony and file transfer) to highly ATM will undoubtedly continue to proliferate.
bursty traffic (e.g., interactive data and video). Because
of its support for such a wide range of traffic, ATM was ATM STANDARDS
designated by the telecommunication standardization sec-
tor of the International Telecommunications Union (ITU-T,
The telecommunication standardization sector of the ITU,
formerly CCITT) as the multiplexing and switching tech- the international standards agency commissioned by the
nique for Broadband, or high-speed, ISDN (B-ISDN) (1). United Nations for the global standardization of telecom-
ATM is a form of packet-switching technology. That is, munications, has developed a number of standards for ATM
ATM networks transmit their information in small, fixed- networks. Other standards bodies and consortia (e.g., the
length packets called cells, each of which contains 48 octets ATM Forum, ANSI) have also contributed to the develop-
(or bytes) of data and 5 octets of header information. The ment of ATM standards. This section presents an overview
small, fixed cell size was chosen to facilitate the rapid of the standards, with particular emphasis on the protocol
processing of packets in hardware and to minimize the reference model used by ATM (2).
amount of time required to fill a single packet. This is
particularly important for real-time applications such as
Protocol Reference Model
voice and video that require short packetization delays.
ATM is also connection-oriented. In other words, a The B-ISDN protocol reference model, defined in ITU-T
virtual circuit must be established before a call can take recommendation I.321, is shown in Fig. 1(1). The purpose of
place, where a call is defined as the transfer of information the protocol reference model is to clarify the functions that
between two or more endpoints. The establishment of a ATM networks perform by grouping them into a set of
virtual circuit entails the initiation of a signaling process, interrelated, function-specific layers and planes. The refer-
during which a route is selected according to the call’s ence model consists of a user plane, a control plane, and a
quality of service requirements, connection identifiers at management plane. Within the user and control planes is a
each switch on the route are established, and network hierarchical set of layers. The user plane defines a set of
resources such as bandwidth and buffer space may be functions for the transfer of user information between
reserved for the connection. communication endpoints; the control plane defines control
Another important characteristic of ATM is that its functions such as call establishment, call maintenance, and
network functions are typically implemented in hardware. call release; and the management plane defines the opera-
With the introduction of high-speed fiber optic transmis- tions necessary to control information flow between planes
sion lines, the communication bottleneck has shifted from and layers and to maintain accurate and fault-tolerant
the communication links to the processing at switching network operation.
nodes and at terminal equipment. Hardware implementa- Within the user and control planes, there are three
tion is necessary to overcome this bottleneck because it layers: the physical layer, the ATM layer, and the ATM
minimizes the cell-processing overhead, thereby allowing adaptation layer (AAL). Figure 2 summarizes the functions
the network to match link rates on the order of gigabits per of each layer (1). The physical layer performs primarily bit-
second. level functions, the ATM layer is primarily responsible for
Finally, as its name indicates, ATM is asynchronous. the switching of ATM cells, and the ATM adaptation layer is
Time is slotted into cell-sized intervals, and slots are responsible for the conversion of higher-layer protocol
assigned to calls in an asynchronous, demand-based man- frames into ATM cells. The functions that the physical,
ner. Because slots are allocated to calls on demand, ATM ATM, and adaptation layers perform are described in more
can easily accommodate traffic whose bit rate fluctuates detail next.
over time. Moreover, in ATM, no bandwidth is consumed
unless information is actually transmitted. ATM also gains Physical Layer
bandwidth efficiency by being able to multiplex bursty The physical layer is divided into two sublayers: the phy-
traffic sources statistically. Because bursty traffic does sical medium sublayer and the transmission convergence
not require continuous allocation of the bandwidth at its sublayer (1).
1
Wiley Encyclopedia of Computer Science and Engineering, edited by Benjamin Wah.
Copyright # 2008 John Wiley & Sons, Inc.
2 ASYNCHRONOUS TRANSFER MODE NETWORKS
error control is the insertion of an 8-bit CRC in the ATM cell
header to protect the contents of the ATM cell header. Cell
delineation is the detection of cell boundaries. Transmis-
sion frame adaptation is the encapsulation of departing
cells into an appropriate framing structure (either cell-
based or SDH-based).
ATM Layer
The ATM layer lies atop the physical layer and specifies the
functions required for the switching and flow control of
ATM cells (1).
There are two interfaces in an ATM network: the user-
network interface (UNI) between the ATM endpoint and
the ATM switch, and the network-network interface (NNI)
Figure 1. Protocol reference model for ATM. between two ATM switches. Although a 48-octet cell pay-
load is used at both interfaces, the 5-octet cell header differs
slightly at these interfaces. Figure 3 shows the cell header
structures used at the UNI and NNI (1). At the UNI, the
Physical Medium Sublayer. The physical medium (PM)
header contains a 4-bit generic flow control (GFC) field, a
sublayer performs medium-dependent functions. For
24-bit label field containing virtual path identifier (VPI)
example, it provides bit transmission capabilities including
and virtual channel identifier (VCI) subfields (8 bits for the
bit alignment, line coding and electrical/optical conversion.
VPI and 16 bits for the VCI), a 2-bit payload type (PT) field, a
The PM sublayer is also responsible for bit timing (i.e., the
1-bit cell loss priority (CLP) field, and an 8-bit header error
insertion and extraction of bit timing information). The PM
check (HEC) field. The cell header for an NNI cell is
sublayer currently supports two types of interface: optical
identical to that for the UNI cell, except that it lacks the
and electrical.
GFC field; these four bits are used for an additional 4 VPI
bits in the NNI cell header.
Transmission Convergence Sublayer. Above the physical
The VCI and VPI fields are identifier values for virtual
medium sublayer is the transmission convergence (TC)
channel (VC) and virtual path (VP), respectively. A virtual
sublayer, which is primarily responsible for the framing
channel connects two ATM communication endpoints. A
of data transported over the physical medium. The ITU-T
virtual path connects two ATM devices, which can be
recommendation specifies two options for TC sublayer
switches or endpoints, and several virtual channels may
transmission frame structure: cell-based and synchronous
be multiplexed onto the same virtual path. The 2-bit PT
digital hierarchy (SDH). In the cell-based case, cells are
field identifies whether the cell payload contains data or
transported continuously without any regular frame struc-
control information. The CLP bit is used by the user for
ture. Under SDH, cells are carried in a special frame
explicit indication of cell loss priority. If the value of the
structure based on the North American SONET (synchro-
CLP is 1, then the cell is subject to discarding in case of
nous optical network) protocol (3). Regardless of which
congestion. The HEC field is an 8-bit CRC that protects the
transmission frame structure is used, the TC sublayer is
contents of the cell header. The GFC field, which appears
responsible for the following four functions: cell rate decou-
only at the UNI, is used to assist the customer premises
pling, header error control, cell delineation, and transmis-
network in controlling the traffic flow. At the time of writ-
sion frame adaptation. Cell rate decoupling is the insertion
ing, the exact procedures for use of this field have not been
of idle cells at the sending side to adapt the ATM cell
agreed upon.
stream’s rate to the rate of the transmission path. Header
Figure 2. Functions of each layer in the protocol reference model.
ASYNCHRONOUS TRANSFER MODE NETWORKS 3
Figure 3. ATM cell header structure.
ATM Layer Functions bit rate (UBR). ITU-T defines four service categories,
namely, deterministic bit rate (DBR), statistical bit rate
The primary function of the ATM layer is VPI/VCI transla-
(SBR), available bit rate (ABR), and ATM block transfer
tion. As ATM cells arrive at ATM switches, the VPI and VCI
(ABT). The first of the three ITU-T service categories
values contained in their headers are examined by the
correspond roughly to the ATM Forum’s CBR, VBR, and
switch to determine which outport port should be used to
ABR classifications, respectively. The fourth service cate-
forward the cell. In the process, the switch translates the
gory, ABT, is solely defined by ITU-T and is intended for
cell’s original VPI and VCI values into new outgoing VPI
bursty data applications. The UBR category defined by the
and VCI values, which are used in turn by the next ATM
ATM Forum is for calls that request no quality of service
switch to send the cell toward its intended destination. The
guarantees at all. Figure 4 lists the ATM service categories,
table used to perform this translation is initialized during
their quality of service (QoS) parameters, and the traffic
the establishment of the call.
descriptors required by the service category during call
An ATM switch may either be a VP switch, in which case
establishment (1,4).
it translates only the VPI values contained in cell headers,
The constant bit rate (or deterministic bit rate) service
or it may be a VP/VC switch, in which case it translates the
category provides a very strict QoS guarantee. It is targeted
incoming VPI/VCI value into an outgoing VPI/VCI pair.
at real-time applications, such as voice and raw video,
Because VPI and VCI values do not represent a unique end-
which mandate severe restrictions on delay, delay variance
to-end virtual connection, they can be reused at different
(jitter), and cell loss rate. The only traffic descriptors
switches through the network. This is important because
required by the CBR service are the peak cell rate and
the VPI and VCI fields are limited in length and would be
the cell delay variation tolerance. A fixed amount of band-
quickly exhausted if they were used simply as destination
width, determined primarily by the call’s peak cell rate, is
addresses.
reserved for each CBR connection.
The ATM layer supports two types of virtual connec-
The real-time variable bit rate (or statistical bit rate)
tions: switched virtual connections (SVC) and permanent,
service category is intended for real-time bursty applica-
or semipermanent, virtual connections (PVC). Switched
tions (e.g., compressed video), which also require strict QoS
virtual connections are established and torn down dyna-
guarantees. The primary difference between CBR and
mically by an ATM signaling procedure. That is, they exist
VBR-rt is in the traffic descriptors they use. The VBR-rt
only for the duration of a single call. Permanent virtual
service requires the specification of the sustained (or aver-
connections, on the other hand, are established by network
age) cell rate and burst tolerance (i.e., burst length) in
administrators and continue to exist as long as the admin-
addition to the peak cell rate and the cell delay variation
istrator leaves them up, even if they are not used to trans-
mit data.
Other important functions of the ATM layer include cell
multiplexing and demultiplexing, cell header creation and
extraction, and generic flow control. Cell multiplexing is
the merging of cells from several calls onto a single trans-
mission path, cell header creation is the attachment of a 5-
octet cell header to each 48-octet block of user payload, and
generic flow control is used at the UNI to prevent short-
term overload conditions from occurring within the net-
work.
ATM Layer Service Categories
The ATM Forum and ITU-T have defined several distinct
service categories at the ATM layer (1,4). The categories
defined by the ATM Forum include constant bit rate (CBR),
real-time variable bit rate (VBR-rt), non-real-time variable
bit rate (VBR-nrt), available bit rate (ABR), and unspecified Figure 4. ATM layer service categories.
4 ASYNCHRONOUS TRANSFER MODE NETWORKS
tolerance. The ATM Forum also defines a VBR-nrt service
category, in which cell delay variance is not guaranteed.
The available bit rate service category is defined to
exploit the network’s unused bandwidth. It is intended
for non-real-time data applications in which the source is
amenable to enforced adjustment of its transmission rate. A
minimum cell rate is reserved for the ABR connection and
therefore guaranteed by the network. When the network Figure 5. Service classification for AAL.
has unused bandwidth, ABR sources are allowed to
increase their cell rates up to an allowed cell rate (ACR),
a value that is periodically updated by the ABR flow control
mechanism (to be described in the section entitled ‘‘ATM AAL service class A corresponds to constant bit rate
Traffic Control’’). The value of ACR always falls between services with a timing relation required between source
the minimum and the peak cell rate for the connection and and destination. The connection mode is connection-
is determined by the network. oriented. The CBR audio and video belong to this class.
The ATM Forum defines another service category for Class B corresponds to variable bit rate (VBR) services.
non-real-time applications called the unspecified bit rate This class also requires timing between source and desti-
(UBR) service category. The UBR service is entirely best nation, and its mode is connection-oriented. The VBR audio
effort; the call is provided with no QoS guarantees. The and video are examples of class B services. Class C also
ITU-T also defines an additional service category for non- corresponds to VBR connection-oriented services, but the
real-time data applications. The ATM block transfer ser- timing between source and destination needs not be
vice category is intended for the transmission of short related. Class C includes connection-oriented data transfer
bursts, or blocks, of data. Before transmitting a block, such as X.25, signaling, and future high-speed data ser-
the source requests a reservation of bandwidth from the vices. Class D corresponds to connectionless services. Con-
network. If the ABT service is being used with the immedi- nectionless data services such as those supported by LANs
ate transmission option (ABT/IT), the block of data is sent and MANs are examples of class D services.
at the same time as the reservation request. If bandwidth is Four AAL types (Types 1, 2, 3/4, and 5), each with a
not available for transporting the block, then it is simply unique SAR sublayer and CS sublayer, are defined to
discarded, and the source must retransmit it. In the ABT support the four service classes. AAL Type 1 supports
service with delayed transmission (ABT/DT), the source constant bit rate services (class A), and AAL Type 2 sup-
waits for a confirmation from the network that enough ports variable bit rate services with a timing relation
bandwidth is available before transmitting the block of between source and destination (class B). AAL Type 3/4
data. In both cases, the network temporarily reserves was originally specified as two different AAL types (Type 3
bandwidth according to the peak cell rate for each block. and Type 4), but because of their inherent similarities, they
Immediately after transporting the block, the network were eventually merged to support both class C and class D
releases the reserved bandwidth. services. AAL Type 5 also supports class C and class D
services.
ATM Adaptation Layer
AAL Type 5. Currently, the most widely used adaptation
The ATM adaptation layer, which resides atop the ATM
layer is AAL Type 5. AAL Type 5 supports connection-
layer, is responsible for mapping the requirements of
oriented and connectionless services in which there is no
higher layer protocols onto the ATM network (1). It oper-
timing relation between source and destination (classes C
ates in ATM devices at the edge of the ATM network and is
and D). Its functionality was intentionally made simple in
totally absent in ATM switches. The adaptation layer is
order to support high-speed data transfer. AAL Type 5
divided into two sublayers: the convergence sublayer (CS),
assumes that the layers above the ATM adaptation layer
which performs error detection and handling, timing, and
can perform error recovery, retransmission, and sequence
clock recovery; and the segmentation and reassembly
numbering when required, and thus, it does not provide
(SAR) sublayer, which performs segmentation of conver-
these functions. Therefore, only nonassured operation is
gence sublayer protocol data units (PDUs) into ATM cell-
provided; lost or corrupted AAL Type 5 packets will not be
sized SAR sublayer service data units (SDUs) and vice
corrected by retransmission.
versa.
Figure 6 depicts the SAR-SDU format for AAL Type 5
In order to support different service requirements, the
(5,6). The SAR sublayer of AAL Type 5 performs segmenta-
ITU-T has proposed four AAL-specific service classes.
tion of a CS-PDU into a size suitable for the SAR-SDU
Figure 5 depicts the four service classes defined in recom-
payload. Unlike other AAL types, Type 5 devotes the entire
mendation I.362 (1). Note that even though these AAL
48-octet payload of the ATM cell to the SAR-SDU; there is
service classes are similar in many ways to the ATM layer
no overhead. An AAL specific flag (end-of-frame) in the
service categories defined in the previous section, they are
not the same; each exists at a different layer of the protocol
reference model, and each requires a different set of func-
tions.
Figure 6. SAR-SDU format for AAL Type 5.
ASYNCHRONOUS TRANSFER MODE NETWORKS 5
the CS-PDU, and handling of lost and misinserted cells. At
the time of writing, both the SAR-SDU and CS-PDU for-
mats for AAL Type 2 are still under discussion.
AAL Type 3/4. AAL Type 3/4 mainly supports services
that require no timing relation between the source and
destination (classes C and D). At the SAR sublayer, it
defines a 48-octet service data unit, with 44 octets of
user payload; a 2-bit payload type field to indicate whether
the SDU is at the beginning, middle, or end of a CS-PDU; a
4-bit cell sequence number; a 10-bit multiplexing identifier
that allows several CS-PDUs to be multiplexed over a single
Figure 7. CS-PDU format, segmentation and reassembly of AAL VC; a 6-bit cell payload length indicator; and a 10-bit CRC
Type 5. code that covers the payload. The CS-PDU format allows for
up to 65535 octets of user payload and contains a header
and trailer to delineate the PDU.
The functions that AAL Type 3/4 performs include seg-
ATM PT field of the cell header is set when the last cell of a mentation and reassembly of variable-length user data and
CS-PDU is sent. The reassembly of CS-PDU frames at the error handling. It supports message mode (for framed data
destination is controlled by using this flag. transfer) as well as streaming mode (for streamed data
Figure 7 depicts the CS-PDU format for AAL Type 5 transfer). Because Type 3/4 is mainly intended for data
(5,6). It contains the user data payload, along with any services, it provides a retransmission mechanism if neces-
necessary padding bits (PAD) and a CS-PDU trailer, which sary.
are added by the CS sublayer when it receives the user
information from the higher layer. The CS-PDU is padded ATM Signaling
using 0 to 47 bytes of PAD field to make the length of the CS-
ATM follows the principle of out-of-band signaling that was
PDU an integral multiple of 48 bytes (the size of the SAR-
established for N-ISDN. In other words, signaling and data
SDU payload). At the receiving end, a reassembled PDU is
channels are separate. The main purposes of signaling are
passed to the CS sublayer from the SAR sublayer, and CRC
(1) to establish, maintain, and release ATM virtual con-
values are then calculated and compared. If there is no
nections and (2) to negotiate (or renegotiate) the traffic
error, the PAD field is removed by using the value of length
parameters of new (or existing) connections (7). The ATM
field (LF) in the CS-PDU trailer, and user data is passed to
signaling standards support the creation of point-to-point
the higher layer. If an error is detected, the erroneous
as well as multicast connections. Typically, certain VCI and
information is either delivered to the user or discarded
VPI values are reserved by ATM networks for signaling
according to the user’s choice. The use of the CF field is
messages. If additional signaling VCs are required, they
for further study.
may be established through the process of metasignaling.
AAL Type 1. AAL Type 1 supports constant bit rate
services with a fixed timing relation between source and ATM TRAFFIC CONTROL
destination users (class A). At the SAR sublayer, it defines a
48-octet service data unit (SDU), which contains 47 octets of The control of ATM traffic is complicated as a result of
user payload, 4 bits for a sequence number, and a 4-bit CRC ATM’s high-link speed and small cell size, the diverse
value to detect errors in the sequence number field. AAL service requirements of ATM applications, and the diverse
Type 1 performs the following services at the CS sublayer: characteristics of ATM traffic. Furthermore, the configura-
forward error correction to ensure high quality of audio and tion and size of the ATM environment, either local or wide
video applications, clock recovery by monitoring the buffer area, has a significant impact on the choice of traffic control
filling, explicit time indication by inserting a time stamp in mechanisms.
the CS-PDU, and handling of lost and misinserted cells that The factor that most complicates traffic control in ATM
are recognized by the SAR. At the time of writing, the CS- is its high-link speed. Typical ATM link speeds are 155.52
PDU format has not been decided. Mbit/s and 622.08 Mbit/s. At these high-link speeds, 53-
byte ATM cells must be switched at rates greater than one
AAL Type 2. AAL Type 2 supports variable bit rate cell per 2.726 ms or 0.682 ms, respectively. It is apparent
services with a timing relation between source and desti- that the cell processing required by traffic control must
nation (class B). AAL Type 2 is nearly identical to AAL Type perform at speeds comparable to these cell-switching rates.
1, except that it transfers service data units at a variable bit Thus, traffic control should be simple and efficient, without
rate, not at a constant bit rate. Furthermore, AAL Type 2 excessive software processing.
accepts variable length CS-PDUs, and thus, there may Such high speeds render many traditional traffic control
exist some SAR-SDUs that are not completely filled with mechanisms inadequate for use in ATM because of their
user data. The CS sublayer for AAL Type 2 performs the reactive nature. Traditional reactive traffic control
following functions: forward error correction for audio and mechanisms attempt to control network congestion by
video services, clock recovery by inserting a time stamp in responding to it after it occurs and usually involves sending
6 ASYNCHRONOUS TRANSFER MODE NETWORKS
feedback to the source in the form of a choke packet. call at the time of call set-up. This decision is based on the
However, a large bandwidth-delay product (i.e., the traffic characteristics of the new call and the current net-
amount of traffic that can be sent in a single propagation work load. Usage parameter control enforces the traffic
delay time) renders many reactive control schemes ineffec- parameters of the call after it has been accepted into the
tive in high-speed networks. When a node receives feed- network. This enforcement is necessary to ensure that the
back, it may have already transmitted a large amount of call’s actual traffic flow conforms with that reported during
data. Consider a cross-continental 622 Mbit/s connection call admission.
with a propagation delay of 20 ms (propagation-bandwidth Before describing call admission and usage parameter
product of 12.4 Mbit). If a node at one end of the connection control in more detail, it is important to first discuss the
experiences congestion and attempts to throttle the source nature of multimedia traffic. Most ATM traffic belongs to
at the other end by sending it a feedback packet, the source one of two general classes of traffic: continuous traffic and
will already have transmitted over 12 Mb of information bursty traffic. Sources of continuous traffic (e.g., constant
before feedback arrives. This example illustrates the inef- bit rate video, voice without silence detection) are easily
fectiveness of traditional reactive traffic control mechan- handled because their resource utilization is predictable
isms in high-speed networks and argues for novel and they can be deterministically multiplexed. However,
mechanisms that take into account high propagation-band- bursty traffic (e.g., voice with silence detection, variable bit
width products. rate video) is characterized by its unpredictability, and this
Not only is traffic control complicated by high speeds, kind of traffic complicates preventive traffic control.
but it also is made more difficult by the diverse QoS require- Burstiness is a parameter describing how densely or
ments of ATM applications. For example, many applica- sparsely cell arrivals occur. There are a number of ways to
tions have strict delay requirements and must be delivered express traffic burstiness, the most typical of which are the
within a specified amount of time. Other applications have ratio of peak bit rate to average bit rate and the average
strict loss requirements and must be delivered reliably burst length. Several other measures of burstiness have
without an inordinate amount of loss. Traffic controls also been proposed (8). It is well known that burstiness
must address the diverse requirements of such applica- plays a critical role in determining network performance,
tions. and thus, it is critical for traffic control mechanisms to
Another factor complicating traffic control in ATM net- reduce the negative impact of bursty traffic.
works is the diversity of ATM traffic characteristics. In
ATM networks, continuous bit rate traffic is accompanied Call Admission Control. Call admission control is the
by bursty traffic. Bursty traffic generates cells at a peak process by which the network decides whether to accept
rate for a very short period of time and then immediately or reject a new call. When a new call requests access to the
becomes less active, generating fewer cells. To improve the network, it provides a set of traffic descriptors (e.g., peak
efficiency of ATM network utilization, bursty calls should rate, average rate, average burst length) and a set of quality
be allocated an amount of bandwidth that is less than their of service requirements (e.g., acceptable cell loss rate,
peak rate. This allows the network to multiplex more calls acceptable cell delay variance, acceptable delay). The net-
by taking advantage of the small probability that a large work then determines, through signaling, if it has enough
number of bursty calls will be simultaneously active. This resources (e.g., bandwidth, buffer space) to support the new
type of multiplexing is referred to as statistical multiplex- call’s requirements. If it does, the call is immediately
ing. The problem then becomes one of determining how best accepted and allowed to transmit data into the network.
to multiplex bursty calls statistically such that the number Otherwise it is rejected. Call admission control prevents
of cells dropped as a result of excessive burstiness is network congestion by limiting the number of active con-
balanced with the number of bursty traffic streams allowed. nections in the network to a level where the network
Addressing the unique demands of bursty traffic is an resources are adequate to maintain quality of service guar-
important function of ATM traffic control. antees.
For these reasons, many traffic control mechanisms One of the most common ways for an ATM network to
developed for existing networks may not be applicable to make a call admission decision is to use the call’s traffic
ATM networks, and therefore novel forms of traffic control descriptors and quality of service requirements to predict
are required (8,9). One such class of novel mechanisms that the ‘‘equivalent bandwidth’’ required by the call. The
work well in high-speed networks falls under the heading of equivalent bandwidth determines how many resources
preventive control mechanisms. Preventive control need to be reserved by the network to support the new
attempts to manage congestion by preventing it before it call at its requested quality of service. For continuous,
occurs. Preventive traffic control is targeted primarily at constant bit rate calls, determining the equivalent band-
real-time traffic. Another class of traffic control mechan- width is simple. It is merely equal to the peak bit rate of the
isms has been targeted toward non-real-time data traffic call. For bursty connections, however, the process of deter-
and relies on novel reactive feedback mechanisms. mining the equivalent bandwidth should take into account
such factors as a call’s burstiness ratio (the ratio of peak bit
Preventive Traffic Control rate to average bit rate), burst length, and burst interarri-
val time. The equivalent bandwidth for bursty connections
Preventive control for ATM has two major components: call
must be chosen carefully to ameliorate congestion and cell
admission control and usage parameter control (8). Admis-
loss while maximizing the number of connections that can
sion control determines whether to accept or reject a new
be statistically multiplexed.
ASYNCHRONOUS TRANSFER MODE NETWORKS 7
for violating cells. When traffic is bursty, a large number of
cells may be generated in a short period of time, while
conforming to the traffic parameters claimed at the time of
call admission. In such situations, none of these cells should
be considered violating cells. Yet in actual practice, leaky
bucket may erroneously identify such cells as violations of
admission parameters. A virtual leaky bucket mechanism
(also referred to as a marking method) alleviates these
Figure 8. Leaky bucket mechanism. disadvantages (11). In this mechanism, violating cells,
rather than being discarded or buffered, are permitted to
enter the network at a lower priority (CLP ¼ 1). These
violating cells are discarded only when they arrive at a
Usage Parameter Control. Call admission control is congested node. If there are no congested nodes along the
responsible for admitting or rejecting new calls. However, routes to their destinations, the violating cells are trans-
call admission by itself is ineffective if the call does not mitted without being discarded. The virtual leaky bucket
transmit data according to the traffic parameters it pro- mechanism can easily be implemented using the leaky
vided. Users may intentionally or accidentally exceed the bucket method described earlier. When the queue length
traffic parameters declared during call admission, thereby exceeds a threshold, cells are marked as ‘‘droppable’’
overloading the network. In order to prevent the network instead of being discarded. The virtual leaky bucket method
users from violating their traffic contracts and causing the not only allows the user to take advantage of a light network
network to enter a congested state, each call’s traffic flow is load but also allows a larger margin of error in determining
monitored and, if necessary, restricted. This is the purpose the token pool parameters.
of usage parameter control. (Usage parameter control is
also commonly referred to as policing, bandwidth enforce- Reactive Traffic Control
ment, or flow enforcement.)
Preventive control is appropriate for most types of ATM
To monitor a call’s traffic efficiently, the usage para-
traffic. However, there are cases where reactive control is
meter control function must be located as close as possible
beneficial. For instance, reactive control is useful for service
to the actual source of the traffic. An ideal usage parameter
classes like ABR, which allow sources to use bandwidth not
control mechanism should have the ability to detect para-
being used by calls in other service classes. Such a service
meter-violating cells, appear transparent to connections
would be impossible with preventive control because the
respecting their admission parameters, and rapidly
amount of unused bandwidth in the network changes
respond to parameter violations. It should also be simple,
dynamically, and the sources can only be made aware of
fast, and cost effective to implement in hardware. To meet
the amount through reactive feedback.
these requirements, several mechanisms have been pro-
There are two major classes of reactive traffic control
posed and implemented (8).
mechanisms: rate-based and credit-based (12,13). Most
The leaky bucket mechanism (originally proposed in
rate-based traffic control mechanisms establish a closed
Ref. 10) is a typical usage parameter control mechanism
feedback loop in which the source periodically transmits
used for ATM networks. It can simultaneously enforce the
special control cells, called resource management cells, to
average bandwidth and the burst factor of a traffic source.
the destination (or destinations). The destination closes the
One possible implementation of the leaky bucket mechan-
feedback loop by returning the resource management cells
ism is to control the traffic flow by means of tokens. A
to the source. As the feedback cells traverse the network,
conceptual model for the leaky bucket mechanism is
the intermediate switches examine their current conges-
illustrated in Fig. 5.
tion state and mark the feedback cells accordingly. When
In Fig. 8, an arriving cell first enters a queue. If the
the source receives a returning feedback cell, it adjusts its
queue is full, cells are simply discarded. To enter the net-
rate, either by decreasing it in the case of network conges-
work, a cell must first obtain a token from the token pool; if
tion or increasing it in the case of network underuse. An
there is no token, a cell must wait in the queue until a new
example of a rate-based ABR algorithm is the Enhanced
token is generated. Tokens are generated at a fixed rate
Proportional Rate Control Algorithm (EPRCA), which was
corresponding to the average bit rate declared during call
proposed, developed, and tested through the course of ATM
admission. If the number of tokens in the token pool exceeds
Forum activities (12).
some predefined threshold value, token generation stops.
Credit-based mechanisms use link-by-link traffic con-
This threshold value corresponds to the burstiness of the
trol to eliminate loss and optimize use. Intermediate
transmission declared at call admission time; for larger
switches exchange resource management cells that contain
threshold values, a greater degree of burstiness is allowed.
‘‘credits,’’ which reflect the amount of buffer space available
This method enforces the average input rate while allowing
at the next downstream switch. A source cannot transmit a
for a certain degree of burstiness.
new data cell unless it has received at least one credit from
One disadvantage of the leaky bucket mechanism is that
its downstream neighbor. An example of a credit-based
the bandwidth enforcement introduced by the token pool is
mechanism is the Quantum Flow Control (QFC) algorithm,
in effect even when the network load is light and there is no
developed by a consortium of reseachers and ATM equip-
need for enforcement. Another disadvantage of the leaky
ment manufacturers (13).
bucket mechanism is that it may mistake nonviolating cells
8 ASYNCHRONOUS TRANSFER MODE NETWORKS
HARDWARE SWITCH ARCHITECTURES FOR ATM
NETWORKS
In ATM networks, information is segmented into fixed-
length cells, and cells are asynchronously transmitted
through the network. To match the transmission speed
of the network links and to minimize the protocol proces-
sing overhead, ATM performs the switching of cells in
hardware-switching fabrics, unlike traditional packet
switching networks, where switching is largely performed
in software.
A large number of designs has been proposed and imple-
mented for ATM switches (14). Although many differences
exist, ATM switch architectures can be broadly classified Figure 10. A 8 Â 8 Banyan switch with binary switching ele-
into two categories: asynchronous time division (ATD) and ments.
space-division architectures.
Asynchronous Time Division Switches switched through the fabric. Three typical types of space-
division switches are described next.
The ATD, or single path, architectures provide a single,
multiplexed path through the ATM switch for all cells.
Banyan Switches. Banyan switches are examples of
Typically a bus or ring is used. Figure 9 shows the basic
space-division switches. An N Â N Banyan switch is con-
structure of the ATM switch proposed in (15). In Fig. 6, four
structed by arranging a number of binary switching ele-
input ports are connected to four output ports by a time-
ments into several stages (log2N stages). Figure 10 depicts
division multiplexing (TDM) bus. Each input port is allo-
an 8 Â 8 self-routing Banyan switch (14). The switch fabric
cated a fixed time slot on the TDM bus, and the bus is
is composed of twelve 2 Â 2 switching elements assembled
designated to operate at a speed equal to the sum of the
into three stages. From any of the eight input ports, it is
incoming bit rates at all input ports. The TDM slot sizes are
possible to reach all the eight output ports. One desirable
fixed and equal in length to the time it takes to transmit one
characteristic of the Banyan switch is that it is self-routing.
ATM cell. Thus, during one TDM cycle, the four input ports
Because each cross-point switch has only two output lines,
can transfer four ATM cells to four output ports.
only one bit is required to specify the correct output path.
In ATD switches, the maximum throughput is deter-
Very simply, if the desired output addresses of a ATM cell is
mined by a single, multiplexed path. Switches with N input
stored in the cell header in binary code, routing decisions
ports and N output ports must run at a rate N times faster
for the cell can be made at each cross-point switch by
than the transmission links. Therefore, the total through-
examining the appropriate bit of the destination address.
put of ATD ATM switches is bounded by the current cap-
Although the Banyan switch is simple and possesses
abilities of device logic technology. Commercial examples of
attractive features such as modularity, which makes it
ATD switches are the Fore Systems ASX switch and Digi-
suitable for VLSI implementation, it also has some disad-
tal’s VNswitch.
vantages. One of its disadvantages is that it is internally
blocking. In other words, cells destined for different output
Space-Division Switches
ports may contend for a common link within the switch.
To eliminate the single-path limitation and increase total This results in blocking all cells that wish to use that link,
throughput, space-division ATM switches implement mul- except for one. Hence, the Banyan switch is referred to as a
tiple paths through switching fabrics. Most space-division blocking switch. In Fig. 10, three cells are shown arriving
switches are based on multistage interconnection net- on input ports 1, 3, and 4 with destination port addresses of
works, where small switching elements (usually 2 Â 2 0, 1, and 5, respectively. The cell destined for output port 0
cross-point switches) are organized into stages and provide and the cell destined for output port 1 end up contending for
multiple paths through a switching fabric. Rather than the link between the second and third stages. As a result,
being multiplexed onto a single path, ATM cells are space- only one of them (the cell from input port 1 in this example)
actually reaches its destination (output port 0), while the
other is blocked.
Batcher–Banyan Switches. Another example of space-
division switches is the Batcher–Banyan switch (14).
(See Fig. 11.) It consists of two multistage interconnection
networks: a Banyan self-routing network and a Batcher
sorting network. In the Batcher–Banyan switch, the incom-
ing cells first enter the sorting network, which takes the
cells and sorts them into ascending order according to their
output addresses. Cells then enter the Banyan network,
Figure 9. A 4 Â 4 asynchronous time division switch. which routes the cells to their correct output ports.
ASYNCHRONOUS TRANSFER MODE NETWORKS 9
more cells may still contend for the same output port in a
nonblocking switch, resulting in the dropping of all but one
cell. In order to prevent such loss, the buffering of cells by
the switch is necessary. Figure 13 illustrates that buffers
may be placed (1) in the inputs to the switch, (2) in the
outputs to the switch, or (3) within the switching fabric
itself, as a shared buffer (14). Some switches put buffers in
both the input and output ports of a switch.
Figure 11. Batcher–Banyan switch. The first approach to eliminating output contention is to
place buffers in the output ports of the switch (14). In the
worst case, cells arriving simultaneously at all input ports
As shown earlier, the Banyan switch is internally block- can be destined for a single output port. To ensure that no
ing. However, the Banyan switch possesses an interesting cells are lost in this case, the cell transfer must be per-
feature. Namely, internal blocking can be avoided if the formed at N times the speed of the input links, and the
cells arriving at the Banyan switch’s input ports are sorted switch must be able to write N cells into the output buffer
in ascending order by their destination addresses. The during one cell transmission time. Examples of output
Batcher–Banyan switch takes advantage of this fact and buffered switches include the knockout switch by AT&T
uses the Batcher soring network to sort the cells, thereby Bell Labs, the Siemens & Newbridge MainStreetXpress
making the Batcher–Banyan switch internally nonblock- switches, the ATML’s VIRATA switch, and Bay Networks’
ing. The Starlite switch, designed by Bellcore, is based on Lattis switch.
the Batcher–Banyan architecture (16). The second approach to buffering in ATM switches is to
place the buffers in the input ports of the switch (14). Each
Crossbar Switches. The crossbar switch interconnects N input has a dedicated buffer, and cells that would otherwise
inputs and N outputs into a fully meshed topology; that is, be blocked at the output ports of the switch are stored in
there are N2 cross points within the switch (14). (See input buffers. Commercial examples of switches with input
Fig. 12.) Because it is always possible to establish a con- buffers as well as output buffers are IBM’s 8285 Nways
nection between any arbitrary input and output pair, inter- switches, and Cisco’s Lightstream 2020 switches.
nal blocking is impossible in a crossbar switch. A third approach is to use a shared buffer within the
The architecture of the crossbar switch has some advan- switch fabric. In a shared buffer switch, there is no buffer at
tages. First, it uses a simple two-state cross-point switch the input or output ports (14). Arriving cells are immedi-
(open and connected state), which is easy to implement. ately injected into the switch. When output contention
Second, the modularity of the switch design allows simple happens, the winning cell goes through the switch, while
expansion. One can build a larger switch by simply adding the losing cells are stored for later transmission in a shared
more cross-point switches. Lastly, compared to Banyan- buffer common to all of the input ports. Cells just arriving at
based switches, the crossbar switch design results in low the switch join buffered cells in competition for available
transfer latency, because it has the smallest number of outputs. Because more cells are available to select from, it is
connecting points between input and output ports. One possible that fewer output ports will be idle when using the
disadvantage to this design, however, is the fact that it shared buffer scheme. Thus, the shared buffer switch can
uses the maximum number of cross points (cross-point achieve high throughput. However, one drawback is that
switches) needed to implement an N Â N switch. cells may be delivered out of sequence because cells that
The knockout switch by AT&T Bell Labs is a nonblock- arrived more recently may win over buffered cells during
ing switch based on the crossbar design (17,18). It has N contention (19). Another drawback is the increase in the
inputs and N outputs and consists of a crossbar-based number of input and output ports internal to the switch.
switch with a bus interface module at each output (Fig. 12). The Starlite switch with trap by Bellcore is an example of
the shared buffer switch architecture (16). Other examples
Nonblocking Buffered Switches of shared buffer switches include Cisco’s Lightstream 1010
switches, IBM’s Prizma switches, Hitachi’s 5001 switches,
Although some switches such as Batcher–Banyan and
and Lucent’s ATM cell switches.
crossbar switches are internally nonblocking, two or
CONTINUING RESEARCH IN ATM NETWORKS
ATM is continuously evolving, and its attractive ability to
support broadband integrated services with strict quality of
service guarantees has motivated the integration of ATM
and existing widely deployed networks. Recent additions to
ATM research and technology include, but are not limited
to, seamless integration with existing LANs [e.g., LAN
emulation (20)], efficient support for traditional Internet
IP networking [e.g., IP over ATM (21), IP switching (22)],
Figure 12. A knockout (crossbar) switch.
and further development of flow and congestion control
10 ASYNCHRONOUS TRANSFER MODE NETWORKS
Figure 13. Nonblocking buffered
switches.
algorithms to support existing data services [e.g., ABR flow 12. ATM Forum, ATM Forum Traffic management specification
control (12)]. Research on topics related to ATM networks is version 4.0, af-tm-0056.000, April 1996, Mountain View, CA:
currently proceeding and will undoubtedly continue to ATM Forum.
proceed as the technology matures. 13. Quantum Flow Control version 2.0, Flow Control Consortium,
FCC-SPEC-95-1, [Online], July 1995. http://www.qfc.org
14. Y. Oie et al., Survey of switching techniques in high-speed
BIBLIOGRAPHY networks and their performance, Int. J. Satellite Commun., 9:
285–303, 1991.
1. CCITT Recommendation I-Series. Geneva: International Tele- 15. M. De Prycker and M. De Somer, Performance of a service
phone and Telegraph Consultative Committee. independent switching network with distributed control, IEEE
2. J. B. Kim, T. Suda and M. Yoshimura, International standar- J. Select. Areas Commun., 5: 1293–1301, 1987.
dization of B-ISDN, Comput. Networks ISDN Syst., 27: 1994. 16. A. Huang and S. Knauer, Starlite: A wideband digital switch.
3. CCITT Recommendation G-Series. Geneva: International Tel- Proc. IEEE GLOBECOM’84, 1984.
ephone and Telegraph Consultative Committee. 17. K. Y. Eng, A photonic knockout switch for high-speed packet
4. ATM Forum Technical Specifications [Online]. Available www: networks, IEEE J. Select. Areas Commun., 6: 1107–1116, 1988.
www.atmforum.com 18. Y. S. Yeh, M. G. Hluchyj, and A. S. Acampora, The knockout
5. Report of ANSI T1S1.5/91-292, Simple and Efficient Adapta- switch: A simple, modular architecture for high-performance
tion Layer (SEAL), August 1991. packet switching, IEEE J. Select. Areas Commun., 5: 1274–
6. Report of ANSI T1S1.5/91-449, AAL5—A New High Speed 1283, 1987.
Data Transfer, November 1991. 19. J. Y. Hui and E. Arthurs, A broadband packet switch for
7. CCITT Recommendation Q-Series. Geneva: International Tel- integrated transport, IEEE J. Select. Areas Commun., 5:
ephone and Telegraph Consultative Committee. 1264–1273, 1987.
8. J. Bae and T. Suda, Survey of traffic control schemes and 20. ATM Forum, LAN emulation over ATM version 1.0. AF-LANE-
protocols in ATM networks, Proc. IEEE, 79: 1991. 0021, 1995, Mountain View, CA: ATM Forum.
9. B. J. Vickers et al., Congestion control and resource manage- 21. IETF, IP over ATM: A framework document, RFC-1932, 1996.
ment in diverse ATM environments, IECEJ J., J76-B-I (11): 22. Ipsilon Corporation, IP switching: The intelligence of
1993. routing, The Performance of Switching [Online]. Available
10. J. S. Turner, New directions in communications (or which way www.ipsiolon.com
to the information age?), IEEE Commun. Mag., 25 (10): 1986.
11. G. Gallassi, G. Rigolio, and L. Fratta, ATM: Bandwidth
assignment and bandwidth enforcement policies. Proc. TATSUYA SUDA
GLOBECOM’89. University of California, Irvine
Irvine, California
A
AIRCRAFT COMPUTERS smaller, more powerful, and easier to integrate into multi-
ple areas of aircraft applications.
AIRCRAFT ANALOG COMPUTERS Landau (1) defines a digital computer as a computer for
processing data represented by discrete, localized physical
Early aircraft computers were used to take continuous signals, such as the presence or absence of an electric
streams of inputs to provide flight assistance. Examples current. These signals are represented as a series of bits
of aircraft analog inputs are fuel gauge readings, throttle with word lengths of 16, 32, and 64 bits. See micro-
settings, and altitude indicators. Landau (1) defines an computers for further discussion.
analog computer as a computer for processing data repre- Wakerly (2) shows number systems and codes used to
sented by a continuous physical variable, such as electric process binary digits in digital computers. Some impor-
current. Analog computers monitor these inputs and imple- tant number systems used in digital computers are binary,
ment a predetermined service when some set of inputs calls octal, and hexadecimal numbers. He also shows conver-
for a flight control adjustment. For example, when fuel sion between these and base-10 numbers, as well as simple
levels are below a certain point, the analog computer would mathematical operations such as addition, subtraction,
read a low fuel level in the aircraft’s main fuel tanks and division, and multiplication. The American Standard Code
would initiate the pumping of fuel from reserve tanks or the for Information Interchange (ASCII) of the American
balancing of fuel between wing fuel tanks. Some of the first National Standard Institute (ANSI) is also presented,
applications of analog computers to aircraft applications which is Standard No. X3.4-1968 for numerals, symbols,
were for automatic pilot applications, where these analog characters, and control codes used in automatic data
machines took flight control inputs to hold altitude and processing machines, including computers. Figure 1 shows
course. The analog computers use operational amplifiers to a typical aircraft central computer.
build the functionality of summers, adders, subtracters,
and integrators on the electric signals. Microcomputers
The improvements in size, speed, and cost through compu-
Aircraft Digital Computers ter technologies continually implement new computer con-
As the technologies used to build digital computers evolved, sumer products. Many of these products were unavailable
digital computers became smaller, lighter, and less power- to the average consumer until recently. These same break-
hungry, and produced less heat. This improvement made throughs provide enormous functional improvements in
them increasingly acceptable for aircraft applications. aircraft computing. Landau (1) defines microcomputers
Digital computers are synonymous with stored-program as very small, relatively inexpensive computers whose
computers. A stored-program computer has the flexibility central processing unit (CPU) is a microprocessor. A
of being able to accomplish multiple different tasks simply microprocessor (also called MPU or central processing
by changing the stored program. Analog computers are unit) communicates with other devices in the system
hard-wired to perform one and only one function. Analog through wires (or fiber optics) called lines. Each device
computers’ data, as defined earlier, are continuous physical has a unique address, represented in binary format, which
variables. Analog computers may be able to recognize and the MPU recognizes. The number of lines is also the
process numerous physical variables, but each variable has address size in bits. Early MPU machines had 8-bit
its unique characteristics that must be handled during addresses. Machines of 1970 to 1980 typically had 16-bit
processing by the analog computer. The range of output addresses; modern MPU machines have 256 bits.
values for the analog computer is bounded as a given Common terminology for an MPU is random access
voltage range; if they exceed this range, they saturate. memory (RAM), read only memory (ROM), input-output,
Digital computers are not constrained by physical vari- clock, and interrupts. RAM is volatile storage. It holds both
ables. All the inputs and outputs of the digital computer data and instructions for the MPU. ROM may hold both
are in a digital representation. The processing logic and instructions and data. The key point of ROM is that it is
algorithms performed by the computer work in a single nonvolatile. Typically, in an MPU, there is no operational
representation of the cumulative data. It is not uncommon difference between RAM and ROM other than its volatility.
to see aircraft applications that have analog-to-digital Input-output is how data are transferred to and from the
and digital-to-analog signal converters. This method is microcomputer. Output may be from the MPU, ROM, or
more efficient than having the conversions done within RAM. Input may be from the MPU or the RAM. The clock of
the computers. Analog signals to the digital computer an MPU synchronizes the execution of the MPU instruc-
are converted to digital format, where they are quickly tions. Interrupts are inputs to the MPU that cause it to
processed digitally and returned to the analog device (temporarily) suspend one activity in order to perform a
through a digital-to-analog converter as an analog output more important activity.
for that device to act upon. These digital computers are An important family of MPUs that greatly improved the
performance of aircraft computers is the Motorola M6800
family of microcomputers. This family offered a series of
1
Wiley Encyclopedia of Computer Science and Engineering, edited by Benjamin Wah.
Copyright # 2008 John Wiley & Sons, Inc.
2 AIRCRAFT COMPUTERS
AVIONICS
In the early years of aircraft flight, technological innovation
was directed at improving flight performance through
rapid design improvements in aircraft propulsion and
airframes. Secondary development energies went to areas
such as navigation, communication, munitions delivery,
and target detection. The secondary functionality of
aircraft evolved into the field of avionics. Avionics now
provides greater overall performance and accounts for a
greater share of aircraft lifecycle costs than either propul-
sion or airframe components.
Landau (1) defines avionics [avi(ation) þ (electr)onics] as
the branch of electronics dealing with the development and
use of electronic equipment in aviation and astronautics.
The field of avionics has evolved rapidly as electronics has
improved all aspects of aircraft flight. New advances in
these disciplines require avionics to control flight stability,
which was traditionally the pilot’s role.
Aircraft Antennas
An important aspect of avionics is receiving and transmit-
ting electromagnetic signals. Antennas are devices for
Figure 1. Typical aircraft central computer.
transmitting and receiving radio-frequency (RF) energy
from other aircraft, space applications, or ground applica-
tions. Perry and Geppert (4) illustrate the aircraft electro-
improvements in memory size, clock speeds, functionality, magnetic spectrum, influenced by the placement and usage
and overall computer performance. of numerous antennas on a commercial aircraft. Golden (5)
illustrates simple antenna characteristics of dipole, horn,
Personal Computers cavity-backed spiral, parabola, parabolic cylinder, and
Cassegrain antennas.
Landau (1) defines personal computers as electronic
Radiation pattern characteristics include elevation and
machines that can be owned and operated by individuals
azimuth. The typical antenna specifications are polariza-
for home and business applications such as word proces-
tion, beam width, gain, bandwidth, and frequency limit.
sing, games, finance, and electronic communications.
Computers are becoming increasingly important for the
Hamacher et al. (3) explain that rapidly advancing very
new generation of antennas, which include phased-array
large-scale integrated circuit (VLSI) technology has
antennas and smart-skin antennas. For phased-array
resulted in dramatic reductions in the cost of computer
antennas, computers are needed to configure the array
hardware. The greatest impact has been in the area of small
elements to provide direction and range requirements
computing machines, where it has led to an expanding
between the radar pulses. Smart-skin antennas comprise
market for personal computers.
the entire aircraft’s exterior fuselage surface and wings.
The idea of a personally owned computer is fairly new.
Computers are used to configure the portion of the aircraft
The computational power available in handheld toys today
surface needed for some sensor function. The computer also
was only available through large, costly computers in
handles sensor function prioritization and deinterleaving
the late 1950s and early 1960s. Vendors such as Atari,
of conflicting transmissions.
Commodore, and Compaq made simple computer games
household items. Performance improvements in memory,
Aircraft Sensors
throughput, and processing power by companies such as
IBM, Intel, and Apple made facilities such as spreadsheets Sensors, the eyes and ears of an aircraft, are electronic
for home budgets, automated tax programs, word proces- devices for measuring external and internal environmental
sing, and three-dimensional virtual games common house- conditions. Sensors on aircraft include devices for sending
hold items. The introduction of Microsoft’s Disk Operating and receiving RF energy. These types of sensors include
System (DOS) and Windows has also added to the accep- radar, radio, and warning receivers. Another group of
tance of the personal computers through access to software sensors are the infrared (IR) sensors, which include lasers
applications. Improvements in computer technology offer and heat-sensitive sensors. Sensors are also used to mea-
continual improvements, often multiple times a year. The sure direct analog inputs; altimeters and airspeed indica-
durability and portability of these computers is beginning tors are examples. Many of the sensors used on aircraft
to allow them to replace specialized aircraft computers have their own built-in computers for serving their own
that had strict weight, size, power, and functionality functional requirements such as data preprocessing, filter-
requirements. ing, and analysis. Sensors can also be part of a computer
AIRCRAFT COMPUTERS 3
interface suite that provides key aircraft computers with Aircraft Navigation
the direct environmental inputs they need to function.
Navigation is the science of determining present location,
desired location, obstacles between these locations, and
Aircraft Radar
best courses to take to reach these locations. An interesting
Radar (radio detection and ranging) is a sensor that trans- pioneer of aircraft navigation was James Harold Doolittle
mits RF energy to detect air and ground objects and deter- (1886–1993). Best known for his aircraft-carrier-based
mines parameters such as the range, velocity, and direction bomber raid on Tokyo in World War II, General Doolittle
of these objects. The aircraft radar serves as its primary received his Master’s and Doctor of Science degrees in
sensor. Several services are provided by modern aircraft aeronautics from Massachusetts Institute of Technology,
radar, including tracking, mapping, scanning, and identi- where he developed instrumental blind flying in 1929.
fication. Golden (5) states that radar is tasked either to He made navigation history by taking off, flying a set
detect the presence of a target or to determine its location. course, and landing without seeing the ground. For a
Depending on the function emphasized, a radar system modern aircraft, with continuous changes in altitude, air-
might be classified as a search or tracking radar. speed, and course, navigation is a challenge. Aircraft com-
Stimson (6) describes the decibel (named after Alexander puters help meet this challenge by processing the multiple
Graham Bell) as one of the most widely used terms in the inputs and suggesting aircrew actions to maintain course,
design and description of radar systems. The decibel (dB) is a avoid collision and weather, conserve fuel, and suggest
logarithmic unit originally devised to express power ratios, alternative flight solutions.
but also used to express a variety of other ratios. The power An important development in aircraft navigation is the
ratio in dB is expressed as 10 log10 P2/P1, where P2 and P1 are Kalman filter. Welch and Bishop (7) state that in 1960, R.E.
the power levels being compared. Expressed in terms of Kalman published his famous paper describing a recursive
voltage, the gain is (V2/V1)2 dB provided the input voltage solution to the discrete-data linear filtering problem. Since
V1 and output voltage V2 are across equal resistances. that time, due in large part to advances in digital comput-
Stimson (6) also explains the concept of the pulse repeti- ing, the Kalman filter has been the subject of extensive
tion frequency (PRF), which is the rate at which a radar research and application, particularly in the area of auton-
system’s pulses are transmitted: the number of pulses per omous or assisted navigation. The Kalman filter is a set of
second. The interpulse period T of a radar is given by mathematical equations that provides an efficient compu-
T ¼ 1=PRF. For a PRF of 100 Hz, the interpulse period tational (recursive) implementation of the least-squares
would be 0.01 s. method. The filter is very powerful in several aspects: It
The Doppler Effect, as described by Stimson (6), is a shift supports estimation of past, present, and even future
in the frequency of a radiated wave, reflected or received by states, and it can do so even when the precise nature of
an object in motion. By sensing Doppler frequencies, radar the modeled system is unknown.
not only can measure range rates, but can also separate The global positioning system (GPS) is a satellite refer-
target echoes from clutter, or can produce high-resolution ence system that uses multiple satellite inputs to determine
ground maps. Computers are required by an aircraft location. Many modern systems, including aircraft, are
radar to make numerous and timely calculations with equipped with GPS receivers, which allow the system
the received radar data, and to configure the radar to access to the network of GPS satellites and the GPS ser-
meet the aircrew’s needs. vices. Depending on the quality and privileges of the GPS
receiver, the system can have an instantaneous input of its
Aircraft Data Fusion current location, course, and speed within centimeters of
accuracy. GPS receivers, another type of aircraft computer,
Data fusion is a method for integrating data from multiple
can also be programmed to inform aircrews of services
sources in order to give a comprehensive solution to a
related to their flight plan.
problem (multiple inputs, single output). For aircraft com-
Before the GPS receiver, the inertial navigation systems
puters, data fusion specifically deals with integrating data
(INS) were the primary navigation system on aircraft. Fink
from multiple sensors such as radar and infrared sensors.
and Christiansen (8) describe inertial navigation as the
For example, in ground mapping, radar gives good surface
most widely used ‘‘self-contained’’ technology. In the case of
parameters, whereas the infrared sensor provides the
an aircraft, the INS is contained within the aircraft, and is
height and size of items in the surface area being investi-
not dependent on outside inputs. Accelerometers con-
gated. The aircraft computer takes the best inputs from
stantly sense the vehicle’s movements and convert them,
each sensor, provides a common reference frame to inte-
by double integration, into distance traveled. To reduce
grate these inputs, and returns a more comprehensive
errors caused by vehicle attitude, the accelerometers are
solution than either single sensor could have given.
mounted on a gyroscopically controlled stable platform.
Data fusion is becoming increasingly important as air-
crafts’ evolving functionality depends on off-board data
Aircraft Communications
(information) sources. New information such as weather,
flight path re-routing, potential threats, target assignment, Communication technologies on aircraft are predominately
and enroute fuel availability are communicated to the air- radio communication. This technology allows aircrews to
craft from its command and control environment. The air- communicate with ground controllers and other aircraft.
craft computer can now expand its own solution with these Aircraft computers help establish, secure, and amplify
off-board sources. these important communication channels.
4 AIRCRAFT COMPUTERS
These communication technologies are becoming in microseconds. These camps are now merging, because
increasingly important as aircraft become interoperable. their requirements are converging. MIS increasingly needs
As the dependency of aircraft on interoperability increases, real-time performance, while real-time systems are required
the requirements to provide better, more reliable, secure to handle increased data processing workloads. The
point-to-point aircraft communication also increases. The embedded information system addresses both needs.
aircraft computer plays a significant role in meeting this
challenge by formatting and regulating this increased flow Aircraft and the Year 2000
of information.
The year 2000 (Y2K) was a major concern for the aircraft
computer industry. Many of the embedded computers on
Aircraft Displays
aircraft and aircraft support functions were vulnerable to
Displays are visual monitors in aircraft that present Y2K faults because of their age. The basic problem with
desired data to aircrews and passengers. Adam and Gibson those computers was that a year was represented by its low-
(9) illustrate F-15E displays used in the Gulf War. These order two digits. Instead of the year having four digits,
illustrations show heads-up displays (HUDs), vertical these computers saved processing power by using the last
situation displays, radar warning receivers, and low- two digits of the calendar year. For example, 1999 is repre-
altitude navigation and targeting system (Lantirn) displays sented as 99, which is not a problem until you reach the year
typical of modern fighter aircraft. Sweet (10) illustrates the 2000, represented as 00. Even with this representation,
displays of a Boeing 777, showing the digital bus interface to problems are limited to those algorithms sensitive to calen-
the flight-deck panels and an optical-fiber data distribution dar dates. An obvious problem is when an algorithm divides
interface that meets industry standards. by the calendar date, which is division by 0. Division by 0 is
an illegal computer operation, causing problems such as
Aircraft Instrumentation infinite loops, execution termination, and system failure.
The most commonly mentioned issue is the subtraction of
Instrumentation of an aircraft means installing data col- dates to determine time durations and to compare dates. The
lection and analysis equipment to collect information about problem is not that the computer programs fail in a very
the aircraft’s performance. Instrumentation equipment obvious way (e.g., divide-by-zero check) but rather that the
includes various recorders for collecting real-time flight program computes an incorrect result without any warning
parameters such as position and airspeed. Instruments or indication of error. Lefkon and Payne (11) discuss Y2K
also capture flight control inputs, environmental para- and how to make embedded computers Y2K-compliant.
meters, and any anomalies encountered in flight test or
in routine flight. One method of overcoming this limitation
Aircraft Application Program Interfaces
is to link flight instruments to ground recording systems,
which are not limited in their data recording capacities. A An application programming interface (API) is conven-
key issue here is the bandwidth between the aircraft being tionally defined as an interface used by one program to
tested and its ground (recording) station. This bandwidth is make use of the services of another program. The human
limited and places important limitations on what can be interface to a system is usually referred to as the user
recorded. This type of data link is also limited to the range of interface, or, less commonly, the human–computer inter-
the link, limiting the aircraft’s range and altitude during face. Application programs are software written to solve
this type of flight test. Aircraft computers are used both in specific problems. For example, the embedded computer
processing the data as they are being collected on the aircraft software that paints the artificial horizon on a heads-up
and in analyzing the data after they have been collected. display is an application program. A switch that turns the
artificial horizon on or off is an API. Gal-Oz and Isaacs (12)
Aircraft Embedded Information Systems discuss APIs and how to relieve bottlenecks of software
debugging.
Embedded information system is the latest terminology for
an embedded computer system. The software of the
Aircraft Control
embedded computer system is now referred to as embedded
information. The purpose of the aircraft embedded infor- Landau (1) defines a control as an instrument or apparatus
mation system is to process flight inputs (such as sensor and used to regulate a mechanism or a device used to adjust or
flight control) into usable flight information for further control a system. There are two concepts with control. One
flight system or aircrew use. The embedded information is the act of control. The other is the type of device used to
system is a good example of the merging of two camps of enact control. An example of an act of control is when a pilot
computer science applications. The first, and larger, camp initiates changes to throttle and stick settings to alter flight
is the management of information systems (MIS). The MIS path. The devices of control, in this case, are the throttle
dealt primarily with large volumes of information, with and stick.
primary applications in business and banking. The timing Control can be active or passive. Active control is force-
requirements of processing these large information records sensitive. Passive control is displacement-sensitive.
are measured in minutes or hours. The second camp is the Mechanical control is the use of mechanical devices,
real-time embedded computer camp, which was concerned such as levers or cams, to regulate a system. The earliest
with processing a much smaller set of data, but in a very form of mechanical flight control was wires or cables, used
timely fashion. The real-time camp’s timing requirement is to activate ailerons and stabilizers through pilot stick and
AIRCRAFT COMPUTERS 5
foot pedal movements. Today, hydraulic control, the use of among the pooled processors to decrease the time it takes to
fluids for activation, is typical. Aircraft control surfaces are form solutions. Usually, one of the processors acts as the
connected to stick and foot pedals through hydraulic lines. lead processor, or master, while the other processor(s) act
Pistons in the control surfaces are pushed or pulled by as slave(s). The master processor schedules the tasking and
associated similar pistons in the stick or foot pedal. The integrates the final results, which is particularly useful on
control surfaces move accordingly. aircraft in that processors are distributed throughout the
Electronic control is the use of electronic devices, such as aircraft. Some of these computers can be configured to be
motors or relays, to regulate a system. A motor is turned on parallel processors, offering improved performance and
by a switch, and it quickly changes control surfaces by redundancy. Aircraft system redundancy is important
pulling or pushing a lever on the surface. Automatic control because it allows distributed parallel processors to be
is a system-initiated control, which is a system-initiated reconfigured when there is a system failure. Reconfigur-
response to a known set of environmental conditions. Auto- able computers are processors that can be reprogrammed
matic control was used for early versions of automatic pilot to perform different functions and activities. Before com-
systems, which tied flight control feedback systems to puters, it was very difficult to modify systems to adapt to
altitude and direction indicators. The pilot sets his desired their changing requirements. A reconfigurable computer
course and altitude, which is maintained through the flight can be dynamically reprogrammed to handle a critical
control’s automatic feedback system. situation, and then it can be returned to its original
To understand the need for computers in these control configuration.
techniques, it is important to note the progression of the
complexity of the techniques. The earliest techniques con-
Aircraft Buses
nected the pilot directly to his control surfaces. As the
aircraft functionality increased, the pilot’s workload also Buses are links between computers (processors), sensors,
increased, requiring his (or his aircrew’s) being free to and related subsystems for transferring data inputs and
perform other duties. Additionally, flight characteristics outputs. Fink and Christiansen (8) describe two primary
became more complex, requiring more frequent and instan- buses as data buses and address buses. To complete the
taneous control adjustments. The use of computers helped function of an MPU, a microprocessor must access memory
offset and balance the increased workload in aircraft. The and peripheral devices, which is accomplished by placing
application of computers to flight control provides a means data on a bus, either an address bus or a data bus, depend-
for processing and responding to multiple complex flight ing on the function of the operation. The standard 16-bit
control requirements. microprocessor requires a 16-line parallel bus for each
function. An alternative is to multiplex the address or
Aircraft Computer Hardware data bus to reduce the number of pin connections. Common
buses in aircraft are the Military Standard 1553 Bus (Mil-
For aircraft computers, hardware includes the processors,
Std-1553) and the General-Purpose Interface Bus (GPIB),
buses, and peripheral devices inputting to and outputting
which is the IEEE Standard 488 Bus.
from the computers. Landau (1) defines hardware as appa-
ratus used for controlling a spacecraft; the mechanical,
Aircraft Software
magnetic, and electronic design, structure, and devices of
a computer; and the electronic or mechanical equipment Landau (1) defines software as the programs, routines, and
that uses cassettes, disks, and so on. The computers used so on for a computer. The advent of software has provided
on an aircraft are called processors. The processor takes great flexibility and adaptability to almost every aspect of
inputs from peripheral devices and provides specific com- life, which is especially true in all areas of aerospace
putational services for the aircraft. sciences, where flight control, flight safety, in-flight enter-
There are many types and functions of processors on an tainment, navigation, and communications are continu-
aircraft. The most obvious processor is the central compu- ously being improved by software upgrades.
ter, also called the mission computer. The central computer
provides direct control and display to the aircrew. The Operation Flight Programs. An operational flight pro-
federated architecture (discussed in more detail later) is gram (OFP) is the software of an aircraft embedded com-
based on the central computer directing the scheduling and puter system. An OFP is associated with an aircraft’s
tasking of all the aircraft subsystems. Other noteworthy primary flight processors, including the central computer,
computers are the data processing and signal processing vertical and multiple display processors, data processors,
computers of the radar subsystem and the computer of the signal processors, and warning receivers. Many OFPs in
inertial navigation system. Processors are in almost every use today require dedicated software integrated support
component of the aircraft. Through the use of an embedded environments to upgrade and maintain them as the mission
processor, isolated components can perform independent requirements of their parent aircraft are modified. The
functions as well as self-diagnostics. software integrated support environment [also called avio-
Distributed processors offer improved aircraft perfor- nics integrated support environment (AISE), centralized
mance and, in some cases, redundant processing capability. software support activity (CSSA), and software integration
Parallel processors are two or more processors configured laboratory (SIL)] not only allows an OFP to be updated and
to increase processing power by sharing tasks. The maintained, but also provides capabilities to perform unit
workload of the shared processing activity is distributed
6 AIRCRAFT COMPUTERS
testing, subsystem testing, and some of the integrated discussions of software lifecycle engineering and main-
system testing. tenance are presented, and the concept of configuration
management is presented.
Assembly Language. Assembly language is a machine The package concept is one of the most important devel-
(processor) language that represents inputs and outputs opments to be found in modern programming languages,
as digital data and that enables the machine to perform such as Ada, Modula-2, Turbo Pascal, Cþþ, and Eiffel. The
operations with those data. For a good understanding of the designers of the different languages have not agreed on what
Motorola 6800 Assembler Language, refer to Bishop (13). terms to use for this concept: Package, module, unit, and
According to Seidman and Flores (14), the lowest-level class are commonly used. It is generally agreed, however,
(closest to machine) language available to most computers that the package (as in Ada) is the essential programming
is assembly language. When one writes a program in tool to be used for going beyond the programming of very
assembly code, alphanumeric characters are used instead simple class exercises to what is generally called software
of binary code. A special program called an assembler engineering or building production systems. Packages and
(provided with the machine) is designed to take the assem- package-like mechanisms are important tools used in soft-
bly statements and convert them to machine code. Assem- ware engineering to produce production systems. Feldman
bly language is unique among programming languages in (17) illustrates the use of Ada packages to solve problems.
its one-to-one correspondence between the machine code
statements produced by the assembler and the original Databases. Database are essential adjuncts to computer
assembly statements. In general, each line of assembly programming. Databases allow aircraft computer appli-
code assembles into one machine statement. cations the ability to carry pertinent information (such
as flight plans or navigation waypoints) into their missions,
Higher-Order Languages. Higher-order languages (HOLs) rather than generating them enroute. Databases also allow
are computer languages that facilitate human language the aircrew to collect performance information about the
structures to perform machine-level functions. Seidman aircraft’s various subsystems, providing a capability to
and Flores (14) discuss the level of discourse of a pro- adjust the aircraft in flight and avoid system failures.
gramming language as its distance from the underlying Elmasri and Navathe (18) define a database as a collec-
properties of the machine on which it is implemented. A tion of related data. Data are described as known facts that
low-level language is close to the machine, and hence can be recorded and have implicit meaning. A simple
provides access to its facilities almost directly; a high-level example consists of the names, telephone numbers, and
language is far from the machine, and hence insulated addresses of an indexed address book. A database manage-
from the machine’s peculiarities. A language may provide ment system (DBMS) is a collection of programs that enable
both high-level and low-level constructs. Weakly typed users to create and maintain a database. The DBMS is
languages are usually high-level, but often provide some hence a general-purpose software system that facilitates
way of calling low-level subroutines. Strongly typed lan- the processes of defining, constructing, and manipulating
guages are always high-level, and they provide means for databases for various applications.
defining entities that more closely match the real-world
objects being modeled. Fortran is a low-level language that Verification and Validation. A significant portion of the
can be made to function as a high-level language by use of aircraft computer’s lifecycle cost is system and software
subroutines designed for the application. APL, Sobol, and testing, performed in various combinations of unit-level,
SETL (a set-theoretic language) are high-level languages subsystem-level, integrated-system-level, developmental,
with fundamental data types that pervade their language. and operational testing. These types of tests occur fre-
Pascal, Cobol, C, and PL/I are all relatively low-level lan- quently throughout the life of an aircraft system because
guages, in which the correspondence between a program there are frequent upgrades and modifications to the air-
and the computations it causes to be executed is fairly craft and its various subsystems. It is possible to isolate
obvious. Ada is an interesting example of a language acceptance testing to particular subsystems when minor
with both low-level properties and high-level properties. changes are made, but this is the exception. Usually, any
Ada provides quite explicit mechanisms for specifying change made to a subsystem affects other multiple parts of
the layout of data structures in storage, for accessing the system. As aircraft become increasingly dependent on
particular machine locations, and even for communicating computers (which add complexity by the nature of their
with machine interrupt routines, thus facilitating low-level interdependences), and as their subsystems become
requirements. Ada’s strong typing qualities, however, also increasingly integrated, the impact of change also
qualify it as a high-level language. increases drastically. Cook (19) shows that a promising
High-level languages have far more expressive power technology to help understand the impact of aircraft com-
than low-level languages, and the modes of expression puter change is the Advanced Avionics Verification and
are well integrated into the language. One can write quite Validation (AAV&V) program developed by the Air Force
short programs that accomplish very complex operations. Research Laboratory.
Gonzalez (15) developed an Ada Programmer’s Handbook Sommerville (20) develops the concepts of program ver-
that presents the terminology of the HOL Ada and exam- ification and validation. Verification involves checking
ples of its use. He also highlights some of the common that the program conforms to its specification. Validation
programmer errors and examples of those errors. Sodhi involves checking that the program as implemented meets
(16) discusses the advantages of using Ada. Important the expectations of the user.
AIRCRAFT COMPUTERS 7
mendous opportunities to capture complex representations
of data and then save these representations in reusable
objects. Instead of using several variables and interactions
to describe some item or event, this same item or event is
described as an object. The object contains its variables,
control-flow representations, and data-flow representa-
tions. The object is a separable program unit, which can
be reused, reengineered, and archived as a program unit.
The power of this type of programming is that when large
libraries of OO programming units are created, they can be
called on to greatly reduce the workload of computer soft-
ware programming. Gabel (21) says that OO technology lets
an object (a software entity consisting of the data for an
action and the associated action) be reused in different
parts of the application, much as an engineered hardware
product can use a standard type of resistor or micropro-
cessor. Elmasri and Navathe (18) describe an OO database
Figure 2. An aircraft avionics support bench. as an approach with the flexibility to handle complex
requirements without being limited by the data types and
Figure 2 shows an aircraft avionics support bench, query languages available in traditional database systems.
which includes real components from the aircraft such as
the FCC line replaceable unit (LRU) sitting on top of the Open System Architecture. Open system architecture is
pictured equipment. Additional equipment includes the a design methodology that keeps options for updating sys-
buses, cooling, and power connection interfaces, along tems open by providing liberal interfacing standards.
with monitoring and displays. On these types of benches, Ralston and Reilly (22) state that open architectures per-
it is common to emulate system and subsystem responses tain primarily to personal computers. An open architecture
with testing computers such as the single-board computers is one that allows the installation of additional logic
illustrated. cards in the computer chassis beyond those used with
Figure 3 shows another verification and validation asset the most primitive configuration of the system. The cards
called the workstation-based support environment. This are inserted into slots in the computer’s motherboard—the
environment allows an integrated view of the aircraft’s main logic board that holds its CPU and memory chips. A
performance by providing simulations of the aircraft’s computer vendor that adopts such a design knows that,
controls and displays on computer workstations. The because the characteristics of the motherboard will be
simulation is interfaced with stick and throttle controls, public knowledge, other vendors that wish to do so can
vertical situation displays, and touch-screen avionics design and market customized logic cards. Open system
switch panels. architectures are increasingly important in modern air-
craft applications because of the constant need to upgrade
Object-Oriented Technology. Object-oriented (OO) tech- these systems and use the latest technical innovations. It is
nology is one of the most popular computer topics of the extremely difficult to predict interconnection and growth
1990s. OO languages such as Cþþ and Ada 95 offer tre- requirements for next-generation aircraft, which is exactly
what an open architecture attempts to avoid the need for.
Client-Server Systems. A client-server system is one in
which one computer provides services to another computer
on a network. Ralston and Reilly (22) describe the file-
server approach as an example of client-server interaction.
Clients executing on the local machine forward all file
requests (e.g., open, close, read, write, and seek) to the
remote file server. The server accepts a client’s requests,
performs its associated operation, and returns a response to
the client. Indeed, if the client software is structured
transparently, the client need not even be aware that files
being accessed physically reside on machines located else-
where on the network. Client-server systems are being
applied on modern aircraft, where highly distributed
resources and their aircrew and passenger services are
networked to application computers.
Subsystems. The major subsystems of an aircraft are its
Figure 3. A workstation-based aircraft avionics support envi- airframe, power plant, avionics, landing gear, and controls.
ronment. Landau (1) defines a subsystem as any system that is part of