Ines Slama, Badii Jouaber and Djamal Zeghlache
Wireless Networks and Multimedia Services Department, Institut National des Télécommunications
9, rue Charles Fourier - 91011 Evry, France
Keywords: Large-Scale Wireless Sensor Networks, multiple sinks, design parameters, energy efficiency, graph
Abstract: In this paper, we consider the multiple sinks placement problem in energy constrained large-scale Wireless
Sensor Networks (WSN). First, some fundamental design parameters in WSNs such as nodes deployment,
the network architecture, sink velocity and transmission range, are investigated. Each of these parameters is
analysed and discussed according to its influence on the energy consumption in a WSN. Second, a simple
and efficient approach for the placement of multiple sinks within large-scale WSNs is proposed. The
objective is to determine optimal sinks’ positions that maximize the network lifetime by reducing energy
consumption related to data transmissions from sensor nodes to different sinks. Balanced graph partitioning
techniques are used to split the entire WSN into connected sub-networks. Smaller sub-networks are created,
having similar characteristics and where energy consumption can be optimized independently but in the
same way. Therefore, different approaches and mechanisms that enhance the network lifetime in small-size
WSN can be deployed inside each sub-network. Performance results show that the proposed technique
significantly enhances the network lifetime.
A stationary sensor network is a static ad hoc
network composed of hundreds or thousands of
sensor nodes. Each sensor node is equipped with a
sensing device, a low computational capacity
processor, a short-range wireless transmitter-receiver
and a limited battery-supplied energy. Sensors
monitor some surrounding environmental
phenomenon, process the sensed data and forward it
towards a “close” a sink. This latter collects the data
from the different sensor nodes and transmits this
data to some remote control station where the data
will be exploited at the application level.
Achieving maximum lifetime in stationary
WSNs by optimally using the energy within sensor
nodes has been the subject of significant researches
in the last recent years. In this field, radio
transmission and reception operations are being
identified as the most energy consuming features.
On the other hand, the development of large-
scale sensor networks has drawn a lot of attention.
One of the main challenges is to set up new
architectures and mechanisms that can efficiently
scale up with the growing number of nodes that may
be required to ensure adequate coverage of large
areas of interest. At the same time, these new
architectures and mechanisms should maintain low
energy consumption per node so as to get by with
energy guaranty acceptable network lifetime.
Most of known existing protocols and
mechanisms are not scalable. They are mostly
conceived and adapted to relatively small networks
(i.e. reduced number of nodes) and/or when the
amount of data being gathered and transmitted is
small. In particular, centralised approaches, where
data from each sensor is sent to a central base
station, are not efficient and can not scale for large
wireless sensor networks.
The use of multiple mobile base stations is one
possible solution for large-scale WSNs. The idea is
to shorten the path (distance) between each sensor
node and the nearest base station, leading to save
energy consumption for transmission operations. To
achieve this efficiency, the multiple base stations
should be optimally placed within the sensed area.
Slama I., Jouaber B. and Zeghlache D. (2008).
In Proceedings of the International Conference on Wireless Information Networks and Systems, pages 29-37
DOI: 10.5220/0002024700290037
Our contributions in this paper include:
1. We present an interesting analysis of the
fundamental design parameters in large scale WSNs
with multiple, mobile sinks. A brief related work of
the existing techniques of multiple sinks deployment
in WSNs is also provided.
2. We propose to use graph theory techniques and in
particular graph partitioning in order to determine a
balanced partition of large-scale WSNs and then to
optimize the placement of the different sinks over
the obtained smaller sub-networks to minimize the
energy consumed for data transmissions.
The remaining of this paper is organized as
After discussing the design parameters analysis
and related work in section 2, the proposed scheme
is described and explained in section 3. Simulations,
performance results and analysis are presented and
discussed in section 4. Finally, concluding remarks
are given in section 5.
In the following, we propose to discuss and analyse
some fundamental techniques and parameters that
should be seriously taken into account when
designing WSNs. We show through the analysis
presented below how important is their impact and
influence on the performance of such networks and
the way they should be investigated to face the
energy consumption challenge in WSNs.
2.1 Network Architecture for WSNs
Scalability of methods, algorithms and protocols
used in WSNs mostly depends on the
interconnection topologies of sensor nodes. Two
main architectures are proposed and studied in the
literature: Hierarchical and Flat topologies.
Hierarchical topologies allow easier scalable
mechanisms. A hierarchical (multi-tiered)
architecture comprising multiple tiers is depicted in
The lowest tier: represented by a dense
deployment of low cost and low power static
sensor devices. Each sensor is equipped with a
micro controller, a flash memory and a radio.
Their main task is to collect information about
specific phenomena and send it to a higher tier.
The middle tier: formed by mobile sinks which
have significant computation, memory and
storage resources and no power constraints.
These mobile sinks act as relays for information
The highest tier: it is the application which
queries the sensor network through a query
interface and the final information fusion point
that provides to the manager the data in interest.
Another example of hierarchical topologies is the
clustered two tier architecture where a single cluster
head handles several member nodes in its
neighbourhood (i.e. its cluster). The cluster heads
form a separate top layer communication structure.
Network protocols designed for these architectures
are highly scalable. However, they require the
definition of specific roles and mechanisms for
cluster heads as well as specific signalling
As opposed to these hierarchical topologies, flat
distributed topologies are easier to deploy but more
difficult to scale. Here, nodes are connected in a
complete ad-hoc fashion. All sensing nodes have
equivalent roles with no specific hierarchy between
them. The main advantages of flat topologies are
their easy deployment and reduced cost. However,
such a topology are difficult to scale up since
communications between thousands or perhaps
millions of nodes in a ad hoc fashion lead to
degraded performances and hence higher energy
consumption. For instance, routing protocols are a
prominent factor of the scalability of sensor
networks. In recent researches, the proposed routing
protocols require that some of the sensors have
knowledge of the topology of the entire network at
every point in time. This requires a lot of signalling
and do not scale well with a high number of nodes.
Different solutions are proposed in the literature to
overcome these weaknesses. For instance, a
distributed protocol for large-scale WSNs is
proposed in (Tilak, 2003). It is based on localized
interactions and does not require global knowledge
such as the current network topology. In
(Grossglauser, 2001), authors proposed to use
specific mobility patterns in order to achieve higher
capacity in large scale WSNs.
In these studies, the evaluation of the scalability
of the proposed protocols is mainly based on a well
known metric for WSNs which is the network
lifetime. The objective is to avoid significant
degradations of the network lifetime when the
number of nodes composing the WSN increases.
WINSYS 2008 - International Conference on Wireless Information Networks and Systems
Through the analysis of the above sited studies, it
appears that no single architecture can be adopted to
face the scalability issues. Different solutions should
be envisaged since the performances of the WSN
may also depend on the type of applications for
which they have been conceived.
Figure 1: Architecture of a multi-tiered wireless sensor
2.2 Nodes Deployment
It is important to consider the distribution of the
sensors within the phenomena area according to the
needs of the application. We can distinguish three
deployment strategies. The self organization feature
of sensors makes possible to deploy them randomly
over the observed area. In such a case, sensors might
be distributed in a random uniform way, like
dropped from an aircraft. The sensors can also be
regularly deployed in the field as placed on a
predetermined geometric grid. Finally, they can be
placed in a planned manner in order to provide
higher sensor density in a region where the
phenomenon is concentrated.
Random deployment is usually preferred because
more realistic but regular deployment can offer in
some specific cases some advantages. In fact, many
applications require sensors location information to
achieve the desired functionality and since using a
GPS system is not a feasible solution, we choose to
place the sensors over a grid points to have a prior
knowledge of sensors location information. In some
other researches, location estimation techniques are
proposed to get over the problem (Bulusu, 2000),
(Doherty, 2001), (Savvides, 2002), (Nasipuri, 2002).
2.3 Energy Model
Energy consumption in a sensor node has in general
the following components:
a) Sensing energy:
It represents the energy consumed when
activating the sensing circuitry and collecting data
from the environment. The amount of this energy
depends on the task that is assigned to the sensor.
b) Transmission energy:
It represents the energy consumed by the
transmitter and the receiver circuitry of the sensor.
The energy consumption due to transmissions
between node i and node j can be modelled as
(i,j) is the energy consumed at node i when
transmitting to node j with a bit rate equal to r
(j,i) is the energy consumed at node j when
receiving from node i with a bit rate equal to r
is the distance between node i and node j.
dba +=
c) Computation energy
It represents the energy consumed to activate the
sensor’s processing unit in order to operate the
transmitter and receiver circuitry.
Compared to the transmissions energy, the
sensing and computing energies are relatively low.
2.4 Single Hop vs. Multi Hop
When the sensor nodes use single hop
communication, each node sends its data directly to
the sink. In such case, the sensor nodes located
farthest from the sink have to spend the maximum
amount of energy and hence are the first to be dead
whereas in multi-hop communication, the nodes
located closest from the sink are the first to be dead
because they have to relay the farther nodes data.
Moreover, since the communication is directly
between the sensor nodes and the sink, only one
node transmits at the same time and then a
contention less MAC is used.
Multi-hop communication has been vastly
favoured over long-range single-hop links to provide
a large coverage area. The use of multi-hop is
essentially to combat the rapid decay of the received
signal strength as communication distance increases.
However, multi-hop transmission is not always
better than single-hop transmission in perspective of
energy conservation. In fact, we knew from previous
works that it depends on the distance between the
sensor and the sink node, the number of hop and
each hop distance. A careful investigation ((Chen,
2006), (Bhardwaj, 2001), (Mhatre, 2004)) has
proved that the energy consumption rate is
minimized only when using an optimal hop number
of identical hop distances named characteristic
. We can write
dDN /=
(D the distance between the
source and the sink node). Consequently, the most
energy efficient scheme is to use single hop if the
distance between the sensor and the sink is no
greater than d
, else multi-hop with hop distance
of d
2.5 Multiple Mobile Sinks
2.5.1 Why Multiple Sinks?
In recent researches, energy efficient usages of
multiple and/or mobile sinks to increase the network
lifetime were proposed (Gandham, 2003), (Kim,
2005), (Oyman, 2004), (Vincze, 2006). The idea
behind this is to decrease the distance between each
sensor node and the nearest sink. In fact, when a
higher number of sinks are distributed within the
WSN, the path lengths from any sensor node to its
nearest sink is decreased leading to lower energy
consumption and therefore to higher network
However the cost of a sink is more expensive
than the sensor and then the number of sink nodes is
financially constrained. In (Oyman, 2004), the
authors proposed to find the minimum number of
sinks while maximizing the network lifetime by
connecting the budget reserved for the sink nodes
with the lifetime of the sensor nodes.
2.5.2 Why Moving Sink?
In a wireless sensor network where a multi-hop
communication is used, the nodes which are one hop
from the sink drain their energy faster than other
nodes because they have to relay messages
originating from many other nodes in addition to
delivering their own messages. In the case of one-
hop communication, the nodes send directly their
messages to the sink and the nodes farthest from the
sink are the first to drain their energy. In doing so
and in the both cases, many sensor nodes will
become quickly unable to communicate with the
base station and the network becomes inoperational.
Several researches have then demonstrated
analytically and with experimental results that using
multiple mobile sinks increases the network lifetime
(Vincze, 2006), (Luo, 2005).
We note that the sink trajectory can be rather
controlled by the application; it can be mounted on a
remote controlled robot and can be moved from one
point to an other like in (Gandham, 2003). Else, it
can follow a specific mobility model in which case
an estimation of its position can be computed like in
(Chen, 2006).
2.5.3 Existing Approaches
Deploying multiple mobile base stations in WSNs
has been investigated in a dynamically growing
number of papers. We mention hereafter some of the
most relevant.
In (Luo, 2005), the authors have developed an
analytical model that describes the communication
load distribution in WSNs and proved that base
station mobility is a strategy that deserves to be
considered when optimizing the network lifetime.
They have further shown that the optimum
movement strategy for a mobile base station is to
follow the periphery when the deployment area is
Network lifetime elongation using mobile base
station has also been investigated in (Wang, 2005).
The author gave a novel linear programming
formulation for the joint problem of determining the
movement of the sink and the sojourn time at
different points in the network. The simulations have
shown that lifetime maximizing solutions are
achieved by nonuniform sojourn time distributions
among grid points depending on the shape of the
deployment area.
In (Gandham, 2003), authors propose to divide
time into rounds and to dynamically relocate
multiple sinks, at different positions along the
periphery of the sensed field, at the beginning of
each of these rounds. An integer linear program is
used to determine the new locations of the different
base stations. Results have shown that the energy
consumption of individual sensors is better balanced
and the overall energy consumption of all sensors is
minimized. In (Kim, 2005), authors propose another
approach to find the optimal locations of multiple
stationary sink nodes. The proposed scheme allows
sensor nodes to communicate with one or multiple
sinks through multiple paths in order to improve the
network lifetime. In (Oyman, 2004), authors claim
that finding the optimal placement for a given
WINSYS 2008 - International Conference on Wireless Information Networks and Systems
number of sinks is equivalent to the clustering
problem and should be solved using a clustering
algorithm. Another approach to solve the problem of
multiple mobile base station placements is proposed
in (Vincze, 2006). An electrostatic model is applied
to determine sinks’ locations and to coordinates the
movements of these sinks considering the network
Unfortunately, most of the above strategies are
proposed and evaluated over small to medium size
wireless sensor networks (typically less than 100
nodes). For large scale wireless sensor networks,
where hundreds or thousands of nodes can be
deployed, the placement of multiple sinks still
requires advanced studies. For instance, and as
illustrated on Fig. 2, if we consider the case where
the sinks are located along the periphery as stated in
(Gandham, 2003), the paths between each node and
its nearest sink is relatively short when the number
of nodes is limited. However, the more the area size
increases and/or the number of nodes within it
increases, the longer this path is and the shorter the
sensor nodes lifetime will be.
2.5.4 Sink Velocity Influence
A mobile sink can move in two different regimes
(Luo, 2006), a fast mobility regime and a slow
mobility regime. In the fast mobility regime, the sink
moves in a continuous form with a velocity v along
the time without any stop or pause in a particular
position. In the slow mobility regime, the sink
moves in a discrete form and the sink’s trajectory is
a sequence of anchor points between which the sink
moves with a velocity v and at which it pauses
during a period of time (epoch). The slow mobility
regime is considered more realistic and is adopted in
a big number of researches.
However, it is very important to carefully choose
the value of the sink velocity. In fact, when the
mobile sink velocity is high, the sink will more
frequently change its position and visit more the
different regions over the area in interest during the
network lifetime. Therefore, the energy consumption
is efficiently distributed over the sensors and the
network lifetime extended. This can be much more
efficient in the particular case where the sensors
buffer the data sensed and wait until sink approaches
to deliver it (Chen, 2006) which reduces
unnecessarily packet forwarding actions since
sensors are sure of sink arrival before loosing the
data (because of buffer size limitation or packet
deadline expiration). Besides, the high speed moving
sink produces a tolerable data delivery delay
especially in the case of fast mobility regime, which
can be very important for some specific applications.
However, the mobile sink high velocity can have
negative effects. In fact, it can make the session
interval too short to successfully exchange a long
data packet and hence the packet loss rate will
increase. In slow mobility regime, it is preferred that
the epoch be long enough to guaranty long messages
(a) Single sink in a small network
(b) Multiple sinks
in a small network
(c) Multiple sinks in a large scale network
Figure 2: Multiple sink placement.
2.6 Overhead Problem
Using a moving sink to collect the information in a
static wireless sensor network is a strategy that has
been adopted by several researchers because of its
efficiency in increasing the network lifetime.
However, it seems obvious that the mobility of the
sink will inevitably incur additional overhead in data
exchanges since the nodes will continuously need to
be informed of the sink location. This can be much
more dramatic when the number of sensors is very
large. In (Luo, 2006), the authors evaluated the
performances of MobiRoute, a routing protocol that
supports sink mobility. They proved that when using
a slow mobility regime with an epoch much longer
than the sink moving time, the overhead introduced
by the mobility of the sink became negligible
because amortized across a long epoch. Hence, to
avoid that the overhead offsets the benefits brought
by sink mobility, such a scenario should be adopted.
2.7 Buffering Data
In a static wireless sensor network where the sink is
mobile, the sensors can send the sensed data
immediately to the sink by direct transmission if the
sink is in the transmission range of the sensor or by
multi-hop communication over the different relay
sensors. In some proposals like (Chen, 2006),
(Chakrabarti, 2003) and (Rahul, 2003), the sensors
hold the data they sense in their buffer until they
encounter the mobile sink. The authors proved that it
represents the most economic way in terms of
energy. However, if the time between two
successive sensor-sink meetings is too long, packets
loss is expected because of the buffer size limitation
or packet deadline expiration as well as an
intolerable delay of data delivery. Hence, an optimal
choice of the value for these parameters is very
important to guaranty the best results in terms of
network lifetime.
2.8 Transmission Range Influence
In wireless sensor networks with moving sinks, the
transmission range is one of the fundamental design
parameter. In fact, depending on if it increases or
decreases, the network topology changes since the
number of one hop neighbours of a sensor as well as
the path length from a sensor to the sink varies
(Youssef, 2002). Besides, when the transmission
range increases, the connectivity of the network
increases and hence the network lifetime is
improved. However, there is a threshold of the
transmission range value over which the
connectivity of the network is no more influenced
whereas the energy consumption becomes dominant
resulting in the network lifetime degradation
(Gandham, 2003). This threshold obviously depends
on the other parameters of the network (number of
sinks, size of the network, transmission energy
model, the initial energy of each sensor…).
Analytical studies in (Chen, 2006) have also shown
that the transmission range can also influence the
data delivery delay, the sensor-sink meeting delay
and the outage or unsuccessful packet transmission
probability. The lager it is, the lower they are.
Through the analysis presented above, it is
obvious that deploying multiple, mobile sinks in
WSNs efficiently reduces the energy consumption
level and further lengthens the network lifetime.
However, as stated before, most of the existing
solutions are appropriate and efficient over small to
medium size wireless sensor networks (typically less
than 100 nodes). Therefore, further investigations
should be conducted in order to optimize multiple
mobile sink placement in large scale wireless sensor
networks, where hundreds or thousands of nodes can
be deployed. Moreover, the fundamental design
parameters discussed previously and which have a
serious influence on the energy consumption should
be carefully chosen in order to let them contribute in
optimizing the network behaviour and extending its
lifetime duration.
We propose in this work to enhance sink placement
in large scale WSNs. An intuitively appropriate
solution is to decompose the underlying sensor
network and then optimize energy usage in each of
the sub-networks independently. The objective is to
take advantage of the powerful and efficient sink
placement techniques proposed for small scale
WSNs. In order to apply these techniques over large
scale WSNs, we propose to first divide the network
into sub-networks according to specific criteria. An
adequate sink placement technique can then be
applied independently within each of the defined
Graph partitioning is a promising approach to
split a large sensor network into balanced sub-
networks. In practice, different criteria can be
considered in order to partition a large scale wireless
sensor network. One simple objective is to create
balanced sub-networks (in terms of number of
sensors) that group the sensors according to their
neighbourhood. This allows creating smaller sub-
networks with similar characteristics that can be
easily optimized, independently but in the same way.
In graph theory related literature, different
approaches and techniques are proposed for
balanced graph partitioning.
3.1 Existing Graph Partitioning
In (Even, 1997), a fast approximate graph
partitioning algorithm is proposed. The authors
unified the problems of b-balanced cuts and k-
multiway separators using a new approach called
minimum capacity ρ-separators. They studied the
graph partitioning problems on graphs with edge
capacities and vertex weights and described a simple
approximation algorithm for minimum capacity ρ-
separators leading to a fast approximation algorithm
both for b-balanced cuts and k-multiway separators.
They define a ρ-separator as a sub-set of edges
whose removal partitions the vertex set into
connected components such that the sum of the
WINSYS 2008 - International Conference on Wireless Information Networks and Systems
vertex weights in each component is at most ρ times
the weight of the graph. In (Ito, 2006), authors
considered three problems to find an (l, u)-partition
of a given graph. They proposed to partition a graph
G into connected components by deleting some
edges from G making the total weight of each
component equal at least to l and at most to u. The
minimum partition problem is to find an (l, u)-
partition with the minimum number of components,
the maximum partition problem is defined in the
same way and the p-partition problem is to find an
(l, u)-partition with a fixed number p of components.
Authors proved that the three problems are NP-
complete or NP-hard. In (Chlebikova, 1996), authors
studied the approximation of the Maximally
Balanced Connected Partition problem (MBCP).
They first presented the optimization problem that
finds the maximally balanced connected partition for
a graph G. It results in a partition (V
, V
) of V
composed of disjoint sets V
and V
such that both
sub-graphs of G induced by V
and V
connected, and maximize an objective function
“balance”, B
, V
) = min(w(V
), w(V
Authors proved that the problem is NP-hard.
In this work, this last approach will be adapted
and applied to large scale Wireless Sensor
Networks. Our choice is mainly motivated by the
practical approach provided in (Chlebikova, 1996)
and based on the use of a polynomial-time algorithm
that gives an approximate solution.
In the following the Maximally Balanced
Connected Partition (MBCP) technique (Chlebikova,
1996) is adapted and formulated for partitioning a
large WSN. A corresponding approximate resolution
algorithm is then presented.
3.2 Model Formulation
Assume that G = (V, E) is a connected graph where
V is a set of nodes and E is the set of all links
connecting two nodes of V.
In our case, V represents the set of sensors and E
represents the set of all links connecting two sensors
belonging to V.
The objective is to partition G into connected
balanced sub-graphs (in terms of number of nodes).
We assume that all sensors have the same initial
To achieve this objective, let w be a non-
negative vertex-weight function representing the
balancing criteria. In this case, w will reflect the
number of nodes. Hence w(V’) = |V’|.
This MBCP problem can then be formulated as
Maximize B
, V
) = min ( w(V
), w(V
Subject to
1. (V
, V
) is a partition of V into
nonempty disjoints sets V
and V
that sub-graphs of G induced by V
are connected.
The resolution of this model will result into two
balanced sub-networks. Each of them can be
partitioned again using the same process.
This partitioning technique should be applied as
much as required according to the targeted size for
the sub-networks and taking into account the number
of available sinks to be placed. The final result
should be 2
equivalent smaller sub-networks where
n is the number of partitioning iterations.
3.3 Problem Resolution
To solve this model, we used the polynomial
approximation algorithm presented in (Chlebikova,
1996) that finds an approximate solution for the
MBCP problem.
In order to select neighbouring sensors within the
same sub-networks, we adapted the algorithm by
sorting the list of candidates for each partition
according to their distance (vicinity).
The algorithm can be written as follow:
Input: G = (V, E).
V= {v
, v
, v
… v
} where
0. Initialize V
}, V
= V\V
such v
node near the periphery.
1. If | V
| >= 1/2 | V| then Step 3
else Step 2.
2 .Let V
= {u Є V / (V
U {u}, V
is a connected partition of G}.
Choose u of V
such that u the closest
element to V
If |u| < |V| - 2|V
then V
:= V
U {u}, V
:= V
\ {u},
Step 1
else Step 3
3. Return (V
, V
The effect of the proposed partitioning technique on
the WSN lifetime is investigated using numerical
simulations over Matlab environment. A circular
large scale wireless sensor network, with a radius R
= 500m is considered. 1000 nodes are randomly
(uniformly) deployed over the network area. Sensors
are similar with a communication range r= 80m and
an initial energy of 1000J unit. The cost of sending
and receiving operations is 1mJ per packet. Sinks are
assumed to have no energy constraints because they
have larger batteries or their batteries are
rechargeable. Sensors communicate with the sinks in
a multi-hop manner. We assumed that the shortest
path routing algorithm is used to find the shortest
route to the sink. The network lifetime is defined as
the moment at which the first sensor runs out of
energy. Time is divided into rounds. Each round is
composed of T=100 timeframes. Each sensor node
generates one data packet every timeframe.
To evaluate the efficiency of the proposed graph
partitioning technique in elongating the network
lifetime, the following scenario is considered:
Comparative Scenario:
Case 1: The entire network is considered. N sinks
are deployed randomly on the periphery of the
network. Then, the sinks start to move along the
periphery. In one round each sink moved 60 m.
Case 2: The graph partitioning algorithm is used to
define N smaller sub-networks. One single sink is
randomly deployed on the periphery of each sub
network. Then each sink moves 60m each round on
the periphery.
number of sinks
lifetime duration (rounds)
Figure 3: The network lifetime in the comparative
Several simulations are then run to compare the
network lifetime in the two different cases of
described scenario.
Simulation results are presented in figure 3. First,
let’s notice that the simple use of multiple sinks
enhances the network lifetime (with and without
partitioning). Indeed, the network lifetime increases
proportionally to the number of sinks because the
distance between the sensors and their correspondent
sinks decreases.
Second, enhancements of the network lifetime
can be observed in the case of partitioned large-scale
WSNs compared to non partitioned ones. This was
expected as when one sink is moving along the
periphery of each sub-network, the energy
consumption is obviously much more distributed
over the sensors than when all the sinks are moving
along the periphery of the whole network. The nodes
that are the closest to the sinks are logically the ones
who die first because they not only send their own
data but also relay the data of all the nodes in the
network. In this scenario, the nodes who die first in
the case of non partitioned network are the nodes
situated all along the periphery whereas in the case
of partitioned network, they are the ones situated
along the peripheries of the different sub-networks.
Then, using the graph partitioning technique to
deploy the sinks distributes the load relay and
decreases the average distance between the sensors
and the sinks. Indeed, the improvement of the
network lifetime of the partitioned network is much
more important when the number of sinks (or sub-
networks) increases.
For interested readers, other comparative
scenarios are investigated and provided with all
details in (Slama, 2008).
The use of multiple sinks in large scale wireless
sensor networks is necessary in order to cover large
areas and to minimize energy consumption for data
transmission operations. In this paper, we discussed
the fundamental design parameters considered in
WSNs with multiple sinks and which have a
considerable influence on the energy consumption.
We have shown that they should be carefully chosen
in order to let them contribute in optimizing the
network behaviour and extending its lifetime
duration. We also proposed the use of graph
partitioning techniques to obtain smaller and
balanced sub-networks over which existing sink
placement techniques that are optimized for small to
medium scale WSNs can be used.
Performance results show that the proposed
technique considerably enhances the network
lifetime particularly when the sinks are moving
along the periphery.
This first step using graph partitioning approach
to improve energy consumption in large-scale WSNs
is promising. We will focus in complementary and
future work on more elaborated approaches for
optimal multiple sinks placement and WSN
partitioning. In addition, efficient tools should be
proposed to determine the optimal number of
WINSYS 2008 - International Conference on Wireless Information Networks and Systems
partitions and sinks to be used according to the WSN
characteristics, applications’ requirements and
financial costs.
Tilak, S., Murphy, A. & Heinzelman, W., 2003. Non-
Uniform Information Dissemination for Sensor
Networks. In ICNP '03, Proceedings of the
International Conference on Network Protocols.
Grossglauser, M. & Tse, D., 2001. Mobility increases the
capacity of ad hoc wireless networks. ACM\IEEE
Transactions on Networking; 10(4):477-486.
Bulusu, N., Heidemann, J. & Estrin, D., 2000. GPS-less
Low Cost Outdoor Localization for Very Small
Devices. IEEE Personal Communications Magazine,
Vol. 7, No. 5, pp. 28-34.
Doherty, L., Pister, K. S. J. & Ghaoui, L. E., 2001.
Convex Position Estimation in Wireless Sensor
Networks. In INFOCOM’01, Proceedings of the 20th
Annual Joint Conference of the IEEE Computer and
Communications Societies pp. 1655-1663, Anchorage,
Alaska, USA.
Savvides, A., Park, H., & Srivastava, M. B., 2002. The
Bits and Flops of the N-hop Multilateration Primitive
for Node Localization Problems. In ACM WSNA’02,
Proceedings of the First ACM International Workshop
on Wireless Sensor Networks and Applications, pp.
112-121, Atlanta, USA.
Nasipuri, A., & Li, K., 2002. A Directionality based
Location Discovery Scheme for Wireless Sensor
Networks. In ACM WSNA’02, Proceedings of the First
ACM International Workshop on Wireless Sensor
Networks and Applications, pp. 105-111, Atlanta,
Chen, C., Ma, J., & Yu, K., 2006. Designing Energy-
Efficient Wireless Sensor Networks with Mobile
Sinks. In WSW’06 at Sensys’06, Boulder, Colorado,
Bhardwaj, M., et al., 2001. Upper Bounds on the Lifetime
of Sensor Networks. In ICC’01, Proceedings of the
IEEE international conference on communications,
pp. 785 - 790, Helsinki, Finland.
Mhatre, V. & Rosenberg, C., 2004. Design guidelines for
wireless sensor networks: communication, clustering
and aggregation. In Elsevier Ad Hoc Networks, pp.
45-63, issue 2, 2004.
Gandham, S. R. & al., 2003. Energy Efficient Schemes for
Wireless Sensor Networks With Multiple Mobile Base
Stations. In GLOBECOM’03, Proceeding of the IEEE
Global Communication conference.
Kim, H., Seok, Y., Choi, N., Choi, Y. & Kwon, T., 2005.
Optimal Multi-sink Positioning and Energy-efficient
Routing in Wireless Sensor Networks. In LNCS’05,
Lecture Notes in Computer Science, Vol.3391, pp.264-
Oyman, E. I. & Ersoy, C., 2004. Multiple Sink Network
Design Problem in Large Scale Wireless Sensor
Networks. In ICC’04, Proceeding of the IEE
international conference on communications.
Vincze, K., Fodor, R., Vida, and Vidacs, A., 2006.
Electrostatic Modelling of Multiple Mobile Sinks in
Wireless Sensor Networks. In proceeding of IFIP
Networking Workshop on Performance Control in
Wireless Sensor Networks, pp. 30-37, Coimbra,
Luo, J. & Hubaux, J.-P., 2005. Joint Mobility and Routing
for Lifetime Elongation in Wireless Sensor Networks.
In IEEE Infocom’05, Proceeding of the 28th
Conference on Computer Communications.
Wang, Z. M. & al., 2005. Exploiting Sink Mobility for
Maximizing Sensor Networks Lifetime. In Proceeding
of the 38th Hawaii Intel. Conf. Sys. Sci.
Luo, J., Panchard, J., Piorkowski, M., Grosglausser, M. &
Hubaux, J-P., 2006. Mobiroute: Routing towards a
Mobile Sink for Improving Lifetime in Sensor
Networks. In DCOSS'06, International Conference on
Distributed Computing in Sensor Systems (Systems
Chakrabarti, & al., 2003. Predictable Observer Mobility
for Power Efficient Design of Sensor Networks. In
IPSN’03, Proceedings of ACM/IEEE International
Conference on Information Processing in Sensor
Networks Palo Alto, California, USA.
Rahul, C., Shah, & al., 2003. Data MULEs: Modeling a
Three-tier Architecture for Sparse Sensor Networks. In
Elsevier Ad Hoc Networks, vol. 1, issues 2-3, pp. 215-
Youssef, M.A., Younis, M.F. & Arisha, K.A., 2002. A
Constrained Shortest- Path Energy-Aware Routing
Algorithm for Wireless Sensor Networks. In Wireless
Communication and Networking Conference, 2:794 –
Even, G., Naor, J., Rao, S., & Schieber. B., 1997. Fast
approximate graph partitioning algorithms. In Proc.
8th Ann. ACM-SIAM Symp. on Discrete Algorithms,
ACM-SIAM, pages 639 – 648, 1997.
Ito, T., Zhou, X. & Nishizeki, T., 2006. Partitioning a
graph of bounded tree-width to connected subgraphs
of almost uniform size. Journal of discrete algorithms,
Volume 4, Issue 1, March 2006, Pages 142-154.
Chlebikova, J., 1996. Approximability of the Maximally
balanced connected partition problem in graphs.
Inform. Process. Lett., 60:225 – 230, 1996.
Slama, I., Jouaber, B. & Zeghlache, D., 2008. Energy
Efficient Scheme for Large Scale Wireless Sensor
Networks with Multiple Sinks”, In WCNC’08, Proc.
IEEE Wireless Communications and Networking
conference, Pages 2367-2372.