SIMULATION · Keywords: Applications in computer networks, sensor networks, modeling and simulation...

http://sim.sagepub.com

SIMULATION

DOI: 10.1177/0037549707082330 2007; 83; 587 SIMULATION

Ioannis Chatzigiannakis, Georgios Mylonas and Sotiris Nikoletseas studying Routing Protocol Performance

A Model for Obstacles to be used in Simulations of Wireless Sensor Networks and its Application in

http://sim.sagepub.com/cgi/content/abstract/83/8/587 The online version of this article can be found at:

Published by:

http://www.sagepublications.com

On behalf of:

Society for Modeling and Simulation International (SCS)

can be found at:SIMULATION Additional services and information for

http://sim.sagepub.com/cgi/alerts Email Alerts:

http://sim.sagepub.com/subscriptions Subscriptions:

http://www.sagepub.com/journalsReprints.navReprints:

http://www.sagepub.com/journalsPermissions.navPermissions:

© 2007 Simulation Councils Inc.. All rights reserved. Not for commercial use or unauthorized distribution. by guest on August 2, 2008 http://sim.sagepub.comDownloaded from

http://www.scs.org/

http://sim.sagepub.com/cgi/alerts

http://sim.sagepub.com/subscriptions

http://www.sagepub.com/journalsReprints.nav

http://www.sagepub.com/journalsPermissions.nav


A Model for Obstacles to be used inSimulations of Wireless Sensor Networks andits Application in studying Routing ProtocolPerformanceIoannis ChatzigiannakisGeorgios MylonasSotiris NikoletseasDepartment of Computer Engineering and Informatics, andComputer Technology Institute (CTI)University of Patras26500 [email protected]@[email protected]

We propose a simple obstacle model to be used while simulating wireless sensor networks. To thebest of our knowledge, this is the first time such an integrated and systematic obstacle model forthese networks has been proposed. We define several types of obstacles that can be found insidethe deployment area of a wireless sensor network and provide a categorization of these obstaclesbased on their nature (physical and communication obstacles, i.e. obstacles that are formed out ofnode distribution patterns or have physical presence, respectively), their shape and their change ofnature over time. We make an extension to a custom-made sensor network simulator (simDust) andconduct a number of simulations in order to study the effect of obstacles on the performance of somerepresentative (in terms of their logic) data propagation protocols for wireless sensor networks. Ourfindings confirm that obstacle presence has a significant impact on protocol performance, and alsothat different obstacle shapes and sizes may affect each protocol in different ways. This provides aninsight into how a routing protocol will perform in the presence of obstacles and highlights possibleprotocol shortcomings. Moreover, our results show that the effect of obstacles is not directly relatedto the density of a sensor network, and cannot be emulated only by changing the network density.

Keywords: Applications in computer networks, sensor networks, modeling and simulation environ-ments, parallel and distributed computing, performance evaluation, obstacle avoidance, routing

1. Introduction

Wireless sensor networks are large collections of small,low-power, low-cost sensor devices that collect and re-port detailed information about their surrounding physi-cal environment. Large numbers of such devices can bedeployed in areas of interest (such as inaccessible ter-rains, disaster-struck areas, or embedded in our every-

SIMULATION, Vol. 83, Issue 8, August 2007 587–608c� 2007 The Society for Modeling and Simulation InternationalDOI: 10.1177/0037549707082330

day environment) and form a sensor network using self-organization and collaborative methods. Each one of thesedevices carries a number of sensors and a transceiver, en-abling it to sense its environment and communicate withother network nodes, respectively.

Because of their nature, i.e. the inherent restrictions inenergy, cost and processing power, wireless sensor net-works nodes are more prone to hardware failures and ad-versarial effects than nodes participating in other wirelessnetworks, and should also consume considerably less en-ergy in order to operate for very long periods of time. Fur-thermore, these networks are expected to be deployed in‘difficult’ or even hostile environments. Fault tolerance,

Volume 83, Number 8 SIMULATION 587



Chatzigiannakis, Mylonas, and Nikoletseas

low energy consumption and obstacle presence shouldtherefore be taken into consideration. All of the abovecharacteristics and restrictions make the design of proto-cols for wireless sensor networks a real challenge.

The flexibility, fault tolerance, high sensing fidelity,low cost and rapid deployment features of wireless sen-sor networks helped to create many new and exciting ap-plication areas for remote sensing. This wide range of ap-plications is based on the use of various types of sensors(i.e. thermal, acoustic, magnetic, etc.), in order to moni-tor a wide variety of conditions (e.g. temperature, objecttracking, humidity) and report them to one or more con-trol centers (fixed or mobile). This control center could besome human authorities responsible for taking action uponthe realization of some crucial event or a server forward-ing readings from the sensor network to the Internet forfurther use. Sensor networks can therefore be used in im-portant areas such as (1) environmental applications, suchas fire detection, flood detection and precision agriculture�(2) health applications, such as telemonitoring of humanphysiological data� and (3) home applications, e.g. smartenvironments.

The initial vision of sensor networks has been extendedto include additional characteristics, such as heteroge-neous sensor networks, i.e. networks consisting of nodesthat have different resources (processing power, energysupplies, etc.), or networks with actor nodes, i.e. nodesthat carry actuators apart from sensors. For a basic surveyof wireless sensor networks, see Akyildiz et al. [1].

1.1 The Problem – Motivation

As mentioned above, a wireless sensor network consistsof a large number of nodes that are usually scattered inan area of interest and form a sensor network, as shownin Figure 1. Each of these scattered nodes has the abilityto locally collect and route data to one or more controlcenters. Data is propagated to the control center using amulti-hop data dissemination protocol running on everysensor network node.

Assume the realization of a series of K crucial eventsEi , with each event being sensed by a single node pi�i � 1� 2� � � � � K �. We then define the multiple event prop-agation problem as: how can each node pi , via coopera-tion with the rest, propagate the information info�i� re-garding event Ei to the control center of the network in anefficient and fault-tolerant way.

Due to the nature of wireless sensor networks, i.e. thevery large number of nodes, real network deployments areexpensive and difficult to maintain, and this is one of thereasons few real large sensor network deployments cur-rently exist. Although such deployments are crucial foraccurately testing the actual performance of the proposednetwork protocols, wireless sensor network simulationsprovide many advantages over real world deployments,e.g. the ability to easily test a great variety of parame-ters in protocols and deployment settings. Simulation also

Figure 1. Sensor nodes forming a wireless sensor network

provides researchers with a number of other significantbenefits, including repeatable scenarios, isolation of para-meters and exploration of a variety of metrics. It is gen-erally not feasible or too strenuous to run such extensivetests in this manner over a real wireless network. Further-more, simulation represents a very useful complement torigorous analysis of algorithms in the context of compu-tational complexity. In order to obtain asymptotic resultsand investigate network scalability, such analysis is per-formed on simplified models that do not capture importantaspects, such as detailed technical specifications and net-work conditions. Because of the importance of simulationin wireless sensor networks, a need has arisen for the useof more realistic network models.

A large part of the simulations included in researchdealing with wireless sensor networks make many simpli-fying assumptions in order to conduct experiments withmany nodes over a reasonable time� however, the sim-plifications lower the overall simulation complexity. Eventhough there has been a vast amount of research in simu-lating wireless sensor networks, a common simplificationmade by the majority of the network models is that of un-obstructed areas, i.e. areas without any obstacles presentin the network deployment area. Mainly implicit model-ing assumptions on obstacles (i.e. on the node density)have appeared and such assumptions are certainly unableto model real obstacles realistically. It is our belief thatthe inclusion of obstacles has a great impact on both thedesign of protocols for wireless sensor networks, as wellas on the simulation and evaluation of the performance ofthese protocols. In order to be more realistic, since wire-less sensor networks are expected to be deployed in hos-tile environments and/or outdoor areas with harsh envi-ronmental conditions, obstacles should be taken explicitlyinto consideration while designing a protocol for such net-works and also while evaluating its performance via sim-ulation. In addition to physical obstacles, virtual obstaclessuch as the local absence of functioning nodes can emergeduring the network’s operation (e.g. when the energy ofsensor nodes in an area runs out).

588 SIMULATION Volume 83, Number 8



A MODEL FOR OBSTACLES TO BE USED IN SIMULATIONS OF WIRELESS SENSOR NETWORKS

The need for an obstacle model specifically for sensornetworks arises from the special characteristics of thesenetworks. Because of the inherent locality and simplic-ity of the protocols of sensor networks, the constrainedresources in the network and the lack of global networkknowledge, the presence of obstacles is expected to affectprotocol performance (and even correctness) in a numberof ways and should be taken into account. In this sense,a model adequate for simulating ad hoc networks may beunsuitable for wireless sensor networks, or may need ad-ditional features to produce good results.

The use of obstacles in simulations should, in manycases, have a large impact and could reveal inherent pro-tocol design weaknesses (especially in the case of greedygeographic routing algorithms) that may be difficult todiscover otherwise. Routing protocols that use geoloca-tion information to make forwarding decisions comprisea large part of the proposed routing algorithms for sen-sor networks. This is due to the fact that usually such in-formation is assumed to be available in sensor networknodes, since much of the information extracted from thesenetworks is location-specific. Another attractive featureof these protocols is that they require only a small ‘ta-ble’ of information regarding their neighbors, as opposedto larger routing tables. As an example, consider a sen-sor network with one control center and a large wall in-side the deployment area cutting off communication frommost sensor nodes to the control center. A purely greedyforwarding scheme (i.e. one which forwards data only ifa neighbor nearer to the control center exists) used insidesuch a deployment field will have little chance of over-coming this obstacle. Also, purely geographic routing al-gorithms could be trapped into dead ends formed by ob-stacles, as will be explained in the following section.

In this work, we will emphasize the usefulness of in-cluding obstacles in sensor network simulations and pro-pose a realistic simulation model in order to produce cred-ible results. Our obstacle model, apart from providing in-sight to the effects of obstacle inclusion in routing pro-tocol performance, is basically useful in detecting weak-nesses in routing protocol design and discovering situa-tions where routing protocols may fail.

2. Related Work and Our Contribution

2.1 Models for Obstacles inside Wireless SensorNetworks

Although obstacle avoidance is a well-studied subject indistributed computing in general, there has not been muchattention given to the modeling of obstacles in wirelesssensor networks. The study of various data propagationprotocols and their behaviour against obstacles in the net-work deployment area has also been neglected. We pro-vide a summary of some obstacle-related models.

2.1.1 Obstacles Modeled as Polygonal Objects

There is some related work concerning the placement ofobstacles inside the network deployment area in mobile adhoc networks. An obstacle mobility model for mobile adhoc networks has been proposed and evaluated [2,3], ver-sus the random way-point and other more commonly-usedmobility models for ad hoc networks. This model usespolygonal obstacles that may be present inside the net-work deployment area, influencing the way mobile nodesmove and also wireless transmissions between nodes. Theobstacles model buildings and other structures that pro-vide barriers to the movement of mobile nodes, thus beingmore realistic than other models. Obstacles also have aneffect on the communication between nodes, the extent ofwhich depends on the wireless channel model used in thesimulations (from simple line-of-sight blocking to morecomplex signal propagation models).

This approach is practical, but a little too general, sinceit does not provide specific obstacle templates to be usedin simulations. Also, it is better suited for ad hoc net-works and not wireless sensor networks as it does notcater for the regions of very few or no sensors within thedeployment area, forming a virtual obstacle. This case isquite common in sensor networks for a number of rea-sons: nodes use up their energy sources or are destroyedby external factors. Finally, in this model there is no refer-ence to obstacles that have a limited lifetime or make theirappearance after the network has begun functioning. Allobstacles are therefore considered static and determinis-tic in nature. Jardosh et al. [2] used a projection of somebuildings in the campus of University of California, SantaBarbara, as an example of an obstacle map.

2.1.2 Randomly Covering the Network Area withObstacle Squares

Another approach to modeling obstacles in the environ-ment is to divide the deployment area into equal-sizedsquare cells. Each cell is either occupied by an obstacle(obstacles are static and do not move) or not. The obstacledensity is known in advance and obstacles are placed ina random fashion by assigning an independent probabilityto each cell of whether or not it will be occupied by anobstacle. Alternatively, this probability is relative in someway, leading to clusters of cells occupied by obstacles inorder to produce bigger obstacles.

This approach is useful and easy to implement whilesimulating wireless sensor networks, but in general is notas realistic as the other models presented in this section.Moreover, the shape and size of the obstacles may alsohave an effect on the network, and in this specific modelsuch features are neglected. Also, while this is a proba-bilistic model for creating obstacles, all obstacles are cre-ated before the network starts functioning.





2.1.3 Obstacles Modeled as Routing Holes

An interesting approach is described by Fang et al. [4]where all obstacles are modeled as routing holes, i.e. re-gions without enough sensors to propagate the informa-tion closer to its destination. Generally, routing holes aredefined as simple regions enclosed by a (possibly con-cave) polygonal cycle which contains all the nodes whereinformation can be stuck, ceasing subsequent forwarding.

This definition of obstacles as routing holes is verygeneral and can have many interpretations. Apart from thecase of communication voids (areas with no or very fewsensor nodes), the meaning of routing holes can be definedby specific application needs, or indicate geographicalcharacteristics of the sensor deployment area. For exam-ple, in a forest fire detection application, the sensors in-side the fire region will be destroyed and leave a routinghole in the network. Identifying the area that is coveredby the routing hole is equivalent to detecting the area al-ready covered by the fire. Massive drop-off of sensors onirregular terrain leaves holes that correspond to mountainsor shadows. Holes can also be used to detect regions withlow sensor density due to the depletion of the energy sup-plies of the nodes. Holes can therefore be thought of asareas where adding new nodes will significantly improvethe overall connectivity of the network.

Although this approach gives an insight to the sig-nificance of obstacles in wireless sensor networks, andalso what could be modeled as an obstacle, it is a bitgeneral and gives no reference obstacles to use in simula-tions. This is critical, since in simulations there is alwaysthe need to repeat experiments and compare a number ofmethods against each other. It is, however, useful as a the-oretical analysis tool.

2.1.4 Obstacles Modeled as Straight Walls

Another method of incorporating obstacles in wirelesssensor network simulations is that of Rao et al. [5]. Ob-stacles are modeled as straight walls that are parallel tothe x or y axis with varying length and number (this no-tion of obstacles modeled as walls is also used by Fon-seca et al. [6]). In addition to walls, the case of voids, i.e.areas of regular shape which contain no nodes, are alsoexplored [5]. However, all obstacles are placed before thenetwork starts to function, and so communication voidsthat appear during the operation of the network cannot beset explicitly.

A similar approach is followed by Kim et al. [7], whereobstacles (walls) are of fixed length, the center of eachwall is positioned randomly in the network field and it isequally possible for each obstacle to be parallel to the x ory axis. Shah et al. [8] considered one wall in the middleof the network. Its length was varied as 0�5, 1 and 2 timesthe radio range of the nodes.

This approach to modeling obstacles is useful, espe-cially in the case of only one control center in the sensor

network and geographic routing as in this case walls ef-fectively block greedy forwarding schemes. It is howeversimplistic, and does not take into account parameters suchas the obstacle size and shape. This is important, as rout-ing protocols may perform differently in scenarios withdifferent types of obstacles, as explained in the followingmodel.

2.2 Solutions to the Obstacle Avoidance Problem inWireless Sensor Networks – Simulation Studies

In order to simulate the effect of obstacles on the commu-nication between nodes inside the deployment area, theGloMoSim and ns-2 network simulators [2] models wereextended. These provide a mechanism for the placementof obstacles within the simulation terrain, allowing theuser to define the position, shape and size of obstacles.A calculation of the signal-blocking regions in which thewireless signal is blocked due to the obstacles present inthe deployment area was utilized, and in [3] the authorsused a signal propagation model that simulates the fadingof a radio signal as it propagates through obstacles that liebetween a pair of nodes. A series of experiments was con-ducted using again the GloMoSim and ns-2 network sim-ulators, and their results in both works indicate that theinclusion of obstacles in their network model has a sig-nificant effect on the performance of the ad hoc protocolsthey chose to study. This is partly due to the fact that theirmobility model also considers obstacles, whereas othercommonly-used mobility models do not.

Generally, the proposed solutions for providing ob-stacle avoidance in wireless sensor networks have somephase for locating the obstacles inside the sensor networkdeployment area, and use this information to build a net-work topology improving network connectivity. A largepart of such algorithms uses face routing on a planar sub-graph, such as the algorithm in Bose et al. [9]. Dependingon the approach, this can occur as a setup phase before thenetwork starts operating, or during its operation.

The solution BoundHole has been presented [4] toidentify routing holes using a greedy algorithm. When en-countering a situation where information cannot be prop-agated further using a greedy logic, the main aim is tofind a directed closed cycle that encloses the routing hole.This cycle starts from the node where the forwarding stuckand, following a local rule at each node, goes back to thesame node, identifying the obstacle. Assuming geoloca-tion abilities, this algorithm is local and simple, and eachnode only stores information about its one-hop neighbors.

If after identifying the routing holes inside the networkand the information under propagation becomes stuck at anode while using greedy forwarding, the routing schemechanges and the boundary of the hole is chosen as thepropagation path. When a node that is closer to the desti-nation than the original stuck node is encountered, greedyforwarding is chosen as the forwarding scheme again.





Another solution, without assuming geolocation abili-ties (no GPS), is virtual coordinates [5]. The authors intro-duce a protocol for assigning virtual coordinates to nodes,reflecting network connectivity, in order to use a greedyforwarding scheme. The algorithm for assigning these co-ordinates is iterative. Different geolocation abilities as-sumptions lead to different protocol complexity results(more nodes with GPS abilities implies less iterations toreach the best solution to the problem and message broad-casts). A simplified version of this algorithm (as the orig-inal version was deemed too complex) was implementedin a real environment and tested while placing some smallobstacles in the network field [10].

Trading off the cost of obtaining a limited extra knowl-edge of the network with the benefit of subsequently op-timizing routing toward the control center using this extraknowledge has been proposed [11]. This idea, althoughgeneral, can be used in order to locate obstacles earlyenough and adjust accordingly. Finally, the obstacle modelpresented here has also been used to evaluate the perfor-mance of a geographic routing protocol under ‘difficult’network conditions [12].

2.3 Other Related Work

In addition to modeling the obstacles that are present inthe network area, there is also a great deal of related workin terms of signal attenuation and fading due to the pres-ence of obstacles [13]. In a wireless network deploymentarea with obstacles present, it is obviously possible fortwo network nodes to be able to communicate with eachother via multipath signal propagation, even although anobstacle is blocking the line of sight between them. Mech-anisms such as reflection, distraction and scattering havebeen studied both analytically and experimentally. A num-ber of models for signal propagation have been proposedto enable calculation of the strength of the received sig-nal from a source to a point in the network field. Thesemodels use formulae which include constants dependentupon a number of parameters, such as the frequency ofthe radio signal, the surrounding environment and the sizeof the antennas, calculated experimentally. Also, in somenetwork simulators such as TOSSIM [14], obstacles canbe described indirectly by specifying available links be-tween pairs of nodes, but no specific model is providedto the user. Boukerche et al. [15,16] present the designand implementation of the SwimNet network simulator.This simulator concerns wireless networks in general, andproposes a scalable parallel simulation testbed for wire-less and mobile networks. This is in contrast to the workpresented here, which focuses on sensor networks and es-pecially on the presence and impact of obstacles.

In general, models are provided for average use cases(e.g. indoor office spaces) and do not consider the ef-fect of the various physical obstacles or obstacles that areformed out of node distribution patterns. Although using

such models may be more realistic for the average case,it is not easy to trace specific protocol weaknesses dueto obstacles. Also, in the case of simulations, the adapta-tion of such schemes introduces additional computationaloverheads which may lead to prohibitive execution timesfor experiments with a very large number of nodes.

2.4 Our Contribution

We propose a systematic and generic obstacle model tobe used in simulations of wireless sensor networks andprovide a categorization of the obstacle types, based ona variety of criteria. We believe that the inclusion of ob-stacles in wireless sensor network simulators will lead tointeresting and important findings and that the categoriza-tion of obstacles is necessary in order to study the effectof the various types of obstacles on the behavior of datadissemination protocols for wireless sensor networks. Ourmodel caters for both physical and communication obsta-cles and deterministic and probabilistic obstacles (definedin Section 4). Furthermore, we include obstacles of var-ious shapes that are expected to appear in more realisticdeployment scenarios. Also, obstacles of various shapesin our model can be combined to produce more complexshapes.

The advantages of employing our obstacle model are:

� such a model can be easily implemented and used�

� it covers a number of different obstacle types�

� it can be easily randomized and used to produce ran-dom network instances in a network simulator�

� probabilistic obstacles may be used to produce morerealistic simulations, and can also be useful in casessuch as delay tolerant networks� and

� it produces more realistic network topologies, in-stead of assuming homogeneous randomly de-ployed nodes.

We have implemented our obstacles model in the sim-Dust network simulator [17–22] in order to incorporatethe proposed obstacle model. In this way, we have createda simulation environment that integrates a variety of net-work topologies, protocols and obstacles. We provide ex-perimental results comparing the performance of severalrepresentative protocols for data propagation in wirelesssensor networks in various settings of obstacles and studytheir behavior. Our findings demonstrate the crucial im-pact of obstacles in protocol performance in general, aswell as the particular effect of certain obstacles to eachprotocol.

Our main goal in this paper was to propose and imple-ment a new model for obstacles and also investigate itssuitability and applicability to the evaluation of the per-formance of protocols used in wireless sensor networks,





under the presence of obstacles. We focused on the prob-lem of data propagation (routing), since it is severely af-fected by obstacles and also because routing functionalityis fundamental for higher level services and applications.Because of this fact, we believe that our findings also pro-vide significant insight to the impact of obstacles on otherproblems� we plan to use our model to explicitly investi-gate this impact.

Regarding routing, we chose to evaluate a few pro-tocols that evolve in different, characteristic ways (i.e. agreedy, single-path protocol and its adaptive extension, aswell as an optimized redundant protocol that uses multiplepaths). Our purpose was to investigate whether our frame-work is able to demonstrate the different impact of variousobstacles to each protocol. Other routing protocols fromthe state-of-the-art could have served this purpose equallywell. Again, we plan to investigate the behavior of otherprotocols in our future work.

This paper is an extended version of the work includedin Chatzigiannakis et al. [23]. We have extended the moti-vation section in the introduction, describing the necessityof using obstacles particularly in sensor network simula-tions, especially in the case of greedy geographic rout-ing algorithms. The related work section has been rewrit-ten, and now includes a detailed presentation of all obsta-cle models (to our knowledge) previously used in simula-tions of wireless sensor networks. Each model is presentedand analyzed, along with its advantages and its shortcom-ings. Finally, the section describing simulation results fea-tures additional figures, which describe experimental re-sults concerning multiple circular obstacles, and greaterdiscussion on the interpretation of these results.

3. A Model for Sensor Networks

Based on the technological specifications of existing wire-less sensor systems, each node is a fully-autonomous com-puting and communication device with constrained re-sources, equipped with a set of monitors (e.g. sensors fortemperature, humidity, etc.) and characterized mainly byits available power supply (battery) and the energy costof computation and transmission of data. The communi-cation equipment broadcasts messages to nearby deviceswithin transmission range �, and can also use a directedtransmission of angle � around a certain line, possibly us-ing some special kind of antenna (Figure 2). The transmis-sion range � can vary (i.e. the transmission power can beset at appropriate levels), while the transmission angle isfixed and cannot change throughout the operation of thenetwork since this would require a modification or move-ment of the antenna used. Note that the protocols consid-ered in this work (see Section 5) can operate even underthe broadcast communication mode i.e. � � 2� .

We consider a simple sensor network for the remotesurveillance of a region or for data collection in an ambi-ent intelligence setting. In practice, such a network may

Figure 2. A directed transmission of angle �

consist of several hundreds or thousands of sensor devicesdeployed within that region. Let n be the total numberof sensor devices that are present in an area of size �.In some cases, the devices may be deployed in a regu-lar fashion (e.g. a 2-dimensional lattice, or a linear array)within that region. More generally, however, communica-tion and networking protocols cannot assume structuredsensor fields. In our remote surveillance network we as-sume that the sensor devices do not move and that theyare not able to change their physical position.

A particular user of this remote surveillance system,which we refer to as the control center � and which is notresource-constrained as in the other nodes of the network,may contact the sensor devices (e.g. via a long-range ra-dio link) in order to acquire information regarding the en-vironmental conditions. In this manner, the user injectssensing tasks in the network, i.e. by broadcasting mes-sages with a task description� the system can support avariety of task types [24]. Those sensor devices match thetask description report to the control center using multi-hop wireless communication and routing mechanisms de-scribed in Section 5. In this work, we assume a single,static (not mobile) control center.

In this paper we focus on some of the physical char-acteristics of the deployment area and their effect on theperformance of the data propagation mechanism used.Our approach addresses some fundamental issues that arepresent in sensor networks and is presented in detail in thefollowing section.

4. A Model for Obstacles

4.1 Physical and Communication Obstacles

Initially, we define two classes of obstacles: physical ob-stacles and communication obstacles.





Defenition 1 (Physical Obstacle �phy) A physical ob-stacle corresponds to an obstacle that prevents the phys-ical presence of sensor devices – a device that is posi-tioned over the obstacle is automatically ‘destroyed’. Inthis sense, the network area that is occupied by the obsta-cle is virtually empty of sensor devices.

Defenition 2 (Communication Obstacle�com) A com-munication obstacle corresponds to an obstacle thatcauses disruption to the wireless communication medium.In particular, we assume that if the line of sight betweentwo devices crosses the obstacle, then communication isblocked. Even although some sensor devices may be de-ployed on top, or even inside the obstacle, no communica-tion with other devices can take place.

Essentially, the class of physical obstacles relates to sit-uations where there is a lack of sensor devices in specificparts of the area of deployment, i.e. the local network den-sity is zero. Those devices that are on the boundaries ofthe physical obstacle are still able to communicate as longas they can overcome it by possibly increasing the trans-mission power. On the other hand, for the class of com-munication obstacles, although some sensor devices canbe located inside (or on top of) the area affected, no com-munication with the rest of the network is possible. Ofcourse, an obstacle can fall within both classes, blockingany physical presence of devices and communication ac-tivity.

4.2 Obstacle Shape

We present now a collection of geometric elements thatcan be used to describe the shape of an obstacle. Thesebasic elements can be used to represent simple real ob-jects, or can be combined to produce more complex ob-jects (Figure 3).

Rectangular (Orthogonal) Obstacles: this type of obstacleroughly corresponds to buildings and large vehicles. Ob-stacles of this shape are not meant to be too big comparedto the overall network plane dimensions. Obstacles of thistype can be found mostly in urban environments.

Defenition 3 (Rectangular Obstacle �rect�p� l� ��) Arectangular-shaped obstacle is positioned on point p ofthe deployment area with dimensions l � � (where l islength and � is width). The position of the obstacle isdefined in terms of its upper left corner.

Circular Obstacles: this type of obstacle roughly corre-sponds to craters, large rocks, lakes, ponds and (large) treelogs. Objects of this type can be found in outdoor environ-ments, countryside and battlefields, etc.

Defenition 4 (Circular Obstacle �circ�p� r�) Acircular-shaped obstacle is centered on point p of thedeployment area with radius r .

Crescent (Boomerang) Obstacles: this type of obstacleroughly corresponds to a lake with a shape that resem-bles that of a crescent or a boomerang. Such an obstacleis defined by a circle and an ellipse, as shown in Figure 4.Although obstacles of this exact shape may be hard to en-counter in real environments, they represent a challengefor routing protocols. This is due to the fact that such anobstacle formulates a ‘loose’ dead-end, in the sense thatdata propagation tends to be trapped in the concave partof the crescent.

Defenition 5 (Crescent Obstacle �cres�p� r� s�) Acrescent-shaped obstacle is centered on point p of the de-ployment area with radius r of the circle in which the ob-stacle is inscribed and width s of the enclosing ellipse.

Ring Obstacles: this type of obstacle can represent areasof the network that are somewhat isolated from the restof the nodes, i.e. the nodes on the inside of the ring areseparated from the rest of the network by the outer part ofthe ring obstacle.

Defenition 6 (Ring Obstacle �ring�p� r�m�) A ring-shaped obstacle is centered on point p of the deploymentarea with radius r of the outer circle and radius m of theinner circle, as depicted in Figure 5.

Stripe Obstacles: this type of obstacle roughly corre-sponds to e.g. a river (communication obstacle) crossingthe network deployment area or a long wall (physical ob-stacle) situated in the network deployment area. The maindifference from the rectangle obstacles is their size, i.e.they are much bigger than rectangle obstacles should be,and the way these obstacles are defined by the user.

Defenition 7 (Stripe Obstacle�strp�p� �� a�) A stripe-shaped obstacle is positioned on point p of the deploy-ment area with width � and angle a with respect to thehorizontal boundary of the area. The position of the ob-stacle is defined in terms of its upper left corner.

4.3 Stochastic Presence of Obstacles

The third criterion in our obstacle model is whether ran-domness characterizes the presence of obstacles.

Deterministic Obstacles: this category consists of all theobstacles that are present throughout the duration of anexperiment (i.e. from the beginning to the end) and donot change in any manner. They are defined by the user





Figure 3. Combination of obstacle shapes in order to represent complex obstacles

before the beginning of the experiment. Deterministic ob-stacles can belong to any type and shape of obstacles, asdescribed previously.

Probabilistic Obstacles: this category, on the contrary,consists of obstacles that are not present throughout theduration of a simulation experiment, but appear in a ran-dom fashion for a period of time and even disappear. For

example, consider a train passing through the network de-ployment area, or a even a road inside the deployment areathat is crossed by cars. These obstacles have a temporaryeffect on the nodes situated in the area. Those nodes ceaseto function throughout the life span of the probabilisticobstacle by which they are capped. Furthermore, stochas-tic obstacles may capture areas whose density drops sig-nificantly over time (due to physical faults, permanent or





Figure 4. A crescent-like obstacle

temporary, as well as software decisions, i.e. power-savingschemes that put sensors to sleep). Probabilistic obstaclescan also belong to any type and shape.

In order to define a probabilistic obstacle, a simpledefinition procedure is as follows. The user first definesthe type and shape of the obstacle, and then assigns to it astart time and a duration. The duration can be defined tobe until the end of the simulation.

5. Protocols for Data Propagation

In this paper, we focus on distributed algorithms for thenetwork layer. We present below three representative (interms of their logic) protocols which aim to avoid floodingthe network, achieve good performance (with respect totime and energy) and are robust. All three protocols use tosome extent geographic information to make greedy for-warding decisions.

5.1 The PFR Protocol

The Probabilistic Forwarding protocol (PFR) [19] is in-spired by the probabilistic multi-path design choice forDirected Diffusion [25]. The basic idea of PFR is to min-imize energy consumption by probabilistically favoringcertain paths of local data transmissions toward the con-trol center.

The protocol avoids flooding by favoring, in a prob-abilistic manner, data propagation along sensor nodeswhich lie ‘close’ to the optimal transmission line E S thatconnects the node detecting the event E and the controlcenter �. This is implemented by locally calculating theangle � � ��E P S�, whose corner point P is the sensornode currently running the local protocol, having receiveda transmission from a nearby node, previously possessingthe event information. If � is equal to or greater than a pre-determined threshold (�threshold), then p will transmit andthus propagate the event information further. Otherwise,it decides whether to transmit with a probability equal to

Figure 5. A ring obstacle

�� . Because of the probabilistic nature of data propa-gation decisions and in order to prevent the data propaga-tion process from early failing, we initially use (for a shorttime period which we evaluate) a flooding mechanism thatleads to a sufficiently large ‘front’ of sensors possessingthe data under propagation. When such a front is created,we perform probabilistic forwarding.

Note that transmission along this line is energy opti-mal. However, it is not always possible to achieve this op-timality, for a variety of reasons. Essentially, PFR capturesthe intuitive, deterministic idea ‘if the distance from E Sis small then send, else do not send’. This idea was en-hanced by random decisions (above a threshold) to allowsome local flooding to happen with small probability andcope with local sensor failures.

5.2 The LTP Protocol

In this subsection we give a short description of the Lo-cal Target protocol (LTP) [21]. The basic idea is to try tosearch for all active neighboring nodes and then use the in-formation retrieved in order to forward i.e. propagate thedata towards the neighbor that is closer to the control cen-ter. In this protocol, each node p� that has received info(�)from p via, possibly, other nodes, goes through the fol-lowing phases.

Phase 1: The Search Phase. It uses a periodic low en-ergy broadcast of a beacon in order to discover a nodenearer to control center than itself. Among the nodes re-turned, p� selects a unique node p�� that is ‘best’ with re-spect to progress toward the control center. That is, thenode p��E which achieves the greatest progress on the p�Sline, among all nodes, should be selected.

Phase 2: The Direct Transmission Phase. p� then sendsinfo(�) to p�� and sends a success message to p (i.e. to thenode that it originally received the information from).

Phase 3: The Backtrack Phase. If consecutive repeti-tions of the search phase fail to discover a node nearer





to control center, then p� sends fail message to the nodethat it originally received the information from.

In the above procedure, propagation of info(�) is donein two steps: (i) node p� locates the next node (p��) andtransmits the information� and (ii) node p� waits until thenext node (p��) succeeds in propagating the message fur-ther toward the control center. This is done to speed up thebacktrack phase in case p�� does not succeed in discover-ing a node nearer to control center.

5.3 The VTRP Protocol

The Varying Transmission Range (VTRP) [17] basicallyworks in a search and forward way similar to LTP, andalso by varying the range of transmissions in order toachieve better performance. This is compared to fixedtransmission range data propagation in some frequentlyoccurring situations such as (1) the case of low densities ofsensor nodes� and (2) when VTRP performs better in thepresence of obstacles or faulty/sleeping sensors because ofthe possibility to increase transmission range. It also by-passes certain critical sensors (such as those close to thecontrol center) that tend to be overused, thus prolongingthe network lifetime.

When a node p� receives some info(�) from node p,VTRP works in three phases. Phases 1 and 2 are identicalto LTP’s first two phases.

Phase 3: The Transmission Range Variation Phase. Ifphase 1 fails to discover a node nearer to control center,p� enters the transmission range variation phase. Morespecifically, each node maintains a local counter withinitial value � 0. Every time the search phase fails, thiscounter is increased by 1. Thus is an indication of thenumber of failures to locate an active node. Based on ,the node modifies its transmission range� according to achange function �� .

We consider here only one of the originally four differ-ent change functions for varying the transmission rangedefined in [17]: that of Multiplicative Progress. In thiscase, the transmission range of the node is increased moredrastically. The change function is defined

�� new � �init ��init � m � �where m is a small constant (m � 3).

This relatively drastic change leads to a bigger proba-bility of finding an active node� however, it leads to higherenergy consumption.

6. The Simulation Environment

This section provides a description of the componentsneeded for conducting our simulation experiments, in-cluding the simulation environment, the metrics used to

evaluate the experimental results and the simulation sce-narios.

To evaluate the performance of the proposed protocolswe have conducted an extensive simulation analysis. Inthis work, we use simDust [17–22], a lightweight networksimulator. simDust is implemented using C++ in Linux,with the aid of the LEDA library, and provides efficientdata types and visualization capabilities. We chose C++for its efficiency and LEDA because it provides a largenumber of efficient data structures using a unified inter-face, making their use sufficient for our purposes.

In contrast to other, more detailed, network simulatorssuch as the Network Simulator, simDust makes an ab-straction of the physical and MAC layers. We make theassumption that all packet collisions and other related is-sues are dealt with in an underlying MAC layer, and donot take them into account in our simulations. Althoughthis is fairly simplifying and simDust cannot provide de-tailed measurements on parameters such as the numberof dropped packets or the precise execution time, this ab-straction of the lower network levels allows simDust toreduce the execution time of simulation experiments, en-abling the study of very large network instances (in thescale of tens of thousands).

One basic idea in this simulator is that a number ofgeneral classes are provided to the programmer to imple-ment new routing protocols for wireless sensor networks.The whole network operates in a synchronized fashion,specifically in what we refer to as simulation rounds. Ineach round, each node of the network can do some in-ternal processing, send or receive messages or sleep (i.e.turn off transceivers and CPU). The programmer defineswhat each node is supposed to do in the course of thesesimulation rounds, i.e. what to do if an event occurs, ifa message is received, etc. Events in the network can betriggered by the simulator which cause nodes in the area toproduce messages that need to be forwarded to the controlcenter of the network. In the current implementation ofsimDust, each event is detected by only one node. Nodesare placed in a rectangular-shaped field, with dimensionsdefined by the programmer. Nodes are placed using somerandom uniform distribution or are placed in a grid.

Regarding connectivity issues inside the network, eachnode maintains a list of its neighbors, i.e. the nodes thatare inside a disk with the current node in the center andwith a radius equal to its transmission range. We make theassumption that if a node is inside, this disk will receiveall messages transmitted from the current node. The trans-mission range of each node can change independently, andin this case neighborhoods are recalculated in the nextsimulation round.

Apart from the simulation part of simDust, there is alsoa visualization part that enables the programmer to havea visual understanding of the simulated network’s opera-tion. The programmer, using the class defined for the datapropagation algorithms and extending them to use somedata types provided by LEDA, can produce a visualiza-





tion of the network’s inner workings. This capability wasimplemented to provide easier understanding of each al-gorithm’s functions and check the validity of our imple-mentations.

Regarding the energy consumption of the devices, sim-Dust implements a rather detailed energy cost model thatenables relatively detailed measurements of the energydissipation. Generally, each node in the sensor networkcan be in one of three different modes at any given timeregarding its energy consumption. These modes are: (1)transmission of a message� (2) reception of a message� and(3) sensing of events.

Following Heinzelman et al. [26] for the case of trans-mitting and receiving a message we assume the followingsimple model where the radio dissipates Eelec to run thetransmitter and receiver circuitry, and eamp for the trans-mitter amplifier to achieve acceptable signal to noise ratio.We also assume an r2 energy consumption when transmit-ting a signal with a range of distance r . Thus, to transmita k-bit message at distance r in our model, the radio ex-pends

ET�k� r� � ET-elec�k�� ET-amp�k� r��

ET�k� r� � Eelec � k � eamp � k � r2�

and to receive this message, the radio expends

ER�k� � ER-elec�k��

ER�k� r� � Eelec � k�where ET-elec, ER-elec represent the energy consumed bythe electronics of the transmitter and the receiver, respec-tively. Concluding, there are three different kinds of en-ergy dissipation:

� ET: energy dissipation for transmission�

� ER: energy dissipation for receiving� and

� Eidle: energy dissipation for idle state.

For the idle state, we assume that the energy consumedfor the circuitry is constant for each time unit and equalsEelec (the time unit is 1 simulation round).

Regarding the blocking of signal transmission betweennodes, recall that we assume that two nodes can commu-nicate directly (that is, if one node is in the communica-tion range of the other) only in the case that there exists aline-of-sight path between them. If no line-of-sight existsbetween two random nodes, then they cannot communi-cate directly. In our model, line-of-sight is blocked onlyby physical obstacles.

Regarding the implementation of obstacles in our sim-ulator, it was partly based on some data types provided byLEDA, in particular the Circle and Segment data types

which correspond to circle and line segments, respec-tively. These data types simplify the procedure of allocat-ing space to each obstacle and positions for the nodes in-side the network area. They also simplify the procedure ofdetecting if an obstacle blocks the line-of-sight betweentwo random nodes in the network. Obstacles are definedin text files that serve as input to the simulator. Such a textfile defines the type of the obstacle and its dimensions.Each file can contain multiple obstacles.

In order to check the availability of line-of-sight be-tween two random nodes, we can define a line segmentfor each two such nodes and check whether this segmentcrosses through any obstacle. If so, we decide that thereis no line-of-sight between the two nodes and that theyare not able to communicate with each other directly. Thischeck is performed for each pair of nodes and for eachobstacle in the network field, in order to determine thenetwork neighborhood of each node (i.e. the set of nodesthat can be contacted directly by the specific node).

Moreover, we make the assumption that the controlcenter is located on the middle of the right edge of thenetwork field. Because of this assumption, the directionof crescent obstacles in the network field is always thesame.

Regarding the simulation parameters used for each ex-periment, the programmer defines the total number ofnodes, the data propagation algorithm that will be used,the total simulation time (in rounds), the total number ofevents taking place inside the network, the possibility ofan event happening in a random simulation round and theparameters regarding the network field and node place-ment.

Regarding the output of the simulator, all results aregathered in text files which contain information both forthe whole experiment and in greater detail: specifically,for each 10 rounds. The statistics that are noted include,among others, success rates (events that have successfullyreached the control center), the energy available in thenetwork, the number of alive nodes, the total number oftransmission and receptions and average number of hopsto reach the control center. As defined earlier, in the multi-ple event propagation problem we have a series of K cru-cial events Ei �E1� E2� � � � � EK � inside the network de-ployment area. Let l be the number of events that weresuccessfully reported to the control center �.

Success Rate: The success rate Ps is defined as the frac-tion of the number of events successfully propagated tothe control center over the total number of events, i.e.Ps � lK .

Total available energy: A straightforward way to com-pare the performance of different protocols for wirelesssensor networks is to study the available energy over time,both in each node and overall in the network. Let Ei bethe available energy for node i . We define the total energyavailable in the sensor network as Etot � �n

i Ei , where





Figure 6. Success rate of LTP, PFR and VTRP for various numbers of nodes (n � [500� 3500]), with initial transmission range � � 50 m

n is the number of nodes in the network. Clearly, the lessenergy a protocol consumes, the better.

Number and distribution of ‘alive’ nodes: Another met-ric of the protocols’ efficiency is the total number of alivenodes in the network and the way they are distributed. Byalive nodes, we refer to nodes which haven’t run out ofenergy supplies. As in the previous cases, the greater thenumber of alive nodes, the better. Furthermore, the distri-bution of these nodes is as important as their number, andparticularly the condition of critical sensor nodes, such asnodes lying close to the control center.

7. Experimental Scenarios and Discussion ofResults

This section provides a description of the setup used inour simulation experiments, regarding node density anddistribution, event and obstacle generation. We provide ashort description of each set of experiments and then dis-cuss the results of that specific set of experiments. All ex-periments were conducted in a network field of dimen-sions 1000 units, using deterministic obstacles. (Experi-ments using probabilistic obstacles are planned for the fu-ture). We make the assumption that the lower left edge of

the field has coordinates �0� 0� and that the control cen-ter of the sensor network is fixed and its coordinates are�999� 499�. Finally, the nodes were placed in the networkfield using a uniformly random distribution.

7.1 Experiments without Obstacles

Our first set of experiments involves sensor networks ofvarying size (number of nodes) in a field with no obsta-cles. More specifically, we position �500� 3500� nodes inthe network field in steps of 500 nodes, and generate 250events to be propagated to the control center each time.Our main purpose for this set of experiments was to de-termine a network size at which all three protocols per-form very well, in order to make a fair comparison for allprotocols for the following sets of experiments. Our onlycriterion for this decision was the success rate of each pro-tocol.

As it can be seen in Figure 6, the performance of thethree protocols increases with the network density andVTRP achieves a high success rate earlier than the othertwo protocols. Moreover, we notice that after the numberof nodes of the network reaches 2500, all three protocolsachieve very high success rate. For this reason, it is fairto make a comparison between these protocols under thiscondition (n � 2500).





Figure 7. Success rate of LTP, PFR and VTRP in the presence of circular obstacles (�phycirc�500� 500� ��) where � � [126� 488] is the radius

of the obstacle, for variable obstacle size, �xed node density and initial transmission range � � 50 m

7.2 Experiments with Circular Obstacles

For this set of experiments we created sensor networkswith varying size, containing one circular obstacle in thecenter of the field, i.e. with coordinates �499� 499�. Thearea covered from this circular obstacle ranged from 5% to70% of the total network area, in steps of 5%. The size ofthe sensor network for each obstacle size was varying, inorder to maintain the same network density as in the caseof no obstacles in the network. Also, our experiments wereconducted for both cases of physical and communicationcircular obstacles.

For this set of experiments only one event was gener-ated in each run of the simulator, more specifically in anode with coordinates �0� 499�, in order to investigate thecase of the propagation of an event generated at a nodeplaced diametrically to the control center. Also, each runof the simulator lasted for a specific number of simula-tor rounds, meaning a specific amount of time for whichwe wait for the report of the single event generated. Thisevent setup was used for 100 runs of the simulator, withthe position of the rest of the nodes in the field being re-calculated each time.

The results for this set of experiments can be seen inFigure 7. It is clear that LTP does not perform as well as

the other two protocols, especially VTRP, which performsexcellently regardless of obstacle size. PFR achieves goodperformance for many obstacle sizes, but after a certainpoint fails to propagate data to the control center. Thisis probably due to the fact that the obstacle blocks thecreation of the protocol’s propagation front (as describedin Section 5), and it cannot overcome the obstacle any-more. Moreover, communication obstacles prove to beharder to overcome than physical obstacles, as expected.Also, we notice that although we keep the network densityunchanged, i.e. the connectivity of the network remainsroughly the same, obstacles have a significant effect onthe performance of the protocols.

7.3 Experiments with Rectangular Obstacles

We created sensor networks of varying size, containing arectangular obstacle in the center of the field at coordi-nates �499� 499�. We considered rectangular obstacles oftwo different dimensions. In the first case, the length ofthe obstacle is 32 times its width and the area coveredby it ranged from 5% to 50% of the total network area,in steps of 5%. In the second case, we created a thin rec-tangular obstacle with a fixed width of 50 units and vari-able length, proportional to the length of the network field,





Figure 8. Success rate of LTP, PFR and VTRP in the presence of rectangular obstacles (top: �comrect �500� 500� �� 2

3��, bottom:

�phyrect�500� 500� �� 50�), where � � [274� 822] is the length of the obstacle, for variable obstacle size, fixed node density and initial transmis-

sion range � � 50 m





ranging from 5% of the field’s length to 70%, in 5% steps.In both cases, we used the same event setup as in the cir-cular obstacle case. We tested both cases of physical andcommunication rectangular obstacles.

The results for this experiment set can be seen in Fig-ure 8. On the one hand, LTP starts dropping its successrate very early, due to the fact that when it reaches theobstacle it cannot find any nodes to forward data, both inphysical and communication obstacles. On the other hand,PFR and VTRP, especially in the case of physical obsta-cles, achieve much higher success rates than LTP. Onceagain, the presence of obstacles in the network area has asignificant impact on the operation of the sensor network,although the network density remains the same.

7.4 Experiments with Ring Obstacles

For this set of experiments, we created a ring obstaclewith its center in the center of the field, as in the previ-ous cases. We studied three different widths of the ringsector of the obstacle, setting it to 25%, 33% and 50% ofthe inner circle radius (see Figure 5). In all three cases, weused the same event setup as in the circular obstacle case.We tested only the case of physical ring obstacles, as thecommunication obstacle case is essentially the same withthe circular communication obstacle case.

The results for this set of experiments are depicted inFigure 9. The general picture is the same for the threering width cases: LTP stops propagating data relativelyquickly, PFR is better than LTP and VTRP achieves highsuccess rate for all cases. From the point where the ringwidth goes beyond the fixed transmission range of LTPand PFR, this type of obstacle is identical to the physicalcircular obstacle case for these two protocols.

7.5 Experiments with Stripe Obstacles

As mentioned earlier in the definition of stripe obstacles,an obstacle of this type could be a river crossing the net-work field. We placed such an obstacle with its center inthe center of the field, its length equal to the length ofthe network area and its width proportional to the widthof the network area. Its width ranged from 5% to 45% ofthe width of the network area. As in previous sets of ex-periments, we used the same event setup as in the circularobstacle case. This set of experiments concerns only phys-ical stripe obstacles.

The results for this set of experiments can be seen inFigure 10. As expected, LTP and PFR do not perform wellagainst stripe obstacles, because their fixed transmissionrange cannot overcome the gap created by the obstacle.Another interesting result is that VTRP does not achievehigh success rate when the stripe obstacle reaches a certainsize. This is due to the fact that the protocol requires a cer-tain amount of time to alter the transmission range which,apparently, is longer than the running time assigned to ourexperiment set.

7.6 Experiments with Crescent Obstacles

Similar to the ring obstacle set of experiments, we placeda crescent obstacle in the center of the field and examinedtwo cases with the crescent width set to 75% and 50%of the radius of the obstacle (Figure 4). In all three caseswe used the same event setup as in the circular obstaclecase. In this set of experiments we tested only the case ofphysical crescent obstacles.

The results for this experiment set are depicted in Fig-ure 11. From these results, it is apparent that this is theharder type of obstacle for all protocols to overcome, as itforms a loose dead-end. LTP, as expected, performs poorlyand PFR’s performance drops very early, compared to theother obstacle cases. VTRP performs very well for thecase where the crescent width is large, as it can overcomethe obstacle by increasing the transmission range. How-ever, it begins to fail on the other case.

7.7 Experiments with Multiple Circular Obstacles

We also conducted a number of experiments with 2 and3 circular obstacles in the network field. The obstacleswhere placed with their centers on the line from the con-trol center to the node where the single event in the net-work is generated. More specifically, centers are locatedat (333, 500) and (667, 500) in the 2-circle case and (250,500), (500, 500) and (750, 500) in the 3-circle case. Weused the same event setup with the other experiment sets.

The results for this experiment set are displayed in Fig-ure 12. LTP fails to produce any positive success rate,while PFR’s performance is similar to that of the singlecircular obstacle. Also, VTRP performs worse than thesingle circular obstacle case. As we can see from theseresults, changing the number of obstacles in the field, butkeeping the same network density and obstacle type (cir-cular) produces totally different results.

This remark can also be deduced from Figure 13, wherewe can see the performance of LTP and PFR for the vari-ous types of obstacles. The type and number of obstaclesclearly affects the performance of protocols and also, net-work density is not directly related to the success rate ofthe protocols when there are obstacles present in the net-work field.

8. Concluding Remarks

In this work we have proposed a systematic obstaclemodel to be used in simulations of wireless sensor net-works. We have also extended the simDust network sim-ulator to incorporate this obstacle model and conducteda series of experiments to study the effect of obstacles inthe performance of three representative protocols for wire-less sensor networks. Our results indicate that the presenceof obstacles in the deployment area of a wireless sensornetwork has, in certain cases, a significant impact on the





Figure 9. Success rate of LTP, PFR and VTRP in the presence of ring obstacles (top: �phyring�500� 500� �� 1

4��, bottom:

�phyring�500� 500� �� 1

3��, next page:�phyring�500� 500� �� 1

2��) where � � [126� 488] is the radius of the circle enclosing the obstacle, for variableobstacle size, fixed node density and initial transmission range � � 50 m





Figure 9. (continued)

Figure 10. Success rate of LTP, PFR and VTRP in the presence of stripe obstacles (�phystrp�1000� 500� ��) where � � [50� 450] is the

width of the obstacle, for variable obstacle size, fixed node density and initial transmission range � � 50 m





Figure 11. Success rate of LTP, PFR and VTRP in the presence of crescent obstacles (top: �phycres�1000� 500� �� 3

4��, bottom:

�phycres�500� 500� �� 1

2��) where � � [126� 488] is the radius of the circle enclosing the obstacle, for variable obstacle size, fixed node densityand initial transmission range � � 50 m





Figure 12. Success rate of LTP, PFR and VTRP in the presence of multiple circular obstacles (top: two �phycirc�500� 500� �1�, bottom: three

�phycirc�500� 500� �2�) where � � [89� 309] and �2 � 73� 230 is the radius of each of the circles in the network field, for variable obstacle size,

fixed node density and initial transmission range � � 50 m





Figure 13. Success rate of LTP (top) and PFR (bottom) for different numbers of nodes (n � 500� 3500) when no obstacles exist and in thepresence of various types of obstacles, respectively, for fixed transmission range � � 50 m





performance of protocols for such networks. Furthermore,different obstacle shapes and sizes may affect each pro-tocol in a different way� this provides useful informationon which protocol is best to use for each obstacle case.Moreover, our results show that the effect of obstacles isnot directly related to the density of such a network, andthus cannot be emulated by simply changing the densityof the network.

Regarding our future work, we plan to apply our modelto other sensor network simulators and also conduct simu-lation experiments with other routing protocols for sensornetworks and study the effect of obstacles on their per-formance. In particular, it seems relatively challenging todesign protocols capable of efficiently handling the cres-cent and large rectangular types of obstacles. We also planto extend our model and add an attenuation factor to ob-stacles. Mobility schemes are also considered as a meansto extend our model. Finally, we are considering using aplethora of experimental scenarios for our obstacle modeland studying the effect of probabilistic obstacles in depth.

9. Acknowledgements

This work has been partially supported by the IST Pro-gramme of the European Union under contract num-ber IST-2005-15964 (AEOLUS). Further support was re-ceived from the program PENED under contract num-ber 03ED568, co-funded 75% by European Union Euro-pean Social Fund (ESF) and 25% by the Greek Govern-ment Ministry of Development, General Secretariat of Re-search and Technology (GSRT). Additional support wasprovided under Measure 8.3 of O.P. Competitiveness, 3rdCommunity Support Framework (CSF).

10. References

[1] I.F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. 2002.Wireless sensor networks: a survey. Journal of Computer Net-works 38, 393–422.

[2] A. Jardosh, E. Belding-Royer, K. Almeroth, and S. Suri. 2003. To-wards realistic mobility models for mobile ad hoc networks.In Proceedings of 9th ACM/IEEE Annual International Confer-ence on Mobile Computing (MOBICOM 2003), San Diego, CA,pp. 217–229.

[3] A. Jardosh, E. Belding-Royer, K. Almeroth, and S. Suri. 2005. Realworld environment models for mobile ad hoc networks. IEEEJournal on Special Areas in Communications Special Issue onWireless Ad hoc Networks 23(3), 622–632.

[4] Q. Fang, J. Gao and L. Guibas. 2004. Locating and BypassingRouting Holes in Sensor Networks. In Proceedings of the 23rdConference of the IEEE Communications Society (Infocom), pp.2458–2468.

[5] A. Rao, S. Ratnasamy, C. Papadimitriou, S. Shenker and I. Stoica.2003. Geographic routing without location information. In Pro-ceedings of 9th ACM/IEEE Annual International Conference onMobile Computing (MOBICOM 2003), San Diego, CA, pp. 96–108.

[6] R. Fonseca, S. Ratnasamy, J. Zhao, C. T. Ee, D. Culler, S. Schenker,and I. Stoica. 2005. Beacon vector routing: Scalable point-to-point routing in wireless sensornets. In Proceedings of the 2nd

Symposium on Networked Systems Design and Implementation(NSDI 2005), Boston, USA.

[7] Y. Kim, R. Govindan, B. Karp and S. Schenker. 2005. GeographicRouting made practical. In Proceedings of 2nd Symposium onNetworked Systems Design and Implementation NSDI 2005.

[8] R.C. Shah, A. Wolisz, and J.M. Rabaey. 2005. On the Performance ofGeographical Routing in the Presence of Localization Errors. InProceedings of the IEEE International Conference on Communi-cations (ICC 2005).

[9] P. Bose, P. Morin, I. Stojmenovic, and J. Urrutia. 1999. Routing withguaranteed delivery in ad hoc wireless networks. In Proceedingsof the ACM Workshop on Discrete Algorithms and Methods forMobile Computing and Communications, pp. 48–55.

[10] M. O’ Dell, R. O’ Dell, M. Wattenhofer and R. Wattenhofer. 2005.Lost in Space or Positioning in Sensor Networks. In Proceedingsof the Workshop on Real-World Wireless Sensor Networks (RE-ALWSN’05).

[11] I. Chatzigiannakis, A. Kinalis, and S. Nikoletseas. 2006. Efficientand Robust Data Dissemination using Limited Extra NetworkKnowledge. In Proceedings of the IEEE International Confer-ence on Distributed Computing in Sensor Networks (DCOSS),Lecture Notes in Computer Science (LNCS), pp. 218–233,Springer Verlag.

[12] O. Powell and S. Nikoletseas. 2007. Simple and Efficient Geo-graphic Routing around Obstacles for Wireless Sensor Networks.In Proceedings of the 6th Workshop on Experimental Algorithms(WEA07).

[13] T. Rappaport. 2000. Wireless communications: Principles and prac-tice. Prentice Hall.

[14] TOSSIM. Simulating TinyOS Networks. http://www.cs.berkeley.edu/pal/research/tossim.html.

[15] A. Boukerche, S. K. Das, A. Fabbri, and O. Yildiz. 1999. Exploit-ing Model Independence for Parallel PCS Network Simulation.In Proceedings of ACM Workshop on Parallel and DistributedSimulation, 166–173.

[16] A. Boukerche, S. K. Das, and A. Fabbri. 2001. SWiMNet: A Scal-able Parallel Simulation Testbed for Wireless and Mobile Net-works. ACM/Kluwer Wireless Networks 7, 467–486.

[17] T. Antoniou, A. Boukerche, I. Chatzigiannakis, G. Mylonas, andS. Nikoletseas 2004. A new energy efficient and fault-tolerantprotocol for data propagation in smart dust networks using vary-ing transmission range. In Proceedings of 37th Annual Simula-tion Symposium (ANSS 2004), IEEE Press, pp. 43–52.

[18] I. Chatzigiannakis, T. Dimitriou, M. Mavronicolas, S. Nikoletseas,and P. Spirakis. 2003. A comparative study of protocols forefficient data propagation in smart dust networks. Journal of Par-allel Processing Letters PPL 13(4), 615–627.

[19] I. Chatzigiannakis, T. Dimitriou, S. Nikoletseas, and P. Spirakis.2006. A Probabilistic Algorithm for Efficient and Robust DataPropagation in Smart Dust Networks. In Proceedings of 5th Eu-ropean Wireless Conference on Mobile and Wireless Systems be-yond 3G (EW 2004), pp. 344–350.

[20] I. Chatzigiannakis and S. Nikoletseas. 2003. A sleep-awake proto-col for information propagation in smart dust networks. In Pro-ceedings of 3rd International Workshop on Mobile, Ad-hoc andSensor Networks (WMAN 2003), IPDPS Workshops, p. 225.

[21] I. Chatzigiannakis, S. Nikoletseas, and P. Spirakis. 2002. Smart dustprotocols for local detection and propagation. In Proceedings of2nd ACM International Annual Workshop on Principles of Mo-bile Computing (POMC 2002), pp. 9–16.

[22] S. Nikoletseas, I. Chatzigiannakis, H. Euthimiou, A. Kinalis, T. An-toniou, and G. Mylonas. 2004. Energy efficient protocols forsensing multiple events in smart dust networks. In Proceedingsof 37th Annual Simulation Symposium (ANSS 2004), IEEE Press,pp. 15–24.

[23] I. Chatzigiannakis, G. Mylonas, and S. Nikoletseas. 2006. Modelingand Evaluation of the Effect of Obstacles on the Performance ofWireless Sensor Networks. In Proceedings of ANSS ’06, pp. 50–60.





[24] A. Boukerche, R.W.N. Pazzi, and R.B. Araujo. 2004. A support-ing protocol to periodic, event-driven and query-based applica-tion scenarios for critical conditions surveillance. In Proceedingsof 1st International Workshop on Algorithmic Aspects of Wire-less Sensor Networks (ALGOSENSORS 2004), Springer-Verlag,Lecture Notes in Computer Science, LNCS 3121, pp. 137–146.

[25] C. Intanagonwiwat, R. Govindan, and D. Estrin. 2000. Directed dif-fusion: A scalable and robust communication paradigm for sen-sor networks. In Proceedings of 6th ACM/IEEE Annual Interna-tional Conference on Mobile Computing, pp. 56–67.

[26] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan. 2000.Energy-efficient communication protocol for wireless microsen-sor networks. In Proceedings of 33rd IEEE Hawaii InternationalConference on System Sciences, pp. 3005–3014.

Ioannis Chatzigiannakis is adjunct faculty member as assistantprofessor at the University of Patras, Computer Engineering andInformatics Department, Patras, Greece and is also the manag-ing director of Research Unit 1 (Foundations of Computer Sci-ence, Relevant Technologies and Applications) at the ResearchAcademic Computer Technology Institute (RACTI), Greece.

Georgios Mylonas is currently a PhD student at the ComputerEngineering and Informatics Department at the University ofPatras, Greece and is also a researcher at the Research Acad-emic Computer Technology Institute (RACTI), Greece.

Sotiris Nikoletseas is an Assistant Professor at the ComputerEngineering and Informatics Department of Patras University,Greece and was also a Visiting Professor of the Univeristy ofGeneva, Switzerland.




SIMULATION · Keywords: Applications in computer networks, sensor networks, modeling and simulation...

Documents

Transcript of SIMULATION · Keywords: Applications in computer networks, sensor networks, modeling and simulation...