link availabilty

7/29/2019 link availabilty

1/14

Link-Stability and Energy Aware RoutingProtocol in Distributed Wireless Networks

Floriano De Rango, Member , IEEE , Francesca Guerriero, Member , IEEE , and

Peppino Fazio, Member , IEEE

Abstract Energy awareness for computation and protocol management is becoming a crucial factor in the design of protocols andalgorithms. On the other hand, in order to support node mobility, scalable routing strategies have been designed and these protocolstry to consider the path duration in order to respect some QoS constraints and to reduce the route discovery procedures. Often energysaving and path duration and stability can be two contrasting efforts and trying to satisfy both of them can be very difficult. In this paper,a novel routing strategy is proposed. This proposed approach tries to account for link stability and for minimum drain rate energyconsumption. In order to verify the correctness of the proposed solution a biobjective optimization formulation has been designed and anovel routing protocol called Link-stAbility and Energy aware Routing protocols (LAER) is proposed. This novel routing scheme hasbeen compared with other three protocols: PERRA, GPSR, and E-GPSR. The protocol performance has been evaluated in terms ofData Packet Delivery Ratio, Normalized Control Overhead, Link duration, Nodes lifetime, and Average energy consumption.

Index Terms MANET, scalable routing, biobjective optimization, link stability, energy consumption.

1 INTRODUCTION

THE aim of this contribution is the proposal of a novelrouting protocol able to account for a joint metric of linkstability and minimum energy drain rate in mobile ad hocnetwork (MANET). This protocol was enhanced by theintegration of a multiobjective integer linear programmingoptimization model, whose solution was calculated throughthe LINGO tool [1].

Energy is an important resource that needs to bepreserved in order to extend the lifetime of the network[2], [3], [4], [5], [6], [7], [8], [9], [10], [11]; on the other hand,the link and path stability among nodes allows thereduction of control overhead and can offer some benefitsalso in terms of energy saving over ad hoc networks [12],[13], [14], [15], [16], [17], [18], [19]. However, as will beshown in this contribution, the selection of more stableroutes under nodes mobility can lead to the selection of shorter routes. This is not always suitable in terms of energyconsumption. On the other hand, sometimes, trying tooptimize the energy can lead to the selection of more fragileroutes. Thus, it is evident that both the aforementionedparameters (i.e., link stability associated with the nodesmobility and energy consumption) should be considered indesigning routing protocols, which allow right tradeoff between route stability and minimum energy consumptionto be achieved [20], [21].

The main aim of this work is to propose an optimizationrouting model within a MANET. The model attempts tominimize simultaneously the energy consumption of the

mobile nodes and maximize the link stability of thetransmissions, when choosing paths for individual trans-missions. The idea of considering, at the same time, energyconsumption and link stability is motivated by the observa-tion that most routing protocols tend to select shorterroutes, in this way high efficiency in using wireless bandwidth and increase path stability are ensured. How-ever, such routes may suffer from a higher energyconsumption, since higher transmission ranges are needed.

Consequently, in order to take into account the energyconsumption and link stability of mobile nodes, a biobjec-tive integer programming (BIP) model was formulated.Moreover, a greedy approach to find the solution to the BIPmodel was adopted and the found suboptimal solution waspreviously verified in [22] by using the software packageLINGO [1]. In this manuscript, on the basis of the biobjective optimization model presented in [22], a routingprotocol is proposed and its validity is experimentallyinvestigated through simulations. A comparison with otherknown approaches, such as Power Efficient ReliableRouting Protocol for Mobile Ad Hoc Networks (PERRA)[21] and geographic routing [2], [23], [24], [25] (in particularGPSR [25]), is also carried out.

The rest of the paper is organized as follows: the nextsection gives a brief overview of the existing link/pathstability and energy aware metrics and routing protocols;the biobjectives optimization problem is described andformulated in Section 3; the proposed routing protocol with joint optimized metrics is presented in Section 4; finally,simulation results and conclusions are summarized, respec-tively, in Sections 5 and 6.

2 RELATED

WORK

The description of some works related to the link stability,energy metrics and the respective routing protocols is givenin this section. In particular, after introducing some recent

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 23, NO. 4, APRIL 2012 713

. The authors are with the Department of Electronic, Informatic and Systems(DEIS), University of Calabria, Via Pietro Bucci, Rende 87036, Italy.E-mail: {derango, guerriero, pfazio}@deis.unical.it.

Manuscript received 1 Mar. 2009; revised 7 Aug. 2009; accepted 23 Oct. 2009;

published online 24 Aug. 2010.Recommended for acceptance by M. Ould-Khaoua.For information on obtaining reprints of this article, please send e-mail to:[email protected], and reference IEEECS Log Number TPDS-2009-03-0093.Digital Object Identifier no. 10.1109/TPDS.2010.160.

1045-9219/12/$31.00 2012 IEEE Published by the IEEE Computer Society


2/14

contributions that separately account for path or linkstability and energy consumption, a few papers on jointenergy-path stability metrics are summarized [20], [21], [22]and the specific contributions of this manuscript are listed.

2.1 Path Stability Aware Metrics and RoutingProtocols

In the literature, many metrics focusing on the link or pathstability have been defined. Among them, some have been based on the definition of the route breakage probabilityand some others on the link duration distribution. How-ever, most of them have considered some parametersassociated with the specific mobility model in order toestimate the stability metric.

In [12], [19] the authors make use of statistical prediction based on the node movement. In this approach, a linkstability probability has been defined on the basis of therandom mobility model.

A formal model to predict the lifetime of a routing path, based on the random walk mobility and on the prediction

technique, was proposed in [18]. It considers a probabilitymodel derived through the subdivision into cells of the areawhere mobile nodes move and on the observations of nodemovements in these cells. Transition probabilities arecalculated and a state-based model of the movement amongthe cells is considered. Each connection between a mobilenode in a cell and the other mobile nodes among itsneighbor cells is considered as the state of the wireless link.In this way, the wireless link dynamic is determined between a mobile node and its neighbors, permitting thecalculation of the link lifetime. After this, through theassumption of independent link failure, the route breakage

probability is derived. More details can be found in [18].There have been many papers published over the lastfew years where the link stability probability is determined by considering the signal stability [26]. However, thisapproach cannot be suitable because it assumes that thesignal strength can be affected by environmental conditionsand its value can change a lot also for the same nodesdistance. This determines many fluctuations in the radiosignal measurement, producing erroneous considerationson the link stability.

Other techniques rely on the use of special devices, suchas the Global Positioning System (GPS), to detect the exactposition of the mobile nodes [23]. Each node can calculateits position and a protocol is applied, which disseminates orrequests the position for the other nodes. This approach isalso criticized, because in some environments, such asindoors or where the mobile nodes are greatly limited inenergy, the GPS is not functional. Some enhanced versionsof GPSR have been presented in literature such as [39], [40],[41], where some movement prediction is applied in orderto reduce the effect of bad location information; however,these fully location-based routing schemes do not account jointly for other metrics such as energy and stability.

In [14], [15] the authors propose a novel approach to inferresidual link lifetime basing the computation on the current

link age and on the previous link observations. Fivedifferent metrics, for stable path selection, have beenproposed in the literature: the first technique is based onthe local choice of the oldest link as the most stable link; the

second class of metrics concerns the selection of theyoungest links, because they are considered more resilientto breakage; the third criterion is based on the selection of the link with the highest average residual lifetime value; thefourth one makes selection of the link with the highestpersistence probability; finally, the fifth metric focuses onthe connection failure probability. The latter approach has

been shown to be robust because it is based on themonitoring of the links lifetime of the mobile nodes in thewireless network, in the past and in the present, to predictits behavior, in the future without considering directlyparameters depending by underlying mobility model suchas node speed or direction.

The path stability, in terms of the number of routetransitions a routing protocol incurs to continue the dataexchange, is considered in [20]. End-to-end delay of asource destination session is another considered perfor-mance metric, particularly for real-time applications. In thiswork, the idea of stability-delay tradeoff (SDT), as ameasure of the efficiency of an MANET routing protocol,was introduced.

In [41], the authors propose a prediction location-basedrouting scheme in order to increase the delivery ratio of GPSR and select the more stable route. However, such as forthe previous listed contributions, energy is not consideredin the packet forwarding.

2.2 Energy Aware Metrics and Routing ProtocolsMany contributions concerning energy consumption have been proposed in the literature. In this section, the focus ison the network layer because it is interesting to show somerecent papers on energy aware routing protocols and onenergy-based metrics. Moreover, some additional metricshave tried to consider also the battery life cycle [8], or theenergy drain rate [27]. In the following, after brieflydescribing some energy related metrics, energy awarerouting protocols paradigms are also summarized.

2.2.1 Minimum Drain Rate (MDR) Cost Energy saving mechanisms based only on metrics related tothe remaining energy cannot be used to establish the bestroute between source and destination nodes. If a node iswilling to accept all route requests only because it currentlyhas enough residual battery capacity, much traffic load will be injected through that node. In this sense, the actual drainrate of energy consumption of the node will tend to be high,resulting in a sharp reduction of battery energy. As aconsequence, it could exhaust the node energy supply veryquickly, causing the node soon to halt. To mitigate thisproblem, other metrics, based on the traffic load character-istics, could be employed. To this end, techniques tomeasure accurately traffic load at nodes should be devised.In particular, the Minimum Drain Rate will be consideredand explained in Section 3.

2.3 Energy-Based Routing Protocols

In addition to energy aware metrics such as those describedin the previous section, also routing strategies and differentstate info management through routing protocols have beenproposed in the literature.

714 IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED SYSTEMS, VOL. 23, NO. 4, APRIL 2012


3/14

In [10], a distributed power control has been designed asa way to improve the energy efficiency of routing algorithmsin ad hoc networks. Each node in the network estimates thepower necessary to reach its own neighbors, and this powerestimate is used for tuning the transmission power (therebyreducing interference and energy consumption).

In [28], an energy efficient Optimization Link StateRouting was proposed. This approach is based on theproactive info management and on the selection of Multi-point Relay (MPR) based on energy metrics, such asMMBCR and MDR.

In [27], the authors proposed an on-demand protocol based on the MDR metric and using a route discoverymechanism and route maintenance, similar to DynamicSource Routing (DSR).

In [19], a Life-time Prediction-based Routing (LPR),focused on the minimization of the variances of the nodesremaining energies in the network, is proposed. In thisprotocol, each node tries to predict the future energyexpenditure, but its estimation depends on many factorsuch as node distances, residual power, hop count, andnode mobility.

2.4 Link Stability and Energy Aware RoutingProtocols

There are few multiple metrics aware routing protocols,over distributed wireless systems, in the literature. How-ever, in the context of novel distributed wireless systemsand multimedia applications, where the system complexityis increasing, the chance of controlling and evaluating morenetwork parameters becomes an important issue.

In this context, the use of multiobjective formulation andmultiple metrics plays a crucial role. To the best of ourknowledge, only two published works consider simulta-neously link stability and energy consumption for pathselection, which is the focus of this study (i.e., [20], [21], [22]).

Specifically, a routing protocol called Power EfficientReliable Routing protocol for mobile Ad hoc networks wasproposed in [21]. This algorithm applies the following threemetrics for path selection: 1) the estimated total energy totransmit and process a data packet; 2) the residual energy;3) the path stability. Route maintenance and route discoveryprocedures are similar to the DSR protocol [21], but with theroute selection based on the three aforementioned metrics.

In [20], [32] theauthors evaluatedthe performance ofsomepath stability-based routing protocols, that is the Associativ-

ity-Based Routing (ABR), Flow-Oriented Routing Protocol(FORP)androute-lifetimeassessment-basedroutingprotocol(RABR). They consider the energy consumption of the abovelisted protocols in order to see the best candidate from anenergetic point of view. However, in the aforementionedcontributions, no novel metric is proposed and the pathselection is performed by considering only the route stabilitymetric. In [22], the authors gave a first contribution on themultiobjectivemathematicalformulationofaroutingscheme,that considers two metrics, that is stability and energy.However, in this first contribution, only the optimizationproblem wasformulated,whereas noanalysis on theprotocolmanagement andprotocol performance hasbeen carried out.

2.5 Contribution of the ProposalIn our proposal, after applying one of the metrics presentedin [14], [15] a heuristic, based on the local knowledge of the

neighborhood, is applied. Following this heuristic the nexthop toward destination is selected among the neighbornodes, that maximize (minimize) the joined link-stability-energy metric.

This local criterion permits a high scalability to be offeredto the routing algorithm in terms of state info storage andcontrol packets transmission sent by any underlying routing

protocol to maintain the network state knowledge.On the basis of previous considerations, the maincontributions of this manuscript are the following:

1. A multiobjective mathematical formulation for the joint stability and energy problem is presented.

2. The proposed protocol is based on a geographicparadigm [23], [24], [25], different by other routingprotocolsaccountingfor jointmetrics,suchas PERRA.

3. Adoption of a novel stability metric based on theresidual link lifetime concept. This metric is con-sidered more robust than the metric proposed in [21] because it is independent on the transmission radiusand node speed parameters, that can be affected bymeasurement errors.

4. A novel energy aware-metric, adopted in ourprevious contributions [28], has been introduced inthe proposed optimization model in order toconsider not only the residual energy but also itstime variation associated with the traffic load.

5. The multiobjective routing algorithm is integratedin the scalable routing protocol and its performanceis tested through simulations and comparison withPERRA [21], GPSR [25], and an enhanced versionof GPSR [41] called Ellipsoid algorithm-based

GPSR (E-GPSR).

3 MATH FORMULATIONThe problem of finding an optimal path in a MANET wasformulated as a bicriteria constrained optimal path model[22], which turns out to be intractable. Indeed, the number of efficient solutions may be exponential in the problem size.Consequently, it is not possible to define efficient methods todetermineall efficientpaths[33],[34], [35], [36].Symbols usedfor the mathematical modeling are listed below in Table 1.

Let xi;j denote the binary variable associated with the

link i; j 8i; j 2 A that is set to one if the link belongs tothe path and zero otherwise.Before introducing the overall math formulation of the

bicriteria-based path selection, the single metric definitionsare provided.

3.1 Link-Stability Aware MetricIn this paper, differently from [18], a link stability metricrather than a path stability metric is considered. This is dueto the protocol scalability properties that we tried to offer tothe routing scheme. As will be shown in next sections, anode with the best tradeoff between link stability and

energy consumption is adopted through a local forwardingcriterion. Before explaining the method adopted to estimatethe link stability grade, the definition of link stability isprovided in what follows:

DE RANGO ET AL.: LINK-STABILITY AND ENERGY AWARE ROUTING PROTOCOL IN DISTRIBUTED WIRELESS NETWORKS 715


4/14

Definition 1. A link between two nodes i and j with

transmission range R is established at time instant t in whenthe distance between both nodes is such thatd i; j < R .Definition 2. A link between two nodes i and j with

transmission range R is broken at instant time t fin when thedistance between both nodes verify the conditiond i; j > R .

Definition 3. A link age a between two nodesi and j is theduration ai; j a i;j tfin t in .

In order to consider a robust stability estimation index,that is independent on the single calculated measure and onthe adopted mobility model, the use of parameters to

evaluate the stability variables such as node speeds,direction change frequency, pause time, etc., is avoided.Moreover, a statistical-based approach has been adopted

in order to discriminate among several links which are more

stable, meaning they are the most likely of all to stayavailable for some periods of time, without exactly predict-ing the residual link lifetime of each link. Thus, to enablemobile devices to make smart decisions in relationship to thestability, a practical method is used, based exclusively onobservations related to the link, in previous time instants. Asa result, this analysis produces an evaluation of the link

residual lifetimeof the link, since the stability of a link is given by its probability of persisting for a certain time span [15].The link residual lifetimerepresents the potential remainingtime that the link can exist before breaking.

By following the strategy outlined in [15], in theproposed mathematical model, the expected residual life-time R i;j a i;j of a link i; j of age a ij is determined from thecollected statistical data as follows:

R i;j a i;j PAmaxaa ij a d a

PAmaxaa ij d a a i; j 8i; j 2 A; 1

where amax represents the maximum observed age of thelinks and d is an array of length amax 1 used to store theobserved data.

In particular, d is determined through a sampling of thelink ages every fixed time interval and its genericcomponent d a represents the number of links with ageequal to a.

The coefficient R i;j a i;j (see (1)) is defined as the ratio between the sum, on all links with age equal or greater thana i;j , of the products of the age a and the number of linkswith age equal to a (that is d[a]), over the total number of links with age greater or equal to a ij . The main disadvan-tage of using the coefficient R i;j a i;j for path selection isrelated to the fact that it does not allow discriminationamong links of the same age.

In order to overcome this drawback, the averagetraveled distance d avg i;j should be taken into account. Itrepresents the average among the movement distances of each node under the Random Way Point (RWP) mobilitymodel. Each node selects a target position to reach with arandomly selected speed and after reaching this position itselects a new one. In our approach we store the crosseddistances and we determine the average. The rationale isthat if two links have the same residual lifetime, a shorteraverage distance is preferable to a longer distance in termsof link stability [22].

Indeed, the transmission and reception operations,associated with data and control packets, are morecomputational expensive in comparison with the operationsassociated with the links ordering. This is particularly truein the case of MANETs, where the node density, defined asthe number of neighbor nodes, is very limited (i.e., ranging between 5 and 10) even in the case of very high-densitynetworks [37]. For this reason, this work does not take intoaccount the number of reordering operations of links with

different expected residual lifetimes.On the basis of the previous considerations, the stabilityof the link i; j 8i; j 2 A, at time t , has been represented by the coefficient s i;j t, defined as follows:


TABLE 1Symbols Adopted in the Math Formulation


5/14

s i;j d avg i;j

R i;j a i;j k8i; j 2 A; 2

where k is a scaling factor, defined in such a way that thelink stability can be compared to the energy consumption.The coefficient s i;j t defined in (2) can be interpreted as areciprocal measure of the stability. It has been defined inthis way for the following reasons: first of all, it is assumed

that the second objective function of the proposed mathe-matical model has to be minimized; in addition, the higherthe residual lifetime of a link, the higher the reliability of thelink. Finally, the higher the average traveled distance between two nodes, the higher the likelihood that theirdistance exceeds the transmission radius with a consequentthe link breaks up.

3.2 Energy-Aware MetricIn this study, it is assumed that each wireless node has thecapability of forwarding an incoming packet to one of itsneighboring nodes and to receive information from a

transmitting node. In addition, each node is able to identifyall its neighbors through protocol messages (this issue will be explained in Section 4). It is assumed that each nodedoes not enter in standby mode and each node canoverhear the packet inside its transmission range and it isnot addressed to itself.

The energy needed to transmit a packet p from node i is:E tx p; i I v tb Joules, where I is the current (inAmpere), v the voltage (in Volt), and tb the time taken totransmit the packet p (in seconds).

The energy E(p, i) spent to transmit a packet from node ito node j is given by

E p; i E tx p; i E rx p; j; 3where E tx and E rx denote, respectively, the amount of energy spent to transmit the packet from node i to node jand to receive the packet at node j ; to the energy spent tooverhear the packet has been avoided in this context such asreferred in [28]. The power dissipated by mobile nodes toexchange beacowning messages and/or to remain always inactive modality is also considered.

In the proposed model, the power dissipation isdetermined by considering both the power consumptionat a transmitter P i;j t and the power consumption at areceiver Qi;j t.

In particular, the transmission power is modeled asfollows [3], [4], [32]:

P i;j t ci;j f i;j ; 4

where P i;j trepresents the power dissipated to transmit ata given instant of time t and f i;j is the rate of the data streamsent from node i to j .

The coefficient ci;j represents the power consumptioncost per bit associated with the link (i, j) and it is modeledas follows:

ci;j b ~ d i;j ; 5

where b is a distance-independent parameter accounting forthe network characteristics; represents the path loss indexand 2 4 [3], [4], [32];d i;j denotes the last observation

of the distance between node i and node j ; ~ is a coefficientassociated with the distance-dependent term, which has been defined in such a way as to take into account thereciprocal movement among nodes.

In particular, ~ is modeled as follows:

~ d avg i;jd 0i;j

; 6

where d 0i;j represents the distance between node i and node j observed at the time t0, when the link is formed for thefirst time, d avg i;j is the average traveled distance betweennodes i and j at time t , whereas is the parametertraditionally associated with the physical distance betweennodes in static networks [3], [4], [32]. The power wasted inthe reception is movement and time independent and it ismodeled in the following way:

Qi;j f i;j ; 7

where represents the energy dissipated to receive one bitof information. In our model, it is assumed that is constantand the same for every node. f i;j is the data rate from node ito node j .

It is important to point out that, starting from node i, thegeneric nondestination neighboring node j can be selected if and only if both the following conditions are satisfied:

1. j has enough energy to receive the information sentfrom node i,

2. j is able (in terms of energy) to transmit theinformation toward another relay node.

In order to guarantee that conditions (a) and (b) are fulfilled,

two constraints have been introduced in the proposedmodel.Let T i;j be the time required to send a packet of

information from node i to node j and let E res i denote theresidual energy of the node i 8i 2 N ; the first constraint isstated as follows:

T i;j Qi;j E res j 8i; j 2 A; j 6 D; 8

where is a parameter between 0 and 1 (i.e., 0 < < 1), thathas been introduced to avoid the selection of a node forwhich the energy required to receive a message is equal toits residual energy.

It is also important to guarantee that node i is able tosend the information toward a neighbor j without goingdown. Indeed, the following constraint has to be fulfilled:

T i;j P i;j t E res i 8i; j 2 A: 9

Constraint (9) means that the energy to transmit theinformation from node i toward node j should be loweror equal to the residual energy of node i. It is assumed thateach node is able to estimate its residual energy.

The Minimum Drain Rate, such as in [27] has beenapplied with a cost function that takes into account thedrain rate index (DR) and the residual energy ( E res ) to

measure the energy dissipation rate in a given node.Each node i monitors its energy consumption caused bythe transmission, reception and overhearing activities andcomputes the energy drain rate, denoted by DR i , for every



6/14

T seconds sampling interval by averaging the amount of energy consumption and estimating the energy dissipationper second during the past T seconds. The actual value of DR i is calculated by utilizing the well-known exponentialweighted moving averagemethod applied to the drain ratevalues DR it 1 and DR curr ; i, which represent theprevious and the newly calculated values

DR curr;i DR i t ; 10

DR it DR it 1 1 DR curr;i : 11

In addition to drain rate DR i , also another parameter calledPropensity (PR) is applied. The P R i term expresses thepropensity of a node i to receive information from anothernode. Given a generic instant of time t , the coefficientPR j t 8 j 2 N i , is defined as follows:

PR j t E res jE 0 j

; 12

where E 0

j represent the initial energy of node j .The propensity PR j t is used, in conjunction with thepower transmission P i;j tand drain rate DR j to define thecoefficient ei;j tthat is used in the first objective function of the proposed model, to characterize a node from anenergetic point of view.

More specifically, ei;j t can be represented as follows:

ei;j t P i;j tPR j t

DR jt

E 0 jT i;j : 13

For each link i; j 2 A and for each instant of time t , thecoefficient ei;j t is defined as the ratio between the power

P i;j tdissipated to send information from node i toward tonode j and the propensity PR jt of node j .

The rationale of defining the coefficient ei;j t as in (13)is that the objective function has to be minimized. Inaddition, the higher the value of PR jt, the higher thepropensity of node j to receive information, whereas thehigher the value of P i;j t, the lower the advantage of selecting node j . Moreover, the higher is the energy drainrate DR i associated with node i, the higher is theconsumption in the time of energy and, consequently, thehigher is the link cost ei;j .

3.3 Multiobjective Problem Formulation

The problem of selecting the best path (in the Pareto sense[33], [34], [35], [36]) connecting node s to node d , accountingfor energy consumption and link stability can be mathema-tically stated as follows:

min f 1 Xi;j 2A

ei;j t x i;j ; 14

min f 2 Xi;j 2A

s i;j t x i;j ; 15

subject to

xi;j T i;j P i;j t E res i 8i; j 2 A; 16

xi;j T i;j Qi;j t E res j 8i; j 2 A; 17

Xi;j 2A

x i;j Xi;j 2A

x j;i 1 if i S;0 if i 2 N nf S; D g;

1 if i D;

8> p 2), whereas for applications, where it is important toreduce the link breakage and the queuing delay, moreimportance could be given to the stability weight p2 (i.e., p2 > p 1). Moreover, all metrics are calculated on a local basiseach t time interval. This permits to update the parametersmodel inorder tocapture thenetworkdynamics.Moredetailsabout this sampling procedure are given in [21].



7/14

4 LINK S TABILITY AND ENERGY AWARE ROUTING(LAER) P ROTOCOL

The LAER algorithm requires each node i to advertise itslocation ( xi , yi , z i), rate of energy consumption ( MDR i), andlink stability index for each link outgoing by node i. Wewill insert the information mentioned above in LAER HELLO packet.

Each node broadcasts HELLO packets to all its neighborsthat are in its communication range; each node in LAERmaintains the table of its direct neighbors.

When a node receives the HELLO packet, it updates theinformation of the neighbor, if neighbor ID is alreadypresent in table or adds a neighbor information, if it is anew neighbor.

4.1 Forwarding StrategyThe data forwarding strategy of LAER is based on a greedytechnique such as GPSR. However, differently by GPSR, thenext hop selection tries to minimize the joint energy-stability metric. LAER packet forwarding presents highscalability property because only the neighborhood anddestination knowledge are necessary for the greedytechnique. The flexibility of energy-stability-based greedyforwarding is offered through the capability to weight thestability and the energy consumption on the basis of theinterest of the application layer. This means that if anapplication is more sensitive to the path stability and,consequently, the link stability, it is possible to give moreimportance to the s i;j index. On the other hand, anapplication that needs to prolong the network lifetime andto reduce the energy consumption also selecting longerroute with higher data packet end-to-end delay, the ei; jterms is more considered, as underlined in (10).

In Fig. 1, it is shown the packet forwarding under thegreedy techniquebased on theeuclideandistance (see GPSR)and the forwarding scheme with the joint stability andenergyaware metric. In particular, in the figurethe followingsituation is depicted: S falls in thetransmissionrange of noden1 (and vice versa), n1 in the transmission range of nodes n2and n4 , n2 in that of n3 and n5 , and n3 in that of node D . It is

possible to observe as the selected path can be differentdependingon themetric considered andon the weights usedwhen the joint metric is applied. In particular, the GPSRscheme selects the path S-n1-n2-n3-D because all neighbor

nodes that minimize the distance toward the destination Dare selected (maximum progress), whereas the LEARforwarding scheme selects S n1 n4 n2 n3 D path.

This means that LAER selects a longer path but with higherresidual energy.In order to avoid either routing loop or long packet

detour and to offer always a progress direction, acombined euclidean distance-based forwarding and a jointstability-energy metric for the next hop selection areadopted. In particular, it is selected as next hop theneighbor node j of current node i with the highest f tot , anda distance from destination equal or lower than the currentnode i.

This approach guarantees, as GPSR, a progress directionin the application of greedy technique but, differently from

GPSR, it permits also to select the best candidate for the joint metric rather than the node with only the highesteuclidean progress direction. In the following, it is shownthe pseudocode of modified LAER Greedy-Forward fromnode i at the arrival of packet p.

LAER GREEDY-FORWARD (p, i) jbest ncurr ;smin max si; j; 8 j 2 N id best Distance jbest ;p:d ;For each j 2 N i

Calculate f tot i; j ;distance Distance j; p:d ;

if distance < d best then f for each j 0 2 N iif s i;j < s min then fsmin s i;j ; jbest j 0;g

gif j best ncurr then return LAER Greedy failure;elsef forward p to j best ; return LAER Greedy success;g

where N i is the set of neighbors of the node i.The LAER algorithm fails when no neighbors with

progress direction in the euclidean sense is found and theLAER greedy failure is returned such as expressed in the

pseudocode above. This is the case when for the currentnode n1 where the euclidean distance of all its neighborsfrom D is greater than distance n1D such as shown in Fig. 2.Like in greedy routing, at this point node n1 thinks that


Fig. 1. Greedy forwarding (a) GPSR greedy approach based on theeuclidean distance; (b) LAER greedy based on the joint metric f tot . Fig. 2. Perimeter forwarding in GPSR and LAER protocols.


8/14

there is a hole (or local maximum) in the geographicaldistribution of nodes and a recovery procedure is appliedsuch as will be explained below. On the other hand, if alsothe recovery procedure fails the data packet is dropped.

In our implementation of LAER, we recovered from thissituation by applying Karp and Tung solution (GPSR) [25]that uses a planar subgraph of the wireless networks graphto route around holes. More details about this recoverytechnique are provided in the next section.

4.2 Local Maximum Recovery StrategyDuring the greedy technique, it is possible to meet a voidor local maximum in the GPSR. Local maximum representsa point in the network where it is not possible to find anyneighbor node that leads to the minimization of thedistance toward the destination in comparison with thecurrent node (see Fig. 2). In this case, the protocol assumesto use a recovery mode called Perimeter Forwarding. Thistechnique permits the go out from local maximumselecting other neighbor nodes among the perimeter of

the polygon face such as explained also in [24], [25], [39].In LAER, this situation can happen due to greedy routingstrategy based on the minimization of the joint metric f totassociated with each link (i,j). In our case it is met a localmaximum if a node i cannot find any neighbor node j 8 j 2 N i , that minimizes f tot , where N i is the set of isneighbors. In this case, LAER uses an approach similar toGPSR but the joint metric is used to select the set of neighbor candidates for the perimeter mode. An exampleis shown in Fig. 2 where a node n1 is in a local maximumand it has to select a neighbor on the basis of the perimeterforwarding. A set of possible candidates is f n2; n6; n11 g

that represents n1s neighbors on distinct faces of thepolygons obtained after the graph planarization. In ourapproach, the Gabriel planarization is applied accordingwith [25] to avoid loop in the perimeter modeand also theright hand rule. However, differently from authors in [25], itis not selected the node on the face with the minimumangle formed by the direction followed by the data packetentering in n1 (from s to n1) and the line connecting s anda specific neighbor on the planarised graph. In our case,we applied the joint metric with just one metric todiscriminate the neighbor node to select in the perimeter forwarding mode. It is important to observe that it isassumed that a local maximum is a network point thatdoes not allow progress to be made in terms of the jointf tot metric. However, this condition can be avoided if thesingle metrics are applied. Thus, in LAER recovery mode,first of all a n1 neighbor node with the lowest si;j metric isselected. If no neighbor is found, the other metric isconsidered ei;j and in the last case the approach of theoriginal GPSR is applied. The motivation to select in primisthe link stability aware metric in the recovery mode is topromote the selection of shorter and more stable routes toreduce the long packet detour determined by the classicalperimeter forwarding [37].

Let us consider now a node i with a list of its neighbors

indicated with N i and their relative positions. Let iin bethe ingress angle calculated on the basis of the direction of the data packet arriving on the node i (it is possible tocalculate it through the knowledge of position of node

sending the packet and the node i receiving the packet);iout is the angle formed by two lines: the first one

connecting node i and its predecessor and the secondconnecting node i and j 2 N i . The GPSR right-hand-ruleselects the node on the polygon face that minimizes thedifference i iout iin 8 j 2 N i . In order to determine

iin and iout it is necessary to calculate the ATAN ( tan

1) of the angle formed by the lines, as expressed above. Theinformation about the coordinates contained in the packetarriving on the node ( self .l.x, self.l.y), the coordinates of thenode i sending the packet ( i.x, i.y), and the coordinates of neighbor node j with j 2 N i permit to calculate these twoangles through a simple inverse trigonometric formula.

Right-Hand-Forward (p, i)iin NORM ATAN self:l:y i:y;self:l:x i:x min 3 ; smin max s i;j ; emin max ei;j ; 8 j 2 N i

For each j 2 N i If s i;j < s min then f smin s i;j ; j min jg else if s i;j smin &&ei;j < e min then femin ei;j ; j min j

g else f jout NORM ATAN self:l:y j:y;self:l:x j:x curr NORM jout

jin

If curr < min then curr min ; j min j greturn j min ;

4.3 Packet FormatsConcerning the data packets and HELLO packets adopted by LAER, it is necessary a packet modification andextension because we need to update the info related toenergy index and stability index of neighbor nodes and because also the weights p1 and p2 can be determined byapplication layer on the basis of the importance given to theenergy consumption or to link stability. For this reason, amodified version of HELLO and DATA packet formats wasadopted such as presented below. In particular, LAERHELLO packet information is shown in Table 2.

The data packet format, instead, is presented in Table 3.In this case also the weights info is carried by data packets.

It is observed that L p is important to establish when toswitch from perimeter to greedy mode. When, in the perimetermode, a node with distance from destination lower than L pis reached, it is possible to switch in greedy forwarding.

Concerning e0, it represents that location of the first edge onthe new face of the polygon traversed. This last variable isimportant to apply the right-hand-rule according with acriterion similar to that adopted in [25].


TABLE 2Fields Adopted in LAER HELLO Packet


9/14

5 P ERFORMANCE EVALUATIONIt is assumed that all mobile nodes are equipped with IEEE802.11a network interface card, with data rates of 11 Mbps.In our simulations, the voltage V is chosen as 5 V and it isassumed that the packet transmission time t p is calculated by ph6106

pd 54106 ph=6:10

6 pd =54:106seconds, where phis the packet header size in bits and pd the payload size.

In order to validate the effectiveness of the LAERprotocol, some simulations and comparisons with otherenergy aware protocols have been assessed. In the follow-ing, it will be shown how LAER represents a good tradeoff in terms of protocol overhead, data packet delivery ratio(DPDR), and average energy consumption in comparisonwith the other protocols.

5.1 Protocols Considered for ComparisonProtocols adopted for comparison purpose are, respectively,PERRA as representative of on-demand routing protocolsaccounting multiple metrics and E-GPSR and GPSR asrepresentative of scalable- and location-based routing.

5.1.1 PERRAPERRA is an on-demand routing protocol that providesnew features achieving power efficiency and reliable datatransmission. Some basic functions are listed below:

1. PERRA uses a route discovery procedure throughthe RREQs propagation that involves just nodes thatmeet the sources energy requirements beforetransmitting data packets.

2. Data packets are transmitted through the optimumpath on the basis of the minimum residual energy,path stability, and total estimated energy to transmitand process a data packet.

3. Alternative routes are prepared in case of link breakand used before an actual break occurs.

The objective function in PERRA is the following:

f tot w1 E t w2 min E res w3 min LL ; 23

where E t is the energy spent in the transmission and in theprocessing of a packet, E res is the residual energy, and LLis the link lifetime. This approach selects the minimum

E res and LL among nodes belonging to the path fromsource to destination.

The main differences with our proposal are the applica-tion of an on-demand strategy, the flooding of the routerequest for the path discovery, and the different definitionof the metric. However, because this protocol is an exampleof a routing protocol using multiple metrics in the path

establishment, it has been considered a good candidate forcomparison with LAER.

5.1.2 Ellipsoid Algorithm-Based GPSR In order to compare LAER protocol with a novel GPSRversion, we considered an enhanced GPSR proposed in [41]called E-GPSR. The main features of E-GPSR are brieflylisted below:

1. Calculation of future position of neighbor nodes onthe basis of a prediction technique based on theLeast Squares Lattice filter and time series.

2. Selection of the next node to reach the destination based on the ellipsoid algorithm. Through thisapproach is selected the neighbor node that mini-mizes the difference distance between current totaldistance d and the future total distance d fromcurrent node to destination node. In particularconsidering a source node S , an intermediate nodeR, and a destination node D the ellipsoid algorithmselects the node that minimizes d d 0, whered d SR d RD and d 0 d 0SR d 0RD .

For more details about this approach, please refer to [41].

5.2 Simulation Parameters

To evaluate the LAER protocol, the ns-2 network simulatorwas used [38]. A wireless network is simulated, with50 nodes mo ving in a 870 870 m 2 area. Each node movesrandomly in this area, with a speed selected in a range [0,vmax ] with no pause time. Between mobile hosts there are 8and 16 CBR/UDP sources generating 8 packets/s (with apacket size of 512 bytes). The duration of each simulation is700 seconds. To extract average values, we simulated eachscenario five times.

Simulation output variables that have been considered inour simulator are:

.

Data packet delivery ratio: it is the number of packets received at destination on data packets sent by source.

. Protocol overhead: it is calculated as the number of HELLO packets sent in the LAER and GPSRprotocols and the number of RREQ, RREP, andRERR in the PERRA protocol.

To have detailed energy-related information over asimulation, the ns-2 code was modified to obtain theamount of energy consumed (energy spent in transmitting,receiving) over time. In this way, accurate information wasobtained about energy at every simulation time. We used

these data to evaluate the protocols from the energeticpoint of view.Simulation output variables considered in the evaluation

of the energy and link stability metrics are the following:


TABLE 3Fields Adopted in LAER Data Packet


10/14

. Average link stability: this parameter is adoptedrather than path stability because for protocols suchas GPSR, E-GPSR, and LAER the path stabilitycannot be considered due to the absence of a pathestablishment phase;

. Average energy consumption: this parameter allowsto make considerations about energy wastageassociated with the route maintenance and routediscovery and it accounts for energy consumptionduring transmission and reception of control anddata packets;

. Average node residual energy: it can be useful toevaluate the remaining energy in order to have anidea of the network lifetime;

. Variance of node residual energy: this parameter isconsidered to evaluate the distribution of energyamong nodes. The greater is the dispersion aroundthe average residual node energy, the higher is theunfairness in the node usage and in the energydissipation among nodes.

Simulation parameters adopted in the performance eva-luation campaigns are listed in Tables 4, 5, and 6. Table 4lists the common parameters adopted in the simulator

regardless the specific considered protocols. Tables 5 and6 present simulation parameters adopted, respectively, forthe PERRA and LAER protocols. PERRAs parameters arefixed as in [21].

5.3 Simulation ResultsTwo simulation campaigns are shown in the followingsections. The first one exploits the performance of theproposed protocol against PERRA, GPSR, and E-GPSR

considering the standard statistics such as DPDR andcontrol overhead. The second campaign focuses on the linkstability and energy consumption.

5.3.1 Data Packet Delivery Ratio and Control Overhead Evaluation

The DPDR for different number of connections is shown inFig. 3. GPSR, E-GPSR, PERRA, and LAER present similarperformance when the traffic load is not heavy withpercentage value about 99 percent for very low mobility.However, for higher traffic load and high mobility (10-20 m/s) the low scalability of PERRA is visible and LAER,

GPSR, and E-GPSR perform better. PERRA wastes band-width for control overhead and the reactive management of the protocol leads to a degradation of performance reducingthe DPDR to 85-90 percent. The curve depicted in Fig. 4testifies the increase in the normalized control overhead forhigher speed. It is possible to observe the good scalability of protocol based on the local topology knowledge such asLAER, GPSR, and E-GPSR. The greedy technique applied to both protocols and the only local control packets exchange(HELLO pkts) determines a similar performance of LAER,GPSR, and E-GPSR, differently by PERRA that is forced tostart new route discovery procedure that increases the

control overhead.Obviously, the changing of p1 and p2 parameters, asshown in Fig. 5, do not affect the performance of PERRA,GPSR, and E-GPSR. However, it is interesting to observe as


TABLE 4Common Parameters Adopted in the Simulations

TABLE 5PERRA Parameters Adopted in the Simulations

TABLE 6GPSR, E-GPSR, and LAER Parameters

Adopted in the Simulations

Fig. 3. Data packet delivery ratio for different maximum node speed andnumber of connections.


11/14

the coefficients associated with stability and energy metricsdo not affect also the performance of LAER. This is due tothe specific scenario where the initial node energy is so highto not see any node energy exhaustion. The differentperformance associated with the different metrics weightare more evident in the next section.

5.3.2 Link Stability and Energy Evaluation

In Table 7, it is possible to observe the increasing linkduration trend for decreasing node mobility such asexpected. The higher node mobility determines more link breakage reducing the average link duration.

Both PERRA, E-GPSR and LAER increase the linkduration because specific link aware metrics permit toselect the most appropriate nodes. However, it is possible tosee that LAER can increase the average link duration forfixed p1 and p2 values (they are fixed both to 0.5 in thegraphics where the nodes speed changes). This means thatthe link stability aware metric can better discriminate theneighbor nodes through the adoption of the history of the

link lifetime and the statistical behavior to infer considera-tion on the residual link duration, whereas PERRA throughconsideration of only node speed is not always able todiscriminate the longer link from a lifetime point of view.

The imprecision in the link stability metric of PERRA isaccentuated when more independent movements are made by a mobile node and the RREP message of PERRA is notable to predict these fast variations. On the other hand, astatistical characterization of the link duration can avoid tosend often control packets on the built path in order torefresh the previously discovered info. Moreover, in thesecond table (Table 8) the effect of the coefficients p1 and p2

is also evaluated. As expected, a higherp

1 value determinesthe selection of more stable nodes and consequently linkwith a longer lifetime. It is interesting to observe as theadvantage of link stability metric of LAER is more evidentfor lower speed. This is due to the fact that when the moststable node is selected and discriminated among otherneighbor nodes its link lasts more times due to the lowernode mobility and this increases a lot the average linkduration. Moreover, it is interesting also to see theimprovement of E-GPSR over GPSR on the basis of themobility prediction and ellipsoid algorithm implementa-tion. However, also LAER is able to perform well in termsof link duration with the addition of reduced energy

consumption as it is possible to see later.The average energy consumption for decreasing nodesspeed and different stability weight values are shown,respectively, in Figs. 6 and 7. It is interesting to observe


Fig. 4. Normalized control overhead versus maximum node speed.

Fig. 5. Normalized control overhead versus stability weights.

TABLE 7Average Link Duration for Decreasing Nodes Speed

TABLE 8Average Link Duration for Different Stability Weights

Fig. 6. Average energy consumption versus maximum node speeds.

Fig. 7. Average energy consumption versus stability weights.


12/14

how both GPSR, E-GPSR and LAER consume lowerenergy: this is due to the simplest topology managementand to the absence of route discovery procedures that areenergy consuming. Moreover, LAER improves further theperformance reducing the energy consumption about 15-20 percent in comparison with GPSR for 0.1-10 m/s andabout 30 percent in comparison with PERRA.

When coefficients p1 and p2 are changed it is possible toimprove even more the performance of LAER in comparisonwith GPSR, E-GPSR, and LAER. Increasing the stabilityweights leads to the selection of shorter and more stableroutes such as testified also in [22] and this increasesthe energy consumption of nodes. On the other hand, theincrease of the stability weight and the decrease of thecoefficient associated with the energy aware metric reducethe remaining residual energy on nodes such as highlightedinFig.9.ItisinterestingtoobservehowinFig.8theGPSRandE-GPSR perform better than PERRA for higher node speeds.

This can seem to be strange, but it is important to observe asforhigherspeed (15-20m/s) themobilityoffersa natural load balancing effect and this means that also even if GPSR is notconscious of the residual energy, different nodes are selecteddueto thelinkbreakageandlocaltopologychange.However,for medium and low speed the capability to discriminatenode with higher residual energy can become important andPERRAandLAERaremoreeffectiveandoutperformE-GPSRand GPSR.

Moreover, LAER presents better performance prolong-ing more the network lifetime because the local topologymanagement is less energy consuming than route discovery

procedure of PERRA. E-GPSR and GPSR show performancesimilar to that achieved with LAER, when node mobility ishigh, but the better next hop selection criterion of LAER isstill visible.

The last evaluated parameter is the variance of residualnode energy such as depicted in Figs. 10 and 11. Thisparameter has been adopted to consider the load balancingcapacity of the three protocols. It is interesting to see howGPSR is the worst in terms of load balancing because it

selects the node on the basis on the only minimumeuclidean distance from destination without consideringother parameters. On the other hand, PERRA, E-GPSR, andLAER present a similar trend due to usage of the residualenergy metrics (PERRA and LAER) and to the use of thelink stability of E-GPSR. However, LAER reduces thevariance permitting a lower dispersion of node energyaround the average, because the use of an energy awaremetric is able to consider not only the residual energy butalso the drain rate trend and the traffic load on each singlenode. A higher traffic load on a specific node implies ahigher drain rate and faster energy consumption. Thismeans that also the energy metric of LAER is better than

PERRA permitting to discriminate between nodes with thesame residual energy but with different traffic load.Moreover, to confirm our previous assertion the GPSRand E-GPSR energy variance are close to the PERRA andLAER energy variance when the nodes mobility is high totestify the benefic advantage of mobility that determines anatural traffic load balancing. It is important to notice alsoas the energy variance of E-GPSR is better than GPSR forlower node mobility because a criterion based on thestability (node position prediction) is more effective thanonly a topological criterion such as maximum progress(minimum distance from destination).

6 CONCLUSIONSA scalable routing protocol called LAER, based on the jointmetric of link stability and energy drain rate, has been


Fig. 8. Average node residual energy versus maximum node speeds.

Fig. 9. Average node residual energy versus stability weights.

Fig. 10. Variance of node residual energy versusmaximum nodespeeds.

Fig. 11. Variance of node residual energy versus stability weights.


13/14

proposed. It is based on the local topology knowledge and itmakes use of a greedy technique based on a joint metric anda modified perimeter forwarding strategy for the recoveryfrom local maximum. Its performances have been comparedwith other three protocols proposed in literature such asGPSR, E-GPSR, and PERRA. LAER protocol inherits thescalability of GPSR and E-GPSR, improving the perfor-

mance in terms of node selection with higher link durationwhen a higher weight is given to the stability index and ahigher residual energy is given to energy aware index.LAER outperforms PERRA in terms of control overheadand in terms of a higher capability to balance traffic loaddue to the minimum drain rate metric included in the jointmetric. Moreover, also the average link duration can belonger in comparison with PERRA and E-GPSR, due to thecapability to better discriminate the node behavior asso-ciated not only with the current node condition but alsowith the history of link lifetime.

REFERENCES[1] L. Schrage, Optimization Modeling with LINGO. Lindo Publishing,

2003.[2] C. Taddia, A. Giovanardi, G. Mazzini, and M. Zorzi, Energy

Efficient Unicast Routing Protocols over 802.11b, Proc. IEEEGlobal Telecomm. Conf. (GLOBECOM 05),pp. 555-560, Nov./Dec.2005.

[3] S. Lindsey, K. Sivalingam, and C.S. Raghavendra, PowerOptimization in Routing Protocols for Wireless and MobileNetworks, Handbook of Wireless Networks and Mobile Computing,I. Stojmenovic, ed., Wiley, 2001.

[4] L. Feeney and M. Nilsson, Investigating the Energy Consump-tion of a Wireless Network Interface in an Ad Hoc NetworkingEnvironment, Proc. IEEE INFOCOM, pp. 1548-1557, 2001.

[5] L.M. Feeney, Energy Efficient Communication in Ad HocWireless Networks, technical report, Swedish Inst. of ComputerScience (SICS), Draft Chapter, 2004.

[6] Q. Zhao and L. Tong, Energy Efficiency of Large-Scale WirelessNetworks: Proactive versus Reactive Networking, IEEE J. Selected Areas in Comm.,vol. 23, no. 5, pp. 1100-1113, May 2005.

[7] F. De Rango et al., OLSR versus DSR: A Comparative Analysis of Proactive and Reactive Mechanisms from an Energetic Point of View in Wireless Ad Hoc Networks, Computer Comm. J.,vol. 31,pp. 3843-3854, Oct. 2008.

[8] C.-K. Toh, Maximum Battery Life Routing to Support UbiquitousMobile Computing in Wireless Ad Hoc Networks, IEEE Comm. Magazine, vol. 39, no. 6, pp. 138-147, June 2001.

[9] C.E. Jones et al., A Survey of Energy Efficient Network Protocolsfor Wireless Networks, Wireless Networks J.,vol.7, no. 4, pp. 343-

58, Aug. 2001.[10] P. Bergamo, D. Maniezzo, A. Travasoni, A. Giovanardi, G.Mazzini, and M. Zorzi, Distributed Power Control for EnergyEfficient Routing in Ad Hoc Networks, Wireless Networks J.,vol. 10, no. 1, pp. 29-42, 2004.

[11] S. Jang, D. He, and J. Rao, A Prediction-based LinkAvailability Estimation for Mobile Ad Hoc Networks, Proc.IEEE INFOCOM 01, pp. 22-26, 2001.

[12] G. Lim, K. Shin, S. Lee, H. Yoon, and J. Soo Ma, Link Stability andRoute Lifetime in Ad-Hoc Wireless Networks, Proc. Intl Conf.Parallel Processing Workshops (ICPPW 02),p. 116, 2002.

[13] M. Gerharz, C. de Waal, M. Frank, and P. Martini, Link Stabilityin Mobile Wireless Ad Hoc Networks, Proc. IEEE 27th Conf. LocalComputer Networks (LCN), pp. 30-39, Nov. 2002.

[14] M. Gerharz, C. de Waal, P. Martini, and P. James, Strategies forFinding Stable Paths in Mobile Wireless Ad Hoc Networks, Proc.

IEEE 28th Ann. Conf. Local Computer Networks (LCN 03),pp. 130-139, 2003.[15] N. Meghanathan and A. Farago, Looking at Protocol Efficiency

from a New Angle: Stability-Delay Analysis, Proc. Second IntlConf. Mobile Computing and Networking,pp. 51-55, 2004.

[16] X. Wu et al., A Hybrid View of Mobility in MANETs: AnalyticalModels and Simulation Study, Computer Comm., vol. 31, no. 16,pp. 3810-3821, Oct. 2008.

[17] Y.-C. Tseng, Y.-F. Li, and Y.-C. Chang, On Route Lifetime inMultihop Mobile Ad Hoc Networks, IEEE Trans. Mobile Comput-ing, vol. 2, no. 4, pp. 366-376, Oct.-Dec. 2003.

[18] M. Maleki, K. Dantu, and M. Pedram, Lifetime PredictionRouting in Mobile Ad Hoc Networks, Proc. IEEE Wireless Comm.and Networking (WCNC 03), pp. 1185-1190, Mar. 2003.

[19] N. Meghanathan, Stability-Energy Consumption Tradeoff amongMobile Ad Hoc Network Routing Protocols, Proc. Third Intl Conf.Wireless and Mobile Comm. (ICWMC 07),Mar. 2007.

[20] K.-J. Kim and S.-J. Yoo, Power-Efficient Reliable Routing Protocolfor Mobile Ad-Hoc Networks, IEICE Trans. Comm., vol. E88-B,pp. 4588-4597, 2005.

[21] F. Guerriero et al., A Biobjective Optimization Model for Routingin Mobile Ad Hoc Networks, Applied Math. Modeling, vol. 33,pp. 1493-1512, Mar. 2008.

[22] F. De Rango et al., A Multi-Objective Approach for EnergyConsumption and Link Stability Issues in Ad Hoc Networks,IEEE Comm. Letters,vol. 10, no. 1, pp. 28-30. Jan. 2006.

[23] M. Mauve, J. Widmer, and H. Hartenstein, A Survey on Position-Based Routing in Mobile Ad-Hoc Networks, IEEE Network,vol. 15, no. 6, pp. 30-39. Nov./Dec. 2001.

[24] I. Stojmenovic, Position-Based Routing in Ad Hoc Networks,

IEEE Comm. Mag., vol. 40, no. 7, pp. 128-134, July 2002.[25] B. Karp and H.T. Kung, Greedy Perimeter Stateless Routing forWireless Networks, Proc. ACM/IEEE MobiCom 00,pp. 243-254,Aug. 2000.

[26] C.-K-. Toh, Associativy-Based Routing for Ad-Hoc MobileNetworks, Wireless Personal Comm., vol. 4, no. 2, pp. 103-139,Mar. 1997.

[27] K. Dongkyun;, J.J. Garcia-Luna-Aceves, K. Obraczka, J.-C. Cano,and P. Manzoni, Routing Mechanisms for Mobile Ad HocNetworks Based on the Energy Drain Rate, IEEE Trans. MobileComputing, vol. 2, no. 2, pp. 161-173, Apr.-June 2003.

[28] F. De Rango, M. Fotino, and S. Marano, EE-OLSR: EnergyEfficient OLSR Routing Protocol for Mobile Ad-Hoc Networks,Proc. IEEE Military Comm. Conf. (Milcom 08),pp. 1-7, Nov. 2008.

[29] P. Santi, Topology Control in Wireless Ad Hoc and SensorNetworks, ACM Computing Surveys, vol. 37, no. 2, pp. 164-194, June 2005.

[30] E.L. Lloyd, R. Liu, M.V. Marathe, R. Ramanathan, and S.S. Ravi,Algorithmic Aspects of Topology Control Problems for Ad HocNetworks, Mobile Networks and Applications,vol. 10, pp. 19-34,2005.

[31] V. Rodoplu and T.H. Meng, Minimum Energy Mobile WirelessNetworks, IEEE J. Selected Areas in Comm.,vol. 17, no. 8, pp. 1333-1344, Aug. 1999.

[32] Y. Hou, Y. Shi, J. Pan, A. Efrat, and S. Midkiff, Maximizing theLifetime of Wireless Sensor Networks through Single-SessionFlow Routing, IEEE Trans. Mobile Computing, vol. 5, no. 9,pp. 1255-1266, Sept. 2006.

[33] Y. Colletter and P. Siany, Multiobjective Optimization: Principles andCase Studies, Decision Engineering. Springer-Verlag, 2003.

[34] A.J.V. Skriver and K.A. Andersen, A Label Correcting Approachfor Solving Bicriterion Shortest Path Problems, Computer andOperations Research, vol. 27, pp. 507-524, 2000.

[35] T. Tanino, T. Tanaka, and M. Inuiguchi, Multi-Objective Program-ming and Goal Programming: Theory and Applications. Springer,2003.

[36] F. De Rango, M. Gerla, and S. Marano, A Scalable RoutingScheme with Group Motion Support in Large and Dense WirelessAd Hoc Networks, Elsevier Computers and Electrical Eng. J.,vol. 32,nos. 1-3, pp. 224-240, 2006.

[37] K. Fall and K. Varadhan,(eds.), ns Notes and Documents,TheVINT Project, UC Berkeley, LBL, USC/ISI, and Xerox PARC,http://www.isi.edu/~salehi/ns_doc/, Feb. 2000.

[38] D. Son, A. Helmy, and B. Krishnamachari, The Effect of Mobility-Induced Location Errors on Geographic Routing in Mobile Ad hocSensor Networks: Analysis and Improvement Using MobilityPrediction, IEEE Trans. Mobile Computing, vol. 3, no. 3, pp. 233-

245, July/Aug. 2004.[39] V. Naumov, R. Baumann, and T. Gross, An Evaluation of Inter-Vehicle Ad Hoc Networks Based on Realistic Vehicular Traces,Proc. Seventh ACM Intl Symp. Mobile Ad Hoc Networking andComputing, pp. 108-119, May 2006.



14/14

[40] W. Creixell and K. Sezaki, Routing Protocol for Ad Hoc MobileNetworks Using Mobility Prediction, Intl J. Ad Hoc andUbiquitous Computing, vol. 2, pp. 149-156, 2007.

[41] K. Yamazaki and K. Sezaki, A Proposal of Geographic RoutingProtocols for Location-Aware Services, Electronics and Commu-nications in Japan, vol. 87, no. 4, pp. 26-34, 2003.

Floriano De Rango received the degree incomputer science engineering in October 2000,and the PhD degree in electronics and tele-communications engineering in January 2005,both at the University of Calabria, Italy. FromJanuary 2000 to October 2000, he worked in theTelecom Research LAB C.S.E.L.T. in Turin as avisiting scholar student. From March 2004 toNovember 2004, he was a visiting researcher atthe University of California at Los Angeles

(UCLA). He is now assistant professor at D.E.I.S. Department,University of Calabria. He was the recipient of Young ResearcherAward in 2007. He served as a reviewer and the TPC member for manyInternational Conferences and reviewer for many journals such as IEEE Communication Letters , JSAC , IEEE Trans. on Vehicular Technology ,Computer Communication , Eurasip JWCN , WINET , etc. His interestsinclude Satellite networks, IP QoS architectures, Adaptive WirelessNetworks, Ad Hoc Networks and Pervasive Computing. He has co-authored more than 130 papers among International Journals andConferences Proceedings. He is a member of the IEEE.

Francesca Guerriero received the PhD degreein engineering of systems and informatics fromthe University of Calabria and is currently work-ing there as an associate professor of Opera-tions Research. Her main research interests liein field of network optimization. Other area ofinterests include logistics and parallel comput-ing. She has published extensively in thesefields, in a variety of journals, including Opera- tions Research , Computational Optimization and

Applications , Optimization Methods and Software , Journal of Optimiza- tion, Theory and Applications and Parallel Computing . She is actuallythe director of the LOGICA Laboratory, University of Calabria. She is amember of the IEEE.

Peppino Fazio received the degree in compu-ter science engineering in May 2004. SinceNovember 2004, he has been working towardthe PhD degree in electronics and communica-tions engineering at the University of Calabriaand received the same in January 2008; at themoment he is a research fellow at DEISDepartment of University of Calabria and he iscollaborating with Department of UniversidadPolitecnica de Valencia. His research interests

include mobile communication networks, QoS architectures andinterworking wireless and wired networks, mobility modeling for WLANenvironments, and mobility analysis for prediction purposes. He is amember of the IEEE.

. For more information on this or any other computing topic,please visit our Digital Library at www.computer.org/publications/dlib.
