Analysis of Processing Architectures for Wireless Sensor Networks
Ijeoma Okeke, Alastair Allen, David Hendry and Fabio Verdicchio
School of Engineering, University of Aberdeen, Aberdeen, Scotland, U.K.
Keywords:
Energy Analysis, Distributed Processing, Architecture, Fairness, Network Lifetime, Wireless Sensor Net-
works.
Abstract:
Wireless Sensor Networks (WSN) are networks of low-cost communication devices with sensing and compu-
tational capabilities enabling remote, real-time measurement, monitoring and control of divers physical and
environmental parameters. As WSNs are typically battery powered, energy-aware techniques are critical for
extending its lifetime. Aside from energy-efficient communication protocols, distributed processing strategies
are being explored whereby,computational capabilities of sensor nodes are utilised to locally process sensed
data in order to reduce communication cost. However, as local processing increases, the impact of processing
energy cost becomes significant creating a need to analyse WSNs under this emergent scenario as previous
work have focused mostly on communication cost. We analysed the energy cost for WSN under different
processing architectures. We used a fairness metric to quantify the fairness of energy cost distribution in the
network. Our results showed a positive correlation between fairness and network lifetime. Hence, we ar-
gue that local processing can be exploited to reduce transmission and improve system performance without
adversely reducing network lifetime. We conclude that although local processing marginally increases node
energy consumption, it improves overall network life time as energy cost is evenly distributed in the network.
Moreover, it enhances network maintenance as nodes have similar lifetimes.
1 INTRODUCTION
Wireless Sensor Networks (WSN) have been applied
in a variety of application areas especially where
wired technologies are impracticable such as agricul-
ture (Baggio, 2005), health (Jafari et al., 2005), en-
vironmental monitoring (Othman and Shazali, 2012)
and infrastructural monitoring (Wang et al., 2007).
Although the success of WSN have been demon-
strated in these varied areas, a major challenge lim-
iting its widespread adoption is energy-efficiency.
Early WSN applications were mainly used for re-
mote data harvesting therefore, existing work focused
mainly on developing energy-efficient communica-
tion protocols (Pantazis et al., 2013) due to the high
energy cost associated with radio communication on
sensor nodes. Recently, WSNs are being used beyond
simple monitoring and data harvesting applications.
For instance, on-chip processing capabilities of sen-
sors are being exploited for energy-aware distributed
processing (Chong et al., 2011). Distributed process-
ing improves the latency of system response in wire-
less sensor and actuator networks because event de-
tection and control decisions can be performed faster,
locally. It also reduces transmission of large sized
raw-data thereby reducing communication cost. Ex-
isting work have focused mostly on the energy cost
of communication activities in WSN but less atten-
tion have been given to analysing the cost of dis-
tributed processing as we have done in this work. We
analyse local and global energy costs for our mod-
els and quantify energy efficiency in terms of a fair-
ness metric. We discuss our results and show that the
advantages of distributed processing in WSN can be
explored without significantly affecting network life-
time.
2 PROCESSING ARCHITECTURE
2.1 Centralised Architecture
A centralised processing architecture is shown in fig-
ure 1. In this case, processing actions such as filtering
and classification are performed on a single central
node called the sink. The leaf nodes only sense and
transmit raw data to the central node for use in pro-
cessing as shown. For all the network diagrams in
this paper, the nodes with reduced functionality are
Okeke, I., Allen, A., Hendry, D. and Verdicchio, F.
Analysis of Processing Architectures for Wireless Sensor Networks.
DOI: 10.5220/0005735701290136
In Proceedings of the 5th Inter national Confererence on Sensor Networks (SENSORNETS 2016), pages 129-136
ISBN: 978-989-758-169-4
Copyright
c
2016 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
129
represented by a circle while nodes with processing
function are denoted by a square.
…
…
…
…
Figure 1: Centralised Processing.
2.2 Distributed Architecture
2.2.1 Node Level
Node level distributed architecture is represented as
shown in figure 2.
…
…
…
…
Figure 2: Node Level Distributed Processing.
In this case, processing is purely distributed with
every sensor node performing localised processing
before transmitting to the sink.
2.2.2 Network Level
In the network level architecture, processing activ-
ities are performed locally on a subset of nodes
(sometimes called cluster heads) which cooperatively
achieve the network’s objective as shown in figure 3.
Figure 3: Network Level Distributed Processing.
Typical monitoring and control objectives require
sensors to cooperate to achieve results. Hence, the
need to investigate the prospects of using energy-
constrained sensor nodes as a distributed processing
resource.
3 ENERGY COST MODELS
The energy cost of sensing is usually the same for ev-
ery node irrespective of it’s role in the network. How-
ever, communication cost depends on the communi-
cation distance between the source node and the sink
node as well as on the size of data packets. Similarly,
processing cost is dependent on the energy costs as-
sociated with switching and leakage currents in the
processor circuit.
The consensus of existing literature on energy
consumption models for WSNs is that radio commu-
nication is the most energy consuming activity for
wireless sensors due to high energy dissipation by
the wireless radio. Therefore, several communica-
tion protocols have been proposed for energy-efficient
communication in WSN. However, most of these pro-
tocols are very small incremental improvements of an
existing protocol based on the radio energy model
proposed in (Heinzelman et al., 2002; Heinzelman
et al., 2000).
Recently, a more comprehensive energy model
presented by (Halgamuge et al., 2009) which ac-
counted for seven different aspects of sensor energy
consumption activities showed that the results ob-
tained using the former model were over-estimated
as the model focused mainly on the communication
energy cost. Thus, instead of assuming a fixed value
for processing energy as most previous work did, pro-
cessing energy, was expressed in terms of switch-
ing and leakage current losses thereby accounting
for other factors which were unaccounted for in pre-
existing models. These two models provided the basis
for further investigation into the effect of increased
distributed in-network processing on the lifetime of
WSNs.
3.1 Communication Energy Cost Model
The communication energy cost E
comm
, is defined as
the energy required to communicate nbits of data over
a communication distance d as follows:
E
comm
= E
T
+ E
R
(1)
where, E
T
and E
R
is the cost of transmitting and re-
ceiving data respectively. E
T
and E
R
is computed
from equations 2 and 3 as follow:
E
T
= E
T
e
(n) + E
T
a
(n,d)
= n∗ (E
elect
+ ε
mp
∗ d
4
)
(2)
and
E
R
= n∗ E
elect
(3)
SENSORNETS 2016 - 5th International Conference on Sensor Networks
130
where, E
T
e
is energy consumed by transmission elec-
tronics, E
T
a
is energy consumed by amplifier electron-
ics; E
elect
is energy consumed by device electronics;
ε
mp
is the distance based energy loss constant.
3.2 Processing Energy Cost Model
The energy cost of processing, E
process
, is defined as
the energy cost of processing n bits of data and is
given by equation 4.
E
process
(n,N
clock
) =
nN
clock
C
avg
V
2
s
|
{z }
switching current
+nV
s
N
clock
f
I
leak
e
V
s
pV
t
|
{z }
leakage current
(4)
where, n remains the data size in bits; N
clock
is the
number of clock cycles required to complete a sin-
gle processing task; C
avg
is the average capacitance
switched per cycle; V
s
is the supply voltage; f is the
processor frequency; I
leak
is the leakage current; p is
a constant depending on the processor and V
t
is the
thermal voltage of the processing circuit.
3.3 Sensor Life Time
We define network lifetime as the time taken for the
first node to die (i.e. run out of battery power) (Chang
and Tassiulas, 2004). A node’s role within the net-
work determines how much energy it expends thereby
affecting its lifetime.
Given that a sensor node has an initial energy in
Joules; the lifetime of the sensor (L
t
) can be estimated
using equation 5.
L
t
=
E
init
E
rmax
∗ T (5)
where E
init
is the initial energy associated with the
node, E
rmax
is the maximum node energy cost for one
event cycle and T is the time in seconds taken to com-
plete one event cycle.
Equation 5 can be used to estimate network life-
time prior to deployment providing useful informa-
tion for planning network maintenance.
3.4 The Fairness Metric
Jain’s fairness index (Jain et al., 1998) is a perfor-
mance criteria used in resource allocation schemes
for shared resources however, in this context, we de-
fine it as a global metric which is a measure of the
equity of energy cost distribution among nodes in a
sensor network. The fairness value lies between 0 &
1 with higher values corresponding to fairer distribu-
tions. The fairness index, F is given by equation 6.
F(x
1
,x
2
,...,x
s
) =
(
∑
s
i=1
x
i
)
2
s.
∑
s
i=1
x
2
i
(6)
where x
i
is computed as
x
i
=
E
round
i
E
global
(7)
s represents the number of nodes in the network;
E
round
i
is the local energy cost for node i and E
global
the network energy cost for one event cycle respec-
tively. A high fairness index implies that energy cost
is evenly spread among the nodes with a more uni-
form lifetime while a low index value shows that only
a few nodes are overburdened by the energy cost of
the system.
4 ENERGY COST ANALYSIS
In this section, we present the energy cost analysis us-
ing our energy cost models. Five scenarios denoted as
S1-S5 representing the main processing architectures
used in literature were defined. For simplicity, we fo-
cused only on communication and processing costs as
we assumed other node energy costs to be fairly con-
stant for all network nodes.
We provide analytical expressions for the local
and global energy cost functions for the different net-
work architectures. The local energy cost captures the
energy expended by a single node in the network and
is a function of the node’s role within the network and
the communication model used while the global en-
ergy cost captures the total energy expended by all
nodes in the network after each complete event cycle.
We classify node functions into : basic roles (com-
mon to all nodes e.g. sensing, logging, communi-
cation) and non-basic roles (nodes can posses either
or none e.g. processing, control, actuation). Sen-
sor nodes can operate either in the full-function (FF)
mode or the reduced-function (RF) mode. RF nodes
perform only basic roles and are represented by a cir-
cle shape while FF nodes perform one or more non-
basic roles in addition to basic roles and are repre-
sented by a square shape.
We carry out analysis and simulation using MAT-
LAB for the different architectures to identify the
best network architecture for distributed processing in
terms of energy-efficiency. Our analysis was based on
the following assumptions:
1. Every data processing activity takes equal number
of clock cycles so that N
clock
is equal to a constant
value (Halgamuge et al., 2009).
Analysis of Processing Architectures for Wireless Sensor Networks
131
2. Carrier access is by TDMA and all node functions
are completed within a full cycle.
3. The packet size n, is equal to a maximum of 133
bytes (1064 bits); 114 bytes for data, 19 bytes for
control over head (Wang et al., 2006).
4. The communication distance between nodes
d
comm
is the same across scenarios.
5. The energy cost of device electronics E
elect
= 50nJ/bit while ∈
mp
= 0.0013pJ/bit/m4
(Heinzelman et al., 2000).
6. Other parameters were substituted for using the
values obtained from data sheets as shown in table
1 (Halgamuge et al., 2009).
Table 1: Table of Parameters for Energy Cost Analysis.
Symbol Description Value
N
clock
clock cycles per task 0.97∗ 10
6
C
avg
average capacitance 22pF
V
s
supply voltage 2.7V
f sensor frequency 191.42 MHz
I
leak
leakage current 1.196mA
V
t
thermal voltage 0.2V
p processor constant 21.26
n size of data packet 1064 bits
d
comm
communication distance 100m
4.1 S1: Centralised Processing with
Direct Communication
The S1 scenario represents the simple sense and send
network design with only direct data communication
allowed between the source and sink. Data processing
occurs centrally at the sink nodeas illustrated in figure
4. All nodes are assumed to be equally spaced from
the sink hence the communication distance d
comm
is
equal for all the nodes in the network. The local en-
ergy costs for S1 is given in equations (8) and (9).
…
…
…
…
Figure 4: S1 Network Diagram.
(i) sink node
E
Snode
=
s−1
∑
i=1
E
R
i
+ E
process
(8)
(ii) other nodes (s-1)
E
node
= E
T
(9)
where E
node
and E
Snode
are source node and sink node
local energy costs respectively.
The global energy costs for S1 is also given by
equations (10)-(12).
E
commT
=
s−1
∑
i=1
E
R
i
+
s−1
∑
i=1
E
T
i
(10)
E
processT
= E
process
(11)
E
global
=
s−1
∑
i=1
E
R
i
+
s−1
∑
i=1
E
T
i
+ E
process
(12)
where E
global
is the total energy cost for the net-
work and E
comm
T
, E
process
T
are global communication
and processing energy costs respectively. The same
nomenclature is assumed for all scenarios.
4.2 S2: Centralised Processing with
Multi-hop Communication
The S2 scenario is similar to S1 but with multi-hop
communication. This represents network architec-
tures used for monitoring long horizontal structures
such as water or oil pipelines and bridges with only 1-
D, multi-hop communication possiblebetween source
and sink nodes. Processing activity such as aggrega-
tion or filtering takes place only at the sink node as
shown in figure 5.
…
Figure 5: S2 Network Diagram.
Equal spacing is assumed between adjacent nodes so
that the communication distance d
comm
is equal for all
nodes. Thus, the local energy costs for S2 is given
by equations (13) - (15) where E
fnode
, is local energy
cost for the first node.
(i) sink node
E
Snode
= E
R
+ E
process
(13)
(ii) first node
E
fnode
= E
T
(14)
(iii) other nodes (s-2)
E
node
= E
T
+ E
R
(15)
Similarly, the global energy costs for S2 is given by
equations (16) - (18).
E
commT
=
s−1
∑
i=1
E
R
i
+
s−1
∑
i=1
E
T
i
(16)
SENSORNETS 2016 - 5th International Conference on Sensor Networks
132
E
processT
= E
process
(17)
E
global
=
s−1
∑
i=1
E
R
i
+
s−1
∑
i=1
E
T
i
+ E
process
(18)
4.3 S3: Decentralised Processing with
Direct Communication
In S3 scenario, processing occurs in a distributed
manner at individual nodes. Additionally, every node
communicates directly with the sink node by a single
hop as shown in figure 6.
…
…
…
…
Figure 6: S3 Network Diagram.
In this case, a single network objective function
maybe divided into smaller tasks which could be per-
formed in a distributed or cooperative manner within
the network. The sink node is then responsible for
aggregating the final results as shown in the network
diagram. Again, the communication distance d
comm
,
is constant as each node is assumed equally spaced
from the sink node for direct communication.
The local energy costs for S3 is given by equations
(19) and (20).
(i) sink node
E
Snode
=
s−1
∑
i=1
E
R
i
+ E
process
(19)
(ii) other nodes (s-1)
E
node
= E
T
+ E
process
(20)
Similarly, the global energy costs for S3 is given
by equations (21) - (23).
E
commT
=
s−1
∑
i=1
E
R
i
+
s−1
∑
i=1
E
T
i
(21)
E
processT
=
s
∑
i=1
E
process
i
(22)
E
global
=
s−1
∑
i=1
E
R
i
+
s−1
∑
i=1
E
T
i
+
s
∑
i=1
E
process
i
(23)
4.4 S4: Decentralised Processing with
Multi-hop Communication
Scenario S4 is similar to the distributed architecture in
S3 but with 1D-multi-hop communication as shown
in figure 7.
Figure 7: S4 Network Diagram.
The nodes are equally spaced apart so that com-
munication distance between adjacent nodes is equal
to d
comm
. Therefore, the local energy costs for S4 is
given by equations (24) - (26).
(i) sink node
E
Snode
= E
R
+ E
process
(24)
(ii) first node
E
fnode
= E
process
+ E
T
(25)
(iii) other nodes (s-2)
E
node
= E
R
+ E
T
+ E
process
(26)
Similarly, the global energy costs for S4 is given
by equations (27) - (29).
E
commT
=
s−1
∑
i=1
E
R
i
+
s−1
∑
i=1
E
T
i
(27)
E
processT
=
s
∑
i=1
E
process
i
(28)
E
global
=
s−1
∑
i=1
E
R
i
+
s−1
∑
i=1
E
T
i
+
s
∑
i=1
E
process
i
(29)
4.5 S5: Decentralised Processing with
Cluster Heads
Scenario S5 represents the processing architecture
obtained in dense WSNs. A dense network is di-
vided into smaller sub-networks called clusters with
a coordinating node in each cluster called a cluster-
head(CH). The end nodes operate in the RF mode and
communicates with the CH by a single hop. The CHs
operate in the FF mode with a single hop communi-
cation to the sink as illustrated in figure 8.
The communication distance within clusters, d
sc
,
is the distance between a leaf node and its CH and is
assumed equal for all leaf nodes. The cluster heads
are also assumed to be equally spaced within the net-
work from the sink node with a communication dis-
tance equal to d
comm
. The number of nodes within
Analysis of Processing Architectures for Wireless Sensor Networks
133
Figure 8: S5 Network Diagram.
a cluster S
npc
, is given by
s
S
clust
; where S
clust
is the
number of desired clusters within the network and s
remains the network size. Similarly, the number of
clusters can be obtained as S
clust
=
s
S
npc
. Hence, the
values of S
npc
and S
clust
can be easily adjusted de-
pending on the desired network parameters.
The local energy costs for S5 is given by equations
(30) - (32).
(i) cluster head nodes (S
clust
)
E
CH
=
S
npc
−1
∑
i=1
E
R
i
+ E
T
+ E
process
(30)
(ii) other nodes [s− S
clust
− 1]
E
node
= E
T
(31)
(iii) sink node
E
Snode
=
S
clust
∑
i=1
E
R
i
+ E
process
(32)
where E
CH
, E
Snode
and E
node
corresponds to local en-
ergy cost for cluster heads, sink node and leaf nodes
respectively. The global energy costs for S5 are given
by equations (33) - (35).
E
commT
= S
clust
"
S
npc
−1
∑
i=1
E
R
i
+ E
T
#
+
"
s−S
clust
−1
∑
i=1
E
T
i
#
+
S
clust
∑
i=1
E
R
i
(33)
E
processT
=
S
clust
+1
∑
i=1
E
process
i
(34)
E
global
= S
clust
"
S
npc
−1
∑
i=1
E
R
i
+ E
T
#
+
S
clust
∑
i=1
E
R
i
+
"
s−S
clust
−1
∑
i=1
E
T
i
#
+
S
clust
+1
∑
i=1
E
process
i
(35)
5 RESULTS AND DISCUSSION
In this section, we present the results of our energy
cost analysis for the different processing architectures
S1-S5. The local and global energy costs were com-
puted separately for the five scenarios in MATLAB. A
network size of s = 100 and communication distance
d
comm
= 100m was used. The fairness metric for each
scenario was computed as a function of the local and
global energy costs. The results section which fol-
lows next, presents and discusses the results of our
analysis in terms of network lifetime and fairness for
distributed processing architectures in WSNs.
5.1 Local Analysis
The local analysis results are shown in table 2. In the
central processing architectures (S1 & S2), the max-
imum energy cost corresponds to the energy cost of
the sink node which operates in FF mode. For the
distributed architectures (S3 & S4), the maximum lo-
cal energy cost is nearly the same for all nodes in the
network because all the nodes operate in FF mode
thereby resulting in a higher local energy cost per
node. In the cluster scenario S5, the maximum local
energy is expended by CHs which run in FF mode.
Table 2: Local analysis for S1-S5, s = 100 nodes.
Metric S1 S2 S3 S4 S5
Max. Energy (J) 0.204 0.198 0.204 0.199 0.199
Min. Energy (J) 0.000192 0.000192 0.199 0.198 0.000192
The maximum local energy expended by a single
node is ∼ 0.2J for all scenarios while the minimum
varies depending on the degree of distributed process-
ing performed in the network as shown in table 2. The
communication and processing costs is dependent on
packet size n. Thus, a sensor node operating in FF
mode expends 99.91% more energy than a node oper-
ating in RF mode for a given n as shown by the min-
imum and maximum local energy costs. This implies
that n has a high impact on processing energy cost
which is not negligible and so, must be accounted for
in robust energy models.
5.2 Global Analysis
The global communication cost E
comm
T
, remained
fairly constant while global processing cost E
process
T
,
varied across the scenarios as shown in table 3.
This is because our network model restricts com-
munication between source and sink to a single hop in
the first four scenarios S1- S4, irrespective of the po-
sition and role of the node in the network. However,
SENSORNETS 2016 - 5th International Conference on Sensor Networks
134
Table 3: Global analysis for S1-S5, s = 100 nodes.
E
comm
T
(J)
0.0242 0.0242 0.0242 0.0242 0.0243
E
process
T
(J)
0.198 0.198 19.838 19.838 4.166
E
global
(J) 0.223 0.223 19.862 19.862 4.19
Fairness Index 0.0119 0.0126 0.999 0.999 0.212
we observe an increase in communication cost for the
cluster model S5 where the sink node is 2 hops away
from the source. The purely distributed processing ar-
chitectures gave the highest global energy cost across
the scenarios. However, restricting processing to only
a subset of nodes reduced the global energy cost by
∼ 75% in the S5 scenario. This is because distributed
processing increases local energy cost resulting in a
higher global cost.
5.3 Fairness Analysis
The fairness metric gives an indication of the degree
of uniformity of energy cost distribution among the
network nodes. From table 3, we show that central
processing architectures results in an unfair distribu-
tion of energy cost among network nodes as only a
small fraction of the nodes operate in the FF mode
while the rest operate in the RF mode. On the other
hand, distributed processing architectures result in a
fairer distribution of energy cost as most of the nodes
are operating in FF mode. Our results so far show that
although increased processing activity marginally in-
creases the global energy cost, it does not affect the
maximum local cost. However, the cluster model S5,
is a preferred architecture as it combines the advan-
tages of a lower global energy cost and higher fair-
ness index associated with the centralised and purely
distributed architectures respectively.
5.4 Network Lifetime
If we assume a fixed initial amount of energy, E
init
, to
be available to every node in the network, the over-
all network lifetime which is determined by the max-
imum local cost is not significantly affected by dis-
tributed processing since the maximum cost is same
across scenarios (∼ 0.2J). This provides a basis for
exploiting on-node processing capabilities of wireless
sensor nodes for distributed and localised processing
in relevant application areas where they provide the
advantage of improved network performance. Based
on our results, we also argue that the global energy
cost is not a good indication of the lifetime of the
network because it does not provide adequate infor-
mation on the energy cost associated with individual
nodes. As WSNs are made up of individual nodes, lo-
cal energy cost provides a better metric for analysis as
it provides information on the energy costs associated
with every nodes in the network which is critical for
estimating network life time.
5.5 Effect of Network Density on
Energy Cost and Fairness
The effect of network density on energy costs and
fairness for S1-S5 is shown in table 4.
Table 4: Effect of Network Size on Global Energy Cost and
Fairness.
S Model Proc. cost Comm. cost Total cost Fairness
100 S1 0.198379572 0.02422728 0.22260685 0.011947738
S2 0.198379572 0.02422728 0.22260685 0.012583027
S3 19.83795725 0.02422728 19.8621845 0.999993536
S4 19.83795725 0.02422728 19.8621845 0.99999999
S5 4.165971022 0.02428048 4.1902515 0.211523937
500 S1 0.198379572 0.12211528 0.32049485 0.004059142
S2 0.198379572 0.12211528 0.32049485 0.00521334
S3 99.18978625 0.12211528 99.3119015 0.999964859
S4 99.18978625 0.12211528 99.3119015 0.999999998
S5 20.03633682 0.12216848 20.1585053 0.203538262
1000 S1 0.198379572 0.24447528 0.44285485 0.003098163
S2 0.198379572 0.24447528 0.44285485 0.004973211
S3 198.3795725 0.24447528 198.624048 0.999928995
S4 198.3795725 0.24447528 198.624048 0.999999999
S5 39.87429407 0.24452848 40.1188226 0.202538755
In general, our results show a marginal increase
in global energy cost as the network size increases
with the distributed processing scenarios showing the
highest values as before due to higher processing
cost associated with each node. This is not sur-
prising as higher number of nodes means more en-
ergy cost per node. However, we note that local en-
ergy cost was unaffected by network size because al-
though the network size increased, individual node
roles did not change hence local energy cost remained
unchanged. Processing cost remained constant for
scenarios S1 and S2 as the network size increased be-
cause the number of nodes in FF mode remained un-
changed however, scenarios S3-S5 experienced sig-
nificant increase because the number of nodes in FF
mode increased with network size. Communication
cost remained unchanged for scenarios S1-S4 due to
the single-hop communication maintained as the net-
work size increased. However, for the S5 scenario,
the multi-hop communication resulted in higher com-
munication energy cost at CHs for higher network
sizes. The fairness of centralised processing de-
creased as the network size increased but increased
for distributed processing. This was because of the
widening gap between the minimum and maximum
local energy costs as the network size increased.
Analysis of Processing Architectures for Wireless Sensor Networks
135
6 CONCLUSION
In this work, we studied the effect of processing ar-
chitectures on energy cost in WSNs. We performed
local and global cost analysis for 5 configurations and
compared results across scenarios. We showed that
a higher number of FF nodes results in higher global
cost but the maximum local cost remains unaffected.
We identified a positive correlation between network
lifetime and fairness and argue that a marginal in-
crease in global cost does not automatically corre-
spond to a lower network lifetime. Rather, due to the
higher fairness associated with distributed process-
ing, network designers could exploit the on-node pro-
cessing capabilities of sensors for performance im-
provement during the fixed lifetime of the network.
Uniform energy cost distribution aids the planning
of scheduled network maintenance as every node in
the network has a similar lifetime. Thus, our results
support the continued exploration of energy-efficient
strategies for in-network processing in WSNs with
energy constraints.
ACKNOWLEDGEMENTS
Ijeoma Okeke is a commonwealth scholar (PhD)
funded by the UK government.
REFERENCES
Baggio, A. (2005). Wireless sensor networks in precision
agriculture. In ACM Workshop on Real-World Wire-
less Sensor Networks (REALWSN 2005), Stockholm,
Sweden.
Chang, J.-H. and Tassiulas, L. (2004). Maximum life-
time routing in wireless sensor networks. Networking,
IEEE/ACM Transactions on, 12(4):609–619. ID: 16.
Chong, S. K., Gaber, M. M., Krishnaswamy, S., and
Loke, S. W. (2011). Energy-aware data processing
techniques for wireless sensor networks: a review,
pages 117–137. Transactions on large-scale data-and
knowledge-centered systems III. Springer.
Halgamuge, M. N., Zukerman, M., Ramamohanarao, K.,
and Vu, H. L. (2009). An estimation of sensor energy
consumption. Progress In Electromagnetics Research
B, (12):259–295. Cited By :34.
Heinzelman, W. B., Chandrakasan, P., and Balakrishnan, H.
(2002). An application-specific protocol architecture
for wireless microsensor networks. IEEE Trans Wirel
Commun, 1(4):660–670.
Heinzelman, W. R., Chandrakasan, A., and Balakrishnan,
H. (2000). Energy-efficient communication protocol
for wireless microsensor networks. In System Sci-
ences, 2000. Proceedings of the 33rd Annual Hawaii
International Conference on, page 10 pp. vol.2. ID: 1.
Jafari, R., Encarnacao, A., Zahoory, A., Dabiri, F., Noshadi,
H., and Sarrafzadeh, M. (2005). Wireless sensor net-
works for health monitoring. In Mobile and Ubiqui-
tous Systems: Networking and Services, 2005. Mo-
biQuitous 2005. The Second Annual International
Conference on, pages 479–481. IEEE.
Jain, R., Chiu, D.-M., and Hawe, W. (1998). A quantitative
measure of fairness and discrimination for resource al-
location in shared computer systems.
Othman, M. F. and Shazali, K. (2012). Wireless sensor
network applications: A study in environment mon-
itoring system. In Procedia Engineering, volume 41,
pages 1204–1210. Cited By :21.
Pantazis, N. A., Nikolidakis, S. A., and Vergados, D. D.
(2013). Energy-efficient routing protocols in wireless
sensor networks: A survey. IEEE Communications
Surveys and Tutorials, 15(2):551–591. Cited By :34.
Wang, M., Cao, J., Chen, B., Xu, Y., and Li, J. (2007).
Distributed processing in wireless sensor networks for
structural health monitoring, volume 4611 LNCS of
Lecture Notes in Computer Science (including sub-
series Lecture Notes in Artificial Intelligence and Lec-
ture Notes in Bioinformatics). Cited By :5.
Wang, Q., Hempstead, M., and Yang, W. (2006). A re-
alistic power consumption model for wireless sensor
network devices. In Sensor and Ad Hoc Communi-
cations and Networks, 2006. SECON ’06. 2006 3rd
Annual IEEE Communications Society on, volume 1,
pages 286–295. ID: 10.
SENSORNETS 2016 - 5th International Conference on Sensor Networks
136
