![1-2 Smart Dust
1.2 Opportunities
1.2.1 Growing Research and Commercial Interest
Research and commercial interest in the area of wireless sensor networks are currently growing expo-
nentially, which is manifested in many ways:
• The number of Web pages (Google: 26,000 hits for sensor networks; 8000 for wireless sensor
networks in August 2003)
• The increasing number of
• Dedicated annual workshops, such as IPSN (information processing in sensor networks);
SenSys; EWSN (European workshop on wireless sensor networks); SNPA (sensor network
protocols and applications); and WSNA (wireless sensor networks and applications)
• Conference sessions on sensor networks in the communications and mobile computing com-
munities (ISIT, ICC, Globecom, INFOCOM, VTC, MobiCom, MobiHoc)
• Research projects funded by NSF (apart from ongoing programs, a new specific effort now
focuses on sensors and sensor networks) and DARPA through its SensIT (sensor information
technology), NEST (networked embedded software technology), MSET (multisensor exploi-
tation), UGS (unattended ground sensors), NETEX (networking in extreme environments),
ISP (integrated sensing and processing), and communicator programs
Special issues and sections in renowned journals are common, e.g., in the IEEE Proceedings [1] and signal
processing, communications, and networking magazines. Commercial interest is reflected in investments
by established companies as well as start-ups that offer general and specific hardware and software
solutions.
Compared to the use of a few expensive (but highly accurate) sensors, the strategy of deploying a large
number of inexpensive sensors has significant advantages, at smaller or comparable total system cost:
much higher spatial resolution; higher robustness against failures through distributed operation; uniform
coverage; small obtrusiveness; ease of deployment; reduced energy consumption; and, consequently,
increased system lifetime. The main point is to position sensors close to the source of a potential problem
phenomenon, where the acquired data are likely to have the greatest benefit or impact.
Pure sensing in a fine-grained manner may revolutionize the way in which complex physical systems
are understood. The addition of actuators, however, opens a completely new dimension by permitting
management and manipulation of the environment at a scale that offers enormous opportunities for
almost every scientific discipline. Indeed, Business 2.0 (http://www.business2.com/) lists sensor robots
as one of “six technologies that will change the world,” and Technology Review at MIT and Globalfuture
identify WSNs as one of the “10 emerging technologies that will change the world” (http://www.global-
future.com/mit-trends2003.htm). The combination of sensor network technology with MEMS and nan-
otechnology will greatly reduce the size of the nodes and enhance the capabilities of the network.
The remainder of this chapter lists and briefly describes a number of applications for wireless sensor
networks, grouped into different categories. However, because the number of areas of application is
growing rapidly, every attempt at compiling an exhaustive list is bound to fail.
1.2.2 Applications
1.2.2.1 General Engineering
• Automotive telematics. Cars, which comprise a network of dozens of sensors and actuators, are
networked into a system of systems to improve the safety and efficiency of traffic.
• Fingertip accelerometer virtual keyboards. These devices may replace the conventional input
devices for PCs and musical instruments.
• Sensing and maintenance in industrial plants. Complex industrial robots are equipped with up
to 200 sensors that are usually connected by cables to a main computer. Because cables are
7037_C001.fm Page 2 Tuesday, November 1, 2005 12:46 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-17-320.jpg)
![1-4 Smart Dust
• Surveillance and battle-space monitoring. Vibration and magnetic sensors can report vehicle and
personnel movement, permitting close surveillance of opposing forces.
• Urban warfare. Sensors are deployed in buildings that have been cleared to prevent reoccupation;
movements of friend and foe are displayed in PDA-like devices carried by soldiers. Snipers can be
localized by the collaborative effort of multiple acoustic sensors.
• Protection. Sensitive objects such as atomic plants, bridges, retaining walls, oil and gas pipelines,
communication towers, ammunition depots, and military headquarters can be protected by intel-
ligent sensor fields able to discriminate between different classes of intruders. Biological and
chemical attacks can be detected early or even prevented by a sensor network acting as a warning
system.
• Self-healing minefields.The self-healing minefield system is designed to achieve an increased resistance
to dismounted and mounted breaching by adding a novel dimension to the minefield. Instead of a
static complex obstacle, the self-healing minefield is an intelligent, dynamic obstacle that senses
relative positions and responds to an enemy’s breaching attempt by physical reorganization.
1.2.2.5 Health Monitoring and Surgery
• Medical sensing. Physiological data such as body temperature, blood pressure, and pulse are sensed
and automatically transmitted to a computer or physician, where it can be used for health status
monitoring and medical exploration. Wireless sensing bandages may warn of infection. Tiny
sensors in the blood stream, possibly powered by a weak external electromagnetic field, can
continuously analyze the blood and prevent coagulation and thrombosis.
• Microsurgery. A swarm of MEMS-based robots may collaborate to perform microscopic and
minimally invasive surgery.
The opportunities for wireless sensor networks are ubiquitous. However, a number of formidable chal-
lenges must be solved before these exciting applications may become reality.
1.3 Technical Challenges
Populating the world with networks of sensors requires a fundamental understanding of techniques for
connecting and managing sensor nodes with a communication network in scalable and resource-efficient
ways. Clearly, sensor networks belong to the class of ad hoc networks, but they have specific characteristics
that are not present in general ad hoc networks.
Ad hoc and sensor networks share a number of challenges such as energy constraints and routing. On
the other hand,general ad hoc networks most likely induce traffic patterns different from sensor networks,
have other lifetime requirements, and are often considered to consist of mobile nodes [2–4]. In WSNs,
most nodes are static; however, the network of basic sensor nodes may be overlaid by more powerful
mobile sensors (robots) that, guided by the basic sensors, can move to interesting areas or even track
intruders in the case of military applications.
Network nodes are equipped with wireless transmitters and receivers using antennas that may be
omnidirectional (isotropic radiation), highly directional (point-to-point), possibly steerable, or some
combination thereof. At a given point in time, depending on the nodes’ positions and their transmitter
and receiver coverage patterns, transmission power levels, and cochannel interference levels, a wireless
connectivity exists in the form of a random, multihop graph between the nodes. This ad hoc topology
may change with time as the nodes move or adjust their transmission and reception parameters.
Because the most challenging issue in sensor networks is limited and unrechargeable energy provision,
many research efforts aim at improving the energy efficiency from different aspects. In sensor networks,
energy is consumed mainly for three purposes: data transmission, signal processing, and hardware
operation [5]. It is desirable to develop energy-efficient processing techniques that minimize power
requirements across all levels of the protocol stack and, at the same time, minimize message passing for
network control and coordination.
7037_C001.fm Page 4 Tuesday, November 1, 2005 12:46 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-19-320.jpg)
![Opportunities and Challenges in Wireless Sensor Networks 1-5
1.3.1 Performance Metrics
To discuss the issues in more detail, it is necessary to examine a list of metrics that determine the
performance of a sensor network:
• Energy efficiency/system lifetime. The sensors are battery operated, rendering energy a very scarce
resource that must be wisely managed in order to extend the lifetime of the network [6].
• Latency. Many sensor applications require delay-guaranteed service. Protocols must ensure that
sensed data will be delivered to the user within a certain delay. Prominent examples in this class
of networks are certainly the sensor-actuator networks.
• Accuracy. Obtaining accurate information is the primary objective; accuracy can be improved
through joint detection and estimation. Rate distortion theory is a possible tool to assess accuracy.
• Fault tolerance. Robustness to sensor and link failures must be achieved through redundancy and
collaborative processing and communication.
• Scalability. Because a sensor network may contain thousands of nodes, scalability is a critical
factor that guarantees that the network performance does not significantly degrade as the network
size (or node density) increases.
• Transport capacity/throughput. Because most sensor data must be delivered to a single base station
or fusion center, a critical area in the sensor network exists (the gray area in Figure 1.1.), whose
sensor nodes must relay the data generated by virtually all nodes in the network. Thus, the traffic
load at those critical nodes is heavy, even when the average traffic rate is low. Apparently, this area
has a paramount influence on system lifetime, packet end-to-end delay, and scalability.
Because of the interdependence of energy consumption, delay, and throughput, all these issues and
metrics are tightly coupled. Thus, the design of a WSN necessarily consists of the resolution of numerous
trade-offs, which also reflects in the network protocol stack, in which a cross-layer approach is needed
instead of the traditional layer-by-layer protocol design.
1.3.2 Power Supply
The most difficult constraints in the design of WSNs are those regarding the minimum energy consumption
necessary to drive the circuits and possible microelectromechanical devices (MEMS) [5, 7, 8]. The energy
problem is aggravated if actuators are present that may be substantially hungrier for power than the sensors.
When miniaturizing the node, the energy density of the power supply is the primary issue. Current
technology yields batteries with approximately 1 J/mm3 of energy, while capacitors can achieve as much as
1 mJ/mm3. If a node is designed to have a relatively short life span, for example, a few months, a battery is
a logical solution. However, for nodes that can generate sensor readings for long periods of time, a charging
FIGURE 1.1 Sensor network with base station (or fusion center). The gray-shaded area indicates the critical area
whose nodes must relay all the packets.
critical nodes
BS
7037_C001.fm Page 5 Tuesday, November 1, 2005 12:46 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-20-320.jpg)
![1-6 Smart Dust
method for the supply is preferable. Currently, research groups are investigating the use of solar cells to
charge capacitors with photocurrents from the ambient light sources. Solar flux can yield power densities
of approximately 1 mW/mm2. The energy efficiency of a solar cell ranges from 10 to 30% in current
technologies, giving 300 μW in full sunlight in the best-case scenario for a 1-mm2 solar cell operating at
1 V. Series-stacked solar cells will need to be utilized in order to provide appropriate voltages.
Sensor acquisition can be achieved at 1 nJ per sample, and modern processors can perform compu-
tations as low as 1 nJ per instruction. For wireless communications, the primary candidate technologies
are based on RF and optical transmission techniques, each of which has its advantages and disadvantages.
RF presents a problem because the nodes may offer very limited space for antennas, thereby demanding
very short-wavelength (i.e., high-frequency) transmission, which suffers from high attenuation. Thus,
communication in that regime is not currently compatible with low-power operation. Current RF
transmission techniques (e.g., Bluetooth [9]) consume about 100 nJ per bit for a distance of 10 to 100 m,
making communication very expensive compared to acquisition and processing.
An alternative is to employ free-space optical transmission. If a line-of-sight path is available, a well-
designed free-space optical link requires significantly lower energy than its RF counterpart, currently
about 1 nJ per bit. The reason for this power advantage is that optical transceivers require only simple
baseband analog and digital circuitry and no modulators, active filters, and demodulators. Furthermore,
the extremely short wavelength of visible light makes it possible for a millimeter-scale device to emit a
narrow beam, corresponding to an antenna gain of roughly five to six orders of magnitude compared to
an isotropic radiator. However, a major disadvantage is that the beam needs to be pointed very precisely
at the receiver, which may be prohibitively difficult to achieve.
In WSNs, where sensor sampling, processing, data transmission, and, possibly, actuation are involved,
the trade-off between these tasks plays an important role in power usage. Balancing these parameters
will be the focus of the design process of WSNs.
1.3.3 Design of Energy-Efficient Protocols
It is well acknowledged that clustering is an efficient way to save energy for static sensor networks [10–13].
Clustering has three significant differences from conventional clustering schemes. First, data compression
in the form of distributed source coding is applied within a cluster to reduce the number of packets to
be transmitted [14, 15]. Second, the data-centric property makes an identity (e.g., an address) for a
sensor node obsolete. In fact, the user is often interested in phenomena occurring in a specified area
[16], rather than in an individual sensor node. Third, randomized rotation of cluster heads helps ensure
a balanced energy consumption [11].
Another strategy to increase energy efficiency is to use broadcast and multicast trees [6, 17, 18], which
take advantage of the broadcast property of omnidirectional antennas. The disadvantage is that the high
computational complexity may offset the achievable benefit. For sensor networks, this one-to-many
communication scheme is less important; however, because all data must be delivered to a single desti-
nation, the traffic scheme (for application traffic) is the opposite, i.e., many to one. In this case, clearly
the wireless multicast advantage offers less benefit, unless path diversity or cooperative diversity schemes
are implemented [19, 20].
The exploitation of sleep modes [21, 22] is imperative to prevent sensor nodes from wasting energy
in receiving packets unintended for them. Combined with efficient medium access protocols, the “sleep-
ing” approach could reach optimal energy efficiency without degradation in throughput (but at some
penalty in delay).
1.3.4 Capacity/Throughput
Two parameters describe the network’s capability to carry traffic: transport capacity and throughput.
The former is a distance-weighted sum capacity that permits evaluation of network performance.
Throughput is a traditional measure of how much traffic can be delivered by the network [23–30]. In a
7037_C001.fm Page 6 Tuesday, November 1, 2005 12:46 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-21-320.jpg)
![Opportunities and Challenges in Wireless Sensor Networks 1-7
packet network, the (network-layer) throughput may be defined as the expected number of successful
packet transmissions of a given node per timeslot.
The capacity of wireless networks in general is an active area of research in the information theory
community. The results obtained mostly take the form of scaling laws or “order-of” results; the prefactors
are difficult to determine analytically. Important results include the scaling law for point-to-point coding,
which shows that the throughput decreases with for a network with N nodes [23]. Newer results
[28] permit network coding, which yields a slightly more optimistic scaling behavior, although at high
complexity. Grossglauser and Tse [26] have shown that mobility may keep the per-node capacity constant
as the network grows, but that benefit comes at the cost of unbounded delay.
The throughput is related to (error-free) transmission rate of each transmitter, which, in turn, is upper
bounded by the channel capacity. From the pure information theoretic point of view, the capacity is
computed based on the ergodic channel assumption, i.e., the code words are long compared to the
coherence time of the channel. This Shannon-type capacity is also called throughput capacity [31].
However, in practical networks, particularly with delay-constrained applications, this capacity cannot
provide a helpful indication of the channel’s ability to transmit with a small probability of error.
Moreover, in the multiple-access system, the corresponding power allocation strategies for maximum
achievable capacity always favor the “good” channels, thus leading to unfairness among the nodes.
Therefore, for delay-constrained applications, the channel is usually assumed to be nonergodic and the
capacity is a random variable, instead of a constant in the classical definition by Shannon. For a delay-
bound D, the channel is often assumed to be block fading with block length D, and a composite channel
model is appropriate when specifying the capacity. Correspondingly, given the noise power, the channel
state (a random variable in the case of fading channels), and power allocation, new definitions for delay-
constrained systems have been proposed [32–35].
1.3.5 Routing
In ad hoc networks, routing protocols are expected to implement three main functions: determining and
detecting network topology changes (e.g., breakdown of nodes and link failures); maintaining network
connectivity; and calculating and finding proper routes. In sensor networks, up-to-date, less effort has
been given to routing protocols, even though it is clear that ad hoc routing protocols (such as destination-
sequenced distance vector (DSDV), temporally-ordered routing algorithm (TORA), dynamic source rout-
ing (DSR), and ad hoc on-demand distance vector (AODV) [4, 36–39]) are not suited well for sensor
networks since the main type of traffic in WSNs is “many to one” because all nodes typically report to
a single base station or fusion center. Nonetheless, some merits of these protocols relate to the features
of sensor networks, like multihop communication and QoS routing [39]. Routing may be associated with
data compression [15] to enhance the scalability of the network.
1.3.6 Channel Access and Scheduling
In WSNs, scheduling must be studied at two levels: the system level and the node level. At the node level,
a scheduler determines which flow among all multiplexing flows will be eligible to transmit next (the
same concept as in traditional wired scheduling); at the system level, a scheme determines which nodes
will be transmitting. System-level scheduling is essentially a medium access (MAC) problem, with the
goal of minimum collisions and maximum spatial reuse — a topic receiving great attention from the
research community because it is tightly coupled with energy efficiency and throughput.
Most of the current wireless scheduling algorithms aim at improved fairness, delay, robustness (with
respect to network topology changes) and energy efficiency [62, 64, 65, 66]. Some also propose a distrib-
uted implementation, in contrast to the centralized implementation in wired or cellular networks, which
originated from general fair queuing. Also, wireless (or sensor) counterparts of other wired scheduling
classes, like priority scheduling [67, 68] and earliest deadline first (EDF) [69], confirm that prioritization
is necessary to achieve delay balancing and energy balancing.
1/ N
7037_C001.fm Page 7 Tuesday, November 1, 2005 12:46 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-22-320.jpg)
![1-8 Smart Dust
The main problem in WSNs is that all the sensor data must be forwarded to a base station via multihop
routing. Consequently, the traffic pattern is highly nonuniform, putting a high burden on the sensor
nodes close to the base station (the critical nodes in Figure 1.1). The scheduling algorithm and routing
protocols must aim at energy and delay balancing, ensuring that packets originating close and far away
from the base station experience a comparable delay, and that the critical nodes do not die prematurely
due to the heavy relay traffic [40].
At this point, due to the complexity of scheduling algorithms and the wireless environment, most
performance measures are given through simulation rather than analytically. Moreover, medium access
and scheduling are usually considered separately. When discussing scheduling, the system is assumed to
have a single user; whereas in the MAC layer, all flows multiplexing at the node are treated in the same
way, i.e., a default FIFO buffer is assumed to schedule flows. It is necessary to consider them jointly to
optimize performance figures such as delay, throughput, and packet loss probability.
Because of the bursty nature of the network traffic, random access methods are commonly employed
in WSNs, with or without carrier sense mechanisms. For illustrative purposes, consider the simplest
sensible MAC scheme possible: all nodes are transmitting packets independently in every timeslot with
the same transmit probability p at equal transmitting power levels; the next-hop receiver of every packet
is one of its neighbors. The packets are of equal length and fit into one timeslot. This MAC scheme was
considered in Silvester and Kleinrock [41], Hu [42], and Haenggi [43]. The resulting (per-node) through-
put turns out to be a polynomial in p of order N, where N is the number of nodes in the network.
A typical throughput polynomial is shown in Figure 1.2. At p = 0, the derivative is 1, indicating that,
for small p, the throughput equals p. This is intuitive because there are few collisions for small p and the
throughput g(p) is approximately linear. The region in which the packet loss probability is less than 10%
can be denoted as the collisionless region. It ranges from 0 to about pmax/8. The next region, up to pmax,
is the practical region in which energy consumption (transmission attempts) is traded off against through-
put; it is therefore called the trade-off region. The difference p – g(p) is the interference loss. For small
networks, all N nodes interfere with each other because spatial reuse is not possible: If more than one
node is transmitting, a collision occurs and all packets are lost. Thus, the (per-node) throughput is
p(1 – p)N–1, and the optimum transmit probability is 1/N. The maximum throughput is (1 – 1/N)N–1/N.
With increasing N, the throughput approaches 1/(eN), as pointed out in Silvester and Kleinrock [41]
and LaMaire et al. [44]. Therefore the difference pmax – 1/N is the spatial reuse gain (see Figure 1.2).
This simple example illustrates the concepts of collisions, energy-throughput trade-offs, and spatial reuse,
which are present in every MAC scheme.
1.3.7 Modeling
The bases for analysis and simulations and analytical approaches are accurate and tractable models.
Comprehensive network models should include the number of nodes and their relative distribution; their
degree and type of mobility; the characteristics of the wireless link; the volume of traffic injected by the
sources and the lifespan of their interaction; and detailed energy consumption models.
1.3.7.1 Wireless Link
An attenuation proportional to dα, where d is the distance between two nodes and α is the so-called path
loss exponent, is widely accepted as a model for path loss. Alpha ranges from 2 to 4 or even 5 [45],
depending on the channel characteristics (environment, antenna position, frequency). This path loss
model, together with the fact that packets are successfully transmitted if the signal-to-noise-and-inter-
ference ratio (SNIR) is bigger than some threshold [8], results in a deterministic model often used for
analysis of multihop packet networks [23, 26, 41, 42, 46–48]. Thus, the radius for a successful transmission
has a deterministic value, irrespective of the condition of the wireless channel. If only interferers within
a certain distance of the receiver are considered, this “physical model” [23] turns into a “disk model.”
The stochastic nature of the fading channel and thus the fact that the SINR is a random variable are
mostly neglected. However, the volatility of the channel cannot be ignored in wireless networks [5, 8];
7037_C001.fm Page 8 Tuesday, November 1, 2005 12:46 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-23-320.jpg)
![Opportunities and Challenges in Wireless Sensor Networks 1-9
Sousa and Silvester have also pointed out the inaccuracy of disk models [49] and it is easily demonstrated
experimentally [50, 51]. In addition, this “prevalent all-or-nothing model” [52] leads to the assumption
that a transmission over a multihop path fails completely or is 100% successful, ignoring the fact that
end-to-end packet loss probabilities increase with the number of hops. Although fading has been con-
sidered in the context of packet networks [53, 54], its impact on the throughput of multihop networks
and protocols at the MAC and higher layers is largely an open problem.
A more accurate channel model will have an impact on most of the metrics listed in Section 1.3.1. In
the case of Rayleigh fading, first results show that the energy benefits of routing over many short hops
may vanish completely, in particular if latency is taken into account [20, 55, 56]. The Rayleigh fading
model not only is more accurate than the disk model, but also has the additional advantage of permitting
separation of noise effects and interference effects due to the exponential distribution of the received
power. As a consequence, the performance analysis can conveniently be split into the analysis of a zero-
interference (noise-analysis) and a zero-noise (interference-analysis) network.
1.3.7.2 Energy Consumption
To model energy consumption, four basic different states of a node can be identified: transmission,
reception, listening, and sleeping. They consist of the following tasks:
• Acquisition: sensing, A/D conversion, preprocessing, and perhaps storing
• Transmission: processing for address determination, packetization, encoding, framing, and maybe
queuing; supply for the baseband and RF circuitry (The nonlinearity of the power amplifier must
be taken into account because the power consumption is most likely not proportional to the
transmit power [56].)
• Reception: Low-noise amplifier, downconverter oscillator, filtering, detection, decoding, error
detection, and address check; reception even if a node is not the intended receiver
• Listening: Similar to reception except that the signal processing chain stops at the detection
• Sleeping: Power supply to stay alive
Reception and transmission comprise all the processing required for physical communication and net-
working protocols. For the physical layer, the energy consumption depends mostly on the circuitry, the
error correction schemes, and the implementation of the receiver [57]. At the higher layers, the choice
FIGURE 1.2 Generic throughput polynomial for a simple random MAC scheme.
spatial reuse gain pmax
–1/N
maximum throughput gmax
trade-off region p [pmax /8,pmax]
interference loss pmax
–gmax
g
(
p
)
1
0 0.5
pmax
p
1
N
Transmit probability p
t
u
p
h
g
u
o
r
h
T
g
∍
∍
collisionless region p [0,pmax /8]
7037_C001.fm Page 9 Tuesday, November 1, 2005 12:46 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-24-320.jpg)
![1-10 Smart Dust
of protocols (e.g., routing, ARQ schemes, size of packet headers, number of beacons and other infra-
structure packets) determines the energy efficiency.
1.3.7.3 Node Distribution and Mobility
Regular grids (square, triangle, hexagon) and uniformly random distributions are widely used analytically
tractable models. The latter can be problematic because nodes can be arbitrarily close, leading to unre-
alistic received power levels if the path attenuation is assumed to be proportional to dα. Regular grids
overlaid with Gaussian variations in the positions may be more accurate. Generic mobility models for
WSNs are difficult to define because they are highly application specific, so this issue must be studied
on a case-by-case basis.
1.3.7.4 Traffic
Often, simulation work is based on constant bitrate traffic for convenience, but this is most probably not
the typical traffic class. Models for bursty many-to-one traffic are needed, but they certainly depend
strongly on the application.
1.3.8 Connectivity
Network connectivity is an important issue because it is crucial for most applications that the network
is not partitioned into disjoint parts. If the nodes’ positions are modeled as a Poisson point process in
two dimensions (which, for all practical purposes, corresponds to a uniformly random distribution), the
problem of connectivity has been studied using the tool of continuum percolation theory [58, 59]. For
large networks, the phenomenon of a sharp phase transition can be observed: the probability that the
network percolates jumps abruptly from almost 0 to almost 1 as soon as the density of the network is
bigger than some critical value. Most such results are based on the geometric disk abstraction. It is
conjectured, though, that other connectivity functions lead to better connectivity, i.e., the disk is appar-
ently the hardest shape to connect [60]. A practical consequence of this conjecture is that fading results
in improved connectivity. Recent work [61] also discusses the impact of interference. The simplifying
assumptions necessary to achieve these results leave many open problems.
1.3.9 Quality of Service
Quality of service refers to the capability of a network to deliver data reliably and timely. A high quantity
of service, i.e., throughput or transport capacity, is generally not sufficient to satisfy an application’s delay
requirements. Consequently, the speed of propagation of information may be as crucial as the throughput.
Accordingly, in addition to network capacity, an important issue in many WSNs is that of quality-of-
service (QoS) guarantees. Previous QoS-related work in wireless networks mostly focused on delay (see,
for example, Lu et al. [62], Ju and Li [63], and Liu et al. [64]). QoS, in a broader sense, consists of the
triple (R, Pe, D), where R denotes throughput; Pe denotes reliability as measured by, for example, bit
error probability or packet loss probability; and D denotes delay. For a given R, the reliability of a
connection as a function of the delay will follow the general curve shown in Figure 1.3.
FIGURE 1.3 Reliability as a function of the delay. The circles indicate the QoS requirements of different possible
traffic classes.
reliability
delay
3
100%
1
2
7037_C001.fm Page 10 Tuesday, November 1, 2005 12:46 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-25-320.jpg)
![Opportunities and Challenges in Wireless Sensor Networks 1-11
Note that capacity is only one point on the reliability-delay curve and therefore not always a relevant
performance measure. For example, in certain sensing and control applications, the value of information
quickly degrades as the latency increases. Because QoS is affected by design choices at the physical,
medium-access, and network layers, an integrated approach to managing QoS is necessary.
1.3.10 Security
Depending on the application, security can be critical. The network should enable intrusion detection
and tolerance as well as robust operation in the case of failure because, often, the sensor nodes are not
protected against physical mishandling or attacks. Eavesdropping, jamming, and listen-and-retransmit
attacks can hamper or prevent the operation; therefore, access control, message integrity, and confiden-
tiality must be guaranteed.
1.3.11 Implementation
Companies such as Crossbow, Ember, Sensoria, and Millenial are building small sensor nodes with
wireless capabilities. However, a per-node cost of $100 to $200 (not including sophisticated sensors) is
prohibitive for large networks. Nodes must become an order of magnitude cheaper in order to render
applications with a large number of nodes affordable. With the current pace of progress in VLSI and
MEMS technology, this is bound to happen in the next few years. The fusion of MEMS and electronics
onto a single chip, however, still poses difficulties. Miniaturization will make steady progress, except for
two crucial components: the antenna and the battery, where it will be very challenging to find innovative
solutions. Furthermore, the impact of the hardware on optimum protocol design is largely an open topic.
The characteristics of the power amplifier, for example, greatly influence the energy efficiency of routing
algorithms [56].
1.3.12 Other Issues
• Distributed signal processing. Most tasks require the combined effort of multiple network nodes,
which requires protocols that provide coordination, efficient local exchange of information, and,
possibly, hierarchical operation.
• Synchronization and localization. The notion of time is critical. Coordinated sensing and actuating
in the physical world require a sense of global time that must be paired with relative or absolute
knowledge of nodes’ locations.
• Wireless reprogramming. A deployed WSN may need to be reprogrammed or updated. So far,
no networking protocols are available to carry out such a task reliably in a multihop network.
The main difficulty is the acknowledgment of packets in such a joint multihop/multicast
communication.
1.4 Concluding Remarks
Wireless sensor networks have numerous exciting applications in virtually all fields of science and
engineering, including health care, industry, military, security, environmental science, geology, agricul-
ture, and social studies. In particular, the combination with macroscopic or MEMS-based actuators is
intriguing because it permits manipulation of the environment in an unprecedented manner. Researchers
and operators currently face a number of critical issues that need be resolved before these applications
become reality. Wireless networking and distributed data processing of embedded sensing/actuating
nodes under tight energy constraints demand new approaches to protocol design and hardware/software
integration.
7037_C001.fm Page 11 Tuesday, November 1, 2005 12:46 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-26-320.jpg)
![2-1
2
Next-Generation
Technologies to Enable
Sensor Networks*
2.1 Introduction ...................................................................... 2-1
Geolocation and Identification of Mobile Targets • Long-Term
Architecture
2.2 Goals for Real-Time Distributed Network Computing
for Sensor Data Fusion..................................................... 2-5
2.3 The Convergence of Networking and Real-Time
Computing......................................................................... 2-6
Guaranteeing Network Resources • Guaranteeing Storage
Buffer Resources • Guaranteeing Computational Resources
2.4 Middleware ...................................................................... 2-11
Control and Command of System • Parallel Processing
2.5 Network Resource Management .................................... 2-11
Graph Generator • Metrics Object • Graph Search • NRM
Agents • Sensor Interface • Mapping Database • Topology
Database • NRM Federation • NRM Fault Tolerance
2.6 Experimental Results....................................................... 2-16
2.1 Introduction
Several important technical advances make extracting more information from intelligence, surveillance,
and reconnaissance (ISR) sensors very affordable and practical. As shown in Figure 2.1, for the radar
application the most significant advancement is expected to come from employing collaborative and
network centric sensor netting. One important application of this capability is to achieve ultrawideband
multifrequency and multiaspect imaging by fusing the data from multiple sensors. In some cases, it is
highly desirable to exploit multimodalities, in addition to multifrequency and multiaspect imaging.
Key enablers to fuse data from disparate sensors are the advent of high-speed fiber and wireless
networks and the leveraging of distributed computing. ISR sensors need to perform enough on-board
computation to match the available bandwidth; however, after some initial preprocessing, the data will
be distributed across the network to be fused with other sensor data so as to maximize the information
content. For example, on an experimental basis, MIT Lincoln Laboratory has demonstrated a virtual
radar with ultrawideband frequency [1]. Two radars, located at the Lincoln Space Surveillance Complex
*This work is sponsored by the United States Air Force under Air Force contract F19628-00-C-002. Opinions,
interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by the
U.S. government.
Joel I. Goodman
MIT Lincoln Laboratory
Albert I. Reuther
MIT Lincoln Laboratory
David R. Martinez
MIT Lincoln Laboratory
7037_C002.fm Page 1 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-30-320.jpg)
![2-2 Smart Dust
in Westford, Massachusetts, were employed; each of the two independent radars transmitted the data via
a high-speed fiber network. The total bandwidth transmitted via fiber exceeded 1 Gbits/sec (billion bits
per second). One radar was operating at X-band with 1-MHz bandwidth, and the second was operating
at Ku-band with a 2-MHz bandwidth. A synthetic radar with an instantaneous bandwidth of 8 MHz was
achieved after employing advanced ultrawideband signal processing [2].
These capabilities are now being extended to include high-speed wireless and fiber networking with
distributed computing. As the Internet protocol (IP) technologies continue to advance in the commercial
sector, the military can begin to leverage IP formatted sensor data to be compatible with commercial high-
speed routers and switches. Sensor data from theater can be posted to high-speed networks, wireless and
fiber, to request computing services as they become available on this network. The sensor data are processed
in a distributed fashion across the network, thereby providing a larger pool of resources in real time to meet
stringent latency requirements. The availability of distributed processing in a grid-computing architecture
offers a high degree of robustness throughout the network. One important application to benefit from these
advances is the ability to geolocate and identify mobile targets accurately from multiaspect sensor data.
2.1.1 Geolocation and Identification of Mobile Targets
Accurately geolocating and identifying mobile targets depends on the extraction of information from different
sensor data. Typically, data from a single sensor are not sufficient to achieve a high probability of correct
classification and still maintain a low probability of false alarm.This goal is challenging because mobile targets
typically move at a wide range of speeds, tend to move and stop often, and can be easily mistaken for a civilian
target. While the target is moving the sensor of choice is the ground moving target indication (GMTI). If the
target stops, the same sensor or a different sensor working cooperatively must employ synthetic aperture
radar (SAR). Before it can be declared foe, the target must often be confirmed with electro-optical or infrared
(EO/IR) images. The goal of future networked systems is to have multiple sensors providing the necessary
multimodality data to maximize the chances of accurately declaring a target.
FIGURE 2.1 Radar technology evolution.
Front End
Back End
Advanced
Algorithms
Space-time Adaptive
Imaging
Discrimination
Digital Array
Antennas
Filters
Power Devices
Correlation Processing
Pulse Compression
Doppler
Synthetic Aperture Radar
AEGIS
Patriot THAAD
Ground-based
Beale
~ 40s – 60s
~ 70s – 80s
~ 90s – 2000s
>2000s
E-2C
F-15
Chain Home
AWACS
Future
Collaborative/
Network Centric Ultra-Wideband
Multifrequency
Multiaspect
Imaging
7037_C002.fm Page 2 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-31-320.jpg)
![Next-Generation Technologies to Enable Sensor Networks 2-3
A typical sensing sequence starts by a wide area surveillance platform, such as the Global Hawk
unmanned aerial vehicle (UAV), covering several square kilometers until a target exceeds a detection
threshold. The wide area surveillance will typically employ GMTI and SAR strip maps. Once a target has
been detected, the on-board or off-board processing starts a track file to track the target carefully, using
spot GMTI and spot SAR over a much smaller region than that initially covered when performing wide
area surveillance. It is important to recognize that a sensor system is not merely tracking a single target;
several target tracks can be going on in parallel. Therefore, future networked sensor architectures rely
on sharing the information to maximize the available resources.
To date, the most advanced capability demonstrated is based on passing target detections among several
sensors using the Navy cooperative engagement capability (CEC) system. Multisensor tracks are formed
from the detection inputs arriving at a central location.Although this capability has provided a significant
advancement, not all the information available from multimodality sensors has been exploited. The
limitation is with the communication and available distributed computing. Multimodality sensor data
together with multiple look angles can substantially improve the probability of correct classification vs.
false alarm density. In addition to multiple modalities and multiple looks on the target, it is also desirable
to send complex (amplitude and phase) radar GMTI data and SAR images to permit the use of high-
definition vector imaging (HDVI) [3]. This technique permits much higher resolution on the target by
suppressing noise around it, thereby enhancing the target image at the expense of using complex video
data and much higher computational rates.
Another important tool to improve the probability of correct classification with minimal false alarm
is high-range resolution (HRR) profiles. With this tool, the sensor bandwidth or, equivalently, the size
of the resolution cell must be small resulting in a large data rate. However, it has been demonstrated that
HRR can provide a significant improvement [4]. Therefore, next generation sensors depend on available
communication pipes with enough bandwidth to share the individual sensor information effectively
across the network. Once the data are posted on the network, the computational resources must exist to
maintain low latencies from the time data become available to the time a target geoposition and identi-
fication are derived. The next subsection discusses the long-term architecture to implement netting of
multiple sensor data efficiently.
2.1.2 Long-Term Architecture
In the future it will be desirable to minimize the infrastructure (foot print) forwardly deployed in the
battlefield. It is most desirable to leverage high-speed satellite communication links to bring sensor data
back to a combined air operations center (CAOC) established in the continental United States (CONUS).
The technology enablers for the long-term architecture shown in Figure 2.2 are high-speed, IP-based
wireless and fiber communication networks, together with distributed grid computing. The in-theater
commander’s ability to task his organic resources to perform reconnaissance and surveillance of the opposing
forces, and then to relay that information back to CONUS, allows significant reduction in the complexity,
level, and cost of in-theater resources. Furthermore, this approach leverages the diverse analysis resources
in CONUS, including highly trained personnel to support the rapid, accurate identification and localization
of targets necessary to enable the time-critical engagement of surface mobile threats.
Space, air, and surface sensors will be deployed quickly to the battlefield. As shown in Figure 2.3, the
stage in the processing chain at which the sensor data are tapped off to be sent via the network will
dictate the amount of data transferred. For example, in a few applications one needs to send the data
directly out of the analog-to-digital converters (A/D) to exploit coherent data combining from multiple
sensors. Most commonly, it is preferable to perform on-board signal preprocessing to minimize the
amount of data transferred. However, one must still be able to preserve content in the transferred data
that is required to exploit features in the data not available from processing a signal sensor end to end.
For example, one might be interested in transmitting wide area surveillance (WAS) data from SAR with
high resolution to be followed by multiaspect SAR processing (shown in Figure 2.3 as application B).
The data volume will be larger than the second example shown in Figure 2.3 as application A, in which
7037_C002.fm Page 3 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-32-320.jpg)
![2-6 Smart Dust
behind due to a network problem or interruption in the processor, buffering the data will become
a problem quickly as increasing volumes of data must be stored to accommodate the delayed
processors. Section 2.3 addresses technological possibilities to mitigate these resource availability
issues.
• Middleware in the network of processors must be developed to accommodate a heterogeneous
mix of computer and network resources. This middleware consists of a task control interface,
which facilitates the communication between network resource management agents and entities,
and an application programming interface for programming applications executed on the collab-
orative network processors. Section 2.4 will address these middleware interfaces.
• A network resource manager (NRM) system is necessary for orchestrating the execution of the
application components on the computation and communication resources available in the col-
laborative network. Section 2.5 will discuss the components and functionality of the NRM.
2.3 The Convergence of Networking and Real-Time Computing
To date, networking of sensors has been demonstrated primarily using localized- and limited-capacity
data links. As a result, the data available on the network from each sensor node typically represent the
product of extensive prior processing of the radar data carried at the individual sensor. For example, the
Navy CEC system, a relatively advanced current system, uses detection reports from independent sensors
in the network to build composite tracks of targets. Access to raw (or possibly minimally preprocessed)
multisensor data opens the opportunity for more effective exploitation of these data through integrated
sensor data processing. The future network-centric ISR architecture will likely employ worldwide wide-
band communication networks to interconnect sensors with distributed processing and fusion sites. The
resulting distributed database will provide a common operational picture for deployed forces. The sensor
data will return to a CONUS entry point and pass over a wideband fiber network to the various processing
centers where the sensor data will be fused. The data link from the theater to CONUS is expected to be
optical to achieve very high link capacity [5].
This section discusses technologies that will guarantee that wireless and terrestrial network resources,
storage buffer resources, and computational resources are available for sensor signal processing.
2.3.1 Guaranteeing Network Resources
Sensor data will traverse wireless and terrestrial (e.g., optical, twisted-copper) networks in which bit errors,
packet loss, and delay could adversely affect the quality and timeliness of the ultimate result. The goal then
is to choose a network and processing architecture to ameliorate the deleterious effects of data loss and
network delay in the data fusion process. Due to the costs associated with developing, deploying, and
maintaining a fixed terrestrial infrastructure, as well as inventing wholly new modulation protocols and
standards for wireless and terrestrial signaling, it is cost-effective and expedient for military technology to
ride the “commercial wave” of technical investment and progress in communication technologies.
With a fixed network infrastructure consisting primarily of commercial components, combating data
loss and delay in terrestrial networks involves choosing the right protocols so that the network can enforce
quality of service (QoS) demands; in wireless networks, this involves aggressive coding, modulation, and
“lightweight” flow control for efficient bandwidth utilization. With sufficient complexity and bandwidth,
it is possible with today’s IP-based protocols to differentiate high-priority data to impart the mandated
QoS for time-critical applications.
2.3.1.1 Terrestrial Networks
Reserving bandwidth on an IP-based network that is uniformly recognized across administrative domains
involves employing protocols like RSVP-TE [6] or CR-LDP [7]. Although having sufficient communica-
tion bandwidth is an important aspect of processing sensor data in real time on a distributed network
of resources, it does not guarantee real-time performance. For example, time-critical applications mapped
7037_C002.fm Page 6 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-35-320.jpg)
![Next-Generation Technologies to Enable Sensor Networks 2-7
onto networked resources should not have processing interrupted to service unmanaged traffic or be
subject to a computational resource’s resident operating system switching contexts to a lower priority
task. For data that originate from sensors at very high streaming rates, a storage solution, as discussed
in Section 2.3.2, is needed that is capable of recording sensor data in real time as well as robust in the
face of network resource failures; this insures that a high-priority application can continue processing in
the presence of malfunctioning or compromised networked equipment. However, adding a buffering
storage solution only alleviates part of the problem; it does not mitigate the underlying problem of losing
packets during network equipment failures or periods of network traffic that exceed network capacities.
For an IP-based network, one solution to this problem is to use remote agents deployed on primary
compute resources or networked terminals located at switches that can dynamically filter unmanaged
traffic. This is implemented by programming computer hardware specifically tasked with packet filtering
(e.g., next generation gigabit Ethernet card) or dynamically reconfiguring the switch that directly connects
to the compute resource in question by supplying an access control list (ACL) to block all packets except
those associated with time-critical targeting. The formation of these exclusive networks using agents has
been dubbed dynamic private networks (DPNs) — in effect, mechanisms for virtually overlaying a circuit
switch onto a packet-switched network.
2.3.1.2 Wireless Networks
Unlike terrestrial networks, flow control and routing in mobile wireless sensor networks must contend
with potentially long point-to-point propagation delays (e.g., satellite to ground) as well as a constantly
changing topology. In a traditional terrestrial network employing link-state routing (e.g., OSPF), each
node maintains a consistent view of a (primarily) fixed network topology so that a shortest path algorithm
[8] can be used to find desirable routes from source to destination. This requires that nodes gather
network connectivity information from other routers.
If OSPF were employed in a mobile wireless network, the overhead of exchanging network connectivity
information about a transient topology could potentially consume the majority of the available bandwidth
[9]. Routing protocols have been specifically designed to address the concerns of mobile networks [10];
these protocols fall into two general categories: proactive and reactive. Proactive routing protocols keep
track of routes to all destinations, while reactive protocols acquire routes on demand. Unlike OSPF,
proactive protocols do not need a consistent view of connectivity; that is, they trade optimal routes for
feasible routes to reduce communication overhead. Reactive routes suffer a high initial overhead in
establishing a route; however, the overall overhead of maintaining network connectivity is substantially
reduced. The category of routing used is highly dependent upon how the sensors communicate with one
another over the network.
Traditional flow control mechanisms over terrestrial networks that deliver reliable transport (e.g., TCP)
may be inappropriate for wireless networks because, unlike wireless networks, terrestrial networks gen-
erally have a very low bit error rate (BER) on the order of 10–10, so errors are primarily due to packet
loss. Packet loss occurs in heavily congested networks when an ingress or egress queue of a switch or
router begins to fill, requiring that some packets in the queue be discarded [11]. This condition is detected
when acknowledgments from the destination node are not received by the source, prompting the source’s
flow control to throttle back the packet transmit rate [12].
In a wireless network in which BERs are four to five orders of magnitude higher than those of terrestrial
networks, packet loss due to bit errors can be mistakenly associated with network congestion, and source
flow control will mistakenly reduce the transmit rate of outgoing packets. Furthermore, when the source
and destination are far apart, such as the communication between a satellite and ground terminal, where
propagation delays can be on the order of 240 ms, delayed acknowledgments from the destination result
in source flow control inefficiently using the available bandwidth. This is due to source flow control
incrementally increasing the transmit rate as destination acknowledgments are received even though the
entire frame of packets may have already been transmitted before the first packet reaches the receiver
[13]. Therefore, to use bandwidth efficiently in a wireless network for reliable transport, flow control
must be capable of differentiating BER from packet loss and account for long-haul packet transport by
7037_C002.fm Page 7 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-36-320.jpg)
![2-8 Smart Dust
more efficiently using the available bandwidth. Some work in this area is reflected in RFC 2488 [14], as
well as proposals for an explicit congestion warning, where, for example, the destination site would
respond to packet errors with an acknowledgment that it received the source packets with a corruption
notification.
At the physical layer, high data rates for a given BER have been realized by employing low-density
parity check codes, such as turbo codes, in conjunction with bandwidth efficient modulation to achieve
spectral efficiencies to within 0.7 dB of the Shannon limit [15]. Furthermore, extremely high spectral
efficiencies have been demonstrated using multiple input, multiple output (MIMO) antenna systems
whose theoretical channel capacity increases linearly with the number of transmit/receive antenna pairs
[16].Although turbo codes are advantageous as a forward error correction mechanism in wireless systems
when trying to maximize throughput, MIMO systems achieve high spectral efficiencies only when
operating in rich scattering environments [17]. In environments in which little scattering occurs, such
as in some air-to-air communication links, MIMO systems offer very little improvement in spectral
efficiency.
2.3.2 Guaranteeing Storage Buffer Resources
For a variety of reasons, it may be very desirable to record streaming sensor data directly to storage media
while simultaneously sending the data on for immediate processing. For sensor signal processing appli-
cations, this enables multimodality data fusion of archived data with real-time (perishable) data from
in-theatre sensors for improved target identification and visualization [18]. Storage media could also be
used for rate conversion in cases in which the transmission rate exceeds the processing rate and for time-
delay buffering for real-time robust fault tolerance (discussed in the next section). The storage media
buffer reuse is deterministic and periodic so that management of the buffer is straightforward.
A number of possible solutions exist:
• Directly attached storage is a set of hard disks connected to a computer via SCSI or IDE/EIDE/
ATA; however, this technology does not scale well to the volume of streaming sensor data.
• Storage area networks are hard disk storage cabinets attached to a computer with a fast data link
like Fibre Channel. The computer attached to the storage cabinet enjoys very fast access to data,
but because the data must travel through that computer, which presents a single point of failure,
to get to other computers on the network, this option is not a desirable solution.
• Network-attached storage connects the hard disk storage cabinet directly to the network as a file
server. However, this technology offers only midrange performance, a single point of failure, and
relatively high cost.
A visionary architecture in which data storage centers operate in parallel at a wide-area network (WAN)
and local area network (LAN) level is described in Cooley et al. [19]. In this architecture, developed by
MIT Lincoln Laboratory, high-rate streaming sensor data are stored in parallel across a partitioned
network of storage arrays, which affords a highly scalable, low-cost solution that is relatively insensitive
to communications or storage equipment failure. This system employs a novel and computationally
efficient encoding and decoding algorithm using low-density parity check codes [20] for erasure recovery.
Initial system performance measures indicate the erasure coding method described in Cooley et al. [19]
has a significantly higher throughput and greater reliability when compared to Reed–Solomon, Tornado
[21], and Luby [20] codes. This system offers a promising low-cost solution that scales in capability with
the performance gains of commodity equipment.
2.3.3 Guaranteeing Computational Resources
The exponential growth in computing technology has contributed to making viable the implementation
of advanced sensor processing in cost-effective hardware with form factors commensurate with the needs
of military users. For example, several generations of embedded signal processors are shown in Figure 2.5.
7037_C002.fm Page 8 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-37-320.jpg)
![Next-Generation Technologies to Enable Sensor Networks 2-9
In the early 1990s, embedded signal processors were built using custom hardware and software. In the late
1990s, a move occurred from custom hardware to COTS processor systems running vendor-specific
software together with application-specific parallel software tuned to each specific application. Most
recently, the military embedded community is beginning to demonstrate requisite performance employing
parallel and portable software running on COTS hardware.
Continuing technology advances in computation and communication will permit future signal pro-
cessors to be built from commodity hardware distributed across a high-speed network and employing
distributed, parallel, and portable software. These computing architectures will deliver 109 to 1012 floating
point operations per second (GFLOPs to TFLOPs) in computational throughput. The distributed nature
of the software will apply to on-board sensor processing as well as off-board processing. Clearly, on-
board embedded processor systems will need to meet the stringent platform requirements in size, weight,
and power.
Wireless and terrestrial network resources are not the only areas in which delays, failures, and errors
must be avoided to process sensor data in a timely fashion. The system design must also guarantee that
the marshaled compute nodes will keep up with the required computational throughput of streaming
data at every stage of the processing chain. This guarantee encompasses two important facets: (1) keeping
the processors from being interrupted while they are processing tasks and (2) implementing fail-over
that is tolerant of fault.
2.3.3.1 Avoiding Processor Interruption
It is easy to take for granted that laptop and desktop computers will process commands as fast as the
hardware and software are capable of doing so. A fact not generally known is that general computers are
interrupted by system task processes and the processes of other applications (one’s own and possibly
from others working in the background on one’s system). System task processes include keyboard and
mouse input; communications on the Ethernet; system I/O; file system maintenance; log file entries; etc.
When the computer interrupts an application to attend to such tasks, the execution of the application is
temporarily suspended until the interrupting task has finished execution. However, because such inter-
ruptions often only consume a few milliseconds of processing time, they are virtually imperceptible to
the user [22].
Nevertheless, the interruptions are detrimental to the execution of real-time applications. Any delay
in processing these streams of data will instigate a need for buffering the data that will grow to insur-
mountable size as the delays escalate. A solution for these interrupt issues is to use a real-time operating
system on the computation processors.
FIGURE 2.5 Embedded signal processor evolution.
85 GFLOPS
COTS Parallel SW
Adaptive Processor
Gen 1 (1992)
22 GOPS
Custom (Parallel) SW
Adaptive Processor
Gen 2 (1998)
AEGIS & Standard Missile
Test Beds (2000+)
PTCN Network
Test Bed (2002+)
VME Backplane
Custom Boards
RACE Crossbar
Multi-chassis COTS
50+ GFLOPS
Portable, Parallel SW
(VSIPL, MPI, & PVL)
High Speed LANs
Network of Workstations
GFLOPS to TFLOPS
Parallel & Distributed SW
(PVL & CORBA)
High Speed LANs & WANs
Networked Clusters, Servers
Distributed
Network
7037_C002.fm Page 9 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-38-320.jpg)
![2-10 Smart Dust
Simply put, real-time operating systems (RTOS) give priority to computational tasks. They usually do
not offer as many operating system features (virtual memory, threaded processing, etc.) because of the
interrupting processing nature of these features [22]. However, an RTOS can ensure that real-time critical
tasks have guaranteed success in meeting streamed processing deadlines. An RTOS does not need to be
run on typical embedded processors; it can also be deployed on Intel and AMD Pentium-class or Motorola
G-series processor systems. This includes Beowulf clusters of standard desktop personal computers and
commodity servers. This is an important benefit, providing a wide range of candidate heterogeneous
computing resources.
A great deal of press has been generated in the past several years about real-time operating systems;
however, the distinction between soft real-time and hard real-time operating systems is seldom discussed.
Hard real-time systems guarantee the completion of tasks in a deterministic time period, while soft real-
time systems give priority to critical tasks over other tasks but do not guarantee the completion of tasks
in a deterministic time period [22]. Examples of hard real-time operating systems are VxWorks (Wind
River Systems, Inc. [23]); RTLinux/Pro (FSMLabs, Inc. [24]); and pSOS (Wind River Systems, Inc. [23]),
as well as dedicated massively parallel embedded operating systems like MC/OS (Mercury Computer
Systems, Inc. [25]). Examples of soft real-time operating systems are Microsoft Pocket PC; Palm OS;
certain real-time Linux releases [24, 26]; and others.
2.3.3.2 Working through System Faults
When fault tolerance in massively parallel computers is addressed, usually the solution is parallel redun-
dant systems for fail-over. If a power supply or fan fails, another power supply or fan that is redundant
in the system takes over the workload of the failed device. If a hard disk drive fails on a redundant array
of independent disks (RAID) system, it can be hot swapped with a new drive and the contents of the
drive rebuilt from the contents of the other drives along with checksum error correction code information.
However, if an individual processor fails on a parallel computer, it is considered a failure of the entire
parallel computer, and an identical backup computer is used as a fail-over. This backup system is then
used as the primary computer, while the failed parallel computer is repaired to become the backup for
the new primary eventually.
If, however, it were possible to isolate the failed processor and remap and rebind the processes on
other processors in that computer — in real time — it would then be possible to have only a number
of redundant processors in the system rather than entire redundant parallel computers. There are two
strategies for determining the remapping as well as two strategies for handling the remapping and
rebinding; each has its advantages and disadvantages.
To discuss these fail-over strategies, it is necessary to define the concepts of tasks and mappings. A signal
processing application can be separated into a series of pipelined stages or tasks that are executed as part
of the given application.A mapping is the task-parallel assignment of a task to a set of computer and network
resources. In terms of determining the fail-over remapping, it is possible to choose a single remapping for
each task or to choose a completely unique secondary path — a new mapping for each task that uses a set
of processors mutually exclusive from the processors in the primary mapping path. If task backup mappings
are chosen for each task, the fail-over will complete faster than a full processing chain fail-over; however,
the rebinding fail-over for a failed task mapping is more difficult because the mappings from the task before
and the task after the failed task mapping must be reconfigured to send data to and receive data from the
new mapping. Conversely, if a completely unique secondary path is chosen as a fail-over, then fail-over
completion will have a longer latency than performing a single task fail-over. However, the fail-over mechan-
ics are simpler because the completely unique secondary path could be fully initialized and ready to receive
the stream of data in the event of a failure in the primary mapping path.
In terms of handling the remapping and rebinding of tasks, it is possible to choose the fail-over
mappings when the application is initially launched or immediately after a fault occurs. In either case,
greater latency is incurred at launch time or after the occurrence of a fault. For these advanced options,
support for this fault tolerance comes mainly from the middleware support, which is discussed in the
next section, and from the NRM discussed in Section 2.5.
7037_C002.fm Page 10 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-39-320.jpg)
![Next-Generation Technologies to Enable Sensor Networks 2-11
2.4 Middleware
Middleware not only provides a standard interface for communications between network resources and
sensors for plug-and-play operation, but also enables the rapid implementation of high-performance
embedded signal processing.
2.4.1 Control and Command of System
Because many systems use a diverse set of hardware, operating systems, programming languages, and
communication protocols for processing sensor data, the manpower and time-to-deployment associated
with integration have a significant cost. A middleware component providing a uniform interface that
abstracts the lower-level system implementation details from the application interface is the common
object request broker architecture (CORBA) [27]. CORBA is a specification and implementation that
defines a standard interface between a client and server. CORBA leverages an interface definition language
(IDL) that can be compiled and linked with an object’s implementation and its clients. Thus, the CORBA
standard enables client and server communications that are independent of the host hardware platforms,
programming language, operating systems, and so on. CORBA has specifications and implementations
to interface with popular communication protocols such as TCP/IP. However, this architecture has an
open specification, general interORB protocol (GIOP) that enables developers to define and plug in
platform-specific communication protocols for unique hardware and software interfaces that meet appli-
cation-specific performance criteria.
For real-time and parallel embedded computing, it is necessary to interface with real-time operating
systems, define end-to-end QoS parameters, and enact efficient data reorganization and queuing at
communication interfaces. CORBA has recently included specifications for real-time performance and
parallel processing, with the expectation that emerging implementations and specification addendums
will produce efficient implementations. This will enable CORBA to move out of the command and
control domain and be included as a middleware component involved in real-time and parallel processing
of time-critical sensor data.
2.4.2 Parallel Processing
The ability to choose one of many potential parallel configurations enables numerous applications to
share the same set of resources with various performance requirements. What is needed is a method to
decouple the mapping, that is, the parallel instantiation of an application on target hardware, from generic
serial application development. Automating the mapping process is the only feasible way of exploring
the large parameter space of parallel configurations in a timely and cost-effective manner.
MIT Lincoln Laboratory has developed a C++-based library known as the parallel vector library (PVL)
[28]. This library contains objects with parameterized methods deeply rooted in linear algebraic expres-
sions commonly found in sensor signal processing. The parameters are used to direct the object instance
to process data as one constituent part of a parallel whole. The parameters that organize objects in parallel
configurations are run-time parameters so that new parallel configurations can be instantiated without
having to recompile a suite of software. The technology of PVL is currently being incorporated into the
parallel vector, signal, and image processing library for C++ (parallel VSIPL++) standard library [29].
2.5 Network Resource Management
Given the stated goals for distributed network computing for sensor fusion as outlined in Section 2.3,
the associated network communication, storage, and processing challenges in Section 2.3, and the desire
for standard interfaces and libraries to enable application parallelism and plug-and-play integration in
Section 2.4, an integrated solution is needed that bridges network communications, distributed storage,
distributed processing, and middleware. Clearly, it is possible for a development team to implement a
7037_C002.fm Page 11 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-40-320.jpg)
![2-12 Smart Dust
“point” solution, but this is inherently not scalable and very difficult to maintain. Therefore an additional
goal is to fully automate the process of configuring network communication, storage, and computational
resources to process data for sensor fusion applications in real time, provide robust fault tolerance in the
face of network resource failures, and impart this service in a highly dynamic network in the face of
competing interests.
To address these needs, the network resource manager (NRM) was developed. The novelty and potency
of the NRM is its capability of taking a sensor signal processing application designed and tested on single
target processing element (PE) and mapping it in a task- and a data-parallel fashion across a network of
computational resources to achieve real-time performance [30]. Figure 2.6 is an object-oriented model
of the components that constitute the NRM. A high-level overview of the NRM follows, and details will
be provided in the following subsections. The task of building a model from which the NRM launches
parallel applications is broken into three distinct phases:
1. Map generation involves breaking an application into various task- and data-parallel components.
2. Map timing collects performance metric information associated with the components (or tasks)
running on host resources. Using the performance metrics, the NRM creates a weighted graph-
theoretic view of various permutations of an application mapped in parallel across networked
resources.
3. Map selection finds the path through the graph that best meets system and application perfor-
mance requirements.
The graph generator and graph search objects will heavily leverage PVL (discussed earlier) objects in
the instantiation of task- and data-parallel configurations of applications on host resources. It should be
noted, however, that the NRM’s capabilities are fully general and independent from those of PVL and
could work with other applications that are not developed using PVL to instantiate task- and data
parallelism.
2.5.1 Graph Generator
As noted previously, PVL uses run-time parameters to generate new parallel configurations. This enables
the NRM to launch applications in arbitrary parallel configurations using software developed for a single
target PE without having to recompile the application software suite. The central challenge is to select a
subset of the potentially astronomical number of permutations of parallel configurations as candidate
parallel mappings. It is expected that the NRM will receive guidance in the form of performance and
resource utilization bounds to help it avoid choosing undesirable configurations. It will also be given a
FIGURE 2.6 Object model for network resource manager (NRM).
NRM
Sensor
Interface
Graph
Generator
Mapping
Database
Topology
Database
Metrics
Object
NRM
Agent
Task
App
Instance
Graph
Search
Sensor
7037_C002.fm Page 12 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-41-320.jpg)
![Next-Generation Technologies to Enable Sensor Networks 2-13
series of constituent tasks that comprise an application, so that its primary objective is to choose candidate
data-parallel configurations for each of the individual tasks. Using a graph-theoretic model, the appli-
cation space may be broken up as shown in Figure 2.7.
Each column in the graph is populated with vertices; each vertex corresponds to a mapping of the
task corresponding to the given column to a potentially unique set of computational resources in the
system. Each vertex has edges entering and exiting: entering edges correspond to communications with
preceding tasks and exiting edges correspond to communications with succeeding tasks. Sensor signal
processing applications may be represented as a stream signal processing flow, in which data move in
one direction from task to task as they are processed. In this graph-theoretic model, task parallelism is
represented along the horizontal axis of the graph, i.e., pipelined, overlapping execution intervals, while
data parallelism is represented by the mapping of each task in the application onto one or more parallel
computational resources of each vertex. The graph-theoretic representation of data- and task-parallel
applications and the corresponding flow of communication enable the graph generator of the NRM to
capture the potentially astronomical number of combinations of application-to-resource mappings in a
concise and efficient fashion.
Finally, the graph generator is also responsible for launching the executable for each task mapping
(vertex) on target resources so that performance metrics can be collected as discussed in the next
subsection.
2.5.2 Metrics Object
The metrics object (MO) is responsible for collecting performance metrics of tasks launched by the graph
generator. The MO works closely with the graph generator to weight the graph. Each of the resources
that hosts a task is time synchronized; metric agents (see NRM agents in Subsection 2.5.4) on each of
the resources will provide the MO measurements for it to formulate the following performance param-
eters associated with graph weights: throughput; latency; RAM memory; and PE utilization. The MO
will calculate another metric known as processor cost, which is a ratio of compute horsepower used in
the mapping to the overall processing horsepower available in the network.
Link utilization percentages within each mapping are also measured, as well as intertask utilization
percentages. Map generation uses task column pairs to gather performance metrics in order to reduce
the effort and time involved drastically. This is possible because the graph search algorithm will use a
running tabulation of resource utilization percentages to ensure that simple linear superposition of path
weights hold, given that these percentages remain under a given threshold. This is explained further in
the next subsection. Once above the threshold, weight modifiers will be applied to subsequent stages
during search. Finally, the metrics object will calculate a network cost, analogous to processor cost, which
FIGURE 2.7 Sample graph with edge and vertex weights.
E = [e1,e2,..,em]
V = [v1,v2,..,vm]
TASK 1
(Stage 1)
TASK 2
(Stage 2)
TASK c–1
(Stage c–1)
TASK c
(Stage c)
7037_C002.fm Page 13 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-42-320.jpg)
![Next-Generation Technologies to Enable Sensor Networks 2-15
application and the associated host resources using the smallest fraction of overall network bandwidth
available. Even for moderately sized graphs (e.g., 1000 vertices by 10 stages), this is a complex combinatorial
optimization problem; the general problem is NP complete. The authors have developed an iterative
heuristic algorithm that has shown favorable performance for this class of problem in the quality of the
solution and time to solution compared to other popular combinatorial optimization algorithms [31].
2.5.4 NRM Agents
The NRM agents are information and service links between the NRM and each of the resources. Agents
must first register and be authenticated (e.g., using Kerberos [32]) before an NRM will invoke their
services. This registration includes a characterization of the resource capabilities and services. When
registered, the NRM will use these remotely deployed agents on computational resources to download
and launch parameterized executables and modify the access control list (ACL) of switches and routers
under its control in the formation of DPNs. Agents also provide a mechanism for centralized software
maintenance and configuration by acting as transaction managers in the download and installation of
applications, databases, middleware, etc. As stated earlier, the agents also provide a measurement object
that is instantiated by applications to provide the NRM’s MO with performance metrics during graph
generation. Finally, agents give the NRM a view of the network state, periodically sending diagnostic
messages indicating its operational status.
2.5.5 Sensor Interface
Sensors can be thought of as resources much like computational and communication resources, which
are served by the NRM agents; thus, the sensor interface can be thought of as another type of NRM
agent. Because many different sensor platforms could be served by an NRM-managed resource network,
the sensor interface provides a common, abstract mechanism for communication between the NRM and
the sensor platforms.
Sensors will request services through the sensor interface from the NRM using a well-defined middleware
interface such as CORBA. This request for services involves requesting the proper application for the data
stream that the sensor will be delivering to the network of resources as well as a request for the required
metric constraints, such as throughput and latency (discussed in Subsection 2.5.2), needed to process the
sensor data stream effectively. The determination of required constraints could involve negotiations between
the sensor and the NRM through the sensor interface. The NRM uses the sensor interface to direct the
sensor platform to start sending a data stream once the NRM has marshaled the resources that the sensor
will need to satisfy the request. Finally, the sensor interface also facilitates communications between the
sensor platform and the NRM regarding flow control, application shutdown, etc.
2.5.6 Mapping Database
This mapping database is populated with data structures generated by the graph generator and metrics
object; it represents the weighted graph-theoretic characterization of the various parallel permutations
of an application that is mapped to networked resources. Graph search uses the mapping database to
reconstitute a weighted graph for each application for which it is asked to find resources and the degree
and form of parallelism needed to meet real-time constraints.
2.5.7 Topology Database
The topology database stores the current state of each of the resources; the graph generator and graph
search use this database. Graph generator uses the topology database to determine which resources are
available and most appropriate for candidate task-application mappings. Graph search uses this database
to verify that resources are functional before a set of resources is chosen to host an application, as well
as for generating and modifying weights associated with resource utilization. The topology database is
7037_C002.fm Page 15 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-44-320.jpg)
![2-16 Smart Dust
generated during the discovery phase when the NRM first comes online (e.g., see Breitbart et al. [33]
and Astic and Foster [34]). Alternatively, an administrator could choose to generate a topology database
for the NRM that enumerates connectivity and capability among all computation and storage resources
under its control. Agent reports (or lack thereof) will affect state changes in this database indicating
whether the resource is online or offline.
2.5.8 NRM Federation
In a large network with a sizeable number of resources, using a single NRM may not be the most effective
solution. In such a scenario, multiple NRMs are organized in a bilevel hierarchy; wide-area network
(WAN) NRMs interface with sensors and administer backbone communication resources, underneath
which local-area network (LAN) NRMs administer and allocate compute resources for regional compute
centers (RCCs). The primary responsibility of a WAN NRM is to choose a location on the network at
which distributed computing is conducted for each application and to allocate WAN bandwidth for data
flow between sensors and LAN resources. The objective of the WAN NRM is to load balance WAN traffic
and computational load, taking into account the relative overall processing capability of each RCC. Each
LAN NRM advertises its current processing capability using standardized metrics.
Each NRM is a federated collection, using a voting mechanism to elect an executor independently at
the LAN and WAN levels. Each federation monitors the health of its executor by inspecting periodic
diagnostic reports that the executor broadcasts. In response to an executor’s diagnostic report (or lack
thereof), the federation may choose to relieve the current executor of its responsibility and elect a new
one. This prevents any one NRM failure from rendering resources unusable or disabling a sensor from
contracting for network services.
Earlier paragraphs have detailed the LAN NRMs graph-theoretic representation of network resources,
as well as its construction, weighting, and search criteria. The WAN NRM graph-theoretic representation
and weighting are somewhat different from that of a LAN NRM; however, its construction and search
criteria are formulated in an identical manner. The vertices in a WAN graph represent RCCs and each
column corresponds to an application, while the concatenation of applications across the columns in a
WAN NRM graph spans a mission. This is in contrast to a LAN NRM, in which the concatenation of
tasks in its graph spans an application.
2.5.9 NRM Fault Tolerance
The absence of a heartbeat or the delivery of an error report by an agent alerts the NRM to a system
fault. The NRM’s fault tolerance policy is application dependent and is derived from a mandate by the
developer and/or client. The policy is a trade-off between resource usage and seamless fail-over and
includes redundant processing, surgical replacement, or restart of the application. Redundant processing
is the most robust fail-over mechanism; the NRM simply assigns duplicate sets of resources to process
the same data. If one set of resources fails, results are obtained from one of the duplicate sets. Redundant
processing has the highest resource cost of all fault tolerant policies.
Conversely, the NRM may choose to replace the failed component dynamically so that processing is
able to continue. In this case, the NRM may have allocated distributed network storage to act as a time-
delay buffer in the event of resource failure. This would enable the application, if so instrumented, to
pick up processing at the point at which the failure occurred. Finally, the NRM could simply choose to
halt execution of the application and start over with a new set of processing resources, although a certain
amount of data and the corresponding results may be lost irrevocably.
2.6 Experimental Results
A proof-of-concept experiment has been conducted at MIT Lincoln Laboratory in which the NRM
allocates distributed networked resources for a sensor data fusion application in various scenarios [35].
7037_C002.fm Page 16 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-45-320.jpg)
![Next-Generation Technologies to Enable Sensor Networks 2-17
The sensor fusion application is OASIS (operator assisted integrated systems), which is an automatic
target recognition and visualization suite (see Figure 2.8). OASIS processes real-time SAR data and
archived data generated by sensors with different modalities like EO and IR [36]. A block diagram of the
experimental test bed is shown in Figure 2.9. The experimentation resource network consisted of three
FIGURE 2.8 OASIS ATR and visualization.
TABLE 2.2 Synopsis of NRM Expected Performance
Experimental
Configuration
Max Comm BW
Requirement
(MB/s)
Max Throughput
Requirement
(GFLOPS)
Processors
Employed
Result
Turn-Around
Time
1 m data 26 0.7 1 1.6
1 m data with HDVI 26 2.2 2 2.6
1/4 m data 410 2.5 2 2.8
1/4 m data with HDVI 410 10 10 7
TABLE 2.3 Synopsis of NRM Performance
Experimental
Configuration
Comm BW
Measured
(MB/s)
Throughput
Measured
(GFLOPS)
Processors
Employed
Result
Turn-Around
Time
1 m data 26 0.7 1 1.4
1 m data with HDVI 26 2.2 2 2.5
1/4 m data 410 2.5 2 2.7
1/4 m data with HDVI 410 10 8 7.8
OASIS
Archived Data
Provides historical
information for
area delimitation &
change
Real-time
SIGINT Data
provides cuing
Real-time
IMINT Data
provide timely,
day-night, all-
weather data
EO
IR
SAR
SAR GMTI
SIGINT
Screener
Registration
Data Mining
3-D Fusion
Emulated
+
+
+
7037_C002.fm Page 17 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-46-320.jpg)
![2-18 Smart Dust
SGI O2 workstations, an eight-processor SGI Origin, an eight-node, dual Pentium3 class Beowulf cluster,
and a PC workstation, which hosted the NRM.
For this experiment, two SGI O2s were used as sensor surrogates to transmit unprocessed complex
SAR imagery generated with range and cross-range resolutions of 1 and 1/4 m, respectively. The sensor
surrogates fed data into the OASIS processing chain. To keep the complexity of the system manageable,
only the most computationally intensive stage was made remappable. This stage, the HDVI processing
[3] (stage 3 in Figure 2.10), had six options for the NRM ranging from a single SGI processor to six
Pentium3 class cluster processors. The HDVI processing was conducted on targets detected on the two
images at both resolutions, and image formation was conducted on processors in the local area network.
The performance metrics for the OASIS applications were determined with a combination of actual
performance measurements and modeled performance analyses. Table 2.2 is a tabulated synopsis of the
expected performance of the NRM and Table 2.3 shows the actual performance of the NRM. The expected
and actual performance values compared very well.
Because this network was PE resource limited, the objective of the NRM was to use the smallest fraction
of PE bandwidth available across the network while meeting network conduit, PE utilization, latency,
throughput, and network-wide bandwidth usage constraints. It is clear from the results that the NRM
was able to tailor the communication and computation solution it delivered based on the particular
application needs and the constraints imposed. The successful completion of this experiment has initiated
further research and development to give the NRM greater functionality, automation, and flexibility.
FIGURE 2.9 Experimentation resource network.
CONUS
Resources
Theater
Resources
Sim’d SAR sensor 1 Sim’d SAR sensor 2
Parallel Cluster
Visualization and
OASIS Data Exploitation
OASIS Data Exploitation
Network Resource Manager
1000 Mbps
Private network
on GLOWNet
7037_C002.fm Page 18 Tuesday, November 1, 2005 12:20 PM
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-47-320.jpg)
![3-1
3
Sensor Network
Management
3.1 Introduction ...................................................................... 3-1
3.2 Management Challenges................................................... 3-2
3.3 Management Dimensions................................................. 3-3
Dimensions for WSN Management • Management Levels •
WSN Functionalities • Management Functional Areas
3.4 MANNA as an Integrating Architecture........................ 3-15
Management Services, Functions, and Models • Functional
Architecture • Information Architecture • Physical Architecture
3.5 Putting It All Together.................................................... 3-25
3.6 Conclusion....................................................................... 3-25
3.1 Introduction
A wireless sensor network (WSN) consists of a large number of sensor nodes deployed over an area and
integrated to collaborate through a wireless network. WSNs encourage several novel and existing appli-
cations such as environmental monitoring; health care; infrastructure management; public safety; med-
ical; home and office security; transportation; and military [1, 2, 9, 17, 18]. These have been enabled by
the rapid convergence of three technologies: digital circuitry, wireless communications, and the micro-
electromechanical system (MEMS). These technologies have enabled very compact and autonomous
sensor nodes, each containing one or more sensor devices, computation and communication capabilities,
and limited power supply.
Some of the applications foreseen for WSNs will require a large number of devices in the order of tens
of thousands of nodes. Traditional methods of sensor networking represent an impractical, complex, and
expensive demand on cable installation. WSNs promise several advantages over traditional sensing
methods in many ways: better coverage, higher resolution, fault tolerance, and robustness. The ad hoc
nature and deploy-and-leave vision make it even more attractive in military applications and other risk-
associated applications, such as catastrophe, toxic zones, and disasters [2, 9]. Performing the processing
at the source can drastically reduce the computational burden on application, network, and management.
On the other hand, any solution must take into account specific characteristics of this type of network.
WSN management must be autonomic, i.e., self-managed (self-organizing, self-healing, self-optimiz-
ing, self-protecting, self-sustaining, self-diagnostic) with a minimum of human interference, and robust
to changes in network states while maintaining the quality of services. Until now, WSNs and their
applications have been developed without considering an integrated management solution. The task of
building and deploying management systems in environments that will contain tens of thousands of
network elements with particular features and organization and that deal with the aforementioned
Linnyer Beatrys Ruiz
Pontifical Catholic University of
Paraná and Federal University of
Minas Gerais
José Marcos Nogueira
Federal University of Minas Gerais
Antonio A. F. Loureiro
Federal University of Minas Gerais
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-51-320.jpg)
![Sensor Network Management 3-3
know its environment and the context surrounding its activity and act accordingly. The management
entities must find and generate rules to perform the best management of the current state of the network
[22].
A managed WSN with this has various characteristics can be called an autonomic system [1], which
is an approach to self-managed computing systems with a minimum of human interference. This term
derives from the autonomic nervous system of the human body, which controls key functions without
conscious awareness or involvement. The processors in such systems use algorithms to determine the
most efficient and cost-effective way to distribute tasks and store data. Along with software probes and
configuration controls, computer systems will be able to monitor, tweak, and even repair themselves
without requiring technology staff — at least, that is the goal [1].
WSN management must be autonomic, i.e., self-managed and robust to changes in network states
while maintaining the quality of service; that is, it must be capable of self-configuration, self-organization,
self-healing, and self-optimization. However, the computational cost of autonomic processes can be
expensive to some WSN architectures.
Probably, the fundamental issue about the management of a WSN is concerned with how the man-
agement can promote plant and resource productivity,and how it integrates in an organized way functions
of configuration, operation, administration, and maintenance of all elements and services.
The task of building and deploying autonomic management systems in environments in which tens
of thousands of network elements with particular features and organization will be present is very
complex. This task becomes even more involved due to the physical restrictions of the sensor nodes, in
particular energy and bandwidth restrictions. The management application to be built also depends on
the kind of application being monitored. A good strategy is to deal with complex management situations
by using management dimensions.
3.3 Management Dimensions
In general, for traditional networks, management aspects are clearly separated from network common
activities, i.e., from the services they provide to their users. It is also said that an overlap of management
and network functionalities exists, although the implementation can be thought of independently. This
separation can be promoted by using two traditional management dimensions: management functional
areas [14] and management levels [15].
The requirements to be satisfied by systems management activities can be categorized into functional
areas. These facilities have come to be known as the specific management functional areas (SMFAs): fault
management; configuration management; performance management; accounting management; and
security management. This has proved to be a helpful way of partitioning the network management
problem from an application point of view [14].
To deal with the complexity of management, management functionality with its associated information
can be decomposed into a number of logical layers: business management; service management; network
management; and network element management. The architecture that describes this layering is called
the logical layered architecture (LLA) [15]. Management activities can be clustered into layers and decou-
pled by introducing manager and agent roles. A logical layer reflects particular aspects of management
and implies the clustering of management information supporting that aspect. Typically, an interaction
takes place between adjacent layers, but due to operational and management considerations other inter-
actions may also occur between nonadjacent layers.
The use of the management dimensions is a good strategy to deal with complex management situations
by decomposing a problem into smaller subproblems, in successive refinements steps, and to provide a
separation between application and management functionalities through a management architecture.
This will make possible the integration of organizational, administrative, and maintenance activities for
a given network.
WSN management must be simple,adherent to network idiosyncrasies,including its dynamic behavior,
and efficient in its use of scarce resources. The adoption of a strategy based on the traditional framework
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-53-320.jpg)
![3-4 Smart Dust
of functional areas and management levels will permit management integration in the future. However,
for WSN management it is necessary to go further. Using management functional areas and management
levels is not enough because WSNs are application specific.
The following discussion concerns how the traditional management dimensions can be applied in
WSN management. Also, new dimension for WSN management is proposed that considers the general
aspects of the different types of the networks.
3.3.1 Dimensions for WSN Management
WSNs are embedded in applications to monitor the environment and act upon it. Thus, the management
application should try to be “compatible” with the kind of application being monitored. In order to have
better development of WSN management services and functions, it is necessary to characterize the WSN
and establish a novel management dimension. Thus, looking at the characteristics of various WSN
applications, five main WSN functionalities are identified: configuration; sensing; processing; commun-
ication; and maintenance. These functionalities define a novel dimension for the management, as pre-
sented in Figure 3.1 [22]. Configuration is the first functionality before a network starts sensing the
environment, processing, and communicating data. Maintenance treats specific characteristics of WSN
applications during the entire network lifetime.
In this way, WSN management will have an organization that comes from abstractions offered by
management functional areas, management levels, and WSN functionalities (configuration, sensing,
processing, communication, and maintenance). The novel dimension introduced can be observed in the
upper part of Figure 3.1.
The coordination among the three planes can be based on policies. Policy-based network management
(PBNM) [7] is a feasible alternative because it allows the manager to set actions to be carried out by the
network without worrying too much about network details. Managers can define suitable actions in due
time and still have a global or local view of the network. PBNM helps to manage complex networks such
as WSNs. The managers will only inform concerning what is expected, but not how it should be obtained.
The agents will be intelligent to decide what to do as well as how and when to do it. Automatic services
and functions can be executed toward self-management if appropriate conditions, such as residual energy
level, are present.
FIGURE 3.1 Management dimensions for WSNs. (From Ruiz, L.B., Nogueira, J.M., Louriero, A.A., IEEE Commun.
Mag., 41(2), 116–125, 2003. With permission.)
WSN FUNCTIONALITIES
Configuration
Maintenance
Sensing
Processing
Communication
MANAGEMENT LEVELS
Business Management
Service Management
Network Management
Network Element Management
Network Element
FUNCTIONAL AREAS
Configuration Management
Fault Management
Performance Management
Security Management
Accounting Management
© 2006 by Taylor & Francis Group, LLC](https://image.slidesharecdn.com/1160662-250528191231-d9346bf4/85/Smart-Dust-Sensor-Network-Applications-Architecture-And-Design-1st-Edition-Mohammad-Ilyas-54-320.jpg)
Smart Dust Sensor Network Applications Architecture And Design 1st Edition Mohammad Ilyas
- 1. Smart Dust Sensor Network Applications Architecture And Design 1st Edition Mohammad Ilyas download https://ebookbell.com/product/smart-dust-sensor-network- applications-architecture-and-design-1st-edition-mohammad- ilyas-2321324 Explore and download more ebooks at ebookbell.com
- 2. Here are some recommended products that we believe you will be interested in. You can click the link to download. Nanochips And Smart Dust The Dangerous New Face Of The Human Microchipping Agenda Makia Freeman https://ebookbell.com/product/nanochips-and-smart-dust-the-dangerous- new-face-of-the-human-microchipping-agenda-makia-freeman-11606474 Functional Smart Nanomaterials And Their Theranostics Approaches Alle Madhusudhan https://ebookbell.com/product/functional-smart-nanomaterials-and- their-theranostics-approaches-alle-madhusudhan-55063206 Start Your Own Restaurant Business And More Pizzeria Coffeehouse Deli Bakery Catering Business Ipro Dist Product Ii 3rd Edition Jacquelyn Lynn https://ebookbell.com/product/start-your-own-restaurant-business-and- more-pizzeria-coffeehouse-deli-bakery-catering-business-ipro-dist- product-ii-3rd-edition-jacquelyn-lynn-33360836 Dr Ak Usmle Ck Notes Smart https://ebookbell.com/product/dr-ak-usmle-ck-notes-smart-58130918
- 3. Smart Hybrid Acdc Microgrids Power Management Energy Management And Power Quality Control Yunwei Ryan Li https://ebookbell.com/product/smart-hybrid-acdc-microgrids-power- management-energy-management-and-power-quality-control-yunwei-ryan- li-44868658 Smart Manufacturing Masoud Soroush Michael Baldea Thomas F Edgar https://ebookbell.com/product/smart-manufacturing-masoud-soroush- michael-baldea-thomas-f-edgar-44887000 Smart Health Technologies For The Covid19 Pandemic Internet Of Medical Things Perspectives Chinmay Chakraborty https://ebookbell.com/product/smart-health-technologies-for-the- covid19-pandemic-internet-of-medical-things-perspectives-chinmay- chakraborty-44888226 Smart Nanostructure Materials And Sensor Technology Rakesh Kumar Sonker https://ebookbell.com/product/smart-nanostructure-materials-and- sensor-technology-rakesh-kumar-sonker-44938902 Smart Stimuliresponsive Polymers Liang Hu Yongfeng Gao Michael J Serpe https://ebookbell.com/product/smart-stimuliresponsive-polymers-liang- hu-yongfeng-gao-michael-j-serpe-45009058
- 6. Smart Dust: Sensor Network Applications, Architecture, and Design Imad Mahgoub Mohammad Ilyas © 2006 by Taylor & Francis Group, LLC
- 7. The material was previously published in Handbook of Sensor Networks: Compact Wireless and Wired Sensing Systems. © CRC Press LLC 2005. Published in 2006 by CRC Press Taylor & Francis Group 6000 Broken Sound Parkway NW, Suite 300 Boca Raton, FL 33487-2742 © 2006 by Taylor & Francis Group, LLC CRC Press is an imprint of Taylor & Francis Group No claim to original U.S. Government works Printed in the United States of America on acid-free paper 10 9 8 7 6 5 4 3 2 1 International Standard Book Number-10: 0-8493-7037-X (Hardcover) International Standard Book Number-13: 978-0-8493-7037-3 (Hardcover) Library of Congress Card Number 2005022133 This book contains information obtained from authentic and highly regarded sources. Reprinted material is quoted with permission, and sources are indicated. A wide variety of references are listed. Reasonable efforts have been made to publish reliable data and information, but the author and the publisher cannot assume responsibility for the validity of all materials or for the consequences of their use. No part of this book may be reprinted, reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other means, now known or hereafter invented, including photocopying, microfilming, and recording, or in any information storage or retrieval system, without written permission from the publishers. For permission to photocopy or use material electronically from this work, please access www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Center, Inc. (CCC) 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-for-profit organization that provides licenses and registration for a variety of users. For organizations that have been granted a photocopy license by the CCC, a separate system of payment has been arranged. Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and are used only for identification and explanation without intent to infringe. Library of Congress Cataloging-in-Publication Data Smart dust : sensor network applications, architecture, and design / editors Imad Mahgoub, and Mohammad Ilyas. p. cm. Includes bibliographical references and index. ISBN 0-8493-7037-X (9780849370373 : alk. paper) 1. Sensor networks. I. Ilyas, Mohammad, 1953- II. Mahgoub, Imad. TK7872.D48S63 2006 681'.2--dc22 2005022133 Visit the Taylor & Francis Web site at http://www.taylorandfrancis.com and the CRC Press Web site at http://www.crcpress.com Taylor & Francis Group is the Academic Division of Informa plc. 7037_Discl.fm Page 1 Tuesday, November 22, 2005 11:03 AM © 2006 by Taylor & Francis Group, LLC
- 8. v Preface Advances in wireless communications and microelectronic mechanical system technologies have enabled the development of networks of a large number of small inexpensive, low-power multifunctional sensors. These networks nicknamed “Smart Dust” present a very interesting and challenging area and have tremendous potential applications. Wireless sensor networks consist of a large number of sensor nodes that may be randomly and densely deployed. Sensor nodes are small electronic components capable of sensing many types of information from the environment including temperature, light, humidity, radiation, the presence or nature of biological organisms, geological features, seismic vibrations, specific types of computer data, and more. Recent advancements have made it possible to make these components small, powerful, and energy efficient, and they can now be manufactured cost-effectively in quantity for specialized telecommunica- tion applications. The sensor nodes are very small in size and are capable of gathering, processing, and communicating information to other nodes and to the outside world. This handbook is expected to capture the current state of sensor networks, and specifically address the architecture, applications, and design of such networks. This handbook has a total of 17 chapters written by experts from around the world. The targeted audience for this handbook includes professionals who are designers and planners for emerging telecommunication networks, researchers (faculty members and graduate students), and those who would like to learn about this field. Although this handbook is not precisely a textbook, it can certainly be used as a textbook for graduate courses and research-oriented courses that deal with wireless sensor networks. Any comments from the readers will be highly appreciated. Many people have contributed to this handbook in their unique ways. The first and the foremost group that deserves immense gratitude is the group of highly-talented and skilled researchers who have con- tributed to this handbook. All of them have been extremely cooperative and professional. It has also been a pleasure to work with Nora Konopka, Helena Redshaw, and Allison Taub of Taylor & Francis, and we are extremely gratified for their support and professionalism. Our families have extended their uncon- ditional love and strong support throughout this project and they all deserve very special thanks. Imad Mahgoub and Mohammad Ilyas Boca Raton, Florida 7037_C000.fm Page v Tuesday, November 22, 2005 10:13 AM © 2006 by Taylor & Francis Group, LLC
- 9. vii Editors Imad Mahgoub,Ph.D., received his B.Sc. degree in electrical engineering from the University of Khartoum, Khartoum, Sudan, in 1978. From 1978 to 1981, he worked for the Sudan Shipping Line Company, Port Sudan, Sudan, as an electrical and electronics engineer. He received his M.S. in applied mathematics in 1983 and his M.S. in electrical and computer engineering in 1986, both from North Carolina State University. In 1989, he received his Ph.D. in computer engineering from The Pennsylvania State University. Since August 1989, Dr. Mahgoub has been with the College of Engineering at Florida Atlantic Uni- versity, Boca Raton, Florida, where he is currently professor of computer science and engineering. He is the director of the Computer Science and Engineering Department Mobile Computing Laboratory at Florida Atlantic University. Dr. Mahgoub has conducted successful research in various areas, including mobile computing; inter- connection networks; performance evaluation of computer systems; and advanced computer architecture. He has published more than 80 research articles and supervised three Ph.D. dissertations and 22 M.S. theses to completion. He has served as a consultant to industry. Dr. Mahgoub served as a member of the executive committee/program committee of the 1998, 1999, and 2000 IEEE International Performance, Computing and Communications Conferences. He has served on the program committees of several international conferences and symposia. He was the vice chair of the 2003, 2004, and 2005 International Symposium on Performance Evaluation of Computer and Telecommunication Systems. Dr. Mahgoub is a senior member of IEEE and a member of ACM. Mohammad Ilyas, Ph.D., received his B.Sc. degree in electrical engineering from the University of Engineering and Technology, Lahore, Pakistan, in 1976. From March 1977 to September 1978, he worked for the Water and Power Development Authority in Pakistan. In 1978, he was awarded a scholarship for his graduate studies and he completed his M.S. degree in electrical and electronic engineering in June 1980 at Shiraz University, Shiraz, Iran. In September 1980, he joined the doctoral program at Queen’s University in Kingston, Ontario, Canada; he completed his Ph.D. degree in 1983. Dr. Ilyas’s doctoral research was about switching and flow control techniques in computer communication networks. Since September 1983, he has been with the College of Engineering at Florida Atlantic University, Boca Raton, Florida, where he is currently associate dean for graduate studies and research. From 1994 to 2000, he was chair of the department. During the 1993–1994 academic year, he was on sabbatical leave with the Department of Computer Engineering, King Saud University, Riyadh, Saudi Arabia. Dr. Ilyas has conducted successful research in various areas, including traffic management and con- gestion control in broadband/high-speed communication networks; traffic characterization; wireless communication networks; performance modeling; and simulation. He has published one book, three handbooks, and more than 140 research articles. He has supervised 10 Ph.D. dissertations and more than 35 M.S. theses to completion. Dr. Ilyas has been a consultant to several national and international organizations; a senior member of IEEE, he is an active participant in several IEEE technical committees and activities. 7037_C000.fm Page vii Tuesday, November 22, 2005 10:13 AM © 2006 by Taylor & Francis Group, LLC
- 10. ix Contributors Özgür B. Akan Georgia Institute of Technology Atlanta, Georgia Cristian Borcea Rutgers University Piscataway, New Jersey Athanassios Boulis University of California at Los Angeles Los Angeles, California Erdal Cayirci Istanbul Technical University Istanbul, Turkey Anantha Chandrakasan Engim, Inc. Acton, Massachusetts Duminda Dewasurendra Virginia Polytechnic Institute and State University Blacksburg, Virginia Jessica Feng University of California at Los Angeles Los Angeles, California Vicente González–Millán University of Valencia Valencia, Spain Joel I. Goodman MIT Lincoln Laboratory Lexington, Massachusetts Martin Haenggi University of Notre Dame Notre Dame, Indiana Hossam Hassanein Queen’s University Kingston, Ontario, Canada Chi-Fu Huang National Chiao-Tung University Hsin-Chu, Taiwan Liviu Iftode Rutgers University Piscataway, New Jersey Chaiporn Jaikaeo University of Delaware Newark, Delaware Porlin Kang Rutgers University Piscataway, New Jersey Zdravko Karakehayov Technical University of Sofia Sofia, Bulgaria Farinaz Koushanfar University of California at Berkeley Berkeley, California Sheng-Po Kuo National Chiao-Tung University Hsin-Chu, Taiwan Antonio A.F. Loureiro Federal University of Minas Gerais Belo Horizonte, Brazil David R. Martinez MIT Lincoln Laboratory Lexington, Massachusetts Amitabh Mishra Virginia Polytechnic Institute and State University Blacksburg, Virginia José Marcos Nogueira Federal University of Minas Gerais Belo Horizonte, Brazil Miodrag Potkonjak University of California at Los Angeles Los Angeles, California Albert I. Reuther MIT Lincoln Laboratory Lexington, Massachusetts Linnyer Beatrys Ruiz Pontifical Catholic University of Paraná Curitiba, Brazil and Federal University of Minas Gerais Belo Horizonte, Brazil 7037_C000.fm Page ix Tuesday, November 22, 2005 10:13 AM © 2006 by Taylor & Francis Group, LLC
- 11. x Ayad Salhieh Wayne State University Detroit, Michigan Enrique Sanchis-Peris University of Valencia Valencia, Spain Loren Schwiebert Wayne State University Detroit, Michigan Chien-Chung Shen University of Delaware Newark, Delaware Amit Sinha Engim, Inc. Acton, Massachusetts Sasha Slijepcevic University of California at Los Angeles Los Angeles, California Chavalit Srisathapornphat University of Delaware Newark, Delaware Weilian Su Georgia Institute of Technology Atlanta, Georgia Yu-Chee Tseng National Chiao-Tung University Hsin-Chu, Taiwan Quanhong Wang Queen’s University Kingston, Ontario, Canada Brett Warneke Dust Networks Berkeley, California Jennifer L. Wong University of California at Los Angeles Los Angeles, California Kenan Xu Queen’s University Kingston, Ontario, Canada 7037_C000.fm Page x Tuesday, November 22, 2005 10:13 AM © 2006 by Taylor & Francis Group, LLC
- 12. xi Contents 1 Opportunities and Challenges in Wireless Sensor Networks Martin Haenggi 1.1 Introduction......................................................................................................................... 1-1 1.2 Opportunities....................................................................................................................... 1-2 1.3 Technical Challenges ........................................................................................................... 1-4 1.4 Concluding Remarks ......................................................................................................... 1-11 2 Next-Generation Technologies to Enable Sensor Networks Joel I. Goodman, Albert I. Reuther, David R. Martinez 2.1 Introduction......................................................................................................................... 2-1 2.2 Goals for Real-Time Distributed Network Computing for Sensor Data Fusion ............. 2-5 2.3 The Convergence of Networking and Real-Time Computing.......................................... 2-6 2.4 Middleware......................................................................................................................... 2-11 2.5 Network Resource Management....................................................................................... 2-11 2.6 Experimental Results ......................................................................................................... 2-16 3 Sensor Network Management Linnyer Beatrys Ruiz, José Marcos Nogueira, Antonio A. F. Loureiro 3.1 Introduction......................................................................................................................... 3-1 3.2 Management Challenges...................................................................................................... 3-2 3.3 Management Dimensions ................................................................................................... 3-3 3.4 MANNA as an Integrating Architecture........................................................................... 3-15 3.5 Putting It All Together....................................................................................................... 3-25 3.6 Conclusion ......................................................................................................................... 3-25 4 Models for Programmability in Sensor Networks Athanassios Boulis 4.1 Introduction......................................................................................................................... 4-1 4.2 Differences between Sensor Networks and Traditional Data Networks........................... 4-2 4.3 Aspects of Efficient Sensor Network Applications............................................................. 4-2 4.4 Need for Sensor Network Programmability....................................................................... 4-3 4.5 Major Models for System-Level Programmability............................................................. 4-4 4.6 Frameworks for System-Level Programmability................................................................ 4-6 4.7 Conclusions........................................................................................................................ 4-12 7037_C000.fm Page xi Tuesday, November 22, 2005 10:13 AM © 2006 by Taylor & Francis Group, LLC
- 13. xii 5 Miniaturizing Sensor Networks with MEMS Brett Warneke 5.1 Introduction......................................................................................................................... 5-1 5.2 MEMS Basics........................................................................................................................ 5-2 5.3 Sensors.................................................................................................................................. 5-4 5.4 Communication................................................................................................................... 5-5 5.5 Micropower Sources.......................................................................................................... 5-10 5.6 Packaging............................................................................................................................ 5-12 5.7 Systems ............................................................................................................................... 5-13 5.8 Conclusion ......................................................................................................................... 5-15 6 Sensor Network Architecture and Applications Chien-Chung Shen, Chaiporn Jaikaeo, Chavalit Srisathapornphat 6.1 Introduction......................................................................................................................... 6-1 6.2 Sensor Network Applications.............................................................................................. 6-1 6.3 Functional Architecture for Sensor Networks.................................................................... 6-3 6.4 Sample Implementation Architectures............................................................................... 6-4 6.5 Summary ............................................................................................................................ 6-12 7 A Practical Perspective on Wireless Sensor Networks Quanhong Wang, Hossam Hassanein, Kenan Xu 7.1 Introduction......................................................................................................................... 7-1 7.2 WSN Applications................................................................................................................ 7-2 7.3 Classification of WSNs ........................................................................................................ 7-6 7.4 Characteristics, Technical Challenges, and Design Directions.......................................... 7-7 7.5 Technical Approaches........................................................................................................ 7-11 7.6 Conclusions and Considerations for Future Research .................................................... 7-22 8 Sensor Network Architecture Jessica Feng, Farinaz Koushanfar, Miodrag Potkonjak 8.1 Overview............................................................................................................................... 8-1 8.2 Motivation and Objectives .................................................................................................. 8-1 8.3 SNs — Global View and Requirements.............................................................................. 8-3 8.4 Individual Components of SN Nodes................................................................................. 8-4 8.5 Sensor Network Node.......................................................................................................... 8-8 8.6 Wireless SNs as Embedded Systems.................................................................................. 8-13 8.7 Summary ............................................................................................................................ 8-16 9 Power-Efficient Topologies for Wireless Sensor Networks Ayad Salhieh, Loren Schwiebert 9.1 Motivation............................................................................................................................ 9-1 9.2 Background .......................................................................................................................... 9-2 9.3 Issues for Topology Design ................................................................................................. 9-3 9.4 Assumptions......................................................................................................................... 9-8 7037_C000.fm Page xii Tuesday, November 22, 2005 10:13 AM © 2006 by Taylor & Francis Group, LLC
- 14. xiii 9.5 Analysis of Power Usage.................................................................................................... 9-10 9.6 Directional Source-Aware Routing Protocol (DSAP) ..................................................... 9-13 9.7 DSAP Analysis.................................................................................................................... 9-15 9.8 Summary ............................................................................................................................ 9-19 10 Overview of Communication Protocols for Sensor Networks Weilian Su, Erdal Cayirci, Özgür B. Akan 10.1 Introduction...................................................................................................................... 10-1 10.2 Applications/Application Layer Protocols....................................................................... 10-2 10.3 Localization Protocols ...................................................................................................... 10-4 10.4 Time Synchronization Protocols ..................................................................................... 10-5 10.5 Transport Layer Protocols................................................................................................ 10-7 10.6 Network Layer Protocols.................................................................................................. 10-9 10.7 Data Link Layer Protocols.............................................................................................. 10-11 10.8 Conclusion ...................................................................................................................... 10-14 11 Positioning and Location Tracking in Wireless Sensor Networks Yu-Chee Tseng, Chi-Fu Huang, Sheng-Po Kuo 11.1 Introduction...................................................................................................................... 11-1 11.2 Fundamentals.................................................................................................................... 11-2 11.3 Positioning and Location Tracking Algorithms.............................................................. 11-4 11.4 Experimental Location Systems..................................................................................... 11-10 11.5 Conclusions..................................................................................................................... 11-12 12 Comparison of Data Processing Techniques in Sensor Networks Vicente González-Millán, Enrique Sanchis-Peris 12.1 Sensor Networks: Organization and Processing ............................................................. 12-1 12.2 Architectures for Sensor Integration ............................................................................... 12-3 12.3 Example of Architecture Evaluation in High-Energy Physics...................................... 12-18 13 Cooperative Computing in Sensor Networks Liviu Iftode, Cristian Borcea, Porlin Kang 13.1 Introduction...................................................................................................................... 13-1 13.2 The Cooperative Computing Model................................................................................ 13-3 13.3 Node Architecture............................................................................................................. 13-4 13.4 Smart Messages ................................................................................................................. 13-5 13.5 Programming Interface .................................................................................................... 13-7 13.6 Prototype Implementation and Evaluation..................................................................... 13-8 13.7 Applications .................................................................................................................... 13-12 13.8 Simulation Results .......................................................................................................... 13-14 13.9 Related Work................................................................................................................... 13-15 13.10 Conclusions..................................................................................................................... 13-18 7037_C000.fm Page xiii Tuesday, November 22, 2005 10:13 AM © 2006 by Taylor & Francis Group, LLC
- 15. xiv 14 Dynamic Power Management in Sensor Networks Amit Sinha, Anantha Chandrakasan 14.1 Introduction...................................................................................................................... 14-1 14.2 Idle Power Management................................................................................................... 14-2 14.3 Active Power Management............................................................................................... 14-5 14.4 System Implementation.................................................................................................... 14-6 14.5 Results.............................................................................................................................. 14-12 15 Design Challenges in Energy-Efficient Medium Access Control for Wireless Sensor Networks Duminda Dewasurendra, Amitabh Mishra 15.1 Introduction...................................................................................................................... 15-1 15.2 Unique Characteristics of Wireless Sensor Networks..................................................... 15-2 15.3 MAC Protocols for Wireless Ad Hoc Networks.............................................................. 15-4 15.4 Design Challenges for Wireless Sensor Networks......................................................... 15-10 15.5 Medium Access Protocols for Wireless Sensor Networks ............................................ 15-13 15.6 Open Issues ..................................................................................................................... 15-22 15.7 Conclusions..................................................................................................................... 15-24 16 Security and Privacy Protection in Wireless Sensor Networks Sasha Slijepcevic, Jennifer L. Wong, Miodrag Potkonjak 16.1 Introduction...................................................................................................................... 16-1 16.2 Unique Security Challenges in Sensor Networks and Enabling Mechanisms............... 16-2 16.3 Security Architectures....................................................................................................... 16-4 16.4 Privacy Protection........................................................................................................... 16-11 16.5 Conclusion ...................................................................................................................... 16-15 17 Low-Power Design for Smart Dust Networks Zdravko Karakehayov 17.1 Introduction...................................................................................................................... 17-1 17.2 Location............................................................................................................................. 17-1 17.3 Sensing............................................................................................................................... 17-2 17.4 Computation..................................................................................................................... 17-2 17.5 Hardware–Software Interaction....................................................................................... 17-5 17.6 Communication................................................................................................................ 17-7 17.7 Orientation...................................................................................................................... 17-10 17.8 Conclusion ...................................................................................................................... 17-10 7037_C000.fm Page xiv Tuesday, November 22, 2005 10:13 AM © 2006 by Taylor & Francis Group, LLC
- 16. 1-1 1 Opportunities and Challenges in Wireless Sensor Networks 1.1 Introduction ...................................................................... 1-1 1.2 Opportunities .................................................................... 1-2 Growing Research and Commercial Interest • Applications 1.3 Technical Challenges......................................................... 1-4 Performance Metrics • Power Supply • Design of Energy- Efficient Protocols • Capacity/Throughput • Routing • Channel Access and Scheduling • Modeling • Connectivity • Quality of Service • Security • Implementation • Other Issues 1.4 Concluding Remarks....................................................... 1-11 1.1 Introduction Due to advances in wireless communications and electronics over the last few years, the development of networks of low-cost, low-power, multifunctional sensors has received increasing attention. These sensors are small in size and able to sense, process data, and communicate with each other, typically over an RF (radio frequency) channel. A sensor network is designed to detect events or phenomena, collect and process data, and transmit sensed information to interested users. Basic features of sensor networks are: • Self-organizing capabilities • Short-range broadcast communication and multihop routing • Dense deployment and cooperative effort of sensor nodes • Frequently changing topology due to fading and node failures • Limitations in energy, transmit power, memory, and computing power These characteristics, particularly the last three, make sensor networks different from other wireless ad hoc or mesh networks. Clearly, the idea of mesh networking is not new; it has been suggested for some time for wireless Internet access or voice communication. Similarly, small computers and sensors are not innovative per se. However, combining small sensors, low-power computers, and radios makes for a new tech- nological platform that has numerous important uses and applications, as will be discussed in the next section. Martin Haenggi University of Notre Dame 7037_C001.fm Page 1 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 17. 1-2 Smart Dust 1.2 Opportunities 1.2.1 Growing Research and Commercial Interest Research and commercial interest in the area of wireless sensor networks are currently growing expo- nentially, which is manifested in many ways: • The number of Web pages (Google: 26,000 hits for sensor networks; 8000 for wireless sensor networks in August 2003) • The increasing number of • Dedicated annual workshops, such as IPSN (information processing in sensor networks); SenSys; EWSN (European workshop on wireless sensor networks); SNPA (sensor network protocols and applications); and WSNA (wireless sensor networks and applications) • Conference sessions on sensor networks in the communications and mobile computing com- munities (ISIT, ICC, Globecom, INFOCOM, VTC, MobiCom, MobiHoc) • Research projects funded by NSF (apart from ongoing programs, a new specific effort now focuses on sensors and sensor networks) and DARPA through its SensIT (sensor information technology), NEST (networked embedded software technology), MSET (multisensor exploi- tation), UGS (unattended ground sensors), NETEX (networking in extreme environments), ISP (integrated sensing and processing), and communicator programs Special issues and sections in renowned journals are common, e.g., in the IEEE Proceedings [1] and signal processing, communications, and networking magazines. Commercial interest is reflected in investments by established companies as well as start-ups that offer general and specific hardware and software solutions. Compared to the use of a few expensive (but highly accurate) sensors, the strategy of deploying a large number of inexpensive sensors has significant advantages, at smaller or comparable total system cost: much higher spatial resolution; higher robustness against failures through distributed operation; uniform coverage; small obtrusiveness; ease of deployment; reduced energy consumption; and, consequently, increased system lifetime. The main point is to position sensors close to the source of a potential problem phenomenon, where the acquired data are likely to have the greatest benefit or impact. Pure sensing in a fine-grained manner may revolutionize the way in which complex physical systems are understood. The addition of actuators, however, opens a completely new dimension by permitting management and manipulation of the environment at a scale that offers enormous opportunities for almost every scientific discipline. Indeed, Business 2.0 (http://www.business2.com/) lists sensor robots as one of “six technologies that will change the world,” and Technology Review at MIT and Globalfuture identify WSNs as one of the “10 emerging technologies that will change the world” (http://www.global- future.com/mit-trends2003.htm). The combination of sensor network technology with MEMS and nan- otechnology will greatly reduce the size of the nodes and enhance the capabilities of the network. The remainder of this chapter lists and briefly describes a number of applications for wireless sensor networks, grouped into different categories. However, because the number of areas of application is growing rapidly, every attempt at compiling an exhaustive list is bound to fail. 1.2.2 Applications 1.2.2.1 General Engineering • Automotive telematics. Cars, which comprise a network of dozens of sensors and actuators, are networked into a system of systems to improve the safety and efficiency of traffic. • Fingertip accelerometer virtual keyboards. These devices may replace the conventional input devices for PCs and musical instruments. • Sensing and maintenance in industrial plants. Complex industrial robots are equipped with up to 200 sensors that are usually connected by cables to a main computer. Because cables are 7037_C001.fm Page 2 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 18. Opportunities and Challenges in Wireless Sensor Networks 1-3 expensive and subject to wear and tear caused by the robot’s movement, companies are replacing them by wireless connections. By mounting small coils on the sensor nodes, the principle of induction is exploited to solve the power supply problem. • Aircraft drag reduction. Engineers can achieve this by combining flow sensors and blowing/sucking actuators mounted on the wings of an airplane. • Smart office spaces. Areas are equipped with light, temperature, and movement sensors, micro- phones for voice activation, and pressure sensors in chairs. Air flow and temperature can be regulated locally for one room rather than centrally. • Tracking of goods in retail stores. Tagging facilitates the store and warehouse management. • Tracking of containers and boxes. Shipping companies are assisted in keeping track of their goods, at least until they move out of range of other goods. • Social studies. Equipping human beings with sensor nodes permits interesting studies of human interaction and social behavior. • Commercial and residential security. 1.2.2.2 Agriculture and Environmental Monitoring • Precision agriculture. Crop and livestock management and precise control of fertilizer concentra- tions are possible. • Planetary exploration. Exploration and surveillance in inhospitable environments such as remote geographic regions or toxic locations can take place. • Geophysical monitoring. Seismic activity can be detected at a much finer scale using a network of sensors equipped with accelerometers. • Monitoring of freshwater quality. The field of hydrochemistry has a compelling need for sensor networks because of the complex spatiotemporal variability in hydrologic, chemical, and ecological parameters and the difficulty of labor-intensive sampling, particularly in remote locations or under adverse conditions. In addition, buoys along the coast could alert surfers, swimmers, and fishermen to dangerous levels of bacteria. • Zebranet. The Zebranet project at Princeton aims at tracking the movement of zebras in Africa. • Habitat monitoring. Researchers at UC Berkeley and the College of the Atlantic in Bar Harbor deployed sensors on Great Duck Island in Maine to measure humidity, pressure, temperature, infrared radiation, total solar radiation, and photosynthetically active radiation (see http:// www.greatduckisland.net/). • Disaster detection. Forest fire and floods can be detected early and causes can be localized precisely by densely deployed sensor networks. • Contaminant transport. The assessment of exposure levels requires high spatial and temporal sampling rates, which can be provided by WSNs. 1.2.2.3 Civil Engineering • Monitoring of structures. Sensors will be placed in bridges to detect and warn of structural weakness and in water reservoirs to spot hazardous materials. The reaction of tall buildings to wind and earthquakes can be studied and material fatigue can be monitored closely. • Urban planning. Urban planners will track groundwater patterns and how much carbon dioxide cities are expelling, enabling them to make better land-use decisions. • Disaster recovery. Buildings razed by an earthquake may be infiltrated with sensor robots to locate signs of life. 1.2.2.4 Military Applications • Asset monitoring and management. Commanders can monitor the status and locations of troops, weapons, and supplies to improve military command, control, communications, and computing (C4). 7037_C001.fm Page 3 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 19. 1-4 Smart Dust • Surveillance and battle-space monitoring. Vibration and magnetic sensors can report vehicle and personnel movement, permitting close surveillance of opposing forces. • Urban warfare. Sensors are deployed in buildings that have been cleared to prevent reoccupation; movements of friend and foe are displayed in PDA-like devices carried by soldiers. Snipers can be localized by the collaborative effort of multiple acoustic sensors. • Protection. Sensitive objects such as atomic plants, bridges, retaining walls, oil and gas pipelines, communication towers, ammunition depots, and military headquarters can be protected by intel- ligent sensor fields able to discriminate between different classes of intruders. Biological and chemical attacks can be detected early or even prevented by a sensor network acting as a warning system. • Self-healing minefields.The self-healing minefield system is designed to achieve an increased resistance to dismounted and mounted breaching by adding a novel dimension to the minefield. Instead of a static complex obstacle, the self-healing minefield is an intelligent, dynamic obstacle that senses relative positions and responds to an enemy’s breaching attempt by physical reorganization. 1.2.2.5 Health Monitoring and Surgery • Medical sensing. Physiological data such as body temperature, blood pressure, and pulse are sensed and automatically transmitted to a computer or physician, where it can be used for health status monitoring and medical exploration. Wireless sensing bandages may warn of infection. Tiny sensors in the blood stream, possibly powered by a weak external electromagnetic field, can continuously analyze the blood and prevent coagulation and thrombosis. • Microsurgery. A swarm of MEMS-based robots may collaborate to perform microscopic and minimally invasive surgery. The opportunities for wireless sensor networks are ubiquitous. However, a number of formidable chal- lenges must be solved before these exciting applications may become reality. 1.3 Technical Challenges Populating the world with networks of sensors requires a fundamental understanding of techniques for connecting and managing sensor nodes with a communication network in scalable and resource-efficient ways. Clearly, sensor networks belong to the class of ad hoc networks, but they have specific characteristics that are not present in general ad hoc networks. Ad hoc and sensor networks share a number of challenges such as energy constraints and routing. On the other hand,general ad hoc networks most likely induce traffic patterns different from sensor networks, have other lifetime requirements, and are often considered to consist of mobile nodes [2–4]. In WSNs, most nodes are static; however, the network of basic sensor nodes may be overlaid by more powerful mobile sensors (robots) that, guided by the basic sensors, can move to interesting areas or even track intruders in the case of military applications. Network nodes are equipped with wireless transmitters and receivers using antennas that may be omnidirectional (isotropic radiation), highly directional (point-to-point), possibly steerable, or some combination thereof. At a given point in time, depending on the nodes’ positions and their transmitter and receiver coverage patterns, transmission power levels, and cochannel interference levels, a wireless connectivity exists in the form of a random, multihop graph between the nodes. This ad hoc topology may change with time as the nodes move or adjust their transmission and reception parameters. Because the most challenging issue in sensor networks is limited and unrechargeable energy provision, many research efforts aim at improving the energy efficiency from different aspects. In sensor networks, energy is consumed mainly for three purposes: data transmission, signal processing, and hardware operation [5]. It is desirable to develop energy-efficient processing techniques that minimize power requirements across all levels of the protocol stack and, at the same time, minimize message passing for network control and coordination. 7037_C001.fm Page 4 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 20. Opportunities and Challenges in Wireless Sensor Networks 1-5 1.3.1 Performance Metrics To discuss the issues in more detail, it is necessary to examine a list of metrics that determine the performance of a sensor network: • Energy efficiency/system lifetime. The sensors are battery operated, rendering energy a very scarce resource that must be wisely managed in order to extend the lifetime of the network [6]. • Latency. Many sensor applications require delay-guaranteed service. Protocols must ensure that sensed data will be delivered to the user within a certain delay. Prominent examples in this class of networks are certainly the sensor-actuator networks. • Accuracy. Obtaining accurate information is the primary objective; accuracy can be improved through joint detection and estimation. Rate distortion theory is a possible tool to assess accuracy. • Fault tolerance. Robustness to sensor and link failures must be achieved through redundancy and collaborative processing and communication. • Scalability. Because a sensor network may contain thousands of nodes, scalability is a critical factor that guarantees that the network performance does not significantly degrade as the network size (or node density) increases. • Transport capacity/throughput. Because most sensor data must be delivered to a single base station or fusion center, a critical area in the sensor network exists (the gray area in Figure 1.1.), whose sensor nodes must relay the data generated by virtually all nodes in the network. Thus, the traffic load at those critical nodes is heavy, even when the average traffic rate is low. Apparently, this area has a paramount influence on system lifetime, packet end-to-end delay, and scalability. Because of the interdependence of energy consumption, delay, and throughput, all these issues and metrics are tightly coupled. Thus, the design of a WSN necessarily consists of the resolution of numerous trade-offs, which also reflects in the network protocol stack, in which a cross-layer approach is needed instead of the traditional layer-by-layer protocol design. 1.3.2 Power Supply The most difficult constraints in the design of WSNs are those regarding the minimum energy consumption necessary to drive the circuits and possible microelectromechanical devices (MEMS) [5, 7, 8]. The energy problem is aggravated if actuators are present that may be substantially hungrier for power than the sensors. When miniaturizing the node, the energy density of the power supply is the primary issue. Current technology yields batteries with approximately 1 J/mm3 of energy, while capacitors can achieve as much as 1 mJ/mm3. If a node is designed to have a relatively short life span, for example, a few months, a battery is a logical solution. However, for nodes that can generate sensor readings for long periods of time, a charging FIGURE 1.1 Sensor network with base station (or fusion center). The gray-shaded area indicates the critical area whose nodes must relay all the packets. critical nodes BS 7037_C001.fm Page 5 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 21. 1-6 Smart Dust method for the supply is preferable. Currently, research groups are investigating the use of solar cells to charge capacitors with photocurrents from the ambient light sources. Solar flux can yield power densities of approximately 1 mW/mm2. The energy efficiency of a solar cell ranges from 10 to 30% in current technologies, giving 300 μW in full sunlight in the best-case scenario for a 1-mm2 solar cell operating at 1 V. Series-stacked solar cells will need to be utilized in order to provide appropriate voltages. Sensor acquisition can be achieved at 1 nJ per sample, and modern processors can perform compu- tations as low as 1 nJ per instruction. For wireless communications, the primary candidate technologies are based on RF and optical transmission techniques, each of which has its advantages and disadvantages. RF presents a problem because the nodes may offer very limited space for antennas, thereby demanding very short-wavelength (i.e., high-frequency) transmission, which suffers from high attenuation. Thus, communication in that regime is not currently compatible with low-power operation. Current RF transmission techniques (e.g., Bluetooth [9]) consume about 100 nJ per bit for a distance of 10 to 100 m, making communication very expensive compared to acquisition and processing. An alternative is to employ free-space optical transmission. If a line-of-sight path is available, a well- designed free-space optical link requires significantly lower energy than its RF counterpart, currently about 1 nJ per bit. The reason for this power advantage is that optical transceivers require only simple baseband analog and digital circuitry and no modulators, active filters, and demodulators. Furthermore, the extremely short wavelength of visible light makes it possible for a millimeter-scale device to emit a narrow beam, corresponding to an antenna gain of roughly five to six orders of magnitude compared to an isotropic radiator. However, a major disadvantage is that the beam needs to be pointed very precisely at the receiver, which may be prohibitively difficult to achieve. In WSNs, where sensor sampling, processing, data transmission, and, possibly, actuation are involved, the trade-off between these tasks plays an important role in power usage. Balancing these parameters will be the focus of the design process of WSNs. 1.3.3 Design of Energy-Efficient Protocols It is well acknowledged that clustering is an efficient way to save energy for static sensor networks [10–13]. Clustering has three significant differences from conventional clustering schemes. First, data compression in the form of distributed source coding is applied within a cluster to reduce the number of packets to be transmitted [14, 15]. Second, the data-centric property makes an identity (e.g., an address) for a sensor node obsolete. In fact, the user is often interested in phenomena occurring in a specified area [16], rather than in an individual sensor node. Third, randomized rotation of cluster heads helps ensure a balanced energy consumption [11]. Another strategy to increase energy efficiency is to use broadcast and multicast trees [6, 17, 18], which take advantage of the broadcast property of omnidirectional antennas. The disadvantage is that the high computational complexity may offset the achievable benefit. For sensor networks, this one-to-many communication scheme is less important; however, because all data must be delivered to a single desti- nation, the traffic scheme (for application traffic) is the opposite, i.e., many to one. In this case, clearly the wireless multicast advantage offers less benefit, unless path diversity or cooperative diversity schemes are implemented [19, 20]. The exploitation of sleep modes [21, 22] is imperative to prevent sensor nodes from wasting energy in receiving packets unintended for them. Combined with efficient medium access protocols, the “sleep- ing” approach could reach optimal energy efficiency without degradation in throughput (but at some penalty in delay). 1.3.4 Capacity/Throughput Two parameters describe the network’s capability to carry traffic: transport capacity and throughput. The former is a distance-weighted sum capacity that permits evaluation of network performance. Throughput is a traditional measure of how much traffic can be delivered by the network [23–30]. In a 7037_C001.fm Page 6 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 22. Opportunities and Challenges in Wireless Sensor Networks 1-7 packet network, the (network-layer) throughput may be defined as the expected number of successful packet transmissions of a given node per timeslot. The capacity of wireless networks in general is an active area of research in the information theory community. The results obtained mostly take the form of scaling laws or “order-of” results; the prefactors are difficult to determine analytically. Important results include the scaling law for point-to-point coding, which shows that the throughput decreases with for a network with N nodes [23]. Newer results [28] permit network coding, which yields a slightly more optimistic scaling behavior, although at high complexity. Grossglauser and Tse [26] have shown that mobility may keep the per-node capacity constant as the network grows, but that benefit comes at the cost of unbounded delay. The throughput is related to (error-free) transmission rate of each transmitter, which, in turn, is upper bounded by the channel capacity. From the pure information theoretic point of view, the capacity is computed based on the ergodic channel assumption, i.e., the code words are long compared to the coherence time of the channel. This Shannon-type capacity is also called throughput capacity [31]. However, in practical networks, particularly with delay-constrained applications, this capacity cannot provide a helpful indication of the channel’s ability to transmit with a small probability of error. Moreover, in the multiple-access system, the corresponding power allocation strategies for maximum achievable capacity always favor the “good” channels, thus leading to unfairness among the nodes. Therefore, for delay-constrained applications, the channel is usually assumed to be nonergodic and the capacity is a random variable, instead of a constant in the classical definition by Shannon. For a delay- bound D, the channel is often assumed to be block fading with block length D, and a composite channel model is appropriate when specifying the capacity. Correspondingly, given the noise power, the channel state (a random variable in the case of fading channels), and power allocation, new definitions for delay- constrained systems have been proposed [32–35]. 1.3.5 Routing In ad hoc networks, routing protocols are expected to implement three main functions: determining and detecting network topology changes (e.g., breakdown of nodes and link failures); maintaining network connectivity; and calculating and finding proper routes. In sensor networks, up-to-date, less effort has been given to routing protocols, even though it is clear that ad hoc routing protocols (such as destination- sequenced distance vector (DSDV), temporally-ordered routing algorithm (TORA), dynamic source rout- ing (DSR), and ad hoc on-demand distance vector (AODV) [4, 36–39]) are not suited well for sensor networks since the main type of traffic in WSNs is “many to one” because all nodes typically report to a single base station or fusion center. Nonetheless, some merits of these protocols relate to the features of sensor networks, like multihop communication and QoS routing [39]. Routing may be associated with data compression [15] to enhance the scalability of the network. 1.3.6 Channel Access and Scheduling In WSNs, scheduling must be studied at two levels: the system level and the node level. At the node level, a scheduler determines which flow among all multiplexing flows will be eligible to transmit next (the same concept as in traditional wired scheduling); at the system level, a scheme determines which nodes will be transmitting. System-level scheduling is essentially a medium access (MAC) problem, with the goal of minimum collisions and maximum spatial reuse — a topic receiving great attention from the research community because it is tightly coupled with energy efficiency and throughput. Most of the current wireless scheduling algorithms aim at improved fairness, delay, robustness (with respect to network topology changes) and energy efficiency [62, 64, 65, 66]. Some also propose a distrib- uted implementation, in contrast to the centralized implementation in wired or cellular networks, which originated from general fair queuing. Also, wireless (or sensor) counterparts of other wired scheduling classes, like priority scheduling [67, 68] and earliest deadline first (EDF) [69], confirm that prioritization is necessary to achieve delay balancing and energy balancing. 1/ N 7037_C001.fm Page 7 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 23. 1-8 Smart Dust The main problem in WSNs is that all the sensor data must be forwarded to a base station via multihop routing. Consequently, the traffic pattern is highly nonuniform, putting a high burden on the sensor nodes close to the base station (the critical nodes in Figure 1.1). The scheduling algorithm and routing protocols must aim at energy and delay balancing, ensuring that packets originating close and far away from the base station experience a comparable delay, and that the critical nodes do not die prematurely due to the heavy relay traffic [40]. At this point, due to the complexity of scheduling algorithms and the wireless environment, most performance measures are given through simulation rather than analytically. Moreover, medium access and scheduling are usually considered separately. When discussing scheduling, the system is assumed to have a single user; whereas in the MAC layer, all flows multiplexing at the node are treated in the same way, i.e., a default FIFO buffer is assumed to schedule flows. It is necessary to consider them jointly to optimize performance figures such as delay, throughput, and packet loss probability. Because of the bursty nature of the network traffic, random access methods are commonly employed in WSNs, with or without carrier sense mechanisms. For illustrative purposes, consider the simplest sensible MAC scheme possible: all nodes are transmitting packets independently in every timeslot with the same transmit probability p at equal transmitting power levels; the next-hop receiver of every packet is one of its neighbors. The packets are of equal length and fit into one timeslot. This MAC scheme was considered in Silvester and Kleinrock [41], Hu [42], and Haenggi [43]. The resulting (per-node) through- put turns out to be a polynomial in p of order N, where N is the number of nodes in the network. A typical throughput polynomial is shown in Figure 1.2. At p = 0, the derivative is 1, indicating that, for small p, the throughput equals p. This is intuitive because there are few collisions for small p and the throughput g(p) is approximately linear. The region in which the packet loss probability is less than 10% can be denoted as the collisionless region. It ranges from 0 to about pmax/8. The next region, up to pmax, is the practical region in which energy consumption (transmission attempts) is traded off against through- put; it is therefore called the trade-off region. The difference p – g(p) is the interference loss. For small networks, all N nodes interfere with each other because spatial reuse is not possible: If more than one node is transmitting, a collision occurs and all packets are lost. Thus, the (per-node) throughput is p(1 – p)N–1, and the optimum transmit probability is 1/N. The maximum throughput is (1 – 1/N)N–1/N. With increasing N, the throughput approaches 1/(eN), as pointed out in Silvester and Kleinrock [41] and LaMaire et al. [44]. Therefore the difference pmax – 1/N is the spatial reuse gain (see Figure 1.2). This simple example illustrates the concepts of collisions, energy-throughput trade-offs, and spatial reuse, which are present in every MAC scheme. 1.3.7 Modeling The bases for analysis and simulations and analytical approaches are accurate and tractable models. Comprehensive network models should include the number of nodes and their relative distribution; their degree and type of mobility; the characteristics of the wireless link; the volume of traffic injected by the sources and the lifespan of their interaction; and detailed energy consumption models. 1.3.7.1 Wireless Link An attenuation proportional to dα, where d is the distance between two nodes and α is the so-called path loss exponent, is widely accepted as a model for path loss. Alpha ranges from 2 to 4 or even 5 [45], depending on the channel characteristics (environment, antenna position, frequency). This path loss model, together with the fact that packets are successfully transmitted if the signal-to-noise-and-inter- ference ratio (SNIR) is bigger than some threshold [8], results in a deterministic model often used for analysis of multihop packet networks [23, 26, 41, 42, 46–48]. Thus, the radius for a successful transmission has a deterministic value, irrespective of the condition of the wireless channel. If only interferers within a certain distance of the receiver are considered, this “physical model” [23] turns into a “disk model.” The stochastic nature of the fading channel and thus the fact that the SINR is a random variable are mostly neglected. However, the volatility of the channel cannot be ignored in wireless networks [5, 8]; 7037_C001.fm Page 8 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 24. Opportunities and Challenges in Wireless Sensor Networks 1-9 Sousa and Silvester have also pointed out the inaccuracy of disk models [49] and it is easily demonstrated experimentally [50, 51]. In addition, this “prevalent all-or-nothing model” [52] leads to the assumption that a transmission over a multihop path fails completely or is 100% successful, ignoring the fact that end-to-end packet loss probabilities increase with the number of hops. Although fading has been con- sidered in the context of packet networks [53, 54], its impact on the throughput of multihop networks and protocols at the MAC and higher layers is largely an open problem. A more accurate channel model will have an impact on most of the metrics listed in Section 1.3.1. In the case of Rayleigh fading, first results show that the energy benefits of routing over many short hops may vanish completely, in particular if latency is taken into account [20, 55, 56]. The Rayleigh fading model not only is more accurate than the disk model, but also has the additional advantage of permitting separation of noise effects and interference effects due to the exponential distribution of the received power. As a consequence, the performance analysis can conveniently be split into the analysis of a zero- interference (noise-analysis) and a zero-noise (interference-analysis) network. 1.3.7.2 Energy Consumption To model energy consumption, four basic different states of a node can be identified: transmission, reception, listening, and sleeping. They consist of the following tasks: • Acquisition: sensing, A/D conversion, preprocessing, and perhaps storing • Transmission: processing for address determination, packetization, encoding, framing, and maybe queuing; supply for the baseband and RF circuitry (The nonlinearity of the power amplifier must be taken into account because the power consumption is most likely not proportional to the transmit power [56].) • Reception: Low-noise amplifier, downconverter oscillator, filtering, detection, decoding, error detection, and address check; reception even if a node is not the intended receiver • Listening: Similar to reception except that the signal processing chain stops at the detection • Sleeping: Power supply to stay alive Reception and transmission comprise all the processing required for physical communication and net- working protocols. For the physical layer, the energy consumption depends mostly on the circuitry, the error correction schemes, and the implementation of the receiver [57]. At the higher layers, the choice FIGURE 1.2 Generic throughput polynomial for a simple random MAC scheme. spatial reuse gain pmax –1/N maximum throughput gmax trade-off region p [pmax /8,pmax] interference loss pmax –gmax g ( p ) 1 0 0.5 pmax p 1 N Transmit probability p t u p h g u o r h T g ∍ ∍ collisionless region p [0,pmax /8] 7037_C001.fm Page 9 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 25. 1-10 Smart Dust of protocols (e.g., routing, ARQ schemes, size of packet headers, number of beacons and other infra- structure packets) determines the energy efficiency. 1.3.7.3 Node Distribution and Mobility Regular grids (square, triangle, hexagon) and uniformly random distributions are widely used analytically tractable models. The latter can be problematic because nodes can be arbitrarily close, leading to unre- alistic received power levels if the path attenuation is assumed to be proportional to dα. Regular grids overlaid with Gaussian variations in the positions may be more accurate. Generic mobility models for WSNs are difficult to define because they are highly application specific, so this issue must be studied on a case-by-case basis. 1.3.7.4 Traffic Often, simulation work is based on constant bitrate traffic for convenience, but this is most probably not the typical traffic class. Models for bursty many-to-one traffic are needed, but they certainly depend strongly on the application. 1.3.8 Connectivity Network connectivity is an important issue because it is crucial for most applications that the network is not partitioned into disjoint parts. If the nodes’ positions are modeled as a Poisson point process in two dimensions (which, for all practical purposes, corresponds to a uniformly random distribution), the problem of connectivity has been studied using the tool of continuum percolation theory [58, 59]. For large networks, the phenomenon of a sharp phase transition can be observed: the probability that the network percolates jumps abruptly from almost 0 to almost 1 as soon as the density of the network is bigger than some critical value. Most such results are based on the geometric disk abstraction. It is conjectured, though, that other connectivity functions lead to better connectivity, i.e., the disk is appar- ently the hardest shape to connect [60]. A practical consequence of this conjecture is that fading results in improved connectivity. Recent work [61] also discusses the impact of interference. The simplifying assumptions necessary to achieve these results leave many open problems. 1.3.9 Quality of Service Quality of service refers to the capability of a network to deliver data reliably and timely. A high quantity of service, i.e., throughput or transport capacity, is generally not sufficient to satisfy an application’s delay requirements. Consequently, the speed of propagation of information may be as crucial as the throughput. Accordingly, in addition to network capacity, an important issue in many WSNs is that of quality-of- service (QoS) guarantees. Previous QoS-related work in wireless networks mostly focused on delay (see, for example, Lu et al. [62], Ju and Li [63], and Liu et al. [64]). QoS, in a broader sense, consists of the triple (R, Pe, D), where R denotes throughput; Pe denotes reliability as measured by, for example, bit error probability or packet loss probability; and D denotes delay. For a given R, the reliability of a connection as a function of the delay will follow the general curve shown in Figure 1.3. FIGURE 1.3 Reliability as a function of the delay. The circles indicate the QoS requirements of different possible traffic classes. reliability delay 3 100% 1 2 7037_C001.fm Page 10 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 26. Opportunities and Challenges in Wireless Sensor Networks 1-11 Note that capacity is only one point on the reliability-delay curve and therefore not always a relevant performance measure. For example, in certain sensing and control applications, the value of information quickly degrades as the latency increases. Because QoS is affected by design choices at the physical, medium-access, and network layers, an integrated approach to managing QoS is necessary. 1.3.10 Security Depending on the application, security can be critical. The network should enable intrusion detection and tolerance as well as robust operation in the case of failure because, often, the sensor nodes are not protected against physical mishandling or attacks. Eavesdropping, jamming, and listen-and-retransmit attacks can hamper or prevent the operation; therefore, access control, message integrity, and confiden- tiality must be guaranteed. 1.3.11 Implementation Companies such as Crossbow, Ember, Sensoria, and Millenial are building small sensor nodes with wireless capabilities. However, a per-node cost of $100 to $200 (not including sophisticated sensors) is prohibitive for large networks. Nodes must become an order of magnitude cheaper in order to render applications with a large number of nodes affordable. With the current pace of progress in VLSI and MEMS technology, this is bound to happen in the next few years. The fusion of MEMS and electronics onto a single chip, however, still poses difficulties. Miniaturization will make steady progress, except for two crucial components: the antenna and the battery, where it will be very challenging to find innovative solutions. Furthermore, the impact of the hardware on optimum protocol design is largely an open topic. The characteristics of the power amplifier, for example, greatly influence the energy efficiency of routing algorithms [56]. 1.3.12 Other Issues • Distributed signal processing. Most tasks require the combined effort of multiple network nodes, which requires protocols that provide coordination, efficient local exchange of information, and, possibly, hierarchical operation. • Synchronization and localization. The notion of time is critical. Coordinated sensing and actuating in the physical world require a sense of global time that must be paired with relative or absolute knowledge of nodes’ locations. • Wireless reprogramming. A deployed WSN may need to be reprogrammed or updated. So far, no networking protocols are available to carry out such a task reliably in a multihop network. The main difficulty is the acknowledgment of packets in such a joint multihop/multicast communication. 1.4 Concluding Remarks Wireless sensor networks have numerous exciting applications in virtually all fields of science and engineering, including health care, industry, military, security, environmental science, geology, agricul- ture, and social studies. In particular, the combination with macroscopic or MEMS-based actuators is intriguing because it permits manipulation of the environment in an unprecedented manner. Researchers and operators currently face a number of critical issues that need be resolved before these applications become reality. Wireless networking and distributed data processing of embedded sensing/actuating nodes under tight energy constraints demand new approaches to protocol design and hardware/software integration. 7037_C001.fm Page 11 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 27. 1-12 Smart Dust References 1. Sensor networks and applications, IEEE Proc., 8, Aug. 2003. 2. Internet Engineering Task Force, Mobile ad-hoc networks (MANET). See http://www.ietf.org/ html.charters/manet-charter.html. 3. Z.J. Haas et al., Eds., Wireless ad hoc networks, IEEE J. Selected Areas Commun., 17, Aug. 1999. Special ed. 4. C.E. Perkins, Ed., Ad Hoc Networking. Addison Wesley, Reading, MA, 2000. 5. A.J. Goldsmith and S.B.Wicker, Design challenges for energy-constrained ad hoc wireless networks, IEEE Wireless Commun., 9, 8–27, Aug. 2002. 6. A. Ephremides, Energy concerns in wireless networks, IEEE Mag. Wireless Commun., 9, 48–59, Aug. 2002. 7. V. Rodoplu and T.H. Meng, Minimum energy mobile wireless networks, IEEE J. Selected Areas Commun., 17(8), 1333–1344, 1999. 8. A. Ephremides, Energy concerns in wireless networks, IEEE Wireless Commun., 9, 48–59, Aug. 2002. 9. Bluetooth wireless technology. Official Bluetooth site: http://www.bluetooth.com. 10. S. Tilak, N.B. Abu–Ghazaleh, and W. Heinzelman, A taxonomy of wireless micro-sensor network models, ACM Mobile Computing Commun. Rev., 6(2), 28–36, 2002. 11. W.B. Heinzelman, A.P. Chandrakasan, and H. Balakrishnan, An application-specific protocol archi- tecture for wireless microsensor networks, IEEE Trans. Wireless Commun., 1, 660–670, Oct. 2002. 12. J. Kulik, W. Heinzelman, and H. Balakrishnan, Negotiation-based protocols for disseminating information in wireless sensor networks, Wireless Networks, 8, 169–185, March–May 2002. 13. A.B. McDonald and T.F. Znati, A mobility-based framework for adaptive clustering in wireless ad- hoc networks, IEEE J. Selected Areas Commun., 17, 1466–1487, Aug. 1999. 14. S.S. Pradhan, J. Kusuma, and K. Ramchandran, Distributed compression in a dense microsensor network, IEEE Signal Process. Mag., 19, 51–60, Mar. 2002. 15. A. Scaglione and S. Servetto, On the interdependence of routing and data compression in multi- hop sensor networks, in Proc. ACM Int. Conf. Mobile Comp. Networks (MobiCom’02), Atlanta, GA, 140–147, Sept. 2002. 16. C. Intanagowiwat, R. Govindan, and D. Estrin, Directed diffusion: a scalable and robust commun- ication paradigm for sensor networks, in ACM Int. Conf. Mobile Computing Networking (Mobi- Com’00), Boston, MA, 56–67, Aug. 2000. 17. J.E. Wieselthier, G.D. Nguyen, and A. Ephremides, On the construction of energy-efficient broadcast and multicast trees in wireless networks, in IEEE INFOCOM, Tel Aviv, Israel, 585–594, Mar. 2000. 18. J.E. Wieselthier, G.D. Nguyen, and A. Ephremides, An insensitivity property of energy-limited wireless networks for session-based multicasting, in IEEE ISIT, Washington, D.C., June 2001. 19. J. Laneman, D. Tse, and G. Wornell, Cooperative diversity in wireless networks: efficient protocols and outage behavior, IEEE Trans. Inf. Theory. Accepted for publication. Available at: http:// www.nd.edu/jnl/pubs/it2002.pdf. 20. M. Haenggi, A formalism for the analysis and design of time and path diversity schemes in wireless sensor networks, in 2nd Int. Workshop Inf. Process. Sensor Networks (IPSN’03), Palo Alto, CA, 417–431, Apr. 2003. Available at http://www.nd.edu/mhaenggi/ipsn03.pdf. 21. C.S. Raghavendra and S. Singh, PAMAS — power aware multi-access protocol with signaling for ad hoc networks, 1999. ACM Computer Commun. Rev. Available at: http://citeseer.nj.nec.com/ 460902.html. 22. C.-K. Toh, Maximum battery life routing to support ubiquitous mobile computing in wireless ad hoc networks, IEEE Commun. Mag., 39, 138–147, June 2001. 23. P. Gupta and P.R. Kumar, The capacity of wireless networks, IEEE Trans. Inf. Theory, 46, 388–404, Mar. 2000. 24. P. Gupta and P.R. Kumar, Towards an information theory of large networks: an achievable rate region, in IEEE Int. Symp. Inf. Theory, Washington, D.C., 159, 2001. 7037_C001.fm Page 12 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 28. Opportunities and Challenges in Wireless Sensor Networks 1-13 25. L.-L. Xie and P.R. Kumar, A network information theory for wireless communication: scaling laws and optimal operation, Apr. 2002. submitted to IEEE Trans. Inf. Theory. Available at: http:// black1.csl.uiuc.edu/prkumar/publications.html. 26. M. Grossglauser and D. Tse, Mobility increases the capacity of ad-hoc wireless networks, in IEEE INFOCOM, Anchorage, AL, 2001. 27. D. Tse and S. Hanly, Effective bandwidths in wireless networks with multiuser receivers, in IEEE INFOCOM, 35–42, 1998. 28. M. Gastpar and M. Vetterli, On the capacity of wireless networks: the relay case, in IEEE INFO- COM, New York, 2002. 29. G. Mergen and L. Tong, On the capacity of regular wireless networks with transceiver multipacket communication, in IEEE Int. Symp. Inf. Theory, Lausanne, Switzerland, 350, 2002. 30. S. Toumpis and A. Goldsmith, Capacity regions for wireless ad hoc networks, IEEE Trans. Wireless Commun., 2, 736–748, July 2003. 31. D.N.C. Tse and S.V. Hanly, Multiaccess fading channels — part I: polymatroid structure, optimal resource allocation and throughput capacities, IEEE Trans. Inf. Theory, 44(7), 2796–2815, 1998. 32. S.V. Hanly and D.N.C. Tse, Multiaccess fading channels — part II: delay-limited capacities, IEEE Trans. Inf. Theory, 44(7), 2816–2831, 1998. 33. R. Negi and J.M. Cioffi, Delay-constrained capacity with causal feedback, IEEE Trans. Inf. Theory, 48, 2478–2494, Sept. 2002. 34. R.A. Berry and R.G. Gallager, Communication over fading channels with delay constraints, IEEE Trans. Inf. Theory, 48, 1135–1149, May 2002. 35. D. Tuninetti, On multiple-access block-fading channels, Mar. 2002. Ph.D. thesis, Institut EURE- COM. Available at: http://www.eurecom.fr/tuninett/publication.html. 36. J. Broch, D. Maltz, D. Johnson, Y. Hu, and J. Jetcheva, A performance comparison of multi-hop wireless ad hoc network routing protocols, in ACM Int. Conf. Mobile Computing Networking (MobiCom), Dallas, TX, 85–97, Oct. 1998. 37. P. Johansson, T. Larsson, and N. Hedman, Scenario-based performance analysis of routing proto- cols for mobile ad-hoc networks, in ACM MobiCom, Seattle, WA, Aug. 1999. 38. S.R. Das, C.E. Perkins, and E.M. Royer, Performance comparison of two on-demand routing protocols for ad hoc networks, in IEEE INFOCOM, Mar. 2000. 39. C.R. Lin and J.-S. Liu, QoS Routing in ad hoc wireless networks, IEEE J. Selected Areas Commun., 17, 1426–1438, Aug. 1999. 40. M. Haenggi, Energy-balancing strategies for wireless sensor networks, in IEEE Int. Symp. Circuits Syst. (ISCAS’03), Bangkok, Thailand, May 2003. Available at http://www.nd.edu/mhaenggi/ iscas03.pdf. 41. J.A. Silvester and L. Kleinrock, On the capacity of multihop slotted ALOHA networks with regular structure, IEEE Trans. Commun., COM-31, 974–982, Aug. 1983. 42. L. Hu, Topology control for multihop packet networks, IEEE Trans. Commun., 41(10), 1474–1481, 1993. 43. M. Haenggi, Probabilistic analysis of a simple MAC scheme for ad hoc wireless networks, in IEEE CAS Workshop on Wireless Communications and Networking, Pasadena, CA, Sept. 2002. 44. R.O. LaMaire, A. Krishna, and H. Ahmadi, Analysis of a wireless MAC protocol with client–server traffic and capture, IEEE J. Selected Areas Commun., 12(8), 1299–1313, 1994. 45. T.S. Rappaport, Wireless Communications — Principles and Practice, 2nd ed. Prentice Hall, Englewood Cliffs, NJ. 46. H. Takagi and L. Kleinrock, Optimal transmission ranges for randomly distributed packet radio terminals, IEEE Trans. Commun., COM-32, 246–257, Mar. 1984. 47. J.L. Wang and J.A. Silvester, Maximum number of independent paths and radio connectivity, IEEE Trans. Commun., 41, 1482–1493, Oct. 1993. 48. C. Schurgers, V. Tsiatsis, S. Ganeriwal, and M. Srivastava, Optimizing sensor networks in the energy–latency–density design space, IEEE Trans. Mobile Computing, 1(1), 70–80, 2002. 7037_C001.fm Page 13 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 29. 1-14 Smart Dust 49. E.S. Sousa and J.A. Silvester, Optimum transmission ranges in a direct-sequence spread-spectrum multihop packet radio network, IEEE J. Selected Areas Commun., 8, 762–771, June 1990. 50. D.A. Maltz, J. Broch, and D.B. Johnson, Lessons from a full-scale multihop wireless ad hoc network testbed, IEEE Personal Commun., 8, 8–15, Feb. 2001. 51. D. Ganesan, B. Krishnamachari, A. Woo, D. Culler, D. Estrin, and S. Wicker, An empirical study of epidemic algorithms in large scale multihop wireless networks, 2002. Intel Research Report IRB- TR-02-003. Available at www.intel-research.net/Publications/Berkeley/05022002170319.pdf. 52. T.J. Shepard, A channel access scheme for large dense packet radio networks, in ACM SIGCOMM, Stanford, CA, Aug. 1996. Available at: http://www.acm.org/sigcomm/sigcomm96/papers/shep- ard.ps. 53. M. Zorzi and S. Pupolin, Optimum transmission ranges in multihop packet radio networks in the presence of fading, IEEE Trans. Commun., 43, 2201–2205, July 1995. 54. Y.Y. Kim and S. Li, Modeling multipath fading channel dynamics for packet data performance analysis, Wireless Networks, 6, 481–492, 2000. 55. M. Haenggi, On routing in random rayleigh fading networks, IEEE Trans. Wireless Commun., 2003. Submitted for publication. Available at http://www.nd.edu/mhaenggi/routing.pdf. 56. M. Haenggi, The impact of power amplifier characteristics on routing in random wireless networks, in IEEE Global Commun. Conf. (GLOBECOM’03), San Francisco, CA, Dec. 2003. Available at http://www.nd.edu/mhaenggi/globecom03.pdf. 57. H. Meyr, M. Moenecleay, and S.A. Fechtel, Digital Communication Receivers: Synchronization, Channel Estimation, and Signal Processing. Wiley Interscience, 1998. 58. R. Meester and R. Roy, Continuum Percolation. Cambridge University Press, New York, 1996. 59. B. Bollobás, Random Graphs, 2nd ed. Cambridge University Press, New York, 2001. 60. L. Booth, J. Bruck, M. Cook, and M. Franceschetti, Ad hoc wireless networks with noisy links, in IEEE Int. Symp. Inf. Theory, Yokohama, Japan, 2003. 61. O. Dousse, F. Baccelli, and P. Thiran, Impact of interferences on connectivity in ad-hoc networks, in IEEE INFOCOM, San Francisco, CA, 2003. 62. S. Lu, V. Bharghavan, and R. Srikant, Fair scheduling in wireless packet networks, IEEE/ACM Trans. Networking, 7, 473–489, Aug. 1999. 63. J.-H. Ju and V.O.K. Li, TDMA scheduling design of multihop packet radio networks based on Latin squares, IEEE J. Selected Areas Commun., 1345–1352, Aug. 1999. 64. H. Luo, S. Lu, and V. Bharghavan, A new model for packet scheduling in multihop wireless networks, in ACM Int. Conf. Mobile Computing Networking (MobiCom’00), Boston, MA, 76–86, 2000. 65. H. Luo, P. Medvedev, J. Cheng, and S. Lu, A self-coordinating approach to distributed fair queueing in ad hoc wireless networks, IEEE INFOCOM, Anchorage, Apr. 2001. 66. A.E. Gamal, C. Nair, B. Prabhakar, E. Uysal-Biyikoglu, and S. Zahedi, Energy-efficient scheduling of packet transmissions over wireless networks, IEEE INFOCOM, New York, 2002, pp. 1773–1782. 67. S. Bhatnagar, B. Deb, and B. Nath, Service differentiation in sensor networks, Fourth International Symposium on Wireless Personal Multimedia Communications, Sept. 2001. 68. V. Kanodia, C. Li, A. Sabharwal, B. Sadeghi, and E. Knightly, Distributed multi-hop scheduling and medium access with delay and throughput constraints, ACM MobiCom, Rome, July 2001. 69. A. Striegel and G. Manimaran, Best-effort scheduling of (m, k)-firm real-time streams in multihop networks, Workshop of Parallel and Distributed Real-Time Systems (WPDRTS) at IPDPS 2000, Apr. 2000. 7037_C001.fm Page 14 Tuesday, November 1, 2005 12:46 PM © 2006 by Taylor & Francis Group, LLC
- 30. 2-1 2 Next-Generation Technologies to Enable Sensor Networks* 2.1 Introduction ...................................................................... 2-1 Geolocation and Identification of Mobile Targets • Long-Term Architecture 2.2 Goals for Real-Time Distributed Network Computing for Sensor Data Fusion..................................................... 2-5 2.3 The Convergence of Networking and Real-Time Computing......................................................................... 2-6 Guaranteeing Network Resources • Guaranteeing Storage Buffer Resources • Guaranteeing Computational Resources 2.4 Middleware ...................................................................... 2-11 Control and Command of System • Parallel Processing 2.5 Network Resource Management .................................... 2-11 Graph Generator • Metrics Object • Graph Search • NRM Agents • Sensor Interface • Mapping Database • Topology Database • NRM Federation • NRM Fault Tolerance 2.6 Experimental Results....................................................... 2-16 2.1 Introduction Several important technical advances make extracting more information from intelligence, surveillance, and reconnaissance (ISR) sensors very affordable and practical. As shown in Figure 2.1, for the radar application the most significant advancement is expected to come from employing collaborative and network centric sensor netting. One important application of this capability is to achieve ultrawideband multifrequency and multiaspect imaging by fusing the data from multiple sensors. In some cases, it is highly desirable to exploit multimodalities, in addition to multifrequency and multiaspect imaging. Key enablers to fuse data from disparate sensors are the advent of high-speed fiber and wireless networks and the leveraging of distributed computing. ISR sensors need to perform enough on-board computation to match the available bandwidth; however, after some initial preprocessing, the data will be distributed across the network to be fused with other sensor data so as to maximize the information content. For example, on an experimental basis, MIT Lincoln Laboratory has demonstrated a virtual radar with ultrawideband frequency [1]. Two radars, located at the Lincoln Space Surveillance Complex *This work is sponsored by the United States Air Force under Air Force contract F19628-00-C-002. Opinions, interpretations, conclusions, and recommendations are those of the authors and are not necessarily endorsed by the U.S. government. Joel I. Goodman MIT Lincoln Laboratory Albert I. Reuther MIT Lincoln Laboratory David R. Martinez MIT Lincoln Laboratory 7037_C002.fm Page 1 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 31. 2-2 Smart Dust in Westford, Massachusetts, were employed; each of the two independent radars transmitted the data via a high-speed fiber network. The total bandwidth transmitted via fiber exceeded 1 Gbits/sec (billion bits per second). One radar was operating at X-band with 1-MHz bandwidth, and the second was operating at Ku-band with a 2-MHz bandwidth. A synthetic radar with an instantaneous bandwidth of 8 MHz was achieved after employing advanced ultrawideband signal processing [2]. These capabilities are now being extended to include high-speed wireless and fiber networking with distributed computing. As the Internet protocol (IP) technologies continue to advance in the commercial sector, the military can begin to leverage IP formatted sensor data to be compatible with commercial high- speed routers and switches. Sensor data from theater can be posted to high-speed networks, wireless and fiber, to request computing services as they become available on this network. The sensor data are processed in a distributed fashion across the network, thereby providing a larger pool of resources in real time to meet stringent latency requirements. The availability of distributed processing in a grid-computing architecture offers a high degree of robustness throughout the network. One important application to benefit from these advances is the ability to geolocate and identify mobile targets accurately from multiaspect sensor data. 2.1.1 Geolocation and Identification of Mobile Targets Accurately geolocating and identifying mobile targets depends on the extraction of information from different sensor data. Typically, data from a single sensor are not sufficient to achieve a high probability of correct classification and still maintain a low probability of false alarm.This goal is challenging because mobile targets typically move at a wide range of speeds, tend to move and stop often, and can be easily mistaken for a civilian target. While the target is moving the sensor of choice is the ground moving target indication (GMTI). If the target stops, the same sensor or a different sensor working cooperatively must employ synthetic aperture radar (SAR). Before it can be declared foe, the target must often be confirmed with electro-optical or infrared (EO/IR) images. The goal of future networked systems is to have multiple sensors providing the necessary multimodality data to maximize the chances of accurately declaring a target. FIGURE 2.1 Radar technology evolution. Front End Back End Advanced Algorithms Space-time Adaptive Imaging Discrimination Digital Array Antennas Filters Power Devices Correlation Processing Pulse Compression Doppler Synthetic Aperture Radar AEGIS Patriot THAAD Ground-based Beale ~ 40s – 60s ~ 70s – 80s ~ 90s – 2000s >2000s E-2C F-15 Chain Home AWACS Future Collaborative/ Network Centric Ultra-Wideband Multifrequency Multiaspect Imaging 7037_C002.fm Page 2 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 32. Next-Generation Technologies to Enable Sensor Networks 2-3 A typical sensing sequence starts by a wide area surveillance platform, such as the Global Hawk unmanned aerial vehicle (UAV), covering several square kilometers until a target exceeds a detection threshold. The wide area surveillance will typically employ GMTI and SAR strip maps. Once a target has been detected, the on-board or off-board processing starts a track file to track the target carefully, using spot GMTI and spot SAR over a much smaller region than that initially covered when performing wide area surveillance. It is important to recognize that a sensor system is not merely tracking a single target; several target tracks can be going on in parallel. Therefore, future networked sensor architectures rely on sharing the information to maximize the available resources. To date, the most advanced capability demonstrated is based on passing target detections among several sensors using the Navy cooperative engagement capability (CEC) system. Multisensor tracks are formed from the detection inputs arriving at a central location.Although this capability has provided a significant advancement, not all the information available from multimodality sensors has been exploited. The limitation is with the communication and available distributed computing. Multimodality sensor data together with multiple look angles can substantially improve the probability of correct classification vs. false alarm density. In addition to multiple modalities and multiple looks on the target, it is also desirable to send complex (amplitude and phase) radar GMTI data and SAR images to permit the use of high- definition vector imaging (HDVI) [3]. This technique permits much higher resolution on the target by suppressing noise around it, thereby enhancing the target image at the expense of using complex video data and much higher computational rates. Another important tool to improve the probability of correct classification with minimal false alarm is high-range resolution (HRR) profiles. With this tool, the sensor bandwidth or, equivalently, the size of the resolution cell must be small resulting in a large data rate. However, it has been demonstrated that HRR can provide a significant improvement [4]. Therefore, next generation sensors depend on available communication pipes with enough bandwidth to share the individual sensor information effectively across the network. Once the data are posted on the network, the computational resources must exist to maintain low latencies from the time data become available to the time a target geoposition and identi- fication are derived. The next subsection discusses the long-term architecture to implement netting of multiple sensor data efficiently. 2.1.2 Long-Term Architecture In the future it will be desirable to minimize the infrastructure (foot print) forwardly deployed in the battlefield. It is most desirable to leverage high-speed satellite communication links to bring sensor data back to a combined air operations center (CAOC) established in the continental United States (CONUS). The technology enablers for the long-term architecture shown in Figure 2.2 are high-speed, IP-based wireless and fiber communication networks, together with distributed grid computing. The in-theater commander’s ability to task his organic resources to perform reconnaissance and surveillance of the opposing forces, and then to relay that information back to CONUS, allows significant reduction in the complexity, level, and cost of in-theater resources. Furthermore, this approach leverages the diverse analysis resources in CONUS, including highly trained personnel to support the rapid, accurate identification and localization of targets necessary to enable the time-critical engagement of surface mobile threats. Space, air, and surface sensors will be deployed quickly to the battlefield. As shown in Figure 2.3, the stage in the processing chain at which the sensor data are tapped off to be sent via the network will dictate the amount of data transferred. For example, in a few applications one needs to send the data directly out of the analog-to-digital converters (A/D) to exploit coherent data combining from multiple sensors. Most commonly, it is preferable to perform on-board signal preprocessing to minimize the amount of data transferred. However, one must still be able to preserve content in the transferred data that is required to exploit features in the data not available from processing a signal sensor end to end. For example, one might be interested in transmitting wide area surveillance (WAS) data from SAR with high resolution to be followed by multiaspect SAR processing (shown in Figure 2.3 as application B). The data volume will be larger than the second example shown in Figure 2.3 as application A, in which 7037_C002.fm Page 3 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 33. 2-4 Smart Dust FIGURE 2.2 Postulated long-term architecture. FIGURE 2.3 Sensor signal processing flow. Exploitation Cell CAOC–F/R HAE UAV Radar/Illuminator Exploitation Cell Exploitation Cell Archival Data/Info Archival Data/Info Command & Control Computing Resources Computing Resources Small UAV Bistatic Receiver Bistatic Receiver Bistatic Receiver Weapon Platforms UGS UGS UGS EO/IR MC2A UGS Radar/Illuminator a p 7037_C002.fm Page 4 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 34. Next-Generation Technologies to Enable Sensor Networks 2-5 most of the GMTI processing is done on board. In any of these applications, it is paramount that “intelligent” data compression be done on board before data transmission to send only the necessary parts of the data requiring additional processing off board. Each sensor will be capable of generating on-board processed data greater than 100 Mbits/sec (million bits per second). Figure 2.4 shows the trade-off between communication link data rates vs. on-board computation throughputs for different postulated levels of image resolution (for spot or strip map SAR modes). For example, for an assumed 1-m strip map SAR, one can send complex video radar data to then perform super-resolution processing off board. This approach would require sending between 100 to 1000 Mbits/sec. Another option is to perform the super-resolution processing on board, requiring between 100 billion floating-point operations per second (GFLOPS) to 1 trillion floating-point operations per second (TFLOPS). Specialized military equipment, such as the common data link (CDL), can achieve data rates reaching 274 Mb/sec. If higher communication capacity were available, one would much prefer to send the large data volume for further processing off board to leverage information content available from multiple sensor data. As communication rates improve in the forthcoming years, it will not matter to the in- theater commander if the data are processed off board with the benefit of allowing exploitation of multiple sensor data at much rawer levels than is possible to date. 2.2 Goals for Real-Time Distributed Network Computing for Sensor Data Fusion Several advantages can be gained by utilizing real-time distributed network computing to enable greater sensor data fusion processing. Distributed network computing potentially reduces the cost of the signal processing systems and the sensor platform because each individual sensor platform no longer needs as much processing capability as a stove-piped stand-alone system (although each platform may need higher bandwidth communications capabilities). Also, fault tolerance of the processing systems is increased because the processing and network systems are shared between sensors, thereby increasing the pool of available signal processors for all of the sensors. Furthermore, the granularity of managed resources is smaller; individual processors and network resources are managed as independent entities rather than managing an entire parallel computer and network as independent entities. This affords more flexible configuration and management of the resources. To enable collaborative network processing of sensor signals, three technological areas are required to evolve and achieve maturity: • Guaranteed communication, storage buffer, and computation resources must keep up with the high-throughput streams of data coming from the sensors. If any stage of the processing falls FIGURE 2.4 SAR data rate and computational throughput trade. 7037_C002.fm Page 5 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 35. 2-6 Smart Dust behind due to a network problem or interruption in the processor, buffering the data will become a problem quickly as increasing volumes of data must be stored to accommodate the delayed processors. Section 2.3 addresses technological possibilities to mitigate these resource availability issues. • Middleware in the network of processors must be developed to accommodate a heterogeneous mix of computer and network resources. This middleware consists of a task control interface, which facilitates the communication between network resource management agents and entities, and an application programming interface for programming applications executed on the collab- orative network processors. Section 2.4 will address these middleware interfaces. • A network resource manager (NRM) system is necessary for orchestrating the execution of the application components on the computation and communication resources available in the col- laborative network. Section 2.5 will discuss the components and functionality of the NRM. 2.3 The Convergence of Networking and Real-Time Computing To date, networking of sensors has been demonstrated primarily using localized- and limited-capacity data links. As a result, the data available on the network from each sensor node typically represent the product of extensive prior processing of the radar data carried at the individual sensor. For example, the Navy CEC system, a relatively advanced current system, uses detection reports from independent sensors in the network to build composite tracks of targets. Access to raw (or possibly minimally preprocessed) multisensor data opens the opportunity for more effective exploitation of these data through integrated sensor data processing. The future network-centric ISR architecture will likely employ worldwide wide- band communication networks to interconnect sensors with distributed processing and fusion sites. The resulting distributed database will provide a common operational picture for deployed forces. The sensor data will return to a CONUS entry point and pass over a wideband fiber network to the various processing centers where the sensor data will be fused. The data link from the theater to CONUS is expected to be optical to achieve very high link capacity [5]. This section discusses technologies that will guarantee that wireless and terrestrial network resources, storage buffer resources, and computational resources are available for sensor signal processing. 2.3.1 Guaranteeing Network Resources Sensor data will traverse wireless and terrestrial (e.g., optical, twisted-copper) networks in which bit errors, packet loss, and delay could adversely affect the quality and timeliness of the ultimate result. The goal then is to choose a network and processing architecture to ameliorate the deleterious effects of data loss and network delay in the data fusion process. Due to the costs associated with developing, deploying, and maintaining a fixed terrestrial infrastructure, as well as inventing wholly new modulation protocols and standards for wireless and terrestrial signaling, it is cost-effective and expedient for military technology to ride the “commercial wave” of technical investment and progress in communication technologies. With a fixed network infrastructure consisting primarily of commercial components, combating data loss and delay in terrestrial networks involves choosing the right protocols so that the network can enforce quality of service (QoS) demands; in wireless networks, this involves aggressive coding, modulation, and “lightweight” flow control for efficient bandwidth utilization. With sufficient complexity and bandwidth, it is possible with today’s IP-based protocols to differentiate high-priority data to impart the mandated QoS for time-critical applications. 2.3.1.1 Terrestrial Networks Reserving bandwidth on an IP-based network that is uniformly recognized across administrative domains involves employing protocols like RSVP-TE [6] or CR-LDP [7]. Although having sufficient communica- tion bandwidth is an important aspect of processing sensor data in real time on a distributed network of resources, it does not guarantee real-time performance. For example, time-critical applications mapped 7037_C002.fm Page 6 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 36. Next-Generation Technologies to Enable Sensor Networks 2-7 onto networked resources should not have processing interrupted to service unmanaged traffic or be subject to a computational resource’s resident operating system switching contexts to a lower priority task. For data that originate from sensors at very high streaming rates, a storage solution, as discussed in Section 2.3.2, is needed that is capable of recording sensor data in real time as well as robust in the face of network resource failures; this insures that a high-priority application can continue processing in the presence of malfunctioning or compromised networked equipment. However, adding a buffering storage solution only alleviates part of the problem; it does not mitigate the underlying problem of losing packets during network equipment failures or periods of network traffic that exceed network capacities. For an IP-based network, one solution to this problem is to use remote agents deployed on primary compute resources or networked terminals located at switches that can dynamically filter unmanaged traffic. This is implemented by programming computer hardware specifically tasked with packet filtering (e.g., next generation gigabit Ethernet card) or dynamically reconfiguring the switch that directly connects to the compute resource in question by supplying an access control list (ACL) to block all packets except those associated with time-critical targeting. The formation of these exclusive networks using agents has been dubbed dynamic private networks (DPNs) — in effect, mechanisms for virtually overlaying a circuit switch onto a packet-switched network. 2.3.1.2 Wireless Networks Unlike terrestrial networks, flow control and routing in mobile wireless sensor networks must contend with potentially long point-to-point propagation delays (e.g., satellite to ground) as well as a constantly changing topology. In a traditional terrestrial network employing link-state routing (e.g., OSPF), each node maintains a consistent view of a (primarily) fixed network topology so that a shortest path algorithm [8] can be used to find desirable routes from source to destination. This requires that nodes gather network connectivity information from other routers. If OSPF were employed in a mobile wireless network, the overhead of exchanging network connectivity information about a transient topology could potentially consume the majority of the available bandwidth [9]. Routing protocols have been specifically designed to address the concerns of mobile networks [10]; these protocols fall into two general categories: proactive and reactive. Proactive routing protocols keep track of routes to all destinations, while reactive protocols acquire routes on demand. Unlike OSPF, proactive protocols do not need a consistent view of connectivity; that is, they trade optimal routes for feasible routes to reduce communication overhead. Reactive routes suffer a high initial overhead in establishing a route; however, the overall overhead of maintaining network connectivity is substantially reduced. The category of routing used is highly dependent upon how the sensors communicate with one another over the network. Traditional flow control mechanisms over terrestrial networks that deliver reliable transport (e.g., TCP) may be inappropriate for wireless networks because, unlike wireless networks, terrestrial networks gen- erally have a very low bit error rate (BER) on the order of 10–10, so errors are primarily due to packet loss. Packet loss occurs in heavily congested networks when an ingress or egress queue of a switch or router begins to fill, requiring that some packets in the queue be discarded [11]. This condition is detected when acknowledgments from the destination node are not received by the source, prompting the source’s flow control to throttle back the packet transmit rate [12]. In a wireless network in which BERs are four to five orders of magnitude higher than those of terrestrial networks, packet loss due to bit errors can be mistakenly associated with network congestion, and source flow control will mistakenly reduce the transmit rate of outgoing packets. Furthermore, when the source and destination are far apart, such as the communication between a satellite and ground terminal, where propagation delays can be on the order of 240 ms, delayed acknowledgments from the destination result in source flow control inefficiently using the available bandwidth. This is due to source flow control incrementally increasing the transmit rate as destination acknowledgments are received even though the entire frame of packets may have already been transmitted before the first packet reaches the receiver [13]. Therefore, to use bandwidth efficiently in a wireless network for reliable transport, flow control must be capable of differentiating BER from packet loss and account for long-haul packet transport by 7037_C002.fm Page 7 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 37. 2-8 Smart Dust more efficiently using the available bandwidth. Some work in this area is reflected in RFC 2488 [14], as well as proposals for an explicit congestion warning, where, for example, the destination site would respond to packet errors with an acknowledgment that it received the source packets with a corruption notification. At the physical layer, high data rates for a given BER have been realized by employing low-density parity check codes, such as turbo codes, in conjunction with bandwidth efficient modulation to achieve spectral efficiencies to within 0.7 dB of the Shannon limit [15]. Furthermore, extremely high spectral efficiencies have been demonstrated using multiple input, multiple output (MIMO) antenna systems whose theoretical channel capacity increases linearly with the number of transmit/receive antenna pairs [16].Although turbo codes are advantageous as a forward error correction mechanism in wireless systems when trying to maximize throughput, MIMO systems achieve high spectral efficiencies only when operating in rich scattering environments [17]. In environments in which little scattering occurs, such as in some air-to-air communication links, MIMO systems offer very little improvement in spectral efficiency. 2.3.2 Guaranteeing Storage Buffer Resources For a variety of reasons, it may be very desirable to record streaming sensor data directly to storage media while simultaneously sending the data on for immediate processing. For sensor signal processing appli- cations, this enables multimodality data fusion of archived data with real-time (perishable) data from in-theatre sensors for improved target identification and visualization [18]. Storage media could also be used for rate conversion in cases in which the transmission rate exceeds the processing rate and for time- delay buffering for real-time robust fault tolerance (discussed in the next section). The storage media buffer reuse is deterministic and periodic so that management of the buffer is straightforward. A number of possible solutions exist: • Directly attached storage is a set of hard disks connected to a computer via SCSI or IDE/EIDE/ ATA; however, this technology does not scale well to the volume of streaming sensor data. • Storage area networks are hard disk storage cabinets attached to a computer with a fast data link like Fibre Channel. The computer attached to the storage cabinet enjoys very fast access to data, but because the data must travel through that computer, which presents a single point of failure, to get to other computers on the network, this option is not a desirable solution. • Network-attached storage connects the hard disk storage cabinet directly to the network as a file server. However, this technology offers only midrange performance, a single point of failure, and relatively high cost. A visionary architecture in which data storage centers operate in parallel at a wide-area network (WAN) and local area network (LAN) level is described in Cooley et al. [19]. In this architecture, developed by MIT Lincoln Laboratory, high-rate streaming sensor data are stored in parallel across a partitioned network of storage arrays, which affords a highly scalable, low-cost solution that is relatively insensitive to communications or storage equipment failure. This system employs a novel and computationally efficient encoding and decoding algorithm using low-density parity check codes [20] for erasure recovery. Initial system performance measures indicate the erasure coding method described in Cooley et al. [19] has a significantly higher throughput and greater reliability when compared to Reed–Solomon, Tornado [21], and Luby [20] codes. This system offers a promising low-cost solution that scales in capability with the performance gains of commodity equipment. 2.3.3 Guaranteeing Computational Resources The exponential growth in computing technology has contributed to making viable the implementation of advanced sensor processing in cost-effective hardware with form factors commensurate with the needs of military users. For example, several generations of embedded signal processors are shown in Figure 2.5. 7037_C002.fm Page 8 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 38. Next-Generation Technologies to Enable Sensor Networks 2-9 In the early 1990s, embedded signal processors were built using custom hardware and software. In the late 1990s, a move occurred from custom hardware to COTS processor systems running vendor-specific software together with application-specific parallel software tuned to each specific application. Most recently, the military embedded community is beginning to demonstrate requisite performance employing parallel and portable software running on COTS hardware. Continuing technology advances in computation and communication will permit future signal pro- cessors to be built from commodity hardware distributed across a high-speed network and employing distributed, parallel, and portable software. These computing architectures will deliver 109 to 1012 floating point operations per second (GFLOPs to TFLOPs) in computational throughput. The distributed nature of the software will apply to on-board sensor processing as well as off-board processing. Clearly, on- board embedded processor systems will need to meet the stringent platform requirements in size, weight, and power. Wireless and terrestrial network resources are not the only areas in which delays, failures, and errors must be avoided to process sensor data in a timely fashion. The system design must also guarantee that the marshaled compute nodes will keep up with the required computational throughput of streaming data at every stage of the processing chain. This guarantee encompasses two important facets: (1) keeping the processors from being interrupted while they are processing tasks and (2) implementing fail-over that is tolerant of fault. 2.3.3.1 Avoiding Processor Interruption It is easy to take for granted that laptop and desktop computers will process commands as fast as the hardware and software are capable of doing so. A fact not generally known is that general computers are interrupted by system task processes and the processes of other applications (one’s own and possibly from others working in the background on one’s system). System task processes include keyboard and mouse input; communications on the Ethernet; system I/O; file system maintenance; log file entries; etc. When the computer interrupts an application to attend to such tasks, the execution of the application is temporarily suspended until the interrupting task has finished execution. However, because such inter- ruptions often only consume a few milliseconds of processing time, they are virtually imperceptible to the user [22]. Nevertheless, the interruptions are detrimental to the execution of real-time applications. Any delay in processing these streams of data will instigate a need for buffering the data that will grow to insur- mountable size as the delays escalate. A solution for these interrupt issues is to use a real-time operating system on the computation processors. FIGURE 2.5 Embedded signal processor evolution. 85 GFLOPS COTS Parallel SW Adaptive Processor Gen 1 (1992) 22 GOPS Custom (Parallel) SW Adaptive Processor Gen 2 (1998) AEGIS & Standard Missile Test Beds (2000+) PTCN Network Test Bed (2002+) VME Backplane Custom Boards RACE Crossbar Multi-chassis COTS 50+ GFLOPS Portable, Parallel SW (VSIPL, MPI, & PVL) High Speed LANs Network of Workstations GFLOPS to TFLOPS Parallel & Distributed SW (PVL & CORBA) High Speed LANs & WANs Networked Clusters, Servers Distributed Network 7037_C002.fm Page 9 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 39. 2-10 Smart Dust Simply put, real-time operating systems (RTOS) give priority to computational tasks. They usually do not offer as many operating system features (virtual memory, threaded processing, etc.) because of the interrupting processing nature of these features [22]. However, an RTOS can ensure that real-time critical tasks have guaranteed success in meeting streamed processing deadlines. An RTOS does not need to be run on typical embedded processors; it can also be deployed on Intel and AMD Pentium-class or Motorola G-series processor systems. This includes Beowulf clusters of standard desktop personal computers and commodity servers. This is an important benefit, providing a wide range of candidate heterogeneous computing resources. A great deal of press has been generated in the past several years about real-time operating systems; however, the distinction between soft real-time and hard real-time operating systems is seldom discussed. Hard real-time systems guarantee the completion of tasks in a deterministic time period, while soft real- time systems give priority to critical tasks over other tasks but do not guarantee the completion of tasks in a deterministic time period [22]. Examples of hard real-time operating systems are VxWorks (Wind River Systems, Inc. [23]); RTLinux/Pro (FSMLabs, Inc. [24]); and pSOS (Wind River Systems, Inc. [23]), as well as dedicated massively parallel embedded operating systems like MC/OS (Mercury Computer Systems, Inc. [25]). Examples of soft real-time operating systems are Microsoft Pocket PC; Palm OS; certain real-time Linux releases [24, 26]; and others. 2.3.3.2 Working through System Faults When fault tolerance in massively parallel computers is addressed, usually the solution is parallel redun- dant systems for fail-over. If a power supply or fan fails, another power supply or fan that is redundant in the system takes over the workload of the failed device. If a hard disk drive fails on a redundant array of independent disks (RAID) system, it can be hot swapped with a new drive and the contents of the drive rebuilt from the contents of the other drives along with checksum error correction code information. However, if an individual processor fails on a parallel computer, it is considered a failure of the entire parallel computer, and an identical backup computer is used as a fail-over. This backup system is then used as the primary computer, while the failed parallel computer is repaired to become the backup for the new primary eventually. If, however, it were possible to isolate the failed processor and remap and rebind the processes on other processors in that computer — in real time — it would then be possible to have only a number of redundant processors in the system rather than entire redundant parallel computers. There are two strategies for determining the remapping as well as two strategies for handling the remapping and rebinding; each has its advantages and disadvantages. To discuss these fail-over strategies, it is necessary to define the concepts of tasks and mappings. A signal processing application can be separated into a series of pipelined stages or tasks that are executed as part of the given application.A mapping is the task-parallel assignment of a task to a set of computer and network resources. In terms of determining the fail-over remapping, it is possible to choose a single remapping for each task or to choose a completely unique secondary path — a new mapping for each task that uses a set of processors mutually exclusive from the processors in the primary mapping path. If task backup mappings are chosen for each task, the fail-over will complete faster than a full processing chain fail-over; however, the rebinding fail-over for a failed task mapping is more difficult because the mappings from the task before and the task after the failed task mapping must be reconfigured to send data to and receive data from the new mapping. Conversely, if a completely unique secondary path is chosen as a fail-over, then fail-over completion will have a longer latency than performing a single task fail-over. However, the fail-over mechan- ics are simpler because the completely unique secondary path could be fully initialized and ready to receive the stream of data in the event of a failure in the primary mapping path. In terms of handling the remapping and rebinding of tasks, it is possible to choose the fail-over mappings when the application is initially launched or immediately after a fault occurs. In either case, greater latency is incurred at launch time or after the occurrence of a fault. For these advanced options, support for this fault tolerance comes mainly from the middleware support, which is discussed in the next section, and from the NRM discussed in Section 2.5. 7037_C002.fm Page 10 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 40. Next-Generation Technologies to Enable Sensor Networks 2-11 2.4 Middleware Middleware not only provides a standard interface for communications between network resources and sensors for plug-and-play operation, but also enables the rapid implementation of high-performance embedded signal processing. 2.4.1 Control and Command of System Because many systems use a diverse set of hardware, operating systems, programming languages, and communication protocols for processing sensor data, the manpower and time-to-deployment associated with integration have a significant cost. A middleware component providing a uniform interface that abstracts the lower-level system implementation details from the application interface is the common object request broker architecture (CORBA) [27]. CORBA is a specification and implementation that defines a standard interface between a client and server. CORBA leverages an interface definition language (IDL) that can be compiled and linked with an object’s implementation and its clients. Thus, the CORBA standard enables client and server communications that are independent of the host hardware platforms, programming language, operating systems, and so on. CORBA has specifications and implementations to interface with popular communication protocols such as TCP/IP. However, this architecture has an open specification, general interORB protocol (GIOP) that enables developers to define and plug in platform-specific communication protocols for unique hardware and software interfaces that meet appli- cation-specific performance criteria. For real-time and parallel embedded computing, it is necessary to interface with real-time operating systems, define end-to-end QoS parameters, and enact efficient data reorganization and queuing at communication interfaces. CORBA has recently included specifications for real-time performance and parallel processing, with the expectation that emerging implementations and specification addendums will produce efficient implementations. This will enable CORBA to move out of the command and control domain and be included as a middleware component involved in real-time and parallel processing of time-critical sensor data. 2.4.2 Parallel Processing The ability to choose one of many potential parallel configurations enables numerous applications to share the same set of resources with various performance requirements. What is needed is a method to decouple the mapping, that is, the parallel instantiation of an application on target hardware, from generic serial application development. Automating the mapping process is the only feasible way of exploring the large parameter space of parallel configurations in a timely and cost-effective manner. MIT Lincoln Laboratory has developed a C++-based library known as the parallel vector library (PVL) [28]. This library contains objects with parameterized methods deeply rooted in linear algebraic expres- sions commonly found in sensor signal processing. The parameters are used to direct the object instance to process data as one constituent part of a parallel whole. The parameters that organize objects in parallel configurations are run-time parameters so that new parallel configurations can be instantiated without having to recompile a suite of software. The technology of PVL is currently being incorporated into the parallel vector, signal, and image processing library for C++ (parallel VSIPL++) standard library [29]. 2.5 Network Resource Management Given the stated goals for distributed network computing for sensor fusion as outlined in Section 2.3, the associated network communication, storage, and processing challenges in Section 2.3, and the desire for standard interfaces and libraries to enable application parallelism and plug-and-play integration in Section 2.4, an integrated solution is needed that bridges network communications, distributed storage, distributed processing, and middleware. Clearly, it is possible for a development team to implement a 7037_C002.fm Page 11 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 41. 2-12 Smart Dust “point” solution, but this is inherently not scalable and very difficult to maintain. Therefore an additional goal is to fully automate the process of configuring network communication, storage, and computational resources to process data for sensor fusion applications in real time, provide robust fault tolerance in the face of network resource failures, and impart this service in a highly dynamic network in the face of competing interests. To address these needs, the network resource manager (NRM) was developed. The novelty and potency of the NRM is its capability of taking a sensor signal processing application designed and tested on single target processing element (PE) and mapping it in a task- and a data-parallel fashion across a network of computational resources to achieve real-time performance [30]. Figure 2.6 is an object-oriented model of the components that constitute the NRM. A high-level overview of the NRM follows, and details will be provided in the following subsections. The task of building a model from which the NRM launches parallel applications is broken into three distinct phases: 1. Map generation involves breaking an application into various task- and data-parallel components. 2. Map timing collects performance metric information associated with the components (or tasks) running on host resources. Using the performance metrics, the NRM creates a weighted graph- theoretic view of various permutations of an application mapped in parallel across networked resources. 3. Map selection finds the path through the graph that best meets system and application perfor- mance requirements. The graph generator and graph search objects will heavily leverage PVL (discussed earlier) objects in the instantiation of task- and data-parallel configurations of applications on host resources. It should be noted, however, that the NRM’s capabilities are fully general and independent from those of PVL and could work with other applications that are not developed using PVL to instantiate task- and data parallelism. 2.5.1 Graph Generator As noted previously, PVL uses run-time parameters to generate new parallel configurations. This enables the NRM to launch applications in arbitrary parallel configurations using software developed for a single target PE without having to recompile the application software suite. The central challenge is to select a subset of the potentially astronomical number of permutations of parallel configurations as candidate parallel mappings. It is expected that the NRM will receive guidance in the form of performance and resource utilization bounds to help it avoid choosing undesirable configurations. It will also be given a FIGURE 2.6 Object model for network resource manager (NRM). NRM Sensor Interface Graph Generator Mapping Database Topology Database Metrics Object NRM Agent Task App Instance Graph Search Sensor 7037_C002.fm Page 12 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 42. Next-Generation Technologies to Enable Sensor Networks 2-13 series of constituent tasks that comprise an application, so that its primary objective is to choose candidate data-parallel configurations for each of the individual tasks. Using a graph-theoretic model, the appli- cation space may be broken up as shown in Figure 2.7. Each column in the graph is populated with vertices; each vertex corresponds to a mapping of the task corresponding to the given column to a potentially unique set of computational resources in the system. Each vertex has edges entering and exiting: entering edges correspond to communications with preceding tasks and exiting edges correspond to communications with succeeding tasks. Sensor signal processing applications may be represented as a stream signal processing flow, in which data move in one direction from task to task as they are processed. In this graph-theoretic model, task parallelism is represented along the horizontal axis of the graph, i.e., pipelined, overlapping execution intervals, while data parallelism is represented by the mapping of each task in the application onto one or more parallel computational resources of each vertex. The graph-theoretic representation of data- and task-parallel applications and the corresponding flow of communication enable the graph generator of the NRM to capture the potentially astronomical number of combinations of application-to-resource mappings in a concise and efficient fashion. Finally, the graph generator is also responsible for launching the executable for each task mapping (vertex) on target resources so that performance metrics can be collected as discussed in the next subsection. 2.5.2 Metrics Object The metrics object (MO) is responsible for collecting performance metrics of tasks launched by the graph generator. The MO works closely with the graph generator to weight the graph. Each of the resources that hosts a task is time synchronized; metric agents (see NRM agents in Subsection 2.5.4) on each of the resources will provide the MO measurements for it to formulate the following performance param- eters associated with graph weights: throughput; latency; RAM memory; and PE utilization. The MO will calculate another metric known as processor cost, which is a ratio of compute horsepower used in the mapping to the overall processing horsepower available in the network. Link utilization percentages within each mapping are also measured, as well as intertask utilization percentages. Map generation uses task column pairs to gather performance metrics in order to reduce the effort and time involved drastically. This is possible because the graph search algorithm will use a running tabulation of resource utilization percentages to ensure that simple linear superposition of path weights hold, given that these percentages remain under a given threshold. This is explained further in the next subsection. Once above the threshold, weight modifiers will be applied to subsequent stages during search. Finally, the metrics object will calculate a network cost, analogous to processor cost, which FIGURE 2.7 Sample graph with edge and vertex weights. E = [e1,e2,..,em] V = [v1,v2,..,vm] TASK 1 (Stage 1) TASK 2 (Stage 2) TASK c–1 (Stage c–1) TASK c (Stage c) 7037_C002.fm Page 13 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 43. 2-14 Smart Dust is a ratio of communications bandwidth used by a mapping pair with respect to the overall bandwidth available in the network. 2.5.3 Graph Search The NRM must choose a path through the graph that determines the task mappings with which an application is launched on network resources. The choice of a path by the NRM is constrained by the time to result and the mandate to use a minimum set of networked resources. The data rate of the sensor data stream will drive required throughput for each task column in the graph; overall latency, which represents the total pipeline delay, is defined as the time period after which all data have been transmitted that a result is generated. To minimize any one application’s impact on resource consumption, the path through the graph could be chosen to minimize the overall usage of computational or communication resources. This choice will depend upon whether an application is launched in a network that is compute resource or communication bandwidth limited. The graph search problem may be formalized as a discrete and constrained optimization problem: given a set of hard constraints, minimize (or maximize) a given objective function. As described in the metrics object subsection, the NRM may choose constraints and an objective function from the set of weights shown in Table 2.1. Scalar weights are singular — that is, only one is associated with a given vertex or edge; vector weights may include many elements in an edge or vertex association. Because each vertex and edge may represent the combination of many PE and network communication elements associated with a mapping pair, processor and network utilization may constitute weight vectors with many elements. Although all weights tabulated previously may be chosen as constraints, memory, throughput, and network and PE utilization are not parameters that can be chosen as an objective function to optimize. This is because throughput is only a function of data rate; maximizing throughput has no impact on performance. Utilization also has no impact on performance and is only a measure of the validity of the solution. That is, subsequent stages in the graph may include resources from earlier stages, so keeping a running tabulation of utilization gives an indication of the onset of usage exceeding capacity and thereby degrading performance. Network utilization and cost, PE utilization and cost, and memory are weights derived and constrained by the NRM,while data rate (throughput) and latency are application dependent and imposed by the sensor. The objective function that the NRM uses is chosen based on the desire to minimize an application’s impact on resource usage or minimize the latency associated with an application’s execution. For example, in a bandwidth-limited network, the graph search problem may be formulated as follows. While meeting appli- cation latency and throughput constraints, using less than 80% of the bandwidth available in the chosen network conduits and PEs and less than 100% of the available local PE-RAM memory, and using only a fraction of the overall processing bandwidth available network wide, select a parallel configuration for the TABLE 2.1 Graph Weights Associated with Individual Edges and Vertices, and Corresponding Sizes (Types) Weight Type Latency Scalar Throughput Scalar PE utilization Vector Processor cost Scalar Network utilization Vector Network cost Scalar Memory Scalar 7037_C002.fm Page 14 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 44. Next-Generation Technologies to Enable Sensor Networks 2-15 application and the associated host resources using the smallest fraction of overall network bandwidth available. Even for moderately sized graphs (e.g., 1000 vertices by 10 stages), this is a complex combinatorial optimization problem; the general problem is NP complete. The authors have developed an iterative heuristic algorithm that has shown favorable performance for this class of problem in the quality of the solution and time to solution compared to other popular combinatorial optimization algorithms [31]. 2.5.4 NRM Agents The NRM agents are information and service links between the NRM and each of the resources. Agents must first register and be authenticated (e.g., using Kerberos [32]) before an NRM will invoke their services. This registration includes a characterization of the resource capabilities and services. When registered, the NRM will use these remotely deployed agents on computational resources to download and launch parameterized executables and modify the access control list (ACL) of switches and routers under its control in the formation of DPNs. Agents also provide a mechanism for centralized software maintenance and configuration by acting as transaction managers in the download and installation of applications, databases, middleware, etc. As stated earlier, the agents also provide a measurement object that is instantiated by applications to provide the NRM’s MO with performance metrics during graph generation. Finally, agents give the NRM a view of the network state, periodically sending diagnostic messages indicating its operational status. 2.5.5 Sensor Interface Sensors can be thought of as resources much like computational and communication resources, which are served by the NRM agents; thus, the sensor interface can be thought of as another type of NRM agent. Because many different sensor platforms could be served by an NRM-managed resource network, the sensor interface provides a common, abstract mechanism for communication between the NRM and the sensor platforms. Sensors will request services through the sensor interface from the NRM using a well-defined middleware interface such as CORBA. This request for services involves requesting the proper application for the data stream that the sensor will be delivering to the network of resources as well as a request for the required metric constraints, such as throughput and latency (discussed in Subsection 2.5.2), needed to process the sensor data stream effectively. The determination of required constraints could involve negotiations between the sensor and the NRM through the sensor interface. The NRM uses the sensor interface to direct the sensor platform to start sending a data stream once the NRM has marshaled the resources that the sensor will need to satisfy the request. Finally, the sensor interface also facilitates communications between the sensor platform and the NRM regarding flow control, application shutdown, etc. 2.5.6 Mapping Database This mapping database is populated with data structures generated by the graph generator and metrics object; it represents the weighted graph-theoretic characterization of the various parallel permutations of an application that is mapped to networked resources. Graph search uses the mapping database to reconstitute a weighted graph for each application for which it is asked to find resources and the degree and form of parallelism needed to meet real-time constraints. 2.5.7 Topology Database The topology database stores the current state of each of the resources; the graph generator and graph search use this database. Graph generator uses the topology database to determine which resources are available and most appropriate for candidate task-application mappings. Graph search uses this database to verify that resources are functional before a set of resources is chosen to host an application, as well as for generating and modifying weights associated with resource utilization. The topology database is 7037_C002.fm Page 15 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 45. 2-16 Smart Dust generated during the discovery phase when the NRM first comes online (e.g., see Breitbart et al. [33] and Astic and Foster [34]). Alternatively, an administrator could choose to generate a topology database for the NRM that enumerates connectivity and capability among all computation and storage resources under its control. Agent reports (or lack thereof) will affect state changes in this database indicating whether the resource is online or offline. 2.5.8 NRM Federation In a large network with a sizeable number of resources, using a single NRM may not be the most effective solution. In such a scenario, multiple NRMs are organized in a bilevel hierarchy; wide-area network (WAN) NRMs interface with sensors and administer backbone communication resources, underneath which local-area network (LAN) NRMs administer and allocate compute resources for regional compute centers (RCCs). The primary responsibility of a WAN NRM is to choose a location on the network at which distributed computing is conducted for each application and to allocate WAN bandwidth for data flow between sensors and LAN resources. The objective of the WAN NRM is to load balance WAN traffic and computational load, taking into account the relative overall processing capability of each RCC. Each LAN NRM advertises its current processing capability using standardized metrics. Each NRM is a federated collection, using a voting mechanism to elect an executor independently at the LAN and WAN levels. Each federation monitors the health of its executor by inspecting periodic diagnostic reports that the executor broadcasts. In response to an executor’s diagnostic report (or lack thereof), the federation may choose to relieve the current executor of its responsibility and elect a new one. This prevents any one NRM failure from rendering resources unusable or disabling a sensor from contracting for network services. Earlier paragraphs have detailed the LAN NRMs graph-theoretic representation of network resources, as well as its construction, weighting, and search criteria. The WAN NRM graph-theoretic representation and weighting are somewhat different from that of a LAN NRM; however, its construction and search criteria are formulated in an identical manner. The vertices in a WAN graph represent RCCs and each column corresponds to an application, while the concatenation of applications across the columns in a WAN NRM graph spans a mission. This is in contrast to a LAN NRM, in which the concatenation of tasks in its graph spans an application. 2.5.9 NRM Fault Tolerance The absence of a heartbeat or the delivery of an error report by an agent alerts the NRM to a system fault. The NRM’s fault tolerance policy is application dependent and is derived from a mandate by the developer and/or client. The policy is a trade-off between resource usage and seamless fail-over and includes redundant processing, surgical replacement, or restart of the application. Redundant processing is the most robust fail-over mechanism; the NRM simply assigns duplicate sets of resources to process the same data. If one set of resources fails, results are obtained from one of the duplicate sets. Redundant processing has the highest resource cost of all fault tolerant policies. Conversely, the NRM may choose to replace the failed component dynamically so that processing is able to continue. In this case, the NRM may have allocated distributed network storage to act as a time- delay buffer in the event of resource failure. This would enable the application, if so instrumented, to pick up processing at the point at which the failure occurred. Finally, the NRM could simply choose to halt execution of the application and start over with a new set of processing resources, although a certain amount of data and the corresponding results may be lost irrevocably. 2.6 Experimental Results A proof-of-concept experiment has been conducted at MIT Lincoln Laboratory in which the NRM allocates distributed networked resources for a sensor data fusion application in various scenarios [35]. 7037_C002.fm Page 16 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 46. Next-Generation Technologies to Enable Sensor Networks 2-17 The sensor fusion application is OASIS (operator assisted integrated systems), which is an automatic target recognition and visualization suite (see Figure 2.8). OASIS processes real-time SAR data and archived data generated by sensors with different modalities like EO and IR [36]. A block diagram of the experimental test bed is shown in Figure 2.9. The experimentation resource network consisted of three FIGURE 2.8 OASIS ATR and visualization. TABLE 2.2 Synopsis of NRM Expected Performance Experimental Configuration Max Comm BW Requirement (MB/s) Max Throughput Requirement (GFLOPS) Processors Employed Result Turn-Around Time 1 m data 26 0.7 1 1.6 1 m data with HDVI 26 2.2 2 2.6 1/4 m data 410 2.5 2 2.8 1/4 m data with HDVI 410 10 10 7 TABLE 2.3 Synopsis of NRM Performance Experimental Configuration Comm BW Measured (MB/s) Throughput Measured (GFLOPS) Processors Employed Result Turn-Around Time 1 m data 26 0.7 1 1.4 1 m data with HDVI 26 2.2 2 2.5 1/4 m data 410 2.5 2 2.7 1/4 m data with HDVI 410 10 8 7.8 OASIS Archived Data Provides historical information for area delimitation & change Real-time SIGINT Data provides cuing Real-time IMINT Data provide timely, day-night, all- weather data EO IR SAR SAR GMTI SIGINT Screener Registration Data Mining 3-D Fusion Emulated + + + 7037_C002.fm Page 17 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 47. 2-18 Smart Dust SGI O2 workstations, an eight-processor SGI Origin, an eight-node, dual Pentium3 class Beowulf cluster, and a PC workstation, which hosted the NRM. For this experiment, two SGI O2s were used as sensor surrogates to transmit unprocessed complex SAR imagery generated with range and cross-range resolutions of 1 and 1/4 m, respectively. The sensor surrogates fed data into the OASIS processing chain. To keep the complexity of the system manageable, only the most computationally intensive stage was made remappable. This stage, the HDVI processing [3] (stage 3 in Figure 2.10), had six options for the NRM ranging from a single SGI processor to six Pentium3 class cluster processors. The HDVI processing was conducted on targets detected on the two images at both resolutions, and image formation was conducted on processors in the local area network. The performance metrics for the OASIS applications were determined with a combination of actual performance measurements and modeled performance analyses. Table 2.2 is a tabulated synopsis of the expected performance of the NRM and Table 2.3 shows the actual performance of the NRM. The expected and actual performance values compared very well. Because this network was PE resource limited, the objective of the NRM was to use the smallest fraction of PE bandwidth available across the network while meeting network conduit, PE utilization, latency, throughput, and network-wide bandwidth usage constraints. It is clear from the results that the NRM was able to tailor the communication and computation solution it delivered based on the particular application needs and the constraints imposed. The successful completion of this experiment has initiated further research and development to give the NRM greater functionality, automation, and flexibility. FIGURE 2.9 Experimentation resource network. CONUS Resources Theater Resources Sim’d SAR sensor 1 Sim’d SAR sensor 2 Parallel Cluster Visualization and OASIS Data Exploitation OASIS Data Exploitation Network Resource Manager 1000 Mbps Private network on GLOWNet 7037_C002.fm Page 18 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 48. Next-Generation Technologies to Enable Sensor Networks 2-19 Acknowledgments The authors thank the members of the Precision Targeting via Collaborative Networking team at MIT Lincoln Laboratory for formulating many of the concepts discussed in this chapter. The authors also thank Dr. Mari Maeda, formerly of DARPA/ITO, and Dr. Gary Koob of DARPA/IPTO for their encour- agement and support of this project. References 1. Usoff, J., Beavers, W., and Cox, J., Wideband networked sensors processing, in Proc. High Perfor- mance Embedded Computing Workshop, November 2001. 2. Cuomo, K.M., Pion, J.E., and Mayhan, J.T., Ultrawide-band coherent processing, IEEE Trans. Antenna Propagation, 47, 1094, June 1999. 3. Benitz, G.R., High-definition vector imaging, MIT Lincoln Lab. J., Special Issue Super-Resolution, 10:2, 147, 1997. 4. Nguyen, D.H. et al., Super-resolution HRR ATR Performance with HDVI, IEEE Trans. Aerospace Electron. Syst., 37:4, 1267, October 2001. 5. Chan, V.W.S., Optical space communications, IEEE J. Selected Topics Quantum Electron., 6:6, 959, November/December, 2000. 6. Awduche, D. et al., RSVP-TE: extensions to RSVP for LSP tunnels, RFC 3209, http://www.faqs.org/ rfcs/rfc3209.html, December 2001. 7. Ash, J. et al., Applicability statement for CR-LDP, RFC 3213, http://www.faqs.org/rfcs/rfc3213.html, January 2002. 8. Cormen, T.H., Leiserson, C.E., and Rivest, R.L., Introduction to Algorithms. McGraw-Hill, New York, 1993. 9. Strater, J. and Wollman, B., OSPF modeling and test results and recommendations, Mitre Technical Report 96W0000017, Mitre Corporation, 1996. FIGURE 2.10 Graph of OASIS application onto the experimental resources. Or-J Or-J P3- 3 P3- 7,8 P3- 3,4,5,6 P3- 1,2,3,4 P3- 1,2,3,4,5,6 O2 Sensor Surrogate Front-End Processing Classifier Processing Back-End & GUI 1 Stage Number 2 3 4 Or-S 7037_C002.fm Page 19 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 49. 2-20 Smart Dust 10. Perkins, C., Ad Hoc Networking, Addison-Wesley, Boston, 2001. 11. Floyd, S. and Jacobson, V., Random early detection gateways for congestion avoidance, IEEE/ACM Trans. Networking, 1:4, 397, August 1993. 12. Stevens, W., TCP slow start, congestion avoidance, fast retransmit and fast recovery algorithms, RFC 2001, http://www.faqs.org/rfcs/rfc2001.html, January 1997. 13. Stadler, J.S., Performance enhancements for TCP/IP on a satellite channel, in Proc. IEEE Military Commun. Conf. 1998 (MILCOM98), 1, 270, October 1998. 14. Allman, M., Glover, D., and Sanchez, L., Enhancing TCP over satellite channels using standard mechanisms, RFC 2488, http://www.faqs.org/rfcs/rfc2488.html, January 1999. 15. Berrou, C., Glavieux, A., and Thitimajshima, P., Near Shannon limit error-correcting coding and decoding: turbo codes. 1, in Conf. Rec. IEEE Int. Conf. Commun. 1993 (ICC 93), 2, 1064, May 1993. 16. Foschini, G.J., Layered space-time architecture for wireless communication in a fading environment when using multiple antennas, Bell Labs Tech. J., 1:2, 41, Autumn 1996. 17. Raleigh, G.G. and Cioffi, J.M., Spatio-temporal coding for wireless communications, in Proc. IEEE Global Telecommun. Conf. 1996 (GLOBECOM 96), 3, 1405, November 1996. 18. Sisterson, L.K. et al., An architecture for semi-automated radar image exploitation, Lincoln Lab. J., 11:2, 175–204, 1998. 19. Cooley, J.A. et al., Software-based erasure codes for scalable distributed storage, in Proc. 20th IEEE Symp. Mass Storage Syst., 157–164, April 2003. 20. Luby, M.G. et al., Practical loss-resilient codes, in Proc. 29th ACM Symp. Theory Computing, 150–159, 1997. 21. Byers, J.W., Luby, M.G., and Mitzenmacher, M., Accessing multiple mirror sites in parallel: using tornado codes to speed up downloads, in Proc. IEEE INFOCOM 1999, 275–283, March 1999. 22. Silberschatz, A. and Galvin, P., Operating System Concepts, 5th ed., Addison-Wesley, Reading, MA, 1998. 23. Wind River Systems, Inc. http://www.windriver.com/, accessed July 2003. 24. FSMLabs (Finite State Machine Labs), Inc. http://www.fsmlabs.com/, accessed July 2003. 25. Mercury Computer Systems, Inc. http://www.mc.com/, accessed July 2003. 26. Abbott, D., Linux for Embedded and Real-Time Applications, Newnes, Amsterdam, 2003. 27. Object Management Group. http://www.omg.org/, accessed July 2003. 28. Hoffmann, H., Kepner, J., and Bond, R., S3P: Automatic, optimized mapping of signal processing applications to parallel architectures, in Proc. High Performance Embedded Computing Workshop 2001, September 2001. 29. The vector, signal, and image processing library. http://www.vsipl.org/, accessed July 2002. 30. Reuther, A.I. and Goodman, J.I., Resource management for digital signal processing via distributed parallel computing, in Proc. High Performance Embedded Computing Workshop 2002, September 2002. 31. Goodman, J.I. et al., Discrete optimization using decision-directed learning for distributed net- worked computing, in Proc. IEEE Asilomar Conf. Signal, Syst. Computers, 1189–1196, November 2002. 32. Neuman, B.C. and Ts’o, T., Kerberos: an authentication service for computer networks, IEEE Commun., 32:9, 33, September 1994. 33. Breitbart, Y. et al., Topology discover in heterogeneous IP networks, in Proc. IEEE INFOCOM 2000, 265–274, March 2000. 34. Astic, I. and Foster, O., A hierarchical topology discovery service for IPv6 networks, in Proc. 2002 Network Operations Manage. Symp., 497–510, April 2002. 7037_C002.fm Page 20 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 50. Next-Generation Technologies to Enable Sensor Networks 2-21 35. Reuther, A.I. and Goodman, J.I., dynamic resource management for a sensor-fusion application via distributed parallel grid computing, in Proc. High Performance Embedded Computing Work- shop 2003, 2003. 36. Avent, R.K., A multi-sensor architecture for detecting high-value mobile targets, in Proc. 2002 SIAM Conf. Imaging Sci. (IS02), March 2002. 7037_C002.fm Page 21 Tuesday, November 1, 2005 12:20 PM © 2006 by Taylor & Francis Group, LLC
- 51. 3-1 3 Sensor Network Management 3.1 Introduction ...................................................................... 3-1 3.2 Management Challenges................................................... 3-2 3.3 Management Dimensions................................................. 3-3 Dimensions for WSN Management • Management Levels • WSN Functionalities • Management Functional Areas 3.4 MANNA as an Integrating Architecture........................ 3-15 Management Services, Functions, and Models • Functional Architecture • Information Architecture • Physical Architecture 3.5 Putting It All Together.................................................... 3-25 3.6 Conclusion....................................................................... 3-25 3.1 Introduction A wireless sensor network (WSN) consists of a large number of sensor nodes deployed over an area and integrated to collaborate through a wireless network. WSNs encourage several novel and existing appli- cations such as environmental monitoring; health care; infrastructure management; public safety; med- ical; home and office security; transportation; and military [1, 2, 9, 17, 18]. These have been enabled by the rapid convergence of three technologies: digital circuitry, wireless communications, and the micro- electromechanical system (MEMS). These technologies have enabled very compact and autonomous sensor nodes, each containing one or more sensor devices, computation and communication capabilities, and limited power supply. Some of the applications foreseen for WSNs will require a large number of devices in the order of tens of thousands of nodes. Traditional methods of sensor networking represent an impractical, complex, and expensive demand on cable installation. WSNs promise several advantages over traditional sensing methods in many ways: better coverage, higher resolution, fault tolerance, and robustness. The ad hoc nature and deploy-and-leave vision make it even more attractive in military applications and other risk- associated applications, such as catastrophe, toxic zones, and disasters [2, 9]. Performing the processing at the source can drastically reduce the computational burden on application, network, and management. On the other hand, any solution must take into account specific characteristics of this type of network. WSN management must be autonomic, i.e., self-managed (self-organizing, self-healing, self-optimiz- ing, self-protecting, self-sustaining, self-diagnostic) with a minimum of human interference, and robust to changes in network states while maintaining the quality of services. Until now, WSNs and their applications have been developed without considering an integrated management solution. The task of building and deploying management systems in environments that will contain tens of thousands of network elements with particular features and organization and that deal with the aforementioned Linnyer Beatrys Ruiz Pontifical Catholic University of Paraná and Federal University of Minas Gerais José Marcos Nogueira Federal University of Minas Gerais Antonio A. F. Loureiro Federal University of Minas Gerais © 2006 by Taylor & Francis Group, LLC
- 52. 3-2 Smart Dust attributes is not trivial. This task becomes more complex due to the physical restrictions of the unattended sensor nodes, in particular energy and bandwidth restrictions. In this chapter, the focus is on WSN management, which comprises a large number of devices in the order of tens of thousands of nodes. Clearly, the mechanisms associated with traditional management paradigms must be rethought. In this sense, a new paradigm called autonomic management is explored. The rest of this chapter is organized as follows. Section 3.2 presents an overview of network management and discusses the management challenges for WSNs. In Section 3.3, management dimensions (manage- ment levels, WSN functionalities, and management functional areas) are presented and discussed. A management architecture for WSNs called MANNA is presented in Section 3.4, as well as how it works. In Section 3.5, a simple example shows the different aspects together. Finally, Section 3.6 presents con- clusions. 3.2 Management Challenges One of the major goals of network management is to promote productivity of network resources and maintain the quality of the service provided. However, the management of traditional networks and of WSNs has several significant differences. This section discusses important characteristics of WSNs that make their management different from that of other networks. A WSN is a tool for distributed sensing of one or more phenomenon that reports the sensed data to one or more observers. A WSN provides services for observers as well as for itself. It produces and transports application data, so, in this sense, the network provides service to itself. The objective of a WSN is to monitor and, eventually, control a remote environment. Sensor nodes execute a common application in a cooperative way (i.e., a clear, common goal in the overall network), which may not be the case in a traditional network. The traditional computer networks are designed to accommodate a diversity of applications. Network elements are installed, configured by technicians, and connected in a network in a way to provide different kinds of services. Technicians’ maintenance of components or resources is a normal fact. The network tends to follow well-established planning of available resources and the location of each network element is well-known. In a WSN this is not often the case because the network is planned to have unattended operation. In fact, the initial configuration of a WSN can be quite different from what was supposed to be in cases such as throwing the nodes into an ocean, forest, or other remote regions. In unpredictable situations, a configuration error such as a planning error may cause the loss of the entire network even before it starts to operate. Energy is a critical resource in WSNs. Thus, all operations performed in the network should be energy efficient. Topology is dynamic because sensor nodes can become out of service temporarily or perma- nently (nodes can be discarded, lost, destroyed, or even run out of energy). In this scenario, faults are a common fact, which is not expected in a traditional network. Depending on the WSN application, it may be interesting to identify uniquely each node in the network. Furthermore, one may be interested in a value associated to a given region and not to a particular node — for instance, in the temperature at the top of a mountain. A WSN is typically data centric, which is not common in traditional networks. A managed WSN is responsible for configuring and reconfiguring under varying (and, in the future, even unpredictable) conditions. System configuration (“node setup” and “network boot up”) must occur automatically; dynamic adjustments need to be done to the current configuration to best handle changes in the environment and itself. A managed WSN always looks for ways to optimize its functioning; it will monitor its constituent parts and fine-tune workflow to achieve predetermined system goals. It must perform something akin to healing — it must be able to recover from routine and extraordinary events that might cause some of its parts to malfunction. The network must be able to discover problems or potential problems, such as uncovered area, and then find an alternate way of using resources or recon- figuring the system to keep it functioning smoothly. In addition, it must detect, identify, and protect itself against various types of attacks to maintain overall system security and integrity. A managed WSN must © 2006 by Taylor & Francis Group, LLC
- 53. Sensor Network Management 3-3 know its environment and the context surrounding its activity and act accordingly. The management entities must find and generate rules to perform the best management of the current state of the network [22]. A managed WSN with this has various characteristics can be called an autonomic system [1], which is an approach to self-managed computing systems with a minimum of human interference. This term derives from the autonomic nervous system of the human body, which controls key functions without conscious awareness or involvement. The processors in such systems use algorithms to determine the most efficient and cost-effective way to distribute tasks and store data. Along with software probes and configuration controls, computer systems will be able to monitor, tweak, and even repair themselves without requiring technology staff — at least, that is the goal [1]. WSN management must be autonomic, i.e., self-managed and robust to changes in network states while maintaining the quality of service; that is, it must be capable of self-configuration, self-organization, self-healing, and self-optimization. However, the computational cost of autonomic processes can be expensive to some WSN architectures. Probably, the fundamental issue about the management of a WSN is concerned with how the man- agement can promote plant and resource productivity,and how it integrates in an organized way functions of configuration, operation, administration, and maintenance of all elements and services. The task of building and deploying autonomic management systems in environments in which tens of thousands of network elements with particular features and organization will be present is very complex. This task becomes even more involved due to the physical restrictions of the sensor nodes, in particular energy and bandwidth restrictions. The management application to be built also depends on the kind of application being monitored. A good strategy is to deal with complex management situations by using management dimensions. 3.3 Management Dimensions In general, for traditional networks, management aspects are clearly separated from network common activities, i.e., from the services they provide to their users. It is also said that an overlap of management and network functionalities exists, although the implementation can be thought of independently. This separation can be promoted by using two traditional management dimensions: management functional areas [14] and management levels [15]. The requirements to be satisfied by systems management activities can be categorized into functional areas. These facilities have come to be known as the specific management functional areas (SMFAs): fault management; configuration management; performance management; accounting management; and security management. This has proved to be a helpful way of partitioning the network management problem from an application point of view [14]. To deal with the complexity of management, management functionality with its associated information can be decomposed into a number of logical layers: business management; service management; network management; and network element management. The architecture that describes this layering is called the logical layered architecture (LLA) [15]. Management activities can be clustered into layers and decou- pled by introducing manager and agent roles. A logical layer reflects particular aspects of management and implies the clustering of management information supporting that aspect. Typically, an interaction takes place between adjacent layers, but due to operational and management considerations other inter- actions may also occur between nonadjacent layers. The use of the management dimensions is a good strategy to deal with complex management situations by decomposing a problem into smaller subproblems, in successive refinements steps, and to provide a separation between application and management functionalities through a management architecture. This will make possible the integration of organizational, administrative, and maintenance activities for a given network. WSN management must be simple,adherent to network idiosyncrasies,including its dynamic behavior, and efficient in its use of scarce resources. The adoption of a strategy based on the traditional framework © 2006 by Taylor & Francis Group, LLC
- 54. 3-4 Smart Dust of functional areas and management levels will permit management integration in the future. However, for WSN management it is necessary to go further. Using management functional areas and management levels is not enough because WSNs are application specific. The following discussion concerns how the traditional management dimensions can be applied in WSN management. Also, new dimension for WSN management is proposed that considers the general aspects of the different types of the networks. 3.3.1 Dimensions for WSN Management WSNs are embedded in applications to monitor the environment and act upon it. Thus, the management application should try to be “compatible” with the kind of application being monitored. In order to have better development of WSN management services and functions, it is necessary to characterize the WSN and establish a novel management dimension. Thus, looking at the characteristics of various WSN applications, five main WSN functionalities are identified: configuration; sensing; processing; commun- ication; and maintenance. These functionalities define a novel dimension for the management, as pre- sented in Figure 3.1 [22]. Configuration is the first functionality before a network starts sensing the environment, processing, and communicating data. Maintenance treats specific characteristics of WSN applications during the entire network lifetime. In this way, WSN management will have an organization that comes from abstractions offered by management functional areas, management levels, and WSN functionalities (configuration, sensing, processing, communication, and maintenance). The novel dimension introduced can be observed in the upper part of Figure 3.1. The coordination among the three planes can be based on policies. Policy-based network management (PBNM) [7] is a feasible alternative because it allows the manager to set actions to be carried out by the network without worrying too much about network details. Managers can define suitable actions in due time and still have a global or local view of the network. PBNM helps to manage complex networks such as WSNs. The managers will only inform concerning what is expected, but not how it should be obtained. The agents will be intelligent to decide what to do as well as how and when to do it. Automatic services and functions can be executed toward self-management if appropriate conditions, such as residual energy level, are present. FIGURE 3.1 Management dimensions for WSNs. (From Ruiz, L.B., Nogueira, J.M., Louriero, A.A., IEEE Commun. Mag., 41(2), 116–125, 2003. With permission.) WSN FUNCTIONALITIES Configuration Maintenance Sensing Processing Communication MANAGEMENT LEVELS Business Management Service Management Network Management Network Element Management Network Element FUNCTIONAL AREAS Configuration Management Fault Management Performance Management Security Management Accounting Management © 2006 by Taylor & Francis Group, LLC
- 55. Exploring the Variety of Random Documents with Different Content
- 56. men found nothing to admire. But the district was of much military importance as a source of supplies and channel of communication for Richmond and Lee’s army. The 3d, under Gen. Foster, was side by side with the 43d and 44th Regiments, both of which have place in Coast Artillery history. They participated in the “great march” thru Kinston, Whitehall and Goldsboro. June 11, 1863, the regiment embarked for home; and was mustered out June 26. Veterans of the 4th Regiment residing in Taunton organized the Taunton City Guard on Nov. 4, 1865, thus giving that city a competitor to its older Light Guard. The company entered the 3d Regiment in 1866, and today exists as the 9th Company, Mass. C. A. For a few months there was an exciting rivalry between the two Taunton companies, as each claimed to be the rightful owner of certain military property in the city,—camp equipage and a fund of $800 coming down from war days. The property would be first concealed by one company and then captured by the other. The courts were appealed to; but finally the matter was compromised; they divided the money, and the companies became joint owners of the tentage and other equipment. Orders were issued by the State authorities on Aug. 20, 1866, combining the 4th and 3d Regiments in a new 3d Regiment, and on Aug. 31, Col. Mason W. Burt of Taunton was elected commander. Col. Burt had been Captain and Major in the 22d Mass. Volunteers from 1861 to 1864. The new regiment consisted of companies in Halifax (A), Fall River (B), Scituate (C), New Bedford (E), Taunton (F) and (G), and Quincy (H). The Halifax Light Infantry, the New Bedford City Guards, B of Fall River, and, a little later, the revived D of Fall River under Capt. Sierra L. Braley, with a new Scituate company, represented the 3d Regiment; while the Taunton Light Guard and Hancock Light Guards of Quincy came from the 4th Regiment. The new Taunton company entered the 3d at this time; but the Standish Guards remained aloof, as the 87th Unattached Company, until 1868. At the latter date the Plymouth company came in as Co. M. Thomas J. Borden became Colonel June 23, 1868, and Bradford D. Davol
- 57. followed on March 9, 1871, both being residents of Fall River. When on Aug. 2, 1876, the regiment was reduced to a battalion, the “3d Battalion of Infantry,” its only surviving companies were the New Bedford City Guards (E), the Taunton City Guards (F), the Taunton Light Guard (G), and the Standish Guards (now H). All others had been disbanded. Maj. Daniel A. Butler, former Captain of the Standish Guards, commanded the 3d Battalion. Meanwhile the Cunningham Rifles of North Bridgewater or Brockton had been organized in 1869, and named after the Adjutant General, James A. Cunningham. Originally Co. I of the 3d, this command was transferred to the 1st Battalion of Infantry, Lt. Col. Wales, in 1876; and so pioneered the way for the remainder of the “Cape” companies to follow two years later. This company exists today as the 10th Company, Mass. C. A. One cause contributing to the disappearance of the 3d Regiment was the fact that it was called upon to perform two tours of duty for the maintenance of public order in Fall River, first on Aug. 5, 1870, continuing three days, and again Sept. 27, 1875, continuing seven days. Such service in connection with industrial disturbance is exceedingly painful to the feelings of the men. Coming as it did when class sensitiveness was acute, and when the old Civil War veterans were ready to retire permanently from active military service, it did much to break up the command. Happily such a situation can hardly recur today. The 3d Regiment participated in musters with the 1st Brigade from 1866 to 1871, the final one being held at Lovell’s Plain, North Weymouth. In 1872 there was a regimental encampment at their old Civil War mobilization ground, “Camp Joe Hooker,” Lakeville. On Dec. 3, 1878, Major Butler’s four-company battalion was consolidated with the 1st and 4th Battalions as part of the 1st Regiment.
- 59. CHAPTER IX SINCE 1878 Col. Wales’ regiment, when he received his commission on Dec. 30, 1878, consisted of the following twelve companies: 1, The Roxbury Artillery or City Guard. 2, The Boston Light Infantry. 3, The Taunton Light Guard. 4, The New Bedford City Guards. 5, The Standish Guards of Plymouth. 6, The Massachusetts Guards of Cambridge. 7, The Pierce Light Guard of Boston. 8, The West Roxbury Rifles. 9, The Taunton City Guard. 10, The Cunningham Rifles of Brockton. 11, The Maverick Rifles of East Boston. 12, The Fall River Rifles. The Fusiliers and the Chelsea Rifle-Veterans were temporarily detached from the regiment, and the Claflin Guards were gone, never to return so far as we now know. The 1st and 8th Companies were directly from the 1st Regiment. The 2d, 6th, 7th and 11th Companies came from the 4th Battalion; the 3d Company came originally from the 4th Regiment and immediately from the 3d; the 4th, 5th and 9th Companies were from the 3d Regiment; the 10th was originally from the 3d and immediately from the 1st. A new 12th Company was organized on Dec. 12, 1878, with Capt. Sierra L. Braley in command. The new
- 60. company speedily forged to the head in efficiency and has always been one of the three or four leaders in the entire regiment. Boston celebrated the 250th anniversary of its settlement on Sept. 17, 1880, and along with other features included a magnificent military display. Everyone conceded that, while other bodies presented a fine appearance, the feature of the parade was the twelve-company 1st Regiment. That day, for the last time, the companies wore their original uniforms—old 1st Regiment, gray with towering bearskin shakos; 4th Battalion, a semi-Zouave costume with low shakos, double breasted blue coats, light blue bloused knickerbockers, and high leather leggins; and the 3d Regiment, low shakos, short blue coats, single-breasted but with three rows of buttons, and blue trousers. The regiment was received enthusiastically by the people of Boston and the day was one long to be remembered. But changes were projected in the interests of efficiency, and first of all, in that very year, 1880, it was decided to adopt the 4th Battalion uniform for the entire twelve companies. So satisfactory did this prove that the Commonwealth utilized the same costume as a state uniform, and issued it to all the organizations of Massachusetts in 1884. Imitation is the sincerest form of flattery; but it can scarcely be said that the 1st relished sharing their distinctive uniform with all the militia,—they felt that they had paid dearly for this flattery. Thereafter the regiment was to be subjected to a continuous and intensifying process of military improvement, at the hands first of the state authorities, and presently of the “Department of Militia Affairs” or “Militia Bureau” in the War Department. While it was inevitable that there should be a deal of experimentation whose results were not always satisfactory, it remains true that constant progress was made thruout the ensuing years. National Guardsmen, since they are human, are prone to complain; certainly they greeted almost every innovation with a chorus of “kicks.” But as soon as a change had demonstrated its usefulness, it was heartily welcomed. More and more time was demanded of the men; and on the other
- 61. hand part of this increased service was rewarded with increased pay by the State or Nation. The four days of camp duty required in 1873 had stretched to fifteen days in 1916, the twelve armory drills of early days to forty-eight. State and Federal pay were not an adequate recompense for the labor performed; the service was still one of unselfish patriotism. But the money invested by the authorities in camp and “rendezvous drill” pay did unquestionably testify to the higher esteem in which, with the passing years, the Guard came to stand. One noticeable consequence of the increasing military strictness was the gradual lowering of average age amongst the companies. Older men cannot be away from their business or families for so many hours and days, under ordinary circumstances. American armies have always been made up of very young men; and under the stress of increased requirements, the National Guard came to be similarly constituted. One company participated in the exercises connected with the funeral of Pres. James A. Garfield at Cleveland in 1881. Nathaniel Wales was elected Brigadier General on Feb. 21, 1882, and on Feb. 24, Austin C. Wellington became Colonel. The Tiger battalion, during the eight years of Wellington’s command, had become the most prominent military institution in Boston; now the entire 1st Regiment was to profit by the skill of the same man, a skill truly amounting to genius. Peculiar qualities are demanded of one who is to succeed in highest degree as a National Guardsman. He must be a well-trained soldier and a hard worker as a matter of course. He must command respect for his personal character and must be able to impart knowledge to others. He must enforce rigid discipline, and must do it without resorting to regular army methods of punishment. On top of all, there has to be sufficient personal magnetism in his make-up to attract men, and enthusiasm enough to overflow and fire others. This description of a model Guardsman is nothing more or less than a description of Austin C. Wellington. No wonder that during his six years of command, the regiment was to register a new high-water mark of success.
- 62. Now the old companies began to come back. When in 1883 the Standish Guards suffered disbandment, their place was promptly taken by the company which had originally held it, the Chelsea Rifles. The Taunton Light Guard ceased to exist in 1884, and at first, the vacant 3d number was filled by the formation of a new company in Natick. Four years later the Natick organization transferred and became Co. L of the 9th, and then the Fusiliers returned to their proper place as 3d Company. 1882 was notable for the Daniel Webster centennial. Pres. Chester A. Arthur honored Boston with a visit on this occasion, and on Oct. 11, the 1st Regiment served as Presidential escort during the celebration at Marshfield. The habit of visiting distant cities now grew on the regiment, so that on August 8, 1885, they were found in New York participating in the tremendous funeral procession in honor of their old-time commander-in-chief, U. S. Grant. Their fame grew. All Roxbury joined in celebrating the centennial of its favorite corps, the City Guard, in 1884. March 22 of that year will long be remembered for its parade, and other demonstrations of affectionate enthusiasm. In 1886 the 12th Company visited Providence, R. I., as guests of the Light Infantry; and assisted their hosts to celebrate in fitting manner the two hundred fiftieth anniversary of Rhode Island’s settlement. 1887 brought the Fusilier centennial; and was likewise properly observed. In 1887 the United States celebrated the centenary of the signing of its constitution, choosing Philadelphia, where the document had been drafted, as the place for the demonstration. Massachusetts decided to send Gov. Oliver Ames and to provide, as his military escort, the most proficient regiment in the State. It was not necessary to lose any time searching for the regiment—orders were promptly issued to Col. Wellington, that he prepare his command for the Philadelphia trip, the Commonwealth to pay expenses. Sept. 15 found the regiment on its way to Philadelphia, Sept. 16 saw them marching as one of the most brilliant units of the
- 63. great parade under command of Gen. Philip H. Sheridan, while Sept. 17 was signalized by their return to Boston. D. W. Reeves was band- leader that year—no unworthy successor to Fillebrown and Gilmore —and he contributed, as his share in the event, a new march, “The March of the First.” Chaplain Minot J. Savage, who added to his gift of eloquence the rarer talent of poetry, wrote words for Reeves’ music, “We’re brothers of all noble men, Who wear our country’s blue; We brothers find in any race, Where men are brave and true. But we’ve a pride in our own band, And we are all agreed, Whatever grand deeds others do, The ‘Old First’ still shall lead.” The fame of the regiment became nation-wide as a consequence of the Philadelphia trip. Col. Wellington’s most notable innovation was the introduction of artillery instruction, or the re-introduction, as it was for those companies originally in the old First. The change was made for the purpose of rendering drills more interesting. It is easier to maintain the interest of artillerymen—they have their guns as a rallying-point. Moreover the artillery virus was in the 1st Regiment blood and was bound eventually to manifest its presence. That year of Col. Wellington’s accession, 1882, the legislature appropriated $5,000 for the construction of “Battery Dalton” at Framingham. Named in honor of the Adjutant General, Samuel Dalton, it was truly a marvelous work of coast defence. Its mortars had a range of five hundred yards. After firing the projectile, the cannoneers walked over and solemnly dug the same up from its self- made grave, and fired it over again. Artillery practice was economically conducted in those pioneer days. Sept. 13, 1883, the regiment was permitted to hold one day’s practice at Fort Warren, a
- 64. great concession by the War Department, and a long step in artillery progress. Sept. 4, 1885, one month after the Grant funeral, the privilege of artillery practice was repeated. A riot in Cambridge brought the 6th Company into active service for two days on Feb. 21 and 22, 1887. Col. Wellington’s death occurred while he still filled the office of regimental commander, on Sept. 18, 1888. The funeral is said to have been the saddest tour of duty ever performed by the regiment, an expression of heart-felt grief. They were then looking forward to occupying the new South Armory; and everyone contributed the entire pay received for the day toward the expenses of a memorial room in the building. This money equipped and furnished the gymnasium in the tower, the room now devoted to the war-game. Thomas R. Mathews, Colonel from Dec. 10, 1888, until July 19, 1897, had served in the 2d Company during the Civil War, and had subsequently been Captain of the 1st Co. (in 1880). On Oct. 8, 1888, just before Col. Mathews’ election, the regiment took part in a general mobilization of militia in Boston. On Thanksgiving day, Nov. 28, 1889, the Boston companies were assembled at the armories in readiness for service in maintaining public order at a great fire then raging. Fortunately they did not have to leave their stations. Prior to 1890 the Companies had been quartered in various halls and rinks of Boston and the suburbs, Faneuil Hall being the most coveted location, unavailable, however, most of the time, and Boylston Hall, Boylston and Washington Streets, ranking next. 1890 was the date of the South Armory dedication. Massachusetts had entered, after long years of discussion, upon her policy of providing adequate accommodations for her volunteer militia. New York had led the way ten years earlier; and the Massachusetts authorities were especially indebted to the N. Y. 7th for providing an armory after which others could pattern. It is a far cry from the 7th’s building to that on Irvington St., but there is a
- 65. similarity of type. It must be borne in mind that the South Armory was relatively one of the best in the country when the 1st Regiment occupied it in 1890. Nor had the railroad developed into such a nuisance at that time. The South Armory was the first State armory in Massachusetts; and led the way for the entire series, by means of which our troops are quartered as well as any in the land; its dedication was an important event in military history. Fall River followed, and dedicated her State armory in 1895, Cambridge and New Bedford in 1903, Brockton in 1906, Chelsea in 1907, and Taunton in 1917. Chelsea and Brockton subsequently lost their buildings by fire; the structures were rebuilt respectively in 1909 and 1912. Col. Mathews’ command served as personal escort to Gov. William E. Russell, Feb. 29, 1892, at the ceremony of presenting Massachusetts’ first long-service medals. Amongst others, twenty- eight officers and men of the 1st received medals. An artillery tour was held at Fort Warren, Aug. 7 to 13, 1892, when the men had practice on the eight-inch muzzle-loading converted rifles and the fifteen-inch muzzle-loading smooth-bores. Modern coast artillery had not yet “arrived”; but the regiment was making progress. In 1893 they encamped at Framingham and manned “Battery Dalton” once more. In 1895 they had their last experience with these twelve-inch mortars—and the sand-bank five hundred yards away; 1894, 1896 and 1897 saw them at Fort Warren each summer. In 1896 the regulars did not take them seriously and could not “waste time” instructing the militiamen; in 1897, with Lieut. Erasmus M. Weaver temporarily detailed as instructor, the regiment made progress. Thereafter, until 1911, regular officers from the forts added to their other service the duty of visiting the South Armory and coaching the militia regiment. All twelve companies were ordered to be in readiness on March 10, 1893, for service in connection with the disastrous “Lincoln St. fire,” but were not marched out of the armories.
- 66. The state expended $2,500 in 1894 providing a model battery at the South Armory. While crude compared with the huge gun and mortar installed in 1913, to which the name “Battery Lombard” is sometimes given, this earlier artillery installation marked a long advance in drills and instruction. On Oct. 9, 1894, the regiment again participated in a general mobilization of the militia at Boston. The monument to Robert Gould Shaw, on the Common, was formally dedicated May 31, 1897, and the regiment paraded in honor of the event. One feature of the day recalled certain historic processions of thirty years previously—the New York 7th, in which Col. Shaw had once served, came on to have a share in this demonstration of affection. On June 1, 1897, by act of the legislature, the regiment received a new name—it became the 1st Regiment of Heavy Artillery. In point of fact it had begun to separate from the 1st Brigade back in Col. Wellington’s time, and had become increasingly committed to the artillery branch; this act of legislation officially recognized a transition which had already taken place. Now the facings on the uniforms could be changed from the blue of infantry to the brighter and more distinctive scarlet. Massachusetts was the first state to have heavy artillery in its militia—the old regiment was again “first.” Companies were rechristened “batteries” in connection with the change of service. Col. Mathews became Brigadier General on July 19, 1897, and Charles Pfaff succeeded as Colonel on July 28. Col. Pfaff’s military training had been in the Cadets, and as Captain of the 8th Company, Coast Artillery; and he had served four years as Major. To him was to fall the honor of commanding the regiment during its Spanish War service. There was nothing unexpected about the war with Spain. From the day the “Maine” was destroyed until April 25, when war was declared, more than two months elapsed. Members of the command were in constant readiness during this entire period for the summons
- 67. which they knew must come; and it was well understood that instant mobilization would ensue upon receipt of orders. But if we had reason to be in readiness, we also had good cause to anticipate danger and hardship. The United States was notorious for lack of preparedness, both by land and sea. On the other hand the might of the Spanish fleet and the fame of the “Spanish infantry” had been so magnified that much popular trepidation existed. Boston anticipated instant attack; merchants and bankers deposited their treasure with inland banks; while real estate owners were insistent that the national government should afford them protection. Col. Pfaff and his men were to volunteer in the belief that they would meet with instant and active fighting. Beyond question the general public drew a deep sigh of relief as the blue-clad column, on that fateful 26th of April, to the music of the “March of the First,” swung steadily down Huntington Ave. The out-of-town commands had left their home stations early and received Godspeed from newsboys and milkmen only. In Boston, however, the display of enthusiasm left nothing to be desired; and demonstrated not only the city’s dependence upon its heavy artillerymen but also its real affection for the red-legged organization. They were paid from April 25. Besides Col. Pfaff, the regimental officers were: Lt. Col., Charles B. Woodman; Majors, Perlie A. Dyar, George F. Quinby, James A. Frye; Captains, 1st Co., Joseph H. Frothingham; 2d Co., Frederic S. Howes; 3d Co., Albert B. Chick; 4th Co., Joseph L. Gibbs; 5th Co., Walter L. Pratt; 6th Co., Walter E. Lombard; 7th Co., Charles P. Nutter; 8th Co., John Bordman, Jr.; 9th Co., Norris O. Danforth; 10th Co., Charles Williamson; 11th Co., Frederick M. Whiting; 12th Co., Sierra L. Braley. Capt. Braley had been private and corporal in the 3d Reg. during its nine-months service in 1862. He had been 2d Lieutenant in Battery I, 2d Mass. Heavy Art., and in Bat. L, 14th U. S. Colored Art., during 1864 and 1865. From 1866 until 1878 he continuously held commissions in the 3d Reg. and, after 1878, in the 1st, his latest command being the 12th Company. Capt. Braley was
- 68. the only officer of the regiment to serve in both the Civil and Spanish Wars. On April 26 the regiment began active duty at Fort Warren, the orders reading that they would encamp there for eight days. Five more days were added to this; and then the command was taken into the U. S. service “for the war.” Since the thirteen days of state duty is added to the total in computing their record, they were the first regiment of the entire nation to begin war service. The Old First still led. When they left the armory for Fort Warren, there were only six men absent from the command—four sick and two out of the country. Opportunity was later given for men with families to withdraw, if they desired; and all were subjected to a rigid physical examination. Ultimately three per cent. were rejected for disability and eight per cent. excused for family reasons. These vacancies were immediately filled from the throngs of would-be recruits who volunteered. It was a disappointment to the regiment that the War Department never permitted them to increase their numbers to the full war strength; their Spanish War roster bore 751 names. They started out in the rain on April 26, and it seemed as if it would rain until they returned; during their first six weeks, they were blest with sunshine only three days. By and by, when they had ceased to care, the weather changed and they had sunny days. At Warren they were quartered in wooden buildings, originally election booths in the city; prisoners from Deer Island were imported to assist in erecting these; and some humorist promptly designated them the “3d Corps of Cadets.” While in the state service, the regiment was fed by a caterer, after the fashion then prevalent at Framingham. When they became U. S. soldiers, they messed themselves. All thru this war, ammunition was very scarce indeed. The least a self-respecting military post can do is to fire morning and evening guns; this was possible in 1898 only by cutting cartridges in two and using half-charges. Most of the ordnance was of Civil War vintage, or very slightly more modern.
- 69. Spain had been vastly over-rated, and there was very little fight in her. The regiment passed a busy and profitable month at Fort Warren from April 26 to May 30, being mustered into the United States service on May 7. During these weeks the companies or “batteries” attained a high degree of proficiency in both infantry and artillery drill. Shortly after midnight on May 13 the Engineers’ steamer, the “Tourist,” came down the harbor from the Navy Yard to announce that the Spanish fleet had actually been sighted off Nantucket. But men watched in vain for the enemy vessels to appear. On Memorial day, thru the exigency of service conditions, the companies were moved and distributed along the coast at posts ranging from Portsmouth to New Bedford. Maj. Frye and the Cape companies remained at Warren. Lt. Col. Woodman with the 3d and 11th Companies garrisoned the fort at Clark’s Point, New Bedford, a work which had been in existence since 1857 but which awaited July 23, 1898, and these companies as godfathers, before it was christened Fort Rodman. The Colonel, Headquarters, and the remaining six companies proceeded by boat to various points along the North Shore, at some of which militia field artillery batteries had previously been on guard, the Colonel himself being stationed at Salem in command of the entire Essex County district. This transfer of troops was accomplished without peril or even discomfort. The 1st and 7th Companies under Maj. Dyar became the garrison at Salem; Maj. Quinby and the 2d Company were at Gloucester; the 6th Company was on Plum Island near Newburyport, and subsequently at Portsmouth; the 5th Company at Marblehead; and the 8th at Nahant as guard of the mining-casemate. Lieuts. E. Dwight Fullerton of the 8th Company and P. Frank Packard of the 2d were specially detailed to duty with the regulars at Fort Columbus, Governor’s Island, New York, and remained there several months. Lieut. Fullerton was called upon to untangle the snarl into which the War Department had gotten with regard to records of sick soldiers in the New York hospitals.
- 70. It fell to the lot of certain “batteries” to reconstruct and man ancient earthworks whose history ran back many years. At Salem, Fort Pickering was put in commission; at Gloucester, the old Stage Fort where Myles Standish once came near having a battle; near Portsmouth, Forts Constitution and McClary; and at Marblehead, Fort Sewall. This is very romantic to relate. No doubt the renovated works with their armament of obsolete field pieces could have afforded some protection against Spanish raiders. But those who were called upon to occupy works built for seventeenth, eighteenth and nineteenth century warfare, and modernize them so as to render them useful under twentieth century conditions, agree in testifying that the romance is all in the narrative and not any in the fact. The 6th Company had at first been stationed in an earthwork on the Plum Island beach which had been constructed by the field battery, whom they relieved; as Plum Island, in June, is notable chiefly for flies and fleas, this company was glad enough when the transfer to Portsmouth brought the men again on solid ground. Fort Constitution had a long history—it used to be known as Fort William and Mary, and from its ancient magazine came the powder used by the patriots at Bunker Hill; but in 1898 it was a comparatively modern work, and mounted a battery of eight-inch rifles. This Spanish War service is something of which the regiment are justly proud. On April 26, Col. Pfaff led 99 per cent. of the full militia strength of his command into the harbor forts, itself a conclusive demonstration that the National Guard is a dependable force. Foremost were they in the entire United States to assume their post of duty. First of all volunteers were they to be mustered in; the genius of “The Old First” was in control. Thruout the entire two-hundred-three days of duty they maintained the very highest standards of efficiency and discipline. It noway lessened the credit belonging to these volunteer soldiers that the Spaniards were so wise as to keep at a safe distance from the Massachusetts coast; the warmest kind of a welcome was awaiting them, had they come. When on Nov. 14, the command were mustered out of Federal
- 71. service and returned to the militia, they had added a most creditable chapter to the long annals of their organization. In 1899 a tour of duty was performed at Fort Rodman; and so satisfactory did it prove that the post was chosen for the annual coast defence exercises, with one exception, until 1906. In 1902 some companies were stationed at Fort Greble and other Rhode Island posts. The only serious objections to Rodman were the haze and fog, which hang low over Buzzard’s Bay. As a consequence of the Spanish War, the flannel shirt and the khaki suit became part of the regimental uniform. Oct. 14, 1899, the regiment participated in the ovation to Admiral George Dewey, and at the same time turned their Spanish War flags and colors over to the custody of the State. Col. Pfaff retired as Brigadier General Apr. 20, 1900. His loyal and generous interest in the old regiment has been shown in making possible the publication of this history. Col. James A. Frye, who commanded the regiment from May 4, 1900, until Jan. 4, 1906, had served as Major during the Spanish War. Upon relinquishing command of the regiment, he became Adj. Gen. of the State. Col. Frye was the one selected to record the services of the command during the Spanish War; and his history will always stand as a worthy monument to his memory. In 1903 the regiment participated in joint coast defence and naval maneuvers at Portland harbor, of which the chief feature was the long hours. The men were on duty all day and all night, so that sleeping almost became a forgotten art. On June 25, 1903, the Coast Artillery shared in the exercises of dedication around the magnificent statue of their old commander, Gen. Joseph Hooker. Members of the regiment had been foremost in securing the appropriation for the statue; and heartily did they rejoice to see the beautiful bronze by D. C. French which finally crowned their labor. 1903 witnessed the most important national militia legislation since the original militia act of 1792. By the “Dick law,” with amendments added in 1908, the militia really became a national
- 72. force, with clearly defined liability of service; and the name, National Guard, was officially conferred upon it. Nevertheless Massachusetts continued to call her citizen soldiers Volunteer Militia. 1904 brought the adoption of magazine-rifles. On Nov. 1, 1905, the regiment was redesignated as the “Corps of Coast Artillery,” a title which has been used by anticipation at various times in this book. Behind the change lay the fact that the War Department had been testing militia heavy or coast artillery; and the latter, in the estimation of the Washington authorities, were not found wanting. A regiment is a closely united body, and is supposed to operate as a unit. A corps, on the contrary, is a group of smaller units associated for administrative purposes, but acting more or less independently in warfare. Tactically a corps is not a unit; each of its members is. Inasmuch as few forts require so much as a full regiment of coast artillery to garrison them, it was deemed best to organize the artillery in smaller units, in companies, better suited to the needs of the average fort. Companies are combined in fort commands of two or more each. Moreover, by 1905, a clear distinction had arisen between coast artillery and heavy artillery; and it was necessary for organizations to decide which branch of the service they would choose. Heavy artillery follows a mobile army, and is used to batter down fortifications. Coast artillery mans the guns and submarine mines of our coast fortifications, and is not a mobile force. A moment’s consideration will convince anyone that the Massachusetts men chose the more exciting branch, when they became coast artillery. The heavy artillery fire from great distances, while themselves entirely out of range of any answering shots, and fire at fixed targets. The coast artillery fire at ships, moving targets possessing the ability to return our shots, who will certainly and quickly “get us” unless we “get them” first. An increase of interest in the scientific side of artillery work immediately followed, and stimulated every officer and enlisted man to do his best. Companies were no longer termed “batteries,” but were given numbers, the designations indicating seniority of charter. The band continued to wear the old regimental number “1” on their uniforms.
- 73. To the twelve companies of the Corps were, in 1907, assigned regular stations in the fortifications of Boston harbor, to which it would be their duty to repair at once in case of threatened hostilities. As they exercised each summer on the very guns which they would man in actual service, they grew familiar with their work to a degree never before possible. After experimenting at seven different posts, in 1913 the 1st, 2d, 3d, and 6th Companies became part of the garrison of Fort Strong on Long Island (named in honor of Gen. Wm. K. Strong); the 5th, 7th, 8th and 11th Companies were assigned to Fort Andrews; and the 4th, 9th, 10th and 12th Companies to Fort Warren. Col. Charles P. Nutter commanded the Corps from Jan. 23, 1906, until March 10, 1910; he had been Captain of the 7th Company during the Spanish War. In August, 1907, the companies participated in a general mobilization of militia at Boston in connection with the “old home week” celebration. The War Department now determined to make a slight change in the name of the organization, perhaps in the interest of alphabetic symmetry. Whatever the cause may have been, on Nov. 15, 1907, the words were transposed and the “Corps of Coast Artillery” became the “Coast Artillery Corps.” It had been so long since the Boston companies were called out to maintain public order at a great fire, that such a contingency was not regarded seriously. Suddenly, on April 12, 1908, as men were returning from Palm Sunday services, they received word that Chelsea was in the clutch of a mammoth conflagration. Vast clouds of smoke could be seen arising on the north-eastern horizon; Boston’s neighbor was indeed stricken. The 5th Company promptly responded to the call for help; but it was evident that assistance must come from outside; local forces were entirely inadequate to meet the emergency. At 5 p. m. the other companies were assembled at their armories; and at 8.30, after eating a hearty supper, they started for their posts of duty. The work was of the usual sort, rescuing property and saving lives, guarding
- 74. the property from vandals and thieves, and assisting the young, the weak and the aged to places of safety. Only men in uniform command confidence at such a season of disorder; only disciplined men, working together, can accomplish results. Right nobly did the Corps meet its responsibilities during its three days in Chelsea, and many a firm friend did it win for the organization. The 5th Company continued on duty five days longer. Upon the local company fell an especially cruel test. First, their new State armory came in the path of the flames and was swept to ruins—while the troops, on duty in the streets, were aware that their own civilian clothing in the lockers was going up in smoke. Worse yet, the fire spread until it involved the homes of many militiamen. The soldiers could hardly keep their thoughts on their work, while their own loved ones were in danger, and their own household effects in need of removal to places of safety; their minds wandered homeward—but the men themselves quietly kept their posts. There never has been any question about the discipline of the Corps in seasons of emergency; the 5th Company proved true to the ancient traditions. The Author
- 75. Col. George F. Quinby Col. E. Dwight Fullerton Page 151 Page 147 Companies of the Corps had been visiting Washington at inauguration time ever since 1835; and almost the entire command went in honor of T. Roosevelt in 1905; finally, in 1909, the Corps went as a regiment and participated in the inaugural parade of President William H. Taft. Participants in such a parade invite comparison between themselves and troops from many other states —military critics, such as Maj. Gen. J. Franklin Bell and Brig. Gen. E. M. Weaver, were unanimous in asserting that the Mass. Coast Artillery Corps and the West Point Cadets bore off the palm for fine military appearance, not even the N. Y. 7th doing as well. By 1909 the Corps had settled in its custom of holding coast defence exercises at the harbor forts; consequently, it was with disappointment and even resentment that they found themselves ordered to serve as infantry in the so-called Cape maneuvers in August of that year. A difference of opinion had arisen between the Adjutant General of Massachusetts and the Corps officers concerning money matters; and this tour of duty was laid on the latter as a penalty. Soldiers must obey orders; however irksome and unwelcome the service, no one in the “blue army” could truthfully
- 76. say that the “red-legged infantry” fell below their comrades in efficiency. Col. Walter E. Lombard was in command from March 17, 1910, until Feb. 21, 1915. At the latter date he became a Major General on the retired list. Col. Lombard had been Captain of the 6th Company during the Spanish War. In June, 1911, the War Department detailed a regular army officer to the Corps as Inspector-instructor, Capt. Russell P. Reeder being the first to perform that duty; at once the standards of instruction were improved, and the artillery work profited greatly from the presence of such a skilled teacher. Sergeant-instructors, four in number, were presently added as assistants to the commissioned officer who performed the chief duties. An immediate result of the Inspector-instructor’s work was the wonderful shooting done by the 4th, 12th and other companies during the 1911 tour of duty. After that date all officers were required to qualify in the technical part of their work by passing regular War Department examinations. The fourth officer to fill this detail, Capt. William H. Wilson, commenced service in Jan., 1915, and soon succeeded in systematizing the work of drill and instruction to a point far beyond anything previously attempted; so that his term of duty brought about a great increase of Corps efficiency. Capt. Wilson was especially qualified for this service in that he had himself been a National Guardsman, and had entered the U. S. army from a New York regiment. Capt. Wilson not only emphasized the artillery work; he also laid stress upon matters thitherto slighted,—company administration, higher infantry, and gunners’ instruction. Again in March, 1913, the entire Corps made its customary pilgrimage to Washington for the purpose of participating in the Presidential inauguration, this time paying the honor to Woodrow Wilson. As in 1909, so now, they were most enthusiastically praised for their fine military appearance and splendid marching. On May 30, 1913, the Gate City Guard of Atlanta, Ga., visited Boston as guests of the Tigers. 1913 was the fifteenth anniversary of the regiment’s
- 77. service in the Spanish war; and on Sept. 20, Col. Lombard tendered a review on the Common to the veterans. On that occasion active officers marched with the veterans, in the positions which they had filled fifteen years previously. Lt. Col. Woodman was in command of the veterans, and Col. Lombard marched as Captain of the 6th Company; while Maj. Shedd led the actives. After the parade, there was a collation, followed by motion pictures, in the Armory. So well had the 5th Company acquitted themselves at the Chelsea fire that they were one of the commands called out to maintain order at Salem when, on June 25, 1914, that ancient city was threatened with destruction; the emergency was similar to that of 1908. To the Chelsea men fell the duty of organizing a huge camp of refugees at Forest River park; and they remained in service seven days. Joseph Hooker was born Nov. 13, 1814, and exactly one hundred years later, his loyal admirers, among whom were numbered the officers of the Coast Artillery Corps, paraded, and participated in a great meeting at Tremont Temple in honor of his memory. Capt. Isaac P. Gragg, former Captain of the 1st Company, was always the prime mover in organizing celebrations in memory of Hooker, and he justly felt that the event of 1914 was the culmination of his life-work. Alas! Capt. Gragg did not long survive the centennial of his beloved commander. Edward Dwight Fullerton was elected Colonel Feb. 9, 1915, and continued in command until retired as Brigadier General, January 16, 1917; he had served as 1st Lieutenant of the 8th Company during the Spanish War. The “House of Governors” was in session at Boston in Aug., 1915, and Gov. David I. Walsh ordered a mobilization of the militia on Aug. 26, as a compliment to the State’s guests. As the authorized strength of the companies had recently been raised, the Boston papers commented upon the appearance of the Corps, in fifteen
- 78. platoons of twenty files, as “wonderful,” not only for numbers, but for steady marching. President Wilson called the militia out for service on the Mexican border June 18, 1916. Massachusetts shared with New Jersey the honor of placing her full quota of organizations at the post of danger in the shortest time; and since the Massachusetts quota was far larger than that of New Jersey, her record was the more creditable. On the ninth day after the troops were summoned to arms, they started for Texas. Of course the Coast Artillery could not be included in this great national mobilization, as they might not safely be spared from their stations at the forts. But on June 26, the day the mobile troops started south, the officers and non-commissioned officers of the Corps were assembled at the Framingham mobilization camp (“Camp Whitney”) for the purpose of drilling the hundreds of recruits there gathered. The officers and non- commissioned officers of the 6th Inf. also took part in this work of instruction. No recruits for Mass. regiments ever constituted a finer personnel than those eager to have a share in the Mexican service. Coming from all over the state, they were uniformly willing, sober, and quick to learn, in order that they might reach the front as soon as possible. The Corps became responsible for the “2d Provisional Regiment,” consisting of about one thousand men, destined for the 8th and 9th Inf. Regiments, and also for the cavalry, machine-guns, supply companies, field artillery, and even for the regimental bands. Wonderfully rapid progress was made, so that in two weeks, the recruits were equipped, and drilled, and ready to go forward. The Corps’ recent training in company administration stood them in good stead and made possible such rapid work. Certain officers of the Corps were drafted into the U. S. service, in order to accompany the recruits on the southward journey. With grave disorder on the Mexican border, and with the greatest war of the world’s history approaching its crisis abroad, conditions were once more favorable for Congressional action in behalf of the militia. Since threatenings of danger were loud and
- 79. insistent, the legislators were induced to take an additional forward step in rendering America’s citizen-soldiers efficient. The National Defence Act, as the new law was termed, completed the process of federalization by placing the militia fully under War Department control, and also provided a modest rate of remuneration for armory drills, thus making it an object for men to maintain regular attendance. Massachusetts had done what she could to encourage the passage of the law, by herself adopting, during the prolonged debate on the National Defence Act, a State law offering to hand over her militia to the Federal government. Indeed by her provision for remunerating men for attendance at rendezvous drills, the Commonwealth had taken her place beside Ohio five years previously as a pioneer in paying her militia. The legislation became effective on June 3, 1916, and went fully into operation on the first of the ensuing month. Right in the midst of their tour, on June 30, the officers and men were asked to take the new Federal oath, under provisions of this act. To the officers the oath was administered at Framingham, while the enlisted men were assembled in their armories that night, for the purpose of swearing in. Almost without exception, and then always with valid excuse, the members of the Corps assumed this additional obligation and became Federal soldiers. Headquarters, band, enlisted specialists, and twelve companies—the entire Corps—were, on June 30, recognized by the War Department as federalized National Guardsmen and were entered upon the U. S. payrolls. Of all the Massachusetts Volunteer Militia, the Coast Artillery Corps were the only organization to comply fully with the new requirements and be recognized as a unit. Companies of the Corps volunteered their services in connection with exhibitions for the benefit of the Mass. Volunteer Aid Association, which was raising funds to relieve distress amongst the families of National Guardsmen then at the border. An unusually fine military display was given at the ball-grounds in connection with a
- 80. benefit ball-game between the Red Sox and the St. Louis teams on July 17. Many Corps officers were detailed for recruiting duty during the summer and autumn of 1916, in an effort to raise the numbers of the regiments at the border to full war-strength. Consequently the coast defence exercises at the forts in August, 1916, were seriously handicapped. Many men were forced to perform double duty. In spite of this limitation, splendid artillery scores were made by the 2d, the 6th and other Companies, the 6th Company earning the coveted Knox trophy. Successive steps followed rapidly during the summer and autumn of 1916 to render effective the process of federalization. By order of Gov. Samuel W. McCall on July 17, the title “Massachusetts Volunteer Militia” was discontinued, and the force redesignated “National Guard, Massachusetts.” In October the War Department authorized the companies to increase their strength from seventy- eight to one hundred twelve officers and men; new regulations established standards of drill and instruction with which organizations must comply in order to qualify for pay; a National Guard reserve was created by transfer of men who had completed their three years of active service; promotion requirements were established for officers; and an assistant Inspector-instructor was detailed to the Corps, Capt. Hugh S. Brown taking his place beside Capt. Wilson. While the new National Guard regulations raised the standard and “tightened the reins,” it is a tribute to the high grade of efficiency already attained by the Corps that Federal control caused no revolutionary changes of method in the organization. As part of the federalizing process, on Dec. 9, 1916, the Militia Bureau of the War Department redesignated the command, and abolished the word Corps from its title. Thereafter it was the “Massachusetts Coast Artillery, National Guard.” On January 16, 1917, the organization received back its old and well-loved designation, and became the 1st Coast Defense Command, Massachusetts Coast Artillery, N. G.; once
- 81. Welcome to our website – the perfect destination for book lovers and knowledge seekers. We believe that every book holds a new world, offering opportunities for learning, discovery, and personal growth. That’s why we are dedicated to bringing you a diverse collection of books, ranging from classic literature and specialized publications to self-development guides and children's books. More than just a book-buying platform, we strive to be a bridge connecting you with timeless cultural and intellectual values. With an elegant, user-friendly interface and a smart search system, you can quickly find the books that best suit your interests. Additionally, our special promotions and home delivery services help you save time and fully enjoy the joy of reading. Join us on a journey of knowledge exploration, passion nurturing, and personal growth every day! ebookbell.com