The Holes Problem in Wireless Sensor Networks: ASurvey · Sink/Black Holes/Worm Holes Sensor...

The Holes Problem in Wireless Sensor Networks:A Survey

Nadeem Ahmed1,2 Salil S. Kanhere1 Sanjay Jha1,2

[email protected] [email protected] [email protected] of Computer Science and Engineering, UNSW, Sydney, Australia

2National ICT Australia(NICTA∗), Sydney, Australia

Several anomalies can occur in wireless sensor networks that impair their desired function-alities i.e., sensing and communication. Different kinds of holes can form in such networkscreating geographically correlated problem areas such as coverage holes, routing holes,jamming holes, sink/black holes and worm holes, etc. We detail in this paper different typesof holes, discuss their characteristics and study their effects on successful working of asensor network. We present state-of-the-art in research for addressing the holes relatedproblems in wireless sensor networks and discuss the relative strengths and short-comingsof the proposed solutions for combating different kinds of holes. We conclude by highlight-ing future research directions.

I. Introduction

The recent advances in the MEMS (Micro-Electro-Mechanical Systems) technology has augmented re-search in wireless sensor networks. A wireless sen-sor network is composed of tiny sensor nodes eachcapable of sensing some phenomenon, doing somelimited data processing and communicating with eachother [1]. These tiny sensor nodes are deployed in thetarget field in large numbers and they collaborate toform an adhoc network capable of reporting the phe-nomenon to a data collection point called sink or basestation. These networked sensors have many poten-tial civil and military applications i.e., they can beutilized for object tracking, intrusion detection, habi-tat and other environmental monitoring, disaster re-covery, hazard and structural monitoring, traffic con-trol, inventory management in factory environmentand health related applications etc. [2], [3].

These myriad of applications present various de-sign, operational, and management challenges forwireless sensor networks. The challenges becomeeven more demanding if we consider the constraintsof wireless sensor networks such as low processingpower and bandwidth, limited battery life, and shortradio ranges.

Wireless sensor networks differ from ad-hoc net-works in several ways. One of the distinguishing fea-tures is the introduction of the sensing componentin sensor networks. A node in a sensor network isthus performing two demanding tasks simultaneously,sensing and communicating. To accomplish these

∗NICTA is funded through the Australian Government’s Back-ing Australia’s Ability initiative, in part through the AustralianResearch Council

tasks, we normally assume that the node not only per-forms required sensing of the phenomenon but is alsoable to communicate with neighbors for onward trans-mission of the sensed data to sink. But this assumptionis often not true in real world deployment scenarios.

Several anomalies can occur in the wireless sensornetwork that can impair their functionality. The tar-get field that is supposed to be 100% covered by thedensly deployed nodes may have coverage holes, ar-eas not covered by any node, due to random aerialdeployment creating voids, presence of obstructions,and, more likely, node failures etc. Similarly, nodesmay not be able to communicate correctly if routingholes, areas devoid of any nodes, exist in the deployedtopology. Thus the network fails to achieve its objec-tives if some of the nodes cannot sense or report thesensed data. Some of these anomalies may be delib-erately created by adversaries that are trying to avoidthe sensor network. These malicious nodes can jamthe communication to form jamming holes or they canoverwhelm regions in the sensor network by denial ofservice attacks such as sink/black/worm holes [4], [5]to hinder their operation normally based on trust.

We discuss in this paper such exceptional circum-stances with special attention to the phenomenon oc-curring in a region or hole and present state of the artin research related to these problems in wireless sen-sor networks. We group together these holes relatedproblems in four categories namely coverage holes,routing holes, jamming holes and sink/worm holes.

The remainder of this paper is organized as fol-lows. We introduce the holes related problems in Sec-tion II. Proposed solutions for coverage holes arediscussed in Section III. Section IV elaborates pro-posed solutions to avoid the impact of routing holes.

4 Mobile Computing and Communications Review, Volume 9, Number 2

Jamming holes are covered in Section V and SectionVI discusses some of the suggested countermeasuresagainst various denial of service holes related attacksi.e., sink/black and worm holes. We suggest somefuture research directions in Section VII and SectionVIII concludes the paper.

II. Problem Definition

We formally define here various types of holes thatcan occur in a wireless sensor network and discusstheir distinguishing characteristics.

Coverage Holes

Although the coverage problem has been interpretedin a variety of ways in the existing literature, we fol-low [6] for defining the coverage hole problem as fol-lows. Given a set of sensors and a target area, no cov-erage hole exists in the target area, if every point inthat target area is covered by at least k sensors, wherek is the required degree of coverage for a particularapplication (see Fig. 1). It is pertinent to mentionthat the coverage hole problem defined is dependenton application requirements. Some applications mayrequire a higher degree of coverage of a given targetarea for fault tolerance/redundancy or for accurate tar-get localization using triangulation-based positioningprotocols [7] or trilateration based localization [8].

( i ) ( ii )

Figure 1: (i). Coverage hole with unit disk sensing model (ii). Sensorwith dark gray sensing circle is necessary if degree of coverage requiredis 2

The sensing coverage of a sensor node is usuallyassumed uniform in all directions and is representedby unit disc model (Fig. 1). However, this idealizedmodel is based on unrealistic assumption: perfect andsame coverage in a circular disc for all the sensors.Moreover, the coverage not only depends on the sens-ing capability of the sensor but also on the event char-acteristics [9] e.g. target detection of military tanksas compared to detection of movement of soldiers de-pends on the nature and characteristics of event as wellas the sensitivity of the sensors involved.

A sensor network should be connected at all timesso that nodes are able to communicate with each other.As with the multiple coverage requirement discussedearlier, multiple connectivity is also desirable to guard

against single link or node failure partitioning the net-work. For the single coverage requirement, Wang etal. [10] proved that protocols which work on the as-sumption that the communication range of sensors isatleast twice the sensing range, only need to guaranteecoverage and it will satisfy the connectivity constraintas well. We discuss different solutions to minimizethe coverage holes in Section III.

Routing Holes

A routing hole consist of a region in the sensor net-work where either nodes are not available or the avail-able nodes cannot participate in the actual routing ofthe data due to various possible reasons. These holescan be formed either due to voids in sensor deploy-ment or because of failure of sensor nodes due to vari-ous reasons such as malfunctioning, battery depletionor an external event such as fire or structure collapsephysically destroying the nodes.

Routing holes can also exist due to local minimumphenomenon often faced in geographic greedy for-warding. Forwarding here is based on destination lo-cation. In Fig. 2, a node x tries to forward the traffic toone of its 1-hop neighbor that is geographically closerto the destination than the node itself. This forwardingprocess stop when x cannot find any 1-hop neighborcloser to the destination than itself and the only routeto destination requires that packet moves temporarilyfarther from the destination to b or y. This specialcase is referred to as local minimum phenomenon andis more likely to occur whenever a routing hole is en-countered.

Destination

c z

xyb

Figure 2: Local minimum phenomenon in greedy forwarding

We discuss different solutions to detect routingholes and to route around them in greedy forwarding.We also detail various multipath and single path rout-ing solutions that provide fault tolerance against therouting holes in Section IV.

Jamming Holes

An interesting scenario can occur in tracking applica-tions when the object to be tracked is equipped withjammers capable of jamming the radio frequency be-ing used for communication among the sensor nodes[4]. When this happens, nodes will still be able to de-tect the presence of the object in the area but unable to

Mobile Computing and Communications Review, Volume 9, Number 2 5

communicate the occurrence back to the sink becauseof the communication jamming. This zone of influ-ence centered at the jammer is referred to as jamminghole in this paper.

The jamming can be deliberate or unintentional.Unintentional jamming results when one or moreof the deployed nodes malfunction and continuouslytransmits and occupies the wireless channel denyingthe facility to other neighboring nodes. In deliber-ate jamming an adversary is trying to impair the func-tionality of the sensor network by interfering with thecommunication ability of the sensor nodes. This ad-versary can be a laptop-class attacker [5] with moreresources and capable of affecting a larger area of thesensor network or a mote-class attacker [5] i.e., oneof the deployed nodes that has been compromised andis now acting maliciously to create a denial of servicecondition.

Apart from communication jamming, jamming ofsensing capabilities is also possible for certain kindof sensor networks e.g. consider the case of a sensornetwork that relies on acoustic sampling for trackingobjects. If the object that is being tracked can intro-duce random high power acoustic noises, the sensorscannot reliably detect its presence and would be un-able to report the existence of the object. We discussproposals to detect jamming holes in a sensor networkin Section V.

Sink/Black Holes/Worm Holes

Sensor networks are highly susceptible to denial ofservice attacks due to their inherent characteristicsi.e., low computational power, limited memory andcommunication bandwidth coupled with use of inse-cure wireless channel. A sink/black hole attack canbe easily launched by an adversary node in the sensornetwork. The malicious node starts advertising veryattractive routes to data sink. The neighbor nodes se-lect the malicious node as the next hop for messageforwarding considering it a high quality route andpropagate this route to other nodes. Almost all trafficis thus attracted to the malicious node that can eitherdrop it, selectively forward it based on some maliciousfiltering mechanism or change the content of the mes-sages before relaying it. This malicious node has thusformed a sink hole with itself at the center.

The sink hole is characterized by intense resourcecontention among neighboring nodes of the maliciousnode for the limited bandwidth and channel access[11]. This results in congestion and can accelerate theenergy consumption of the nodes involved, leading tothe formation of routing holes due to nodes failure.With sink holes forming in a sensor network, severalother types of denial of service attacks are then possi-

ble [5],[11].Worm hole is another kind of denial of service at-

tack [12]. Here the malicious nodes, located in differ-ent part of the sensor network, create a tunnel amongthemselves. They start forwarding packets receivedat one part of the sensor network to the other endof the tunnel using a separate communication radiochannel. The receiving malicious node then replaysthe message in other part of the network. This causesnodes located in different parts of networks to believethat they are neighbors, resulting in incorrect routingconvergence. Section VI details some of the counter-measures against these denial of service attacks.

III. Coverage Holes

In this section we describe various proposed solutionsfor finding and fixing coverage holes in sensor net-works. The proposed protocols discussed here areclassified based on the number of mobile nodes in thesensor network as (i) Mobile sensor networks (ii) Hy-brid sensor networks (iii) Static sensor networks. Foreach of the protocols discussed in this section, a fur-ther differentiation is made according to whether it isdesigned to address single coverage or the multiplecoverage requirement.

III.A. Mobile Sensor Networks

Several researchers have investigated techniques toobtain maximum single coverage of a target area us-ing mobile sensors [13] [14] [15] [16] [17]. A typicalproblem statement for this scenario is to maximize thecoverage of a given target area with constraints on de-ployment time, the distance the sensors have to travelto maximize coverage and the complexity of the pro-tocol [13].

Hole

ba

c

d

e

x

Figure 3: Voronoi diagram. abcde is the Voronoi ploygon for node x.Circle centered at x is the sensing disk

In one of the proposed solutions [13], Wang et al.used Voronoi diagrams to discover the existence ofcoverage holes once all the sensors have been initiallydeployed in the target area. A node needs to know thelocation of its neighbors to construct its Voronoi dia-gram. This implies that either all the nodes are GPSenabled or they use one of the GPS-less localizationtechniques such as [18]. The diagram partitions thewhole space into Voronoi polygons. Each polygon has


a single node with the property that every point in thepolygon is closer to this node than any other node. Asensor node compares its sensing disk with the areaof its Voronoi polygon to estimate any local coveragehole (Fig. 3).

Three distributed self-deployment algorithmshave been proposed: Vector based(VEC), Voronoibased(VOR) and Minimax algorithm. In VEC a nodecalculates the average distance between the neighborsassuming all nodes are uniformly deployed in thegiven target area and tries to keep the distance withits neighbor approximately equal to this distance. InVOR once the coverage hole is detected, the nodemoves toward the farthest vertex of the Voronoipolygon to cover the local maximum coverage hole.Minimax works like VOR with the additional checkthat while moving toward the farthest Voronoi vertex,it also keeps track of distances to other vertices andfinds a target position inside the polygon from wherethe distance to the farthest vertex is minimized.

Simulation results presented in [13] shows that theMinimax algorithm outperform the other two pro-posed algorithm in achieving maximum coverage withlittle increase in computational overhead. However,on average, Minimax moves the sensors longer dis-tance than the other two algorithms to achieve thishigher level of coverage. Simulated mobility has beensuggested as an alternative to get the final target posi-tions for the nodes at the cost of higher communica-tion overhead. The relative analysis of increased mes-sage complexity of simulated mobility and reductionin energy consumption due to lesser movements is anopen research question.

Ganeriwal et al. in [14] proposed a protocol calledCo-Fi that uses mobility capable sensors for repair-ing the loss of coverage due to energy depletion ina deployed sensor network. Low energy nodes, onpredicting death, broadcast a Panic request message.Nodes with high energy level respond with the Panicreply message if they can move without losing exist-ing coverage. The Panic reply message contains resid-ual energy and the mobility cost (shortest distance thehelper node has to travel to reach its final destina-tion). The dying node receives multiple Panic replymessages and it chooses a node with maximum utility(residual enegy - mobility cost) and notifies the se-lected node to move. As the protocol relies on broad-cast of a Panic request message by a dying node, itwill not work when nodes gets physically destroyedcreating un-repairable coverage holes.

In other significant research Howard et al. [15]proposed a potential fields based approach for self-deployment of mobile sensor networks. Nodes aretreated as virtual particles and the virtual forces dueto potential fields repel the nodes from each other and

obstacles. The authors assume that each sensor is ca-pable of determining the range and bearing of bothits neighbor nodes and the obstacles. This approachdoes not require any communication among the nodesfor movement or localization information. Instead thenodes only use their sensed information in making thedecision to move making it a cost effective solutionto the coverage problem. However, extensive simula-tion/experimental studies have not been conducted totest the sensitivity of the approach to changes in com-munication and sensing ranges and different networksizes etc.

In other related work by the same authors [16],an incremental self deployment algorithm is proposedthat maintains the line of sight relationships amongthe nodes. The line of sight constraint is necessaryfor localizing the nodes by using existing deployednodes as landmarks. The target position of a node iscalculated, using the deployment algorithm, at a highprocessing power base station. Each deployed node isresponsible for communicating its local informationback to the base station for utilization in the next iter-ation of the deployment algorithm. This implies thateach node has to maintain bidirectional communica-tion with the base station at all times or else they aredeployed so that they form a multi-hop connected net-work at all times.

The authors have conducted simulations based onachieved coverage and time. Different deploymentgoal selection policies such as deterministic and sto-chastic etc. have been proposed and evaluated. Re-sults have shown that deterministic goal selection pol-icy outperforms stochastic based policies but it stillcannot guarantee complete coverage of the sensor net-work. As compared to the distributed self deploy-ment proposals, the proposed sequential deploymentscheme will take much longer time to deploy whenthe number of nodes is increased.

Heo et al. [17] proposed two schemes for address-ing single coverage problem. In one scheme calledDistributed Self-Spreading algorithm (DSS) the au-thors propose a deployment scheme similar to [15]and VEC [13]. Sensors are assumed to be randomlydeployed initially. They start moving based on partialforces exerted by the neighbors. Forces exerted oneach node by its neighbors depends on the local den-sity of deployment and on the distance between thenode and the neighbor.

Another scheme called Intelligent Deployment andClustering Algorithm (IDCA) is also proposed in [17]to utilize low energy consumption characteristics oflocal clustering. Local density is compared with den-sity expected when all nodes are uniformly distributedin the target area and for close values a sensor se-lects the clustering mode. In this mode the relative re-


maining energy of the sensors dictates whether a nodeshould move or not. The idea is to reduce the variancein remaining energy once all the nodes are uniformlydistributed in the target area.

To summarize, the solutions discussed in this sub-section either utilizes techniques from computationalgeometry e.g. Voronoi diagrams or uses virtual forces,to attract or repel neighbors, to achieve the desiredcoverage of the area.

III.B. Hybrid Sensor Networks

The coverage scenario when only some of the sensorsare capable of moving has been under active research,especially in the field of robotics for exploration pur-poses. The movement capable sensors can help in de-ployment and network repair by moving to appropri-ate locations within the topology to achieve desiredlevel of coverage and connectivity, and to connect apossibly disconnected network.

Corke et al. [19] address the issue of network de-ployment with adequate connectivity using an Un-manned Aerial Vehicle (UAV). The flying robot, re-ferred as AVATAR by the authors, can deploy thenetwork based on a precomputed network topology.The sensors, once deployed, compute their connectiv-ity map and relay this information to the flying UAV.The existing network connectivity is compared withthe desired topology and the separation regions areidentified. Deployment points within this region ofseparation are computed to repair the network and theUAV again deploys more sensors at desired deploy-ment points. The deployment of additional sensors inthe target area increases the sensor density( achievingmultiple coverage and connectivity) or repairs failingsensor nodes in the network.

The experimental results during the network de-ployment stage show that it is very difficult to achievethe desired network topology using aerial deployment.The experiments to achieve a grid topology with 2mby 2m grid spacing resulted in a median deploymenterror value of 1.2m. This basic limitation makes it in-teresting to evaluate the proposed deployment schemein terms of number of nodes required to achieve a de-sired degree of coverage and connectivity.

In [20], Batalin et al. suggested a combined solu-tion for the exploration and coverage of a given targetarea. The coverage problem is solved with the helpof a constantly moving robot in a given target area.The mobile robot first performs the network deploy-ment in the target area as it explores the unknown en-vironment. The deployed nodes then guide the robot,based on their local measurements, to poorly coveredareas. The mobile robot, using its local sensing dataand the recommended direction acquired from a de-

ployed sensor node, decides about its future directionfor exploration. If the robot does not receive a direc-tion beacon when it has traveled a predetermined dis-tance in one direction, it deploys another sensor nodeto improve the local coverage of the area.

The algorithm does not consider the communica-tions between the deployed nodes. All decisions aremade by the robot by directly communicating with aneighbor sensor node. Distributed computation and innetwork processing is suggested in [20] as a solutionfor the homing problem (when a robot wishes to returnto a specific point in the target area). The deploymentstrategy and especially the network repair policy canalso benefit from the multi hop information derivedout of a communicating sensor network and thus wefeel that this should be explored further.

Wang et al. [21] addressed the single coverageproblem by moving the available mobile sensors in ahybrid network to heal coverage holes. The static sen-sors detect their local coverage holes by using Voronoidiagrams as in [13]. The mobile sensors also calculatecoverage holes formed at their current position if theydecide to leave their current position. The static sen-sors bid for the mobile sensors based on the size oftheir detected coverage hole. A mobile sensor com-pares the bids and decides to move if the highest bidreceived has a coverage hole size greater than the newhole generated in its original location due to its move-ment. The bids are broadcast locally up to two hopsand the static sensors are able to direct neighboringeligible mobile sensors to a point close to the farthestvertex in their Voronoi polygon. However, the localbroadcast may prevent the bid messages reaching mo-bile sensors if they are located farther than two hops.

III.C. Static Sensor Networks

In this section we cover topology/density control pro-tocols that select a minimal number of on-duty nodesthat are active at any time out of the available denselydeployed nodes. This node scheduling is feasible aslong as no coverage holes appear due to nodes beingturned off for energy savings. Efforts like [10], [6],[22] etc. have focused on providing the desired multi-ple level of coverage in a given target area while [23],[24], [25] etc. address single coverage problem.

Authors of [10] Wang et al. presented the Cov-erage Configuration Protocol (CCP) that can provideflexibility in configuration of sensor network to self-configure for different degrees of coverage. The au-thors proved that for boundary nodes, nodes whosesensing circle intersects with the boundary, desiredconnectivity is equal to the degree of coverage whilefor interior nodes the desired connectivity is twice thedegree of coverage.


Each deployed node runs the Ks-coverage eligibil-ity algorithm when Rc ≥ 2Rs, where Rc is the com-munication range and Rs is the sensing range. Given arequested coverage degree Ks, a sensor node is sched-uled to sleep if every location within its coveragerange is already Ks covered by other active nodes inits neighborhood. For cases when Rc < 2Rs, CCPdoes not guarantee connectivity along with the cov-erage. Authors have proposed combining CCP withan existing connectivity maintenance protocol, SPAN[26] which provides the communication connectivity.

Huang et al. [6] proposed polynomial-time algo-rithms to verify whether every point in the target areais covered by at least the required number of nodes.Algorithm k-UC assumes a uniform circular sensingdisk while k-NC assumes a non-disk sensing range foreach sensor node. The proposed algorithm requiresonly the location information of each deployed nodeto evaluate the desired multiple coverage. For k-UC,each node calculates the coverage of its neighbor onlyif the neighbor lies within twice the sensing range ofthe node. For k-NC, different neighbor selection rulesare defined for cases when one of the nodes is withinthe sensing range of other nodes and when one/both ofthe nodes are within sensing range of each other. Thenode then calculates its perimeter coverage by findingthe sector of its coverage area occupied by the neigh-bor sensing range. The node thus verifies whetherits whole perimeter 2π is covered by existing neigh-bors to the required degree or not. To detect coverageholes, the authors suggest a central controller entitythat can collect the details of insufficiently coveredsegments from each such node and can dispatch newnodes to cover that existing hole. However, this cen-tralized approach lacks scalability.

Yan et al. in [22] proposed a distributed densitycontrol algorithm capable of providing differentiatedcoverage based on different requirements in differentparts of the deployed sensor network. The algorithmis based on time synchronization among the neigh-bors. In the initialization phase, nodes get their lo-cation and synchronize with neighbor nodes. In thesensing phase, comprising of several rounds of equalduration, each node divides its whole area into gridsand advertises its reference point and, start and stoptime, defined with respect to that reference point. Thenode sorts the neighbors covering a particular grid inascending order of their reference points in a round.Based on the time sequence obtained, a node can de-cide its on-duty time such that the whole grid still getsthe required degree of coverage. The results from allthe covered grids are merged to find the adopted dutyschedule for the node.

For single coverage requirement, Zhang et al. in[23] have proposed the Optimal Geographical Den-

sity Control (OGDC) protocol. This protocol tries tominimize the overlap of sensing areas of all sensornodes for cases when Rc ≥ 2Rs. The algorithm startswith all the nodes initially in “UNDECIDED” state.A node with sufficient power is randomly chosen tostart the process of node selection. This starting nodebroadcasts a power-on message. Nodes, on recep-tion of this power-on message, check their power leveland the existing coverage of area under their sensingrange. If sufficient power is available, and the area isnot fully covered, the node adds the starting node asthe neighbor, sets its state to “ON” and broadcasts thepower-on message again. This process continues withslightly different behavior for power-on messages re-ceived from starting and non-starting nodes. Simu-lation results presented by the authors of [23] showsthat OGDC protocol cannot always preserve the orig-inal coverage of the network completely.

The issue of coverage with different sensing capa-bilities has been addressed in [24] by Tian et al. Theauthors discussed the topology control for both uni-form and different sensing ranges. A node decidesto turn off after discovering that its neighbors (spon-sors) can completely cover its sensing area. The nodecredits the sponsored area based on sectors instead ofthe actual crescent formed in its sensor range. Oncethe sponsored area becomes equal to the sensing areaunder the node, the node qualifies as a candidate tobe switched off. To avoid the possibility of multipleneighbors turning off and creating a coverage hole, thenodes use a random back-off algorithm before goingto sleep.

Jiang et al. in [25] identified two short-comings inthe sponsored area approach of [24]. Firstly, neigh-bors lying outside the sensing range are not consid-ered although they can contribute to the node cover-age. Secondly, the sector based area calculation forcoverage results in a more conservative estimate ofthe neighbors contributions in covering the area. Theauthors introduced the concept of effective neighbornodes for calculating the nodes coverage accuratelyand to decide upon redundant nodes that could be putto sleep while conserving the original coverage. Theneighbor set now includes all nodes within twice thesensing range of each node. The results presented in[25] show that the proposed protocol is able to out-perform the sponsored area approach by about 30% interms of reducing the actual number of nodes requiredfor maintaining the original coverage.

Table 1 summarizes the proposed solutions for cov-erage holes problem. Most of the proposed solutionsassume simplified circular sensing disc model and thatall the sensors are location aware. The circular sensingdisc model is not appropraite for all kinds of sensorsmeasurements like temperature, humidity, acoustics


Category Approach Proposed Solution Main Assumptions CharacteristicsComputationalGeometry

VEC, VOR, Minmax[13]

Location information Localized, scalable, distributed.

Co-Fi [14] Location information.Nodes can predict theirdeath

Single coverage based. Residual energy consid-erations.

Mobile Sensors Virtual Forces Potential Fields [15] Range and bearing Scalable, distributed. No local communicationrequired for localization or movement.

DSS,IDCA [17] Location information Scalable,distributed,residual energy based.

Sequential Incremental [16] Line of sight for localization Centralized. Bidirectional communication withbase station.

Single MobileSensor

UAV [19] Predetermined topology in-formation

Flying robot for deployment and network re-pair. Inaccuracies using aerial deployment

Hybrid Sensors Single Robot [20] Location information Distributed, no multi-hop communication fornetwork deployment and repair.

Multiple MobileSensors

Bidding Protocol [21] Location information Uses Voronoi diagram for single coverage re-quirement.

CCP [10] Location information, uni-form sensing disk

Configurable degree of coverage, calculated byintersection points of sensing circles.

Multiple Cover-age

k-UC, k-NC [6] Location information Perimeter coverage, non-disk sensing modelsupported.

Static Sensors Differentiated [22] Location information, timesynchronization

Grid based differentiated degree of coverage.

OGDC [23] Location information, uni-form sensing disk

Residual energy consideration.

Single Coverage Sponsored Area [24] Location information Sector based coverage calculations, non-disksensing model supported.

Extended-SponsoredArea [25]

Location information, timesynchronization

Uniform disk sensing model.

Table 1: Comparison of proposed solutions to coverage hole problem

etc. This issue is further discussed in the future re-search directions in Section VII. Also solutions basedon mobile and hybrid sensors have been proposed tosolve only the single coverage requirement. We sug-gest that multiple and differentiated coverage can alsobenefit from mobility capable sensors.

IV. Routing Holes

Information driven and data centric routing has beenthe focus of many research efforts in wireless sensornetworks [27], [28], [29] etc. Two popular fault tol-erance schemes used in proposed protocols are single-path routing with route repair and multipath routing.Geographic greedy routing has also been proposed asa scalable and localized solution to address the routingissues associated with MANETs [30] and is a viablesolution for routing in large sensor networks. In thissection we classify the routing holes problem in threecategories. Issues of routing holes are discussed insingle path routing, multipath routing and geographicgreedy routing.

IV.A. Single Path Routing

This section covers single path routing protocols forwireless sensor networks where resilience is providedby different path repair strategies. Since single pathrouting algorithms for wireless sensor networks is a

well investigated area, we only discuss some of therepresentative proposed routing protocols.

Intanagonwiwat et al. [27] were the first to pro-pose a data centric communication protocol for wire-less sensor networks called Directed Diffusion (DD).In the data centric paradigm of DD, data is named us-ing attribute-value pairs. The sink generates an in-terest for the named data that is diffused through thewhole sensor network. Each node establish a gradi-ent toward its neighbor nodes from which it receivesthe interest. A node capable of producing the desirednamed data starts sending exploratory data to all itsneighbors from which it has received interest for thedata at the rate specified in the interest message. Thisexploratory data follows the established gradients, ina multipath way, to reach the sink that initiated theinterest message. The sink chooses to reinforce oneparticular neighbor to draw data, at higher data rates,by sending a positive reinforcement message that isunicast back to the sender. The sender will now startsending data at higher rates along the reinforced gra-dient to the sink. DD can be tuned to work as a multi-path protocol if alternate paths are kept alive by send-ing reinforcement with different interval attributes todifferent neighbors.

In [31], Heidemann et al. proposed two new dif-fusion algorithms, Push and One Phase Pull(OPP) tocater for different application requirements. Push dif-


fusion reverses the role of senders and sinks as com-pared to the original algorithm proposed in DD whichis referred to as Two Phase Pull(TPP). Senders floodthe exploratory data, sinks on receiving exploratorydata send positive reinforcement to create reinforcedgradient. In OPP the exploratory data flooding phaseof TPP has been removed to reduce the overhead.Senders in OPP selects a preferred gradient when it re-ceives multiple interest messages and instead of send-ing exploratory data, it starts sending data at a higherrate on its selected low latency path.

TPP recovers from failed path and routing holesby periodic flooding of interest and exploratory data.The choice of refresh rates for the interest and the ex-ploratory data (interest refresh rate is 30s, less than theexploratory data refresh rate of 90s) thus controls howquickly a routing hole is discovered and bypassed.The algorithm also provides for local repair by in-termediate nodes by sending reinforcement messages.The recovery part in Push depends on the periodic ex-ploratory data flooding, thus in case of routing hole,Push takes much longer than TPP to select anotherroute .

The interest flooding overhead in DD becomes highwhen the number of sinks increase [31]. Also theevent flooding proposed in push diffusion is only vi-able when number of sinks interested in the event ismuch more than the number of sources. Braginskyet al. in [32] proposed Rumor Routing, to addressthese overheads. The rumor routing protocol createspaths leading to each events, instead of flooding theevent in the entire network, by using long-lived pro-tocol packets called Agents. Whenever a sink createsa query, it goes through random walk until it finds theestablished event path, which it can then follow to theevent. Thus network wide flooding is avoided bothfor event and query as compared to the three vari-ant of directed diffusion, TPP (both interest and ex-ploratory data are flooded), Push diffusion (only ex-ploratory data is flooded) and OPP (only interest isflooded).

In rumor routing if the node that originated thequery finds out that the query did not find any es-tablished event path (through timeout based on TTLvalues), it can retransmit the query (another randomwalk) as chances of finding the event path grows ex-ponentially with the number of retransmission, if sucha path does exist. As a last resort, the query can beflooded to find the event at the cost of high overhead.If case of routing holes, higher TTL values or floodingcan route across the hole to find the event.

Luo et al. in [28], proposed Two Tier Data Dis-semination protocol (TTDD) that divides the wholetopology into cells using a grid structure. The gridstructure is constructed from the perspective of sender

using simple geographical greedy forwarding, so eachsender will form its own grid to disseminate data tosinks. The protocol selects forwarding points calledDissemination Nodes (DN) at the sensors that areclosest to the grid points. A query from a sink nowtravels two tiers to reach the sender. At the lower tier,the sink floods the query within its own cell until itreaches its nearest DN. The second tier consists ofrouting along the DN such that it reaches either thesource or a DN that is already receiving data fromthe source. During query forwarding a reverse pathis formed toward the sink that is used to actually for-ward the data to the sink.

Critical to the performance of the TTDD protocolis the choice of the cell size. A bigger cell size willresult in more flooding area for the query and lessernumber of DN for each source. Queries and conse-quently data may take suboptimal paths traversing theDNs of the grid topology. TTDD uses upstream infor-mation duplication by selecting various neighbors ofDN as recruited nodes to replicate the information ofupstream DN. In case the DN fails, one of its neighborcan take over the role of DN for its cell. The problemturns severe in case a routing hole forms covering allthe recruited nodes as well as the DN.

Tian et al. [33] proposed a local pivot-initiated pathrepairing approach while using single path routing inwireless sensor networks. A node located immedi-ate upstream to a failure point, called pivot node, isresponsible to find alternate paths through local in-teraction. The pivot node broadcast Help Request(HREQ) message to its neighbors. A neighbor thatcan provide an alternate route to the destination replieswith Help Response (HREP) message. In case noHREP message is received, the pivot node returns thepacket to the node from which it initially received thepacket. The upstream node now tries the local repairexcluding the node that returned the packet back. Thisprocess continues until a route to destination is foundor the packet is traced backed to the source withoutfinding any suitable path. The approach seems moreenergy efficient but requires each intermediate for-warding node to incorporate some kind of feedbackmechanism to ensure that data is delivered to its suc-cessor.

We have discussed, in this subsection, single pathrouting protocols that use one or more of the follow-ing path repair strategies: retransmission, source ini-tiated path repair, intermediate node initiated local re-pair, routing information duplication and flooding.

IV.B. Multi-path Routing

The basic idea behind multipath routing is to maintainalternate routes to provide resilience against node fail-


SourceSink

( i )

Sink

( ii )

Source

Figure 4: (i). Braided multipath (ii). Node disjoint multipath

ures or to distribute the traffic among multiple pathsfor load balancing and uniform energy consumptionin the network. Ganesan et al. [34] proposed theuse of multipath routing for data dissemination inwireless sensor networks. They considered two ap-proaches of multipath routing namely, node-disjointmultipath routing and braided multipath routing. Thenode-disjoint multipath routing requires that the alter-nate paths and the original path are mutually exclu-sive i.e., they do not intersect each other. The braidedpaths consists of many interleaving paths (Fig. 4).The node-disjoint multipath can tolerate any numberof nodes failure on the original best primary path butfails if a single node on each alternate path fails whilethe braided multipath can provide an alternate routein this scenario. Simulation results shows that forisolated failures, the braided routing out-performs thenode-disjoint multipath routing, while in case of rout-ing holes both braided and node-disjoint gives compa-rable performance but the braided multipath has aboutone third of the overhead for alternate path mainte-nance.

In [29], Ye et al. proposed a multipath routingprotocol named Gradient Broadcast (GRAB) designedspecifically for robust data delivery. The protocol re-lies on two parameters, the cost field and the credit.The sink propagates advertisement packets in the net-work and each node that receives this advertisementrecords the cost to reach the sink along a path. At theend of the cost field setup phase each node has theminimum cost needed to reach the sink. When a nodesends a packet to the destination, it simply includes itsown cost in the packet and broadcast it to its neigh-bors. Only the neighbor with a smaller cost than thecost contained in the packet are allowed to forward it.

The multipath aspect of the GRAB protocol is con-trolled by credit. Credit is the extra allowance thatallows the use of interleaved paths, each of which hasa cost less than the cost+credit alloted by the sourceof the packet. Thus the amount of credit actually con-trols the width of the mesh. The design of GRAB pro-vides for extra consumption of credit in the beginninghops from the source so that the mesh is wider in thebeginning and nodes closer to the sink consume lesscredit resulting in a narrower mesh. If the routing holeis formed near the sender, there are more chances ofbypassing it due to the extra width of mesh available

near the sender.In [35], Dulman et al. explored the tradeoff be-

tween degree of reliability and amount of traffic over-head for multipath routing. The authors present ascheme for delivering data reliably in spite of routefailures. They propose to split the data packet intok sub-packets using forward error correcting codes(FEC) to add redundancy to the original data. The fac-tor k is dependent on the number of disjointed pathsfrom source to destination and the failing probabilityof the alternate paths. The destination then affordsto miss some sub-packets but can still reconstruct theoriginal message. The simulation results show that in-stead of transmitting the whole message on all the al-ternate disjointed multiple paths, transmitting the sub-packets reduces the overhead significantly.

Rahul et al. [36] proposed the use of multipath rout-ing to uniformly distribute the usage of energy in wire-less sensor networks. The idea is that if the low costpath (lowest energy consuming) is always used, nodeson this path will soon deplete their energy and caneven result in network partition. Their proposed pro-tocol keeps a set of good paths for the same desti-nation and probabilistically choose different paths atdifferent times. The alternate paths maintained by theenergy aware protocol are sub-optimal paths as com-pared to the lowest cost path but distributing the trafficacross different paths results in more even distributionof energy usage and prolongs network lifetime of de-ployed network.

IV.C. Geographic Routing

Geographic routing relies on greedy forwarding toroute packets by only making local decisions. Karpet al. in [30] describes the Greedy Perimeter StatelessRouting (GPSR) for MANETs. The protocol starts ingreedy forwarding mode. GPSR recovers from rout-ing holes due to local minimum phenomenon by us-ing perimeter routing mode. In perimeter routing theright-hand rule is used which states that when arriv-ing at node x from a node y the next edge traversedis the next one sequentially counterclockwise aboutx from edge xy. The right-hand rule requires thatall the edges are non-crossing. GPSR proposes eitherRelative Neighborhood Graph(RNG) [37] or GabrielGraph(GG) [38] to get a planar network graph withno crossing edges. The maintenance of planar graphat each node introduces overhead. While all the nodesmaintain the planar graph all the time, this informa-tion is only used by nodes facing the local minimumphenomenon.

Kranakis et al. in [39] proposed the Compass Rout-ing II algorithm that guarantees that the destination isreached even when local minimum phenomenon oc-


curs in greedy forwarding. They proposed the use offace routing using the least deviation angle from theline joining the node to the destination when trying toroute a packet to the next hop.

Similar to the work in [39], Bose et al. [40] pro-posed the FACE-1 and FACE-2 routing algorithms toguarantee packet delivery in MANETs. The suggestedsolution is also based on getting a connected planarsubgraph by using Gabriel Graph and then traversingthe edges of the graph using right-hand rule. In con-trast to GPSR, all routing is done through the perime-ter of the GG formed at each node. FACE-2 modi-fies FACE-1 in that the perimeter traversal follows thenext edge whenever that edge crosses the line fromthe source to destination. The routing hole problem isaddressed by always using the perimeter mode rout-ing, however the nodes in the perimeter depletes moreenergy.

As an extension to the compass routing II algo-rithm, [41] introduced deterministic fall back mecha-nism to get back in the greedy mode from the perime-ter routing mode without necessarily exploring thecomplete face boundary. The proposed algorithm,called GOAFR+, first constructs the Gabriel Graph(GG) of the network. It starts forwarding in a greedymode and when it reaches a local minimum point, itswitches to the face routing of [39]. The algorithmuses two counters to keep track of how many visitednodes in face routing are nearer to the destination andhow many are far from the destination as compared tothe starting node. Based on the values of these twocounters the algorithm decide whether to continue inface routing mode or fall back to the greedy phase.

In [42], De Couto et al. proposed a probabilisticsolution called Intermediate Node Forwarding (INF)for routing around holes assuming non-uniform radioranges. Negative acknowledgment packets (NAKs)has been proposed in the basic geographic forwardingto provide feedback to the source about packet dropsdue to local minimum occurring at any intermediatenode. Once the sender knows about packet drops,it randomly selects an intermediate location from adisk, of radius one quarter of the distance between thesender and the destination, centered at the midpoint ofthe distance between the sender and the destination. Ifthe selected node drops the packet again, the radius ofthe disk is doubled and another random intermediatelocation is selected. Incorporating the NAKs in thegeographical greedy forwarding will increase the pro-tocol overhead. Furthermore the assumption of NAKsreaching the source is not valid if asymmetric links areconsidered.

Li et al. in [43] proposed an active message trans-mission by relaying scheme for communication in adisconnected mobile ad-hoc wireless network. The

protocol relays messages using mobile agents by mov-ing nodes appropriately to complete a routing path ina disconnected network. Two different algorithms arepresented, one relying on full location knowledge ofall the hosts while the other one uses a location up-date mechanism. For wireless sensor networks theproposed protocol can avoid routing holes formed dueto sparse deployment or node failures but will fail toachieve any significant results if the routing holes areformed due to unreachable terrain or physical obstruc-tions in the region.

Fang et al. in [44] defined stuck nodes as nodes inthe topology where packets can possibly get stuck ingreedy multi-hop forwarding due to local minimumand proposed TENT rule to test whether a node in agiven topology is a stuck node. They also came upwith BOUNDHOLE distributed algorithm to find theboundary of the routing hole.

O

vx

u

Figure 5: Illustration of the TENT rule. Solid circle represents node x’stransmission range

The TENT rule at each node maintains locationinformation of all 1-hop neighbors of the node incounter-clockwise order. Then for each pair of adja-cent nodes u, v for node x the perpendicular bisectorof ux and vx is drawn that intersect at a point O. IfO is inside the communication range of x, then x isnot a stuck node(see Fig 5). A node can thus calculateits stuck directions by applying the local check for allpairs of adjacent neighbors. The BOUNDHOLE al-gorithm uses the right-hand rule, after each node hasidentified its stuck direction, to mark the boundary ofthe routing hole.

Yu et al. [45] proposed Geographic and EnergyAware Routing (GEAR) to route a packet toward aregion of interest. GEAR works in two phases, inphase I it uses energy aware next hop neighbor se-lection to route a packet toward a target region whilephase II involves restricted flooding or recursive geo-graphical forwarding to disseminate the packet insidethe region. For phase I, each node keeps two kinds ofcosts, an estimated cost and a learned cost of reachingthe destination through its neighbor. Each node getsthe learned cost to a region of interest from all of itsone-hop neighbors and it can compute its own learnedcost by adding to the selected neighbor cost, the costto reach that neighbor. The estimated cost depends onthe node’s residual energy and its distance to the desti-


Category Protocol State maintained Fault tolerance with routing holesDD [27] Interest gradient and data path Periodic flooding of interest and exploratory data.

Local repair by reinforcement messages.

OPP, Push [31] Interest/event gradient Periodic flooding of interest/event

Single Path Rumor Routing [32] Event path (Agents) Retransmission. Flooding in extreme case.

TTDD [28] Per source DN information Recruited nodes for each DN

Pivot based [33] Only path repair strategy specified. Usesany existing routing protocol

Local repair attempted followed by notification toupstream node

Braided [34] Multiple paths maintained Braided/node-disjoint alternate paths available

Multipath GRAB [29] Cost field Alternate paths based on available credit

FEC based [35] Multiple paths maintained FEC for message reconstruction at destination

EAR [36] Multiple paths maintained Alternate paths and localized flooding

GPSR [30] Location information and planar graph Right hand rule using planar graph to avoid holes

Compass Routing[39],FACE-I,FACE-II[40],GOAFR+ [41]

Location information and planar graph Face routing to avoid holes using planar graph

Geographic Routing INF [42] Location information NAKs and source initiated repair

Active Message Relay[43]

Location information, Active messages Mobile nodes move to reach disconnected neigh-bors.

TENT rule [44] Location information, boundary ofholes(perimeter nodes only)

TENT rule to identify holes. Boundary informationmaintained to avoid the hole.

GEAR [45] Learned and estimated cost Limited flooding in region, learned cost helps findalternate routes

Table 2: Comparison of some proposed routing protocols

nation and is used as default value in case learned costis not available.

In case there are no routing holes, the estimated costis equal to the learned cost. When a routing hole ap-pears, the change in the path is reflected in the newlearned cost of the node that encounters the routinghole. This new learned cost is also propagated backto be utilized for early avoidance of the routing holeso that nodes surrounding the routing hole do not getdepleted at a fast rate. In phase II, for high-densitynetworks, recursive geographical forwarding is usedas it is more energy efficient than restricted flooding.GEAR works well if the region to be covered is asmall fraction of the total area covered by the sensornetwork but the protocol efficiency tend to diminishfor cases when whole region is the area of interest.

Table 2 lists comparison of the proposed routingprotocols that have been discussed in this Section. Themechanism to achieve fault tolerance against the rout-ing hole problem and the state maintained are high-lighted for each of the protocol.

V. Jamming Holes

A jamming hole differs from other types of holes thatcan exist in the sensor network in that it circumventsthe ability of nodes in a specific area to communi-cate/sense. The schemes that are usually employedto combat jamming include the use of various spreadspectrum techniques for radio communications andusing different transmission media like infra-red oroptical combined with radio. The prohibitive factor in

adapting these countermeasures is the cost and com-plexity. We assume for the rest of this section thatno such counter-jamming technique is available to thedeployed network.

Wood et al. [4] proposed a protocol called JAM, todetect and map jammed regions in a sensor network.The detection part of the protocol applies heuristicsbased on available data, e.g. bit-error rates etc., todistinguish jamming from normal interference. Thecritical part of the detection phase is the selection ofa certain threshold value above which the interferenceis declared as jamming. Once jamming is detected,the protocol assumes a carrier sense overriding mech-anism to send a brief, high-priority broadcast messagereferred as JAMMED to its neighbors.

The mapping part of the protocol starts when a nodeoutside the jammed region receives a JAMMED mes-sage from a node inside the jamming hole. The re-ceiving node now broadcasts a BUILD message to itsneighbors to start building the jamming hole bound-ary. Multiple BUILD messages originating fromneighboring nodes are combined to form the bound-ary of the hole.

The JAM protocol assumes that the location infor-mation and unique ID is known to each node. The pro-tocol also relies on the availability of a carrier-sensedefeating broadcast mechanism to notify the jammingto un-jammed neighbors. This is, in our opinion, themost demanding requirement. The broadcast of theJAMMED message will succeed if the jamming is in-termittent. For continuous channel jamming it is hardto guarantee that the JAMMED message will reach


any of the neighbor node outside the jamming hole.Thus, a protocol that can detect the presence of a jam-ming hole without relying on help from the nodes in-side the jamming hole is more desirable. This is fur-ther discussed in future research directions in SectionVII.

VI. Sink /Black Holes/Worm Holes

This section describes some proposed measures tomitigate the effects of sink/black hole and worm holedenial of service attacks in wireless sensor networks.

Karlof et al. [5] analyzed the resilience of var-ious routing protocols and energy conserving topol-ogy maintenance algorithms against sink holes. Theyshowed that popular routing protocols like directeddiffusion, rumor routing and multi-path variant of di-rected diffusion etc. are all vulnerable to sink holes at-tacks. For geographical greedy forwarding algorithmsit is more difficult to create sink holes because in thiscase a malicious node has to advertise different attrac-tive locations to different neighbors in order to qualifyas next hop.

Authors of [5] suggest authentication and link layerencryption as counter-measures against the sink holeattacks. This is essentially to prevent malicious nodesforming part of the topology and participating in therouting protocol for injecting incorrect routing infor-mation. However, the authorization mechanism andlink layer encryption can fail to protect against wormhole attacks.

Wood et al. in [11] identified four possible de-fenses against the sink holes. In the authorizationsolution, only authorized nodes can exchange rout-ing information with each other. The solution is notscalable due to high computation and communicationoverhead. Also, public key cryptography is not feasi-ble in sensor networks given the capacities and con-straints of the sensor devices. In another proposedsolution, nodes can monitor their neighbor behaviorto verify that the next hop does transmit the messagejust sent by the node. This scheme also fails when amalicious node is simply altering the contents of themessage and forwarding it or forwarding the packetsto another malicious node (in worm holes, if usingthe same communication channel) to avoid monitor-ing. The third proposed solution suggests that redun-dancy introduced by using multi path routing can helpin avoiding the sink holes and worm holes especiallywhen dis-joint multi path routes are maintained andactually utilized for data transmission. In another pro-posed solution in [11], probing by geography basedprotocols is carried out to detect the presence of sinkholes. The nodes can periodically send probes acrossthe network diameter to check the routes.

VII. Future Research Directions

Having surveyed solutions proposed for addressingvarious holes related problems in wireless sensor net-works, we discuss in this section some of the futureresearch directions.

VII.A. Coverage based on Realistic As-sumptions

Sensing model

Much of the research thus far assumes simplifiedboolean sensing model (circular disc) for coveragefor protocol design and evaluation. In this model allevents within the circular disc are assumed to be de-tected with probablity 1. This simplified model isclearly not applicable to all types of sensing mea-surements i.e., measuring spatial distribution of localquantities like temperature where the sensing rangeor disc makes little sense. This model is still overlysimplistic for sensing acoustic signals where the sig-nal strength attenuates with distance like electromag-netic waves. A more general model based on certainsignal to noise ratio thresholds, with proper data fu-sion mechanism to reduce the variance of measure-ment, is more appropriate. There is a need to evaluatethe existing proposed protocols based on realistic gen-eral sensing model and to design hybrid coverage pro-tocols capable of delivering accurate spatio-temporalprofile of different kinds of sensing measurements.

Location inaccuracies

Most of the coverage protocols assume loaction-awarenodes (see Table 1). However, location estimates us-ing available sensor localization protocols is not veryaccurate. There is a need to evaluate the effect of loca-tion in-accuracies on the performance of various cov-erage protocols.

VII.B. Mobility Assisted DifferentiatedCoverage

The basic assumption in static sensor networks thatnodes are available to cover every point in the tar-get region to provide the desired level of coverageis overly simplistic. This is particularly true for hos-tile or harsh environments, like battlefields and fire orchemical spill, where the network cannot be carefullydeployed to a predetermined regular topology. Ran-dom aerial deployment of sensor nodes is a possiblesolution but in this case it is very difficult to guaran-tee that required multiple coverage is achieved. Adap-tive sensor placement based on mobile sensors, simi-lar to the beacon placement for localization suggested


in [46], can help achieve differentiated multiple cov-erage in priority areas of a target region by attractingthe mobile sensors if the existing density of deploy-ment cannot support the application requirements.

VII.C. Fairness of Residual Energy Dis-tribution

Locomotion is the most expensive operation in termsof energy consumption in mobility based coverageprotocols. A sensor network deployed using mobilesensors should have a fair spatial energy distributionfor prolonged network lifetime. It is desirable that aprotocol should give maximum coverage while beingspatially fair in residual energy distribution. Existingmobility based protocols needs to be evaluated basedon these metrics.

VII.D. Efficient Detection of MovingJamming Holes

We discussed the jamming holes and one proposedprotocol, JAM, to detect and map the boundary of thejamming holes in Section V. The assumption in theJAM protocol that nodes located within the jamminghole can actually override the jamming signals is unre-alistic. Also, a moving jammer in the target field willleave the proposed protocol with incomplete informa-tion about the boundary of the hole. It would thus beinteresting to explore the scenario of a moving jam-ming hole changing its position in the wireless sensornetwork. The jamming hole detection protocol shouldnot only be able to report the existence of the jam-ming hole but also predict, in real time, the possibledirection of the movement. We identify the need of anefficient jamming hole detection and reporting proto-col capable of dealing with moving jamming holes.

VII.E. Resilience against DoS Attacks

Almost all of the proposed protocols for wireless sen-sor networks assume a trust relationship among nodes.Security considerations are minimal, if any. Some ap-plications i.e., Health monitoring and battlefield track-ing etc. require a higher degree of security and re-silience against various denial of service attacks. Weagree with authors of [5] that security should be abuilt-in feature of these proposed protocols rather thanan add-on. Existing protocols need to be improved ornew efficient protocols developed to achieve the de-sired degree of resilience against various denial of ser-vice attacks.

VIII. Conclusion

We have provided a snapshot of the current state of theart of research in sensor networks dealing with variousholes related problems. We discussed existing solu-tions and listed their behavior with holes such as cov-erage, routing, jamming, sink/black and worm holes.The survey has also revealed some possible future re-search directions. These include developing cover-age protocols based on realistic assumptions, utiliz-ing mobile sensors for achieving differentiated multi-ple coverage, finding an efficient solution for trackingmoving jamming holes and need for more secure andresilient protocols in wireless sensor networks.

References

[1] Ian F. Akyildiz, W. Su, Y. Sankarasubramaniam,and E. Cayirci. A survey on sensor networks.IEEE Communications Magazine, pages 102–114, 2002.

[2] Chee-Yee Chong and Srikanta P. Kumar. Sen-sor networks:Evolution, opportunities, and chal-lenges. In Proceedings of the IEEE, Volume 91No 8, pages 1247–1256, August 2003.

[3] Deborah Estrin, Ramesh Govindan, John Heide-mann, and Satish Kumar. Next century chal-lenges: Scalable coordination in sensor net-works. In IEEE/ACM MobiCom’99, pages 263–270, August 1999.

[4] Anthony D. Wood, John A. Stankovic, andSang H. Son. JAM: A jammed-area mappingservice for sensor networks. In 24th IEEE RealTime System Symposium (RTSS’03), pages 286–298, Dec 2003.

[5] Chris Karlof and David Wagner. Secure routingin wireless sensor networks: Attacks and coun-termeasures. In 1st IEEE International Work-shop SNPA’03, May 2003.

[6] Chi-Fu Huang and Yu-Chee Tseng. The cover-age problem in a wireless sensor network. InProceedings of the 2nd ACM WSNA’03, Sep2003.

[7] D. Nicules and B. Nath. Ad-hoc positioning sys-tem (APS) using AoA. In Proceedings of theIEEE INFOCOM, 2003.

[8] Jerry D. Gibson. Mobile Communications Hand-book. Springer-Verlag, 1999.

[9] Sachin Adlakha and Mani Srivastava. Criti-cal density thresholds for coverage in wireless


sensor networks. In Proceedings of the IEEEWCNC ’03, 3, pages 1615–1620, March 2003.

[10] Xiaorui Wang, Guoliang Xing, Yuanfang Zhang,Chenyang Lu, Robert Pless, and ChristopherGill. Integrated coverage and connectivity con-figuration in wireless sensor networks. In Pro-ceedings of the ACM SenSys ’03, pages 28–39,Nov 2003.

[11] Anthony D. Wood and John A. Stankovic. De-nial of service in sensor networks. IEEE Com-puter, 35(issue 10):48–56, Oct 2002.

[12] Yih-Chun Hu, Adrian Perrig, and David B. John-son. Wormhole detection in wireless adhocnetworks. Technical Report TR01-384, De-partment of Computer Science, Rice University,June 2002.

[13] Guiling Wang, Guohong Cao, and Tom La Porta.Movement-assisted sensor deployment. In IEEEINFOCOM 2004, June 2004.

[14] Saurabh Ganeriwal, Aman Kansal, and Mani B.Srivastava. Self aware actuation for fault repairin sensor networks. In Proceedings of the IEEEInternational Conference on Robotics and Au-tomation(ICRA), May 2004.

[15] Andrew Howard, Maja J Mataric, and Gaurav SSukhatme. Mobile sensor network deploymentusing potential fields:A distributed, scalable so-lution to the area coverage problem. In 6th Inter-national Symposium on Distributed AutonomousRobotics Systems (DARS02), June 2002.

[16] Andrew Howard, Maja J Mataric, and Gaurav SSukhatme. An incremental self-deploymentalgorithm for mobile sensor networks. Au-tonomous Robots,Special Issue on IntelligentEmbedded Systems, 13(2):113–126, Sep 2002.

[17] Nojeong Heo and Pramod K. Varshney. An intel-ligent deployment and clustering algorithm for adistributed mobile sensor network. In Proceed-ings of the IEEE International Conference onSystems, Man and Cybernetics, volume 5, pages4576–4581, Oct 2003.

[18] Nirupama Bulusu, John Heidemann, DeborahEstrin, and Tommy Tran. Self-configuring local-ization systems: Design and experimental eval-uation. ACM Transactions on Embedded Com-puting Systems, May 2003.

[19] Peter Corke, Steven Hrabar, Ronald Peterson,Daniela Rus, Srikanth Saripalli, and Gaurav

Sukhatme. Autonomous deployment and repairof a sensor network using an unmanned aerialvehicle. In Proceedings of the IEEE 2004 Inter-national Conference on Robotics and Automa-tion, Volume 4, pages 3602–3608, May 2004.

[20] Maxim A. Batalin and Gaurav S. Sukhtame.Coverage, exploration and deployment by amobile robot and communication network.Telecommunication Systems Journal, Special Is-sue on Wireless Sensor Networks, 26(2):181–196, 2004.

[21] Guiling Wang, Guohong Cao, and Tom La Porta.A bidding protocol for deploying mobile sen-sors. In 11th IEEE International Conferenceon Network Protocol ICNP ’03, pages 315–324,Nov 2003.

[22] T. Yan, T. He, and J. Stankovic. Differentiatedsurveillance for sensor networks. In Proceedingsof the ACM SenSys ’03, Nov 2003.

[23] Honghai Zhang and Jennifer C. Hou. Maintain-ing sensing coverage and connectivity in largesensor networks. Technical Report UIUDCS-R-2003-2351, Department of Computer Science,University of Illinois at Urbana Champaign,2003.

[24] Di Tian and Nicolas D. Georganas. A coverage-preserving node scheduling scheme for largewireless sensor networks. In Proceedings of thefirst ACM WSNA’02, Sep 2002.

[25] Jie Jiang and Wenhua Dou. A coverage-preserving density control algorithm for wirelesssensor networks. In ADHOC-NOW’04, pages42–55. Springer-Verlag, July 2004.

[26] B.Chen, K. Jamieson, H. Balakrshnan, andR. Morris. SPAN: An energy-efficient coordi-nation algorithm for topology maintenance in adhoc wireless networks. In Proceedings of theACM/IEEE MobiCom ’01, July 2001.

[27] Chalermek Intanagonwiwat, Ramesh Govindan,and Deborah Estrin. Directed diffusion: A scal-able and robust communication paradigm forsensor networks. In Proceedings of the Sixth An-nual ACM/IEEE MobiCom ’2000, pages 56–67,August 2000.

[28] Haiyun Luo, Fan Ye, Jerry Cheng, Songwu Lu,and Lixia Zhang. Two-tier data dissemina-tion model for large-scale wireless sensors net-works. In Proceedings ofACM/IEEE MobiCom’02, September 2002.


[29] Fan Ye, Gary Zhong, Songwu Lu, and LixiaZhang. Gradient broadcast: A robust data de-livery protocol for large scale sensor networks.In IPSN, April 2003.

[30] Brad Karp and H. T. Kung. GPSR: Greedyperimeter stateless routing for wireless net-works. In Proceedings of the ACM/IEEE Mo-biCom ’2000, pages 243–254, 2000.

[31] John Heidemann, Fabio Silva, and Deborah Es-trin. Matching data dissemination algorithms toapplication requirements. In Proceedings of theACM SenSys ’03, Nov 2003.

[32] David Braginsky and Deborah Estrin. Rumorrouting algorithm for sensor networks. In 1stACM WSNA ’02, Sep 2002.

[33] Di Tian and Nicolas D. Georganas. Energy ef-ficeint routing with guaranteed delivery in wire-less sensor networks. In IEEE WCNC ’03, Apr2003.

[34] Deepak Ganesan, Ramesh Govindan, ScottShenker, and Deborah Estrin. Highly-resilient,energy-efficient multipath routing in wirelesssensor networks. Mobile Computing and Com-munications Review, 2001.

[35] Stefan Dulman, Tim Nieberg, Jian Wu, and PaulHavinga. Trade-off between traffic overhead andreliability in multipath routing for wireless sen-sor networks. In IEEE WCNC ’03, Apr 2003.

[36] Rahul C. Shah and Jan M. Rabaey. Energy awarerouting for low energy ad hoc sensor networks.In IEEE WCNC ’02, 1, pages 350–355, March2002.

[37] G. Toussaint. The relative neighborhood graphof a finite planar set. Pattern Recognition 12,4:261–268, 1980.

[38] K.R. Gabriel and R. R. Sokal. A new statisticalapproach to geographic variation analysis. Sys-tematic Zoology, 18:259–278, 1969.

[39] Evangelos Kranakis, Harvinder Singh, and JorgeUrrutia. Compass routing on geometric net-works. In Proceedings of the 11th CanadianConference on Computational Geometry, pages51–54, 1999.

[40] P. Bose, P. Morin, I. Stojmenovic, and J. Urru-tia. Routing with guaranteed delivery in ad hocwireless networks. Wireless Networks, 7:609–616, 2001.

[41] Fabian Kuhn, Roger Wattenhofer, Yan Zhong,and Aaron Zollinger. Geometric ad-hoc routing:Of theory and practice. In 23rd ACM Symposiumon Principles of Distributed Computing (PODC’03), July 2003.

[42] Douglas S. J. De Couto and Robert Morris. Lo-cation proxies and intermediate node forwardingfor practical geographic forwarding. TechnicalReport MIT-LCS-TR-824, MIT Laboratory forComputer Science, June 2001.

[43] Qun Li and Daniela Rus. Sending messagesto mobile users in disconnected ad-hoc wirelessnetworks. In Proceedings of the ACM MobiCom’00, August 2000.

[44] Qing Fang, Jie Gao, and Leonidas J. Guibas. Lo-cating and bypassing routing holes in sensor net-works. In IEEE INFOCOM 2004, June 2004.

[45] Y. Yu, R. Govindan, and D. Estrin. Geographi-cal and energy aware routing: A recursive datadissemination protocol for wireless sensor net-works. Technical Report UCLA/CSD-TR-01-0023, UCLA Computer Science Department,May 2001.

[46] Nirupama Bulusu, John Heidemann, and Debo-rah Estrin. Adaptive beacon placement. In 21stInternational Conference on Distributed Com-puting Systems ICDCS, April 2001.


The Holes Problem in Wireless Sensor Networks: ASurvey · Sink/Black Holes/Worm Holes Sensor...

Documents

Transcript of The Holes Problem in Wireless Sensor Networks: ASurvey · Sink/Black Holes/Worm Holes Sensor...