Вы находитесь на странице: 1из 6

The Second International Conference on Sensor Technologies and Applications

Configuration of Sensor Networks by Energy Minimisation

Sajeh Zairi1,2 Belhassen Zouari Eric Niel


sajeh.zairi@fst.rnu.tn, 1- LIP2 Lab, University of Tunis 2- AMPERE Lab, INSA-Lyon,
sajeh.zairi@insa-lyon.fr El Manar, 2092 Manar II, Tunis(ia) F 69621, France,
belhassen.zouari@fst.rnu.tn eric.niel@insa-lyon.fr

Abstract sors means that sensing data may travel a large number of
hops to reach intended destinations.
Recent advances in wireless sensor networks have led However, sensor nodes are restricted in terms of both en-
to many new protocols specifically designed for sensor net- ergy supply and bandwidth. Energy awareness is thus re-
works where energy awareness is an essential consideration. quired in all layers of the networking protocol stack. Phys-
Attention has been mostly directed towards routing proto- ical and link layer related issues are generally common to
cols. These protocols generally implement a view embracing every type of sensor application, so research in these areas
a local energy reserve. This paper presents an algorithm has focused on system-level power awareness, for example
for determining optimal paths connecting sensor nodes to such as dynamic voltage scaling, radio communication hard-
the sink. These optimal paths contribute to minimizing inter- ware, low-duty cycle issues, system partitioning, energy-
mediate hops and consider a global energy reserve. Most aware MAC protocols 1 [5, 9] and radio off/on-based proto-
of these sensor networks will require application specific cols [12, 2]. At network layer level, the main aim is to find
functionalities and performance requirements because of the ways of setting up energy-efficient route and reliably relay-
wide range of their applications. Modelling sensor network ing data from sensor nodes to the sink to ensure maximum
behaviour before implementation and deployment is there- network lifetime .
fore crucial to programming efficiently the network applica- To transmit data to the sink, a sensor node must compute
tion and to validating all its protocols. Existing works suf- the optimal path connecting it to the base station. In many
fer from the problem of insufficient formal models for sensor studies, the optimal path is computed based on the number of
network validations. Such models enable formal verification hops (intermediate sensor nodes used for transmitting mes-
of all sensor network properties prior to deployment. This sages) or on the transmission cost. Energy is the must im-
paper will also present a formal model representing sensor portant sensor node characteristic, so we have integrated this
network behaviour. constraint into computing the optimal path, which must not
only feature a minimum intermediate hop, but also a max-
imum global energy reserve in the selected sensor nodes.
1. Introduction Our work therefore differs from existing studies because it
considers multi-objectives. More specifically, we consider
a global energy view for the total path. Optimal paths are
Recent advances in micro-electro-mechanical systems
determined on-line. Each sensor node will indeed indepen-
and highly integrated low-power digital electronics have led
dently determine its optimal path based on local information
to the development of micro sensors [1, 8, 13]. Such sen-
and received data (transmitted in specific messages). This
sors are generally equipped with data processing and com-
method will be presented later.
munication capabilities. They include a minimum number
Most of these networks will require application-specific
of essential elements for operating over long periods of time:
functionalities and performance requirements because the
sensing the physical world for some meaningful data, pro-
wide range of sensor network applications and the exponen-
cessing and communicating information to end-users. The
tial increase of deployed sensor nodes [14]. Modelling sen-
sensing component measures ambient conditions affecting
sor network behaviour before implementation and deploy-
the environment surrounding the sensor. The sensor deliv-
ment is therefore crucial to efficiently programming a net-
ers messages, either directly or through intermediate sensors,
work application. Use of formal models allows formal ver-
sensing data to a command center (sink), usually via wirelles
ification of the network properties prior to deployment. A
radio transmitter. Sensor networks are usually application-
driven and perform specific tasks. 1 (Media Access Control) The protocol that controls access to the physi-

Furthermore, the short transmission range of small sen- cal transmission medium

978-0-7695-3330-8/08 $25.00 © 2008 IEEE 148


141
DOI 10.1109/SENSORCOMM.2008.106

Authorized licensed use limited to: Sun Microsystems- sunlibrary@sun.com. Downloaded on February 11, 2010 at 04:45 from IEEE Xplore. Restrictions apply.
bibliographical study brought to our attention the lack of Before introducing this approach, we define a number of
such models for validating sensor networks. This deficiency concepts to be applied in the developed algorithm.
prompted us to develop a formal model for formal verify- When considering multi-objective optimization problem,
ing sensor networks. This paper proposes a coloured Petri the major difficulty is to be able to determine the optimal
net (CP-net) for sensor network modelling, based on gen- solution. Unlike the mono-objective optimization problem,
eral characteristics and functionalities. Development of this which has only one solution, the multi-objective optimiza-
CP-net model was also justified by the increasing availabil- tion problem has a set of solutions. Each solution in this set
ity of efficient methods and tools based on High-level Petri is optimal because we cannot improve one component x∗i of


net models. Use of such models enable validation or per- the solution vector x∗ without degrading at least one other.
formance analysis of system behaviour through exploring The Pareto dominance concept is normally applied to
model state space or simulation. These models also enable solve this problem; this defines an order relation between
the use of formal verification techniques, such as model- two solution vectors − →1 and −
x →2 .
x
checking, for verifying system properties. Our work em- Definition 1 (Dominance).
braces many interesting verification properties, for example Vector u = (u1 , ..., uk ) dominates vector v = (v1 , ..., vk ),
existence of reliable paths connecting each node to the sink denoted by u  v, if and only if u is partially less than v, i.e.,
and optimality of such paths (with respect to transmission ∀ i ∈ {1, ... , k}, ui ≤ vi ∧ ∃ i ∈ {1, ... , k} : ui < vi .
delay and energy consumption). Definition 2 (Pareto optimal solution).
The paper is organized in seven sections. Section 2 intro- Solution x∗ is a Pareto optimal solution, if and only if
duces the algorithm developed to determine optimal paths. there is no solution x such that Z(x) dominates Z(x∗ ),
Section 3 defines major functionalities of the different sen- where Z(x) represents the cost vector associated with x.
sor network components. Section 4 presents a CP-net mod-
elling the sensor network functionalities. Section 5 discusses Let us assume that Pcs represents all paths connecting
the differences between the proposed algorithm and related sensor node ”c” to the sink. In our problem, each solution
efforts in sensor networking. Section 6 forms the conclusion. is represented by a couple. The first element of the couple
is the hop number (HNb) and the second element represents
2. ENERGY-BASED ROUTING ALGO- the energy difference. Indeed, to connect each sensor node to
the sink, we must determine paths with minimum hop num-
RITHM
ber and maximum total amount of energy in the path (QoE).
The Pareto optimal concept minimizes each vector element,
In this section, we present an algorithm for optimal path so the second element of the path cost vector is defined as the
determination suited to sensor network characteristics. This difference between a maximum energy value (ValEnerMax)
data network process has been the subject of continuous and the total quantity of energy in the path.
research over the last two decades; numerous routing pro- Definition 3 (Pareto Optimal Path).
tocols have been studied and used in practical networks. p(HN b, QoE) ∈ Pcs is a Pareto optimal path
Existing algorithms have invariably considered mono/multi- if and only if there is no path p (HN b , QoE  ) ∈
objectives. Our interest in developing optimal path algo- Pcs , for which (HN b , V alEnerM ax − QoE  ) 
rithms suited to sensor networks was in fact partly motivated (HN b, V alEnerM ax − QoE).
by the problem of the unsuitability of existing algorithms Optimal path can be built in the following straightforward
for sensor characteristics. Indeed, most of the existing al- way. The sink broadcasts a control message announcing a
gorithms capable of solving the optimal path problem are cost of (0,0). At any time, information transmitted by the
unsuitable for sensor networks because of: control message are :
• the need for extensive computing capabilities at each • the sequence number,
sensor node;
• the identity of the sensor node transmitting the message,
• the handling of large network information. Identifica- • the total energy accumulated on the path,
tion of all sensor nodes belonging to an optimal path
and integration of this information into a message to be • the identity of the destination sensor associated with the
transmitted to sensor neighbours exemplifie the fact that sensor node transmitting the message,
it is impossible to perform this kind of operation in a • the hop number separating the sensor node transmitting
sensor node due to limited memory capacity and major the message from the sink.
transmission energy consumption for a large message.

Our developed algorithm considers simultaneously two When hearing a control message containing the cost of
different objectives, namely : energy and intermediate hops. the sender, sensor node can verifies one of these two cases:

142
149

Authorized licensed use limited to: Sun Microsystems- sunlibrary@sun.com. Downloaded on February 11, 2010 at 04:45 from IEEE Xplore. Restrictions apply.
• the received message has a new sequence number, next table:

n Path Hops QoE ValEnerMax


• the received message has a sequence number equals to -QoE
the one recorded by the receiver sensor node. 1 (2,5), (5,0) 2 32 168
2 (2,1), (1,0) 2 30 170
3 (2,3), (3,1), (1,0) 3 55 145
In the first case, the receiver sensor node will directly
4 (2,4), (4,3), (3,1), 4 95 105
records all the information transmitted in the received mes-
(1,0)
sage on its set of optimal paths (Sop ). The sensor node trans-
mitting the received message will be considered as the desti-
To compare these paths, we compare their cost vectors:
nation neighbour. This information will be used to compare
* (2,168) for path 1 * (2,170) for path 2
the subsequent received data to determine the best alterna-
* (3,145) for path 3 * (4,105) for path 4
tives.
The cost vector of path 2 is dominated by the cost vector
In the sencond case, the memorized and received paths of path 1. Message transmission using sensor node 1 or 5 is
must be comparable. Sensor node first checks whether it is equivalent, if we consider the hop number constraint, mes-
not the destination neighbour of the sensor node transmitting sages effectively take two hops. However, using sensor node
the message. This check avoids a cycle. If no cycle, the 5 as the destination neighbour is better in relation to the en-
sensor node must compare the received alternative with its ergy constraint. The Pareto dominance concept allowed us
set of optimal paths (Sop ). to compare these two paths.
Different paths are compared using the Pareto concept, as Path 2 is not a Pareto optimal path and must be elimi-
explained in definition 3. If the cost vector representing the nated. Other paths cannot be compared so they are Pareto
new path is dominated by at least one path belong to Sop , optimal paths. New mechanisms must be introduced to com-
then it will not be considered. If it is a Pareto optimal path, pare Pareto optimal paths.
then it will be added to Sop , and all paths belonging to Sop , To validate the Pareto optimal path determination, we
and dominated by this new path, will be removed. When- have integrated this mechanism into a CP-net, modelling the
ever it determine a Pareto optimal path, the sensor node will generic functions of our sensor network. We have developed
broadcasts a message to its neighbours containing this new this model because there are few formal models representing
alternative. the sensor network functionalities. This type of model per-
mits formal verification of sensor network properties prior to
Example deployment.
In the following section and before introducing this
We consider the example provided by the sensor network model, we describe the considered sensor network functions.
shown in figure 1 to illustrate the optimal path determination. The sink has two different fonctions:

• processing of alert messages,

• periodic broadcast of control message for the synchro-


nisation of all sensor nodes

Each sensor node has three fonctions

• regular monitoring of the environment temperature to


detect fire and transmit an early alert message to the
Figure 1. Example of sensor network sink,

• alert message transmission,

• control message processing for the determination of the


In figure 1, each sensor node is represented by a circle Pareto optimal paths.
defining its identity and its actual energy reserve. The sink
is labelled ”0”. Petri net model representing sensor node behaviour must
The different paths, with no cycles, which can be used by model these three functions; this model is introduced in the
sensor node 2 to transmit messages to the sink are defined in following section.

143
150

Authorized licensed use limited to: Sun Microsystems- sunlibrary@sun.com. Downloaded on February 11, 2010 at 04:45 from IEEE Xplore. Restrictions apply.
3. PETRI NET MODEL FOR SENSOR NET- sensing idle state (”Idle Sensing” place) and the radio use
WORKS idle state (”Idle Radio” place).
X

In this section, we present the CP-net developed to model Sensing


S
B
Control Mes
Broadcast

the sensor network functions. We assume that reader is fa- 1`1++1`2++


1`3++1`4++1`5

miliar with CP-net concept [6] Idle


Sensing
Cap
Idle
Radio
Copie
Energie
X c
X X
Cap
Active Inactive

3.1. Petri Net Modelling Sink Functioning State


s s
State

A
Alert
Transmission
X to
fire X
STRING

As previously presented, the sink has two functions. Inactive


Cap

These functions are modelled, in the Petri net, by the two


transitions ”Receives Alert Msg” and ”Sends Cont Msg” Figure 3. Petri net modelling sensor node
(figure 2). functions
1`0@+0
Cont
Sequence AMsgs Alert
Seq Msg

seq seq+1@+60 (Y,X,temp,srcA)

1`0
Different sensor node fonctions are represented by macro-
Authorisation
Cap
sensorNode() Sends Cont
Msg
X Idel
Sink
X Receives
Alert Msg transitions in the Petri net model. A Petri net is associ-
Cap

[(seq,0,0,0,0)]
ated with each macro-transition, which represents its perfor-
Cont Msg
mance. We describe these sub-nets in the following sections.
to Send
CMsgs

3.3. ”Sensing” Sub-net


Figure 2. Petri net modelling sink functions
The net shown in figure 4 models environment tempera-
ture sensing. In this model, we represent this operation by
The ”Receives Alert Msg” transition represents recep- obtaining this value from an external file. Modelling tem-
tion of an alert message. This message is represented by perature value obtention in this way allows us to synchro-
a quadruple token in the ”Alert Msg” place, which is a mail- nize our model with other model representing environment
box containing all the alert messages transmitted by the sen- behaviour.
sor nodes.
The ”Sends Cont Msg” transition represents periodic X Idle X
Sens
broadcast of control messages. Periodicity is represented by Cap
X
the ”Cont Sequence” temporized place. seq (X,eng)
The ”Sends Cont Msg” transition fires when the model Sens
Period
Sens
Value
Sensor
Energy
seq@+10 (X,eng-EngSens)
time corresponds to the time associated with the token in the Seq
(X,20)
EngCap

“Cont Sequence” place and the sink is idle. When firing this Send (X,temp) (X,temp)
Value
Msg
transition, a new control message token is deposited in the SAMsgs
Sensed
ValuSens
mailbox. This token will trigger the broadcast of the control (X,X,temp)

message by the sensor nodes near the sink. [temp>50] [temp<50]


No
In the following section, we describe the Petri net, which Fire Fire

models the sensor node functionalities.


Figure 4. ”Sensing” Sub-net
3.2. Petri Net Which Models Sensor Node
Functioning
Temperature sensing is periodically performed. This pe-
As previously described, a sensor node has three basic riodicity is represented by the token present in the ”Sens Pe-
fonctions, namely: temperature sensing, alert message trans- riod” temporized place. Tokens of this place indicate when
mission, control message process and broadcast. These func- the sensor node must monitor environment temperature.
tions are respectively represented in the Petri net (figure 3) Following environment temperature sensing, two cases
by the ”Sensing”, ”Alert Transmission” and ”Control Mes arise:
Broadcast” macro transitions.
Sensing and message processing (radio use) are assumed • the measured temperature is lower than the specified
to be independent. A sensor node can simultaneously trans- threshold; no fire is detected, the sensor node returns
mit/receive messages and monitor the environment tempera- immediately to its idle place. This operation is per-
ture. In the Petri net model, we have therefore separated the formed when the ”No Fire” transition occurs.

144
151

Authorized licensed use limited to: Sun Microsystems- sunlibrary@sun.com. Downloaded on February 11, 2010 at 04:45 from IEEE Xplore. Restrictions apply.
• the measured temperature is greater than the specified The generated message will be treated, by the neighbours
threshold; a fire is detected. The sensor node deposits a of the transmettor sensor node, in the same manner as previ-
token representing the alert message in the mailbox and ously described.
returns to its idle place. This is done when the ”Fire”
transition occurs. OPaths
Optimal
Path

We describe the sub-net representing alert message trans- (X,seqa,l) (X,seq+1,l2)


CMsgs
mission in the following section. Idle X Receive (seq,saut,eng,Y,dest)::lop Cont
if (l2<>l) then
((seq,saut,eng,Y,dest)::lop)
Control ^^[(seq,saut+1,eng+e,X,Y)]
Radio Msg
Msgs else ((seq,saut,eng,Y,dest)::lop)
Cap
3.4. ”Alert Transmission” Sub-net
(X,e) (X,Y) CapDest
EngCap
Sensor (seq,saut,e,X,dest)::lop
Alert message transmission is performed in multi-hops. If Sensor
Energy1
Neighbor if(lop<>([]))
then lop
NbNeig

the message has not yet reached the sink, the sensor node re- (Y,Nneig+1) (Y,Nneig)
else empty
(X,Nneig)
Nb Neig
UpdateNeig(X)
ceiving this message will transmit it to its destination neigh-
EngCap
bour. This transmission is represented by the ”Transmit Alert Nb Neighbor
(X,0)
Broadcast
(X,eng)
Sensor
receiving CMsg Cont Msgs Energy
Msg” transition (figure 5). NbNeig
(X,Nneig) (X,eng-EngBroad)

This transition fire, when a token representing the alert


message is deposited in the mailbox place and the sensor Figure 6. ”Control Mes Broadcast” Sub-net
node, which must transmit message, is idle.
Send Idle CapDest
Msg Cap
Radio
SAMsgs
X
(X,Y) Neighbor
Destination
Energy is required to perform each sensor node function.
(X,srcA,temp)

Transmit
(X,eng)
Sensor
Correct estimation of the current energy of each sensor node
Alert Msg Energy

(Y,srcA,temp)
(X,eng-EngATrans)

(X,Y,srcA,temp)
EngCap requires us to integrate an energy place (”Sensor Energy”
Alert
Msg
AMsgs
in figure 4) into the Petri net model . Tokens in this place
(X,Y,srcA,temp) associate the current energy reserve with each sensor node.
Retrans
mission
[Y<>0] This reserve is updated at any time the sensor node fires a
transition representing one of its three basic functions.
Figure 5. ”Alert Transmission” Sub-net The described Petri net allows other properties to be
checked in addition to validation of the developed algorithm.
For example, this net can be used to determine the lifetime of
We describe the ”Control Mes Broadcast” sub-net in the the deployed sensor network or the connectivity of this net-
following section. work. Correct transmission of alert message and validation
of the transmission delays can also be verified.
3.5. ”Control Mes Broadcast” Sub-net The developed Petri net also allows performance analysis
of the studied sensor network prior to its deployment.
We have integrate in the ”Control Mes Broadcast” sub-net
the developed mechanisms, permitting optimal path determi-
nation, for its validation. This sub-net models the process of 4. Related work
control message. When receiving a new path, a node must
choose the best alternative. Each sensor node optimal paths Different routing protocols try to solve the problem of
are memorized in a list structure. This structure is associated routing in ad hoc networks in one way or the other. In re-
with the corresponding sensor node according to the token active routing approach, a routing protocol does not take
in the ”Optimal path” place (figure 6). the initiative for finding a route to a destination, until it
When receiving a new control message (represented by a is required. The protocol attempts to discover routes only
token in place message), the sensor node will first define the on-demand by flooding its query in the network. The ex-
new alternative using the message information. Second, it amples of this kind of protocols are AODV [11], DSR
will compare this alternative to the paths saved on the ”Op- [7] and TORA [10]. On the other hand, proactive proto-
timal path” place. If the new one is a Pareto dominant path, cols are based on periodic exchange of control messages.
then it will be added to this list and all paths dominated Some messages are sent locally to enable a node to know
by this new alternative will be removed. This path will no its local neighbourhood, and some messages are sent in en-
longer considered, when it is dominated. tire network which permit to exchange topology among all
Finally, when a new path is considered, the sensor node the nodes of the network. The proactive protocols imme-
informs its neighbours by transmitting a new control mes- diately provide the required route to the destination when
sage defining the new path. needed. The examples of this kind of protocols are OLSR

145
152

Authorized licensed use limited to: Sun Microsystems- sunlibrary@sun.com. Downloaded on February 11, 2010 at 04:45 from IEEE Xplore. Restrictions apply.
[4], TBRPF [3]. In this section, we describe the basic dif- ternatives. We consider both alternatives to ensures message
ferences between the existing work and the developped al- transmission reliability (when the first neighbour failed it can
gorithm. transmit messages to the second). Correct determination of
In this paper, we have proposed a proactive protocol for optimal paths will permit implementation of a correct con-
routes determination. The developement of such protocol is figuration/reconfiguration method.
justified by the characteristics of the sensor network consid-
ered application : fire detection. Indeed, such application References
cannot tolerate the delay of route discovery which is one of
the reactive protocols problems. In this kind of applications, [1] I. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci.
when fire is detected source must immediately inform the Wireless sensor networks: a survey. Computer Networks,
sink. Therefore, reactive protocols are not so suited, proac- 38:393–422, March 2002.
tive protocols are more adapted. In this kind of application, [2] A. Bachir, S. Plancoulaine, D. Barthel, M. Heusse, and
sensor nodes are deployed in hostile environment and are A. Duda. Micro-frame preamble mac for multihop wireless
supposed to work efficently as long as possible. To realise sensor networks. In Proceedings of ICC 2006, 2006.
this objective, sensor nodes must have an equivalent energy [3] T. Clausen, P. Jacquet, A. Laouiti, P. Muhlethaler,
A. Qayyum, and L. Viennot. Efficient routing protocols for
level at any time. Indeed, when a node fails a set of nodes
packet-radio networks based on treesharing. IEEE Inter-
could be disconnected. To extend the network live time as
national Workshop on Mobile Multimedia Communications
maximum as possible, nodes having minimum energy re- (MoMuC ’99), pages 104–113, 1999.
serve must not be setected as destination neighbour, if there [4] T. Clausen, P. Jacquet, P. Laouiti, A.and Muhlethaler,
are other alternatives. This proposition allows to save mini- A. Qayyum, and L. Viennot. Optimized link state routing
mum energy reserve and to have equitible energy reserve in protocol. IEEE INMIC, 2001.
all the sensor nodes. Using existing proactive protocols can [5] W. R. Heinzelman. Energy-scalable algorithms and proto-
not allows this objective. Indeed, most of these protocols de- cols for wireless sensor networks. in the Proceedings of the
termine routes according to the minimisation of transmission International Conference on Acoustics, Speech, and Signal
cost and are not consider equitible preservation of energy Processing (ICASSP ’00), June 2000.
[6] K. Jensen and G. Rozenberg. High-Level Petri Nets: Theory
reverve in all nodes.
and Application. Springer Verlag, 1991.
Using OLSR or TBRPF protocols, each nodes must main- [7] D. B. Johnson and D. A. Maltz. Dynamic source routing in ad
tain a topology table providing the state of each link in the hoc wireless networks. Mobile Computing, 5:153–181, 1996.
network (or within a cluster if hierarchical routing is used). [8] R. Min, M. Bhardwaj, S.-H. Cho, E. Shih, A. Sinha, A. Wang,
In our algorithm, nodes needs only to record optimal paths. and A. Chandrakasan. Low power wireless sensor networks.
In this case, node state depends on the neighbour density and in the Proceedings of Internation Conference on VLSI De-
not on the all network density as it is the case with OLSR sign, January 2001.
and TBRPF protocols. In the developed protocol, control [9] R. Min, M. Bhardwaj, C. Seong-Hwan, A. Sinha, E. Shih,
messages broadcast information representing only a specific A. Wang, and A. Chandrakasan. An architecture for a power
aware distributed microsensor node. in the Proceedings of
alternative and not all the total topology table as with most
the IEEE Workshop on signal processing systems (SIPS’00),
proactive protocols. This proposition allows also to save en- Octobre 2000.
ergy. Indeed, sensor node needs more energy to transmit [10] V. D. Park and M. S. Corson. A highly adaptive dis-
bigger messages. tributed routing algorithm for mobile wireless networks. In
Proc. IEEE Conference on Computer Communications IN-
5. Conclusion FOCOM’97, pages 1405–1413, 1997.
[11] C. E. Perkins and E. M. Royer. Ad hoc on-demand distance
vector routing. Proceedings of the 2nd IEEE Workshop on
We have defined an algorithm, which determines opti-
Mobile Computing Systems and Applications, pages 90–100,
mal paths enabling each sensor node to be connected to the
1999.
sink. These paths take into account two different constraints, [12] J. Polastre, R. Szewczyk, and D. Culler. Telos: En-
namely: hop number and energy accumulated on the consid- abling ultra-low power wireless research. In Proceedings of
ered path. Different alternatives need to be compared using IPSN/SPOTS, pages 364–369, April 2005.
the Pareto dominance concept. [13] J. Rabaey, M. Ammer, J. da Silva Jr., D. Patel, and S. Round.
To validate this algorithm, we have integrated it into a CP- Picoradio supports ad hoc ultra low power wireless network-
net, which we have developed. The lack of formal models ing. IEEE Computer Magazine, 33:42–48, July 2000.
representing sensor network operation justified the develop- [14] A. Woo, S. Madden, and R. Govindan. Networking support
ment of this formal model. for query processing in sensor networks. Communications of
the ACM, 47:47–52, Junes 2004.
Determination of Pareto optimal paths represents a first
stage comprising the following principal prospects. We will
initially use the risk concept to determine the -two- best al-

146
153

Authorized licensed use limited to: Sun Microsystems- sunlibrary@sun.com. Downloaded on February 11, 2010 at 04:45 from IEEE Xplore. Restrictions apply.

Вам также может понравиться