Wireless Sensor Network, 2009, 1, 257-267
doi:10.4236/wsn.2009.14032 Published Online November 2009 (http://www.scirp.org/journal/wsn).
Wireless Sensor Network Management and
Functionality: An Overview
Dimitrios GEORGOUL AS, Keith BLOW
Adptive Communication Networks Research Group,
EE, Aston University, Aston Triangle, B47ET, United Kingdom
Received March 23, 2009; revised May 5, 2 009; accepted June 12, 2009
Abstract
Sensor networks are dense wireless networks of small, low-cost sensors, which collect and disseminate en-
vironmental data. Wireless sensor networks facilitate monitoring and controlling of physical environments
from remote locations with better accuracy. They have applications in a variety of fields such as environ-
mental monitoring; military purposes and gathering sensing information in inhospitable locations. Sensor
nodes have various energy and computational constraints because of their inexpensive nature and adhoc
method of deployment. Considerable research has been focused at overcoming these deficiencies through
more energy efficient routing, localization algorithms and system design. Our survey presents the funda-
mentals of wireless sensor network, thus providing the necessary background required for understanding the
organization, functionality and limitations of those networks. The middleware solution is also investigated
through a critical presentation and analysis of some of the most well established approaches.
Keywords: Wireless Sensor Networks, Organization, Functionality, Middleware
1. Introduction
Wireless sensor networks have been identified as one of
most important technologies for the 21st century [1]. As
technologies advance and hardware prices drop, wireless
sensor networks will find more prosperous ground to
spread in areas where tradition al networks are inadequ ate.
The foundational conc ept which applies in a vast number
of networks can be identified through the simple notion:
Sensing Capabilities plus CPU Power plus Radio Trans-
mission equals a powerful framework for deploying
thousands of potential app lications.
However, this notion is underlined by some complex
and detailed understanding of each separate network
components capabilities and limitations as well as under-
standing in areas of modern network management and
distributed systems theory.
The primary goal of wireless sensor networks is to
make useful measurements for as long as possible. To do
this it is essential to minimize en ergy use b y reducing the
amount of communication between nodes without sacri-
ficing useful data transmission. Each node is designed in
an interconnected web that will grow upon the deploy-
ment in mind. Wireless sensor networks are highly dy-
namic and susceptible to network failures, mainly be-
cause of the physically harsh environments that they are
deployed in and connectivity interruptions [2].
To make the wireless sensor network dream a reality,
an architecture must be developed that will monitor and
control the node communication in order to optimize and
maintain the performance of the ne twork, ensure that the
network operates properly and control/instruct a set of
cluster nodes without human intervention [3].
In order to develop a system architecture with the
above characteristics, we focus explicitly on the functions
and the roles of wireless network management systems.
Additionally, we present the middleware concept as a
novel solution to the limitations that wireless sensor net-
works inhabit. A number of network systems are pre-
sented, critical reviewed and categorized.
2. Network Management Systems
Around 1980s computer networks began to grow and be
interconnected in a large scale. This growth produced
problems in maintaining and managing those networks,
thus the need of network management was realized. To-
day, networks are far more dynamic and interconnected

258 D. GEORGOULAS ET AL.
than before, especially in the area of wireless sensor
networks, thus a managing infrastructure is one of the
most basic requirement for monitoring and controlling
such networks [4].
A network management system can be defined as a
system with the ability to monitor and control a network
from a central location. Ideally there are four key func-
tional areas that this system must support [5]:
1) Fault Management: This area provides the facilities
that allow the discovery of any kind of faults that the
managed devices of the network will produce, determin-
ing in parallel the possible causes of such errors. Thus,
the fault management function should provide mecha-
nisms for error detection, correction, log reports and di-
agnostics preferably without the user interference.
2) Configuration Management: Responsible for moni-
toring the entire network configuration information and
also having access to all the managed devices in terms of
reconfigure, operate and shut down if necessary.
3) Performance Management: Responsible for meas-
uring the network performance through analysis of sta-
tistical data about the system so that it may be main-
tained at an acceptable level.
4) Security Management: This area provides all those
facilities that will ensure that access to network resources
can not be obtained without the proper authorization. In
order to do so, it provides mechanisms for limiting the
access to network resources and provides the end user
with notifications of security breeches and attempts.
Those four functional areas of network management
are far more challenging and vital for a network that will
consist of tiny sensors which can be supplied to a spe-
cific environment running applications such as habitat
monitoring, microclimate research, medical care and
structural monitoring [6]. For every sensor network
application, the network is presented as a distributed
system consisting of many autonomous nodes that co-
operate and coordinate their actions based on a prede-
fined architecture. Every node is assigned with a specific
role inside the network such as data acquisition and
processing. Also, nodes can be used as data aggregation
and caching points in order to redu ce the communication
overhead [7].
3. The System Organization of a Sensor
Network System
The organization of sensor network systems is based
upon the approach that they will adapt in order to moni-
tor and control the state of the wireless sensor network.
There are four predominant approaches [8]:
 Passive Monitoring: The system role is to collect
data during the lifetime of the network. The data
will identify the state of the network in different
time intervals without any action taking place dur-
ing the data gathering. An analysis of the data will
take place in later stages.
 Fault Detection Monitoring: The system dedicates
its resources to identifying faults and errors during
the lifetime of the network. All the information is
gathered and reported back to the operator whose
responsibility is to correct those problems in later
stages. No action is taken by the system towards the
resolution of those problems in real time.
 Reactive Monitoring: The system has a double role
to accomplish during the lifetime of the network.
Firstly, as we identified in the pr evious approaches,
the collection of data that will provide information
about the states of the network, is the main role.
This time though, the system will be eligible to
identify and detect any events and act upon them in
real time mainly by altering the parameters of the
fixed asset under its control.
 Proactive Monitoring: The system collects and
analyzes all the incoming data concerned with the
state of the network. Then an analysis is taking
place similar to the one of the reactive monitoring
with the big difference that certain events, de-
scribed by the collected information, are stored. The
system is then able to maintain better available
network performance by predicting future events
based on past o nes.
Wireless sensor systems can be categorized according
to their architecture which can be centralized, hierarchi-
cal or distributed [9]. The centralized one identifies the
role of the base station as the most important one in the
whole architecture. The base station will collect the in-
formation from all the nodes and will monitor and con-
trol the entire network. Benefits to this architecture can
be found in areas of processing power and decision
making. A base station with unlimited power resources
can perform complex analysis of data and process a vari-
ety of information, reducing the weight of this energy
consuming task from the nodes of the network.
The distributed architecture focuses on the deployment
of multiple manager stations across the network usually
in a web based format. Thus, each substation can coor-
dinate its actions and co-operate based on knowledge
that it can acquire from a neighboring substations. In
this approach, the communication cost is less than the
centralized one and more energy efficient since all the
workload will be distributed evenly across the network.
However, due to the scalability and complexity of wire-
less sensor networks it is proven quite difficult to man-
age and quite expensive in terms of memory cost.
The hybrid between the centralized and distributed
approach is that of the hierarchical one. In this architec-
ture we have the existence of substations in the network
but this time no communication is allowed between them.
The design tends to be cluster based, with the heads of
the cluster be responsible for a set of network nodes in

D. GEORGOULAS ET AL. 259
terms of processing and transmitting information. All the
cluster heads will report back to a single base station. work. Figure 1 is demonstrating this classification.
4. The Functionality of a Sensor Network
System
The main functionality of sensor network systems is based
on the theory behind network management systems, thus is
focusing on two attributes those of monitor and control. In
this section, we classify some well known sensor systems
in terms of the functionality they provide inside the net- Figure 1. The functionalities of wireless sensor networks in
different categories.
Figure 2. The BOSS architecture, song et al 2005.

260 D. GEORGOULAS ET AL.
Two characteristic examples of wireless sensor net-
works that are based on traditional management sys-
tems are those of MANNA [10] and BOSS [11].
MANNA provides a general architecture for managing
a wireless sensor network by using a multidimensional
plane for the functional, physical and the informational
architecture of the network. The functional plane is
responsible for the configuration of the application
specific entities, the information plane is object ori-
ented and specifies all the syntax and semantics that
will be exchanged between the entities of the network
and lastly the physical plane establishes, according to
the available protocols profiles, the communication
interfaces for the management entities that will be pre-
sent inside the network.
The BOSS architecture, Figure 2, is based on the tra-
ditional method of the standard service discovery pro-
tocol, UpnP. With the UpnP protocol the user does not
need to self configure the network and devices in the
network automatically will be discovered. However,
due to computational power consumption required by
the devices and memory space allocation limitations,
the protocol itself is not suitable for tiny sensor devices.
BOSS architecture is overcoming this problem by act-
ing as a mediator between UpnP networks and sensor
nodes. In order to do that, it combines four different
components: service manager, control manager, service
table and a sensor network management service, under
the same framework.
Routing protocols is another alternative way of moni-
toring and controlling a wireless network when they get
embedded in an application with examples such as
LEACH [12] a nd GAF [1 3].
LEACH is a routing protocol for users that want re-
motely to monitor an environment. The protocol is build
upon two foundational assumptions. The first one ac-
knowledges that the base station is at a fixed poin t and in
a far distance from the network nodes and the second one
assumes that all nodes in the network are homogeneous
and energy constrained. In order to maximize the system
lifetime and coverage, LEACH is using a set of methods
such as distributed cluster formation with randomized
selection of cluster heads and local processing. LEACH
dynamic clustering method, splits time in fixed intervals
with equal length. Also, it does not allow clusters and
cluster heads to be at a fixed point inside the network.
LEACH, dictates that once other sensors of the network
receive a message they will join a cluster with the
stronger signal cluster head.
GAF, which stands for geographic adaptive fidelity,
focuses its architecture on the extension of the lifetime of
the network by exploiting node redundancy. This node
redundancy is achieved by switching off unnecessary
sensor nodes in the network without any effect on the
level of routing fidelity.
Figure 3. The GAF nodes state transitions, xu et al 2001.
GAF recognizes three transition states for the nodes,
Figure 3, active, sleeping and discovery. Initially all
nodes in the network are in a discovery state. This means
that all nodes will turn their radio on and exchange dis-
covery messages in order to identify neighbor nodes in
the same grid. When a node is active it will set a timeout,
Ta, that will determine for how long it will stay in that
state before it returns back to the discovery state. While
active, the node periodically re-broadcasts its discovery
message at time intervals, Td. The sleeping state is regu-
lated by a time interval Ts which is dependent upon the
application. GAF assumes that sensor nodes can identify
their location in the forming virtual grid with the use of
GPS cards.
Systems, such as WinMS [14] and Sympathy [15] fo-
cus more on the importance of fault detection in a wire-
less sensor network. WinMS uses a novel management
function, called systematic resource transfer, in order to
provide automatic self-configuration and self-stabiliza-
tion both locally and globally for the given wireless sen-
sor network. This function allows the network, in case of
a failure, to have a predetermined period of time where
nodes will listen to their environment activities and self-
configure. No prior knowledge of the topology of the
network is necessary. WinMS, uses a TDMA-based
MAC protocol, called FlexiMAC [14], in order to sup-
port resource transfer among nodes in the network.
FlexiMac protocol provides synchronized communica-
tion between the nodes. Thus, it can adaptively adjust the
network by providing local and central recovery mecha-
nisms.
Sympathy is a tool for detecting and debugging fail-
ures in wireless sensor networks, but unlike WinMS it
does not provide any automatic network reconfiguration
incase of a failure. One of the main functionalities of the
system is the collection and an alysis of network informa-
tion metrics such as nodes next hop and neighbors. By
doing so, it is able to iden tify which of the nodes deliver
insufficient data to the sink node or to the base station
and locate the cause by reporting back to the end user.
One of the major advantages of Sympathy is th at it takes
into account interactions upon multiple nodes however,
by doing so it will require nodes to exchange neighbor-
hood lists, something that has proven highly costly in
terms of energy levels.

D. GEORGOULAS ET AL. 261
sualization of the actual netwo rk.
Th
for collecting
da
of wireless
se
t provide any interpretation of the displayed
gr
ms such as the
A
to sleep when they are not active inside
th
ide traffic management functions in
th
single radio nodes to evaluate each of their
in
stems in terms of their organization and func-
tio
ole of Middleware in Wireless
Sensor Networks
identified, that a middleware
side a wireless sensor network can establish a frame-
Another very useful functionality of wireless sensor
systems is that of the vi
is ability of an end user to demonstrate graphical rep-
resentations of the different states of the network at
various time intervals can be found in systems such as
the TinyDB [16] and MOTE-VIEW [17].
TinyDB is a distributed query processor for sensor
networks. It uses an SQL like interface
ta from nodes in the given environment and also pro-
vides aggregation, filtering and routing of the acquired
results back to the end user. With the use of a declarative
language for specific user queries, TinyDB proves to be
flexible in two domains. Firstly, all the queries that are
generated are easy to read and understand and secondly
the underlying system will be responsible for the genera-
tion and the modifications of the query without the query
itself to need any modifications. In the core of the system
we find a metadata catalog that identifies the commands
and attributes that are available for querying.
The MOTE-VIEW system is an interface system be-
tween the end user and the deployed network
nsors. Through this interface the user can make altera-
tions to node characteristics in terms of radio frequency,
sampling frequency and transmission power. The system
architecture is based on four layers; data access abstrac-
tion, node abstraction, conversion abstraction and visu-
alization abstraction layer. The data abstraction layer acts
as the database interface where all the data is been stored.
The node abstraction layer collects and stores all the
nodes metadata which will create relatio nal links with the
database. All the raw data that is going to be collected
from the nodes will be translated into understandable en-
gineering units at th e conversion abstraction layer. Finally,
the visualization abstraction layer will provide to the end
user displays of the data in forms of spreadsheets and
charts.
MOTE-VIEW is a passive monitor system in that it
does no
aphical data on behalf of the user. However, in terms
of network and other failures MOTE-VIEW does not
provide any self-configuration scheme.
The resource management is one of the key aspects of
every wireless sensor network. Syste
gent Based Power Management [18] and SenOS [19]
have been created with that concern in mind. The Agent
Based Power Management is a system that builds its ar-
chitecture upon mobile agents. These intelligent entities
are set responsible for local power management process-
ing by applying energy saving strategies to the nodes of
the network. This agent-based scheme is suitable for ap-
plications where the state of the ne twork is partial visible
at a known time or location. One of the major concerns
of this approach is that by minimizing the transmission
power, the communication range of the nodes will be
reduced accordingly, threatening the network connec-
tivity.
SenOS is managing network power resources by in-
structing nodes
e network. To achieve this, SenOS adapts a dynamic
power management algorithm known as DPM. The
DPM algorithm, by observing events inside the net-
work, can generate a policy for state transitions. Based
on that, all redundant nodes are placed inside a cluster
with only one node awake for a period of time per
cluster while the others are in a sleep mode for con-
serving energy. The SenOS architecture is based on a
finite state machine which consists of three compo-
nents. Firstly, we have the kernel which provides a
state sequencer and an event queue. The second com-
ponent is a transition table and the final component is a
call back library.
Siphon [20] and DSN RM [21] are two representative
systems that prov
eir architecture. Siphon, is based on a Stargate imple-
mentation of virtual sinks in order to prevent congestion
at near base stations inside the network. These virtual
sinks act as intermediates between the actual nodes and
the base stations and they are d istributed rando mly inside
the network. If at any point the rate of generate data in-
creases beyond a level that exists in a predetermined
threshold inside the system then the virtual sinks will
redirect the traffic to other visible nodes. The visibility
of the available nodes by the virtual sinks is one of the
disadvantages of this approach, as there is a high prob-
ability that some nodes will be not cov ered by any virtual
sink.
DSN RM (Distributed sensor network resource man-
agement) uses
coming and outgoing data rate and apply delay schemes
to those nodes when necessary in order to reduce the
amount of the traffic in the network. In every DSN there
are a number of decision stations whose role is to act as
data managers in a hierarchical format. However, the ef-
fectiveness of this technique is tightly bound on finding
reliable data for every decision station inside the wireless
sensor network that from its nature can provide inaccurate
data during its lifetime due to connectivity and radio
problems.
Table 1 presents a tabular evaluation of the currently
available sy
nality.
5. The R
M
inany researchers have
work for bridging the gap between applications and low
levels constructs such as the physical layer of the sensor
nodes.

262 D. GEORGOULAS ET AL.
ms designed criteria.
Wireless sensor Reactivity Architecture Function Energy efficiencyAdaptability Memory Scalability
Table 1. Wireless sensor ne twork systeevaluation based on
Netw ork Systems efficiency
BOSS PrCeManage oactive ntralised ment SystemYes Yes Yes Yes
MANNA
S
ol
IEW
ces
Proactive HierarchicalManagement System N/A N/A N/A N/A
LEACH Proactive Distributed Routing Protocol Yes Yes Yes Yes
GAF Proactive Distributed Routing Protocol Yes Yes Yes Yes
WinMProactive HierarchicalFault Detection Yes Yes Yes Yes
SympathyProactive Centralised Fault Detection Yes Yes Yes No
TinyDB Passive Centralised Visualization ToYes No Yes Yes
MOTE- VPassive Centralised Visualization Tool Yes No Yes Yes
SensOS ReactiveHierarchicalManagement resourYes No Yes No
A. B. P. M Proactive Distributed Management resources Yes Yes Yes No
DSNRM Proactive HierarchicalTraffic Control Yes Yes No No
Siphon Proactive Distributed Traffic Control No Yes Yes No
A middleware can be visualized as a network manag-
in
section identifies some of the most well known
m
machine inside a middleware is a
fle
22] middleware is among those that use a
vi
with content specific rou ting.
This section identifies some of the most well known
ntexts that refer to an equal amount of events:
clock timers, message receptions and message send re-
g software mechanism that will create communication
bonds with the network hardware, the operating system
and the actual application. A fully implemented middle-
ware should provide to the end user a flexible interface
through which actions of coordination and support will
take place for multiple applications preferably in real
time.
This
iddleware approaches that have been developed in re-
cent years and classifies them according to their pro-
gramming paradigm. This classification presents mid-
dlewares as virtual machines based on modular pro-
gramming, virtual database systems and adaptive mes-
sage oriented s ystems.
The use of a virtual
xible approach since it can allow a programmer to
partition a large application into smaller modules. The
middleware will inject and distribute those modules in-
side the wireless sensor network with the use of a prede-
fined protocol that will aim to reduce the overall energy
and resource consumption. The main role of the virtual
machine is to interpret those distributed modules. The
communication protocol can be designed based on
modular programming. The use of mobile code can fa-
cilitate an energy efficient framework for the injection
and the transmission of the application modules inside
the network.
The Mate [
rtual machine in order to send applications inside the
wireless sensor network. The developers having identi-
fied the predominant limitations of wireless sensor net-
works such as energy consumption and limited band-
width propose a new programming paradigm that is
based on a tiny centric virtual machine that will allow
complex programs to be very short. In order to achieve
that, Mate’s virtual machine acts as an abstraction layer
middleware approaches that have been developed in re-
cent years and classifies them according to their pro-
gramming paradigm. This classification presents mid-
dlewares as virtual machines based on modular pro-
gramming, virtual database systems and adaptive mes-
sage oriented s ystems.
The use of a virtual machine inside a middleware is a
flexible approach since it can allow a programmer to
partition a large application into smaller modules. The
middleware will inject and distribute those modules in-
side the wireless sensor network with the use of a prede-
fined protocol that will aim to reduce the overall energy
and resource consumption. The main role of the virtual
machine is to interpret those distributed modules. The
communication protocol can be designed based on
modular programming. The use of mobile code can fa-
cilitate an energy efficient framework for the injection
and the transmission of the application modules inside
the network.
The Mate [22] middleware is among those that use a
virtual machine in order to send applications inside the
or network. The developers having identi-wireless sens
fied the predominant limitations of wireless sensor net-
works such as energy consumption and limited band-
width propose a new programming paradigm that is
based on a tiny centric virtual machine that will allow
complex programs to be very short. In order to achieve
that, Mate’s virtual machine acts as an abstraction layer
with content specific rou ting.
Figure 4 presents Mate architecture and execution
model. This high level architecture will enable the pro-
gramming code to break up into small capsules of 24
instructions each that can self-replicate inside the net-
work.
This architecture enables Mate to begin execution in
response to a specific event such as a packet transmission
or a time out. This is applicable thro ugh the three execu-
tion co

D. GEORGOULAS ET AL. 263
Figure 4. Mate architectural concept, P. Levis and D. Culler
(2002).
quests. Each of the three contexts has an operant stack
and a return address one. The operant stack will be used
stack will handle all the subroutine calls.
very capsule that is sent inside the network includes
suffers
fr
s based on the fact that
ea
work
no
ding and working independently. Their inter-
co
p-
tio
al database system. The
m
ta to be a virtual rela-
tio
for instructions handling all the data while the return
address
E
a type and a version number. Based on that information,
Mate can achieve easy version updates by adding a new
number to the capsule every time a new version of the
program is uploaded. However, Mate middleware
om the overhead that every new message introduces
and also all the messages are transmitted by flooding the
network in order to minimize asynchronous events noti-
fications, raising issues with the energy consumption of
every node inside the network.
Agilla [23] is a middleware that provides a mobile
code paradigm for programming and making effective
use of a wireless sensor network. Agilla applications
consist of mobile agents that can clone or migrate across
the network. The framework i
ch agent is acting as an autonomous entity inside the
network allowing the developer to run parallel processes
at the same time. Agilla is based on the Mate architecture
in terms of the virtual machine specifications but unlike
Mate, which as we described above divides an applica-
tion into capsules flooding the network, Agilla uses mo-
bile agents in order to deploy an application .
Figure 5 presents the Agilla model identifying the
communication principle between two neighbor net
des.
In every network node we can have one or more
agents resi
mmunication and coordination is established by local
tuple spaces that are accessible by all the agents resident
in that node and a neighbor list. The local tuple space is a
shared memory architecture that is addressed by
field-matching. A tuple can be defined as a sequence of
data objects that is inserted into the tuple space of each
node by every agent. These data objects will remain in
the node regardless of the agent status. In due time, an
agent can retrieve an old tuple by template matching. In
order to do so the sending agent must generate a query
for that tuple, matching the exact same sequence of fields.
The neighbor list contains the addresses of neighboring
nodes and is accessible for every agent in the network
that wants to clone or migrate in a different location.
Based on these attributes, Agilla allows network re-
programming thereby eliminating the power consum
n cost of flooding the network. However, the lack of a
hierarchical communication model for the agent society
and the absence of precise real location information of
every node in the network can lead to deadlocks and
memory management problems.
What is known as a database middleware will visual-
ize the whole network as a virtu
iddleware in this case will provide the user with an
interface for sending queries to the sensor nodes of the
system to extract the desired data.
The Cougar middleware adopts the above approach by
considering the extracted sensor da
nal database. The developers by using an SQL-like
language assume that the whole network is the database.
The contents of such a database are stored data and the
sensor data. The stored data is represented as a virtual
relationship between the sensors that participate in the
network and the physical characteristics. The sensor data
which is the outcome of processing functions is repre-
sented as time series that will be adapted towards the
query formul ation.
Figure 5. The Agilla architectural concept, C. Fok and G. Roman (2005).

264 D. GEORGOULAS ET AL.
Figure 6. Cougar architecture for query processing, G. Gehrke and S. Madden (2002).
Figure 6 shows a block diagram that can explain the
Cougar architecture for querying processing. One
block presents the user end with the base station in the
active role of transmitting and receiving queries from
the wireless sensor network. The second block presents
the distributed network query processor that consists of
a number of abstract data types with virtual relation-
ships with the operating system of the network. In an
SQL query format of SELECT-FROM-WHERE-
GROUP-INCLUD
ccess this object relational database which mirrors the
act
odel. This model defines three distinct
fa
ES Cougar can allow the user to
aual network.
The third class of middleware is message oriented.
Their core architecture is based on creating a communi-
cation model that will facilitate message exchanges be-
tween nodes and the sink nodes of the wireless sensor
network. To achieve that most middlewares will adapt a
publish-subscribe mechanism, an asynchronous commu-
nication paradigm which allows a loose coupling be-
tween the sender and the receiver saving precious power
resources.
Mires [24] middleware provides such an asynchronous
communication m
ces for the nodes resident in the network. Initially the
network nodes will advertise their sensed data (topic).
Using a multi-hop routing algorithm Mires will route
those advertisement messages to the sink node. Lastly, a
user application interface will select the desired adver-
tised topics to be monitored.
Figure 7. The Mires architecture, E. Souto and G. Vascon-
celos (2004).
Figure 7 demonstrates the key characteristics of the
Mires architecture. The bottom block consists of the
hardware components of the node which are directly
interfaced and controlled by the operating system. The
middleware is placed on top of the operating system to
implement its publish-subscribe communication model.
This model is able to advertise the sensor data (topics)
provided by the running application while it maintains a
topic list provided by the node application. Mires send
only messages referring to subscribed topics thus reduc-
ing like that the numbers of the transmitted packets and
therefore saving energy.

D. GEORGOULAS ET AL. 265
6. Towards the Design of the In-Motes
Middleware
In order to design and develop a successful middleware
solution for a wireless sensor network th at will be able to
satisfy some if not all the functionalities of a network
management system for monitor and control there are
certain design criteria, which we described in the previ-
ous sections, and must be considered and brought for-
ward in our design. In the following paragraphs we are
going to present those design principles for a substantial
middleware development.
A wireless sensor network consists of tiny devices
that are battery powered and provided with a sml
environments. It is obvious, that after
ing to
hus
to nergy efficient manner which of the
prevention together with a flexible way of
re
n efficiently and as long as possible.
Su
al
CPU processor. Usually, and as we have already men-
tioned, they can be deployed in hundreds and typically
physical harsh in
the deployment a physical contact for replacement or
maintenance is highly unlikely. A middleware should
be able to provide remote access to these nodes making
sure that they will exhaust all their resources in terms
of battery power and memory in a timely manner.
Hence, one of the basic design principles for our mid-
dleware is the ability to manage limited power and
resources.
Our approach [25] in order to satisfy the ab ove design
criterion is based in the creation of a flexible communi-
cation protocol between the nodes of our network and the
base station. Thus, inspired from the GAF protocol that
was described in the previous section, we are aim
devhaving node redundancy in mind, telop a protocol
rgulate in an ee
enods of our wireless sensor network will be active and
which ones will be in a sleep mode. Subject to our trials
and the development of our middleware this protocol
will be introduced both hand written in the middleware
engine as well as it will be introduced as part of the run-
ning application.
Field trials such as the CodeBlue Project [26] and the
Wireless Sensing Vineyards [27] identified that a wire-
less sensor network topology is subject to frequent
changes due to factors such as device failure, interfer-
ence, mobility and moving obstacles. Also, they proved
that it is very possible that an application will grow in
time, therefore mechanisms for a dynamic network to-
pology should be available from the middleware. A mid-
dleware should be able to adapt to parameter changes
caused by unexpected external factors of the environ-
ment and also provide mechanisms for fault tolerance
and self configuration of the nodes inside the wireless
sensor network [28].
Based on the above observations, and inspired from
approaches similar to WinMS and Sympathy our mid-
dleware will incorporate some mechanisms for fault de-
tection and
configuring the network in real time [29]. As Sympa-
thy is a fully automated system, we will be aiming to
provide some kind of automatic mechanisms in our mid-
dleware for the above design criteria without thou gh this
to be our first priority.
Unlike traditional networks, sensor networks and
their applications are real-time phenomena with dy-
namic resources. Upon deployment of an application,
core parameters such as energy usage, bandwidth and
processing power cannot be predefined due to the dy-
namic character of these networks. A middleware
should be designed with mechanisms that should allow
the network to ru
ch mechanisms include resource discovery and loca-
tion awareness for the nodes in the system. Low-level
programming models must be introduced as well in
order to bridge the gap between the running application
and the hardware.
As we mentioned before, mechanisms that are pre-
dominant in traditional networks are not sufficient to
maintain the quality of service of a wireless sensor net-
work because of constraints such as the dynamic topol-
ogy and the power limitations. A middleware should be
able to provide an d maintain the qu ality of service ov er a
long period of time while in parallel to be able to adapt
in changes based on the application and on the perform-
ance metrics of the network like these of energy con-
sumption and data delivery delay.
Inspired both form the work of the Mate and Agilla
middlewares we will introduce a mechanism that will
allow mobile code to be transmitted inside our network
and will be able to allow changes in network parameters
such as the bandwidth of each node as well as applica-
tion parameters such as switching between available
sensors of each node. Thus, an architecture will be de-
veloped adapting technologies such as Linda-like tuple
spaces and agents/mobile code transmission with the
support of a virtual machine engine [30].
Wireless sensor networks can be widely deployed in
areas such as healthcare, rescue and military, all of
these are areas where information has a certain value
and is very sensitive. The environments of those areas
tend to be very active and harsh, increasing the chances
for malicious intrusions and attacks such as denial of
service. Traditional approaches and mechanisms used
to secure the network cannot be applied in this kind of
network since they are heavy in terms of energy con-
sumption. A middleware must be able to provide a se-
cure framework for deploying applications inside the
network. During the life-cycle of the network the mid-
dleware should establish protective mechanisms to
ensure security requirements such as authenticity, in-
tegrity and confidentiality.
Table 2 presents a tabular evaluation of the currently
available middleware systems in terms of their organiza-
tion and func tionality.

266 D. GEORGOULAS ET AL.
e sy
s
Table 2. Wireless sensor network middlewar
Project Main feature
stems evaluation based on designed criteria.
ScalabilityMobilityHeterogeneity
Power
Awareness Easy of use
s Opennes
Mate Mobile active Capsules, TinyQS,
Byte code interpreter Full Full Full Partial Full Little
Agilla Generic Agents, TinyQS Full
Cougar Virtual Database, SQL like language Partial
Mires nesC, TinyQS, message oriented Full
In-Motes nesC, TinyQS, agents, behavioral rules Full
Partial Full Partial Little Average
Partial Partial Partial Partial Full
Full Partial Full Full Full
Full Full Full Full Full
7. Conclusions
This survey presents and demonstrates the wireless sen-
sor networks as one of the predominant technologies for
. The foundational concept behind this
sensor networks,” IEEE Computer, pp. 41–49
[7] I. F. Akyildiz, Y. Sankarasubramaniam, W. Su
the 21st century
techy is explained toge limitation
barat are incorporatd to be addresse
the ss ofetwork appl
ble numcations in modern so
ties. The survting the foundatio
funof a. Funct
thatased hose of mon
and corol anwireless sen
sy nurk syst
re critically reviewed providing an analytical explana-
cture and identifying pros and cons
aw some important lessons for our
http://www.ddirv.lv/doc_upl/sazonovs1.pdf.
[5]
n, and M. Srivastava, “Overview of
, 2004.
, and E.
Cayirci, “Wireless sensor networks: a survey,” Cr
ork 3–
[8] L. Le Datta,. Cardelver, “W:
reless r netwnagemestem,” CSSE
nical rt, UWA-CSSE-06-001, 2006.
[9] . Aky. Sanraman. Su, E.
irci, “Wireless sensor networks: a survey,” Co
tworks.
[10] B. Ruiz J. M. Nra, “MAManagement
architecture for wireless sensor networks,” Communica-
, Vol. 41, No. 2, pp. 116–125, 2003.
im, K. Lee, and J. Sung, “UPnP-based sen-
ience, pp. 10–17,
network-management system,” CSSE
[15] od, and E.
debugger,”
ong, “TinyDB:
ne, and G. M. P. O'Hare,
anagement
nolog
riers thether with th ed and nees and d in
proce
for a making a wireless sensor nica- cie-ber of useful appli
ey opens by presennal
ctions network management systemions
are b
nt upon two main attributes t d are critical for every itor sor
stem. A mber of wireless sensor netwoems
a
tion of their archite
helping us drthus
proposed system. The survey continues by presenting the
middleware solution as a key player in overcoming the
wireless sensor networks limitations and as our main
methodology behind our proposed approach. A classifi-
cation and analysis of the current middleware approaches
for the wireless sensor networks is produced. The survey
concludes by presenting the role and the key design prin-
ciples our middleware approach.
8. References
[1] D. Culler, D. Estrin, and M. Srivastava, “Overview of
sensor networks,” IEEE Computer, pp. 41–49, 2004.
[2] J. Hill, R. Szewczyk, A. Woo, S. Hollar, and D. Culler,
“System architecture directions for networked sensors,”
Architectural Support for Programming Languages and
Operating Systems, pp. 93–104, 2000.
[3] A. Boulis, C. C. Han, and M. B. Srivastava, “Design and
implementation of a framework for efficient and pro-
grammable sensor networks,” International Conference
On Mobile Systems, Applications And Services, pp.
187–200, 2003.
[4] Cisco Systems, “Network management systems organiza-
tion,” 2000,
W. Stallings, “Wireless communications and networks,”
ISBN-10: 0131918354, pp. 45–50, 2004.
6] D. Culler, D. Estri[
ompute
Netws, pp. 39422, 2002.
W.
Wi e, A. senso and R ork-ma l-Oli nt syinMS
Tech Repo
I. F
Cay ildiz, Ykarasubiam, Wand mputer
Ne, pp. 393–422, 2002
L. andogueiNNA:
tions Magazine
11] H. Song, D. K[sor network management architecture,” Proceedings of
the second International Conference on Mobile Comput-
ing and Ubiquitous Networking, pp. 85–92, 2005.
[12] W. Heinzelman, A. Chandrakasan, and H. Balakrishan,
“Energy efficient communication protocol for wireless
microsensors networks,” Proceedings of the Hawaii In-
ternational Conference on System Sc
2000.
[13] Y. Xu, “Geography-informed energy conservation for ad
hoc routing,” Mobicom’01, pp. 203–212, 2001.
[14] W. L. Lee, A. Datta, and R. Cardell-Oliver, “WinMS:
Wireless sensor
Technical Report, UWA- CSSE-06-001, 2006.
N. Ramanathan, K. Chang, R. Kapur, L. Gir
Kohler, “Sympathy for the sensor network
Centre for Embedded Network Sensing, pp. 98, 2005.
[16] S. Madden, J. Hellerstein, and W. H
In-network query processing in TinyOS,” ACM Transac-
tions on Database Systems, pp. 122–173, 2003.
[17] Touron, “Crossbow: moteview interface,” Crossbow, 2005,
http://www.xbow.com/Technology/UserInterface. aspx.
[18] R. Tynan, D. Marsh, D. O'Ka
“Intelligent agents for wireless sensor networks,”
AAMAS 2005: 1179–1180.
[19] T. H. Kim and S. Hong, “Sensor network m

D. GEORGOULAS ET AL. 267
e 3rd
on Architectural Support for Programming Lan-
2005.
Paper Award, Proceedings of the 5th
sensing vineyards,” 2008,
[30]
ational Conference on Telecommunica-
protocol for state-driven execution environment,” Pro-
ceedings of the International Conference on Ubiquitous
Computing, pp. 197–199, 2003.
[20] J. Wan, S. Eisenman, A. Campbell, and J. Crowcroft,
“Siphon: overload traffic management using multi-radio
virtual sinks in sensor networks,” Proceedings of th [26]
International Conference on Embedded Networked Sen-
sor Systems, pp. 116–129, 2005.
[21] Zhang, “Network management in wireless sensor net-
works,” 2001.
http://www.csse.uwa.edu.au/~winnie/Network_Managem-
ent _in_WSNs_.pdf.
[22] P. Levis and D. Culler, “Mate: A tiny virtual machine for
sensor networks,” Proceedings of the International Con-
ference
guages and Operating Systems, San Jose, CA, USA, pp.
100–111, 2002.
[23] C. L. Fok and G. C. Roman, “Mobile agent middleware
for sensor networks: An application case study,” Pro-
ceedings of the 4th International Conference on Informa-
tion Processing in Sensor Networks, pp. 382–387,
[24] S. Eduardo, G. Germano, and V. Glauco, “A mes-
sage-oriented middleware for sensor networks,” Pro-
ceedings of the 2nd workshop on Middleware for perva-t
sive and ad-hoc computing, pp. 127–134, 2004.
[25] D. Georgoulas and K. Blow, “In-Motes: An intelligent
agent based middleware for wireless sensor networks,”
Best Student
WSEAS International Conference SEPADS 06, pp.
225–231, 2006.
Welsh et al, “Wireless medical sensor networks in emer-
gency response: Implementation and pilot results,” IEEE
International Conference on Technologies for Homeland
Security.
[27] Holler, “Wireless
http://camalie.com/WirelessSensing/WirelessSensors.htm.
[28] D. Georgoulas and K. Blow, “Making motes intelligent:
An agent based approach to wireless sensor networks,” In
WSEAS on Communications Journal, pp. 515–522, 2006.
[29] D. Georgoulas and K. Blow, “In-Motes bins: A real time
application for environmental monitoring in wireless sen-
sor networks,” Proceedings of the 9th IEEE/IFIP Interna-
tional Conference on Mobile and Wireless Communica-
tions Networks, pp. 21–26, 2007.
D. Georgoulas and K. Blow, “Intelligent mobile agent
middleware for wireless sensor networks: A real time ap-
plication case study,” AICT 2008, in Proceedings of 4th
Advanced Intern
ions: Programmable networks, active networks and mo-
bile agents, protocol & standards, Athens, pp. 35–43,
2008.