A centrality-based topology control protocol for wireless mesh networks

1

3

4

5

6

7 Q1

8 Q2

9

1 1

1213141516

17181920212223

2 4

48

49

50

51

52

53

54

55

56

Q1

Ad Hoc Networks xxx (2014) xxx–xxx

ADHOC 1083 No. of Pages 30, Model 3G

8 August 2014

Q1

Contents lists available at ScienceDirect

Ad Hoc Networks

journal homepage: www.elsevier .com/locate /adhoc

A centrality-based topology control protocol for wireless meshnetworks

http://dx.doi.org/10.1016/j.adhoc.2014.07.0261570-8705/� 2014 Published by Elsevier B.V.

⇑ Corresponding author. Tel.: +34 93 401 7027.E-mail addresses: [email protected] (A. Vázquez-Rodas),

[email protected] (L.J. de la Cruz Llopis).

Please cite this article in press as: A. Vázquez-Rodas, L.J. de la Cruz Llopis, A centrality-based topology control protocol for wirelesnetworks, Ad Hoc Netw. (2014), http://dx.doi.org/10.1016/j.adhoc.2014.07.026

Andrés Vázquez-Rodas ⇑, Luis J. de la Cruz LlopisDepartment of Telematics Engineering, Universitat Politècnica de Catalunya (UPC), C. Jordi Girona 1-3, 08034 Barcelona, Spain

252627282930313233343536

a r t i c l e i n f o

Article history:Received 28 February 2014Received in revised form 30 June 2014Accepted 28 July 2014Available online xxxx

Keywords:Topology controlWireless mesh networksSocial network analysisCentrality metricsConnected dominating set

37383940414243444546

a b s t r a c t

Nodes in wireless multi-hop networks establish links with their neighbors, which are usedfor data transmission. In general, in this kind of networks every node has the possibility ofacting as a router, forwarding the received packets when they are not the final destinationof the carried data. Due to the routing protocol procedures, when the network is quitedense the overload added by the routing management packets can be very high. To reducethe effects of this overload a topology control mechanism can be used, which can be imple-mented using different techniques. One of these techniques consists of enabling ordisabling the routing functionality in every node. Many advantages result from selectingjust a subset of nodes for routing tasks: reduction of collisions, protocol overhead, interfer-ence and energy consumption, better network organization and scalability. In this paper, anew protocol for topology control in wireless mesh networks is proposed. The protocol isbased on the centrality metrics developed by social network analysts. Our target network isa wireless mesh network created by user hand-held devices. For this kind of networks, weaim to construct a connected dominating set that includes the most central nodes. Theresulting performance using the three most common centrality measures (degree, close-ness and betweenness) is evaluated. As we are working with dynamic and decentralizednetworks, a distributed implementation is also proposed and evaluated. Some simulationshave been carried out to analyze the benefits of the proposed mechanism when reactive orproactive routing protocols are used. The results confirm that the use of the topologycontrol contributes to a better network performance.

� 2014 Published by Elsevier B.V.

47

57

58

59

60

61

62

63

64

65

1. Introduction

Smart environments, smart devices, smart interaction,computing anytime and anywhere, . . ., the accelerateddevelopment of information technologies and mobiledevices results in people and/or ‘‘things’’ increasinglydependent on the online services offered through the Inter-net. In such scenario, Wireless Mesh Networks (WMNs)have evolved as a cost effective possible solution for user

66

67

68

69

uninterrupted access to networking facilities. Valued fea-tures like robustness, reliability, easy deployment andmaintenance, self-forming and self-configuration, makeWMNs an important alternative to achieve an always-onconnectivity. Among the three typical architectures ofWMNs [1], the present work focus on the client meshingone. In this case, the end-user devices are able to simulta-neously provide application interface, routing and networkconfiguration capabilities. Nowadays, the most commonhand-held device used by an increasing number of peopleis the smart phone. The evolution of such mobile deviceswith their variety of embedded sensors results now in anot only communication equipment, but a complete

s mesh

http://dx.doi.org/10.1016/j.adhoc.2014.07.026

mailto:[email protected]

mailto:[email protected]


http://www.sciencedirect.com/science/journal/15708705

http://www.elsevier.com/locate/adhoc


70

71

72

73

74

75

76

77

78

79

80

81

82

83

84

85

86

87

88

89

90

91

92

93

94

95

96

97

98

99

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

2 A. Vázquez-RodasQ1 , L.J. de la Cruz Llopis / Ad Hoc Networks xxx (2014) xxx–xxx


8 August 2014

Q1

sensing system [2]. Numerous applications are emerging inmany fields like health, traffic, human behavior, environ-ment monitoring, social networking and commerce [2].With this perspective it is entirely feasible to consider aWMN composed by mobile phone users moving around acity as a cost-effective complement to commercial cellularnetworks. One of the major concerns in this kind of devicesis related to their energy consumption. Therefore, optimi-zation techniques that aim to reduce it are always required[3].

Topology control techniques have been developed toimprove the energy efficiency and the battery lifetime ofa variety of networks. It also aims to reduce collisions,protocol overhead, and interference by means of a bettercontrol over the network connections and redundancy[4]. In general, there are three main types of topology con-trol approaches [4]. First, power control techniques [5,6],adjust the communication range of the wireless nodes bymeans of the transmission power of their transceivers. Thisway, nodes are able to better manage their neighborhoodsize, interference level, power consumption and connectiv-ity. Secondly, power mode mechanisms [7,8] control theactive or sleep operation modes of the nodes to dispensewith redundant stations and still achieve the desiredconnectivity. Finally, hierarchical clustering approaches[9–11] aim to construct an efficient virtual backbone fordata forwarding by the selection of a connected dominat-ing set (CDS). From graph theory, a CDS of a graph is a con-nected subset in which all the nodes that does not pertainat that subset have at least one adjacent neighbor insidethe subset. Due to the reduced number of nodes develop-ing routing task, the main advantages of this CDS-basedtopology control are: collisions, protocol overhead andenergy consumption reduction, efficient network organiza-tion and scalability improvement. In this work we evaluatean alternative method for this last category of topologycontrol based on social network analysis metrics.

In this context, thanks to the increasing availability ofnetwork maps which depicts the behavior of complex sys-tems and the universality of their characteristics [12], net-work science appears as the renewed study of the structureand the dynamic behavior of a variety of networked sys-tems [13]. Accordingly, social network analysts have devel-oped an important set of measures and metrics whichallow understanding the behavior and quantify the topol-ogy features of a diversity of networks [14]. Specifically,in this work we focus on centrality metrics developed toidentify the most important actors (nodes) in a networkby means of graph theory definitions and concepts.

In summary, the purpose of this work is to present andevaluate an alternative topology control mechanism basedon centrality measures borrowed from social networkanalysis. This topology optimization has been applied toa client wireless mesh network formed with user hand-held devices.

The rest of the paper is organized as follows. In Section2 we report and analyze the related work. Section 3 pro-vides a background on centrality metrics. Section 4 pre-sents and evaluates the proposed topology controlmechanism. A performance evaluation has been carriedout by means of simulations, taking into account both

Please cite this article in press as: A. Vázquez-Rodas, L.J. de la Cruz Llopnetworks, Ad Hoc Netw. (2014), http://dx.doi.org/10.1016/j.adhoc.2014

reactive and proactive routing protocols. The results arepresented in Section 5. Finally, the conclusions and futureworks are summarized in Section 6.

2. Related work

Nowadays, the application of complex networks tech-niques and social network analysis concepts to improvethe performance of wireless ad hoc networks is growingas a fertile research area [15]. Some recent works are sum-marized in the following. [16–18] apply the small worldphenomena, re-popularized by [19], to reduce the averagepath length of the network. The basic idea of these propos-als is to modify the physical topology of the network basedon the social features of the underlying graph. The smallworld property (or low average path length) could beachieved either by the aggregation of long-ranged links[17] or by a combination of rewiring, deletion and/or addi-tion of links/nodes [16]. Authors in [18] combine centralitymeasures with directional beamforming to create long-range links between more central nodes. The same authorsextend their study to sparse highly partitioned networks in[20].

SimBet [21] is a routing protocol designed for delay-tolerant MANETs. It uses two social network analysismetrics (centrality and similarity) for message forwardingdecisions. Betweenness centrality is selected to identifymore suitable bridge nodes, and the similarity measure isused to find nodes that are closer to the destination neigh-borhood. A utility function combines the similarity and thebetweenness utilities and allows adjusting the relativeimportance of them. For performance evaluation both util-ities has been assigned the same importance. For their part,authors in [22] apply social network analysis metrics todetect critical nodes in a WMN. They show how networkreliability substantially degrades when coordinated attacksare directed to highest centrality nodes. Simulations evincethat nodes with high betweenness centrality exhibit agreater impact than nodes with high degree or closenesscentrality. Authors also propose a socially-aware TDMAchannel access scheduling algorithm. The main idea is togive higher priority (assigning more time slots) to nodeswith high closeness centrality values. Simulations showimportant throughput improvements at the expense ofincreased delay.

The time-evolution of the topological characteristics ofvehicular networks from the perspective of graph theoryand social network analysis is the subject addressed in[23]. It is confirmed that relevant and useful informationabout the behavior of the VANETs could be inferred fromthe centrality metrics. The importance of nodes with highcentrality values on the design of more efficient VANETprotocols is also discussed.

A topology control algorithm for WSN based in edgebetweenness centrality [24] is proposed in [25]. This met-ric is used to identify most relevant edges or links betweennodes, regarding energy consumption. For each node, theaim of the proposal is to select a set of logical neighborsthat minimize energy consumption and fulfill QoS require-ments. Simulation results show better performance of thisproposal in comparison with traditional topology control

is, A centrality-based topology control protocol for wireless mesh.07.026


190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241242

244244

245

246

247

248

249

250

251

252

253

254255

257257

258

259

260

261

262

263

264

265

266

267268

270270

271

272

273

274

275

276

277

A. Vázquez-RodasQ1 , L.J. de la Cruz Llopis / Ad Hoc Networks xxx (2014) xxx–xxx 3


8 August 2014

Q1

methods in terms of number of logical neighbors, energyconsumption, latency and hit-ratio (percentage of servedqueries).

Authors in [26] propose a routing protocol based on aconnected dominating set (CDS). A simple marking processis used to establish the initial CDS: a node is designed asgateway if it has at least two unconnected neighbors. Thisinitial CDS is reduced by the application of rules based onthe node IDs. An extension of the selective removal rulesare presented in [9]. In this case the degree and the energylevel of the nodes are considered to reduce the CDS and toachieve balanced energy consumption. In [27] ConnectedDominating Sets with bounded diameters are taken intoaccount. To construct these DS with the smallest size,authors propose and evaluate two centralized algorithmsand one distributed version. On the other hand, given thefact that the transmission ranges of all network nodes arenot necessarily equal, authors in [28] model the networkby means of a Directed Graph and propose two differentsolutions for the case in which all the network links arebidirectional. A recent and extensive survey on energy-aware distributed topology control algorithms is presentedin [4].

Many topology control mechanisms are intended tosensor networks in which the dominant data flow tradi-tionally goes from the sensor nodes to the sink. In thiswork we focus on mesh networks in which it is more com-mon that every node may be origin or destination of thedata flow. On the other hand, we evaluate the use of cen-trality metrics which are calculated both in a centralizedor a distributed way. Besides, we propose and evaluate adistributed implementation of the router selection mecha-nism, as well as different possibilities to achieve total net-work connectivity.

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

299

300

301

302

3. Background on centrality

Centrality is one of the most useful mathematical mea-sures developed by social network analysts to capture thestructural properties of social relationships. It aims to iden-tify the most important actors/vertices within a graph thatrepresents any physical network. Centrality metrics couldbe based mainly on the degree of a vertex (number ofedges connected to it [14]) or on the geodesic distancesbetween them [15]. In the following we present a sum-mary of the three most useful centrality metrics.

Degree Centrality is the simplest centrality metric. It isdefined as the number of edges (links) attached to a vertex(node) [29]. In a WMN the degree centrality of a mesh sta-tion can be viewed as the number of one-hop neighborswith which it has been established a peer link. This central-ity is generally scaled by the number of nodes N in order tobe a measure independent of the network size. Then, thedegree centrality Dci for the node i is computed as:

Dci ¼PN

j¼1 xij

N � 1ð1Þ

where xij ¼ 1 if there is a link between node i and node jand xij ¼ 0 otherwise. In some networks, it could resultlogical to think that a node which has links to many others


(high degree centrality) has greater impact in the networkthan nodes with few links.

Closeness Centrality aims to identify nodes that spreadmessages in shorter time [29]. For that, this metric uses theconcept of geodesic path, which is the path with the short-est distance between two nodes. The closeness centralityCci for node i is computed as:

Cci ¼N � 1PN

j¼1 dij

ð2Þ

where i – j and dij is the geodesic distance between nodes iand j. In social networks, actors with high closeness cen-trality can communicate their ideas faster than actors withlower closeness centrality [14].

Betweenness Centrality measures the proportion ofshortest paths between any pair of nodes passing througha specific node [30]. The control over communications,connections and information flows could be dominatedby nodes with high betweenness centrality. The between-ness centrality Bci for node i is computed as:

Bci ¼XN

j¼1

Xj�1

k¼1

gjkðiÞ

gjkð3Þ

where i – j – k; gjk is the total number of geodesic pathsbetween node j and k, and gjkðiÞ is the number of geodesicpaths between node j and k that pass through node i.

4. Topology control mechanism and centralityevaluation

In this section we propose and evaluate a topology con-trol mechanism based on centrality metrics. We start witha description of the dynamic wireless mesh network sub-ject to analysis. Then, a centralized implementation andevaluation is presented in order to identify the most appro-priate centrality metric for our purpose. At this point, dif-ferent considerations are done for a practical distributedimplementation. Finally, the practical implementation ofthe protocol is presented.

4.1. Scenario under consideration

For the experimental evaluation we have considered adynamic wireless mesh network generated by the ns-3network simulator [31]. The mesh stations (nodes) arebased on the IEEE 802.11s amendment which is currentlyincorporated in the IEEE 802.11-2012 standard [32]. Thesimulation scenario consists of 100 mesh stations uni-formly distributed in a 1040 � 520 m area. According to[33] these values guarantee minimal quality criteria forstringent protocol evaluation. Specifically, the averageshortest-path hop count is greater than four hops (to avoidthat most of the data packets be interchanged just amongone or two-hop neighbors) and the average network parti-tioning is lesser than 5% (to avoid an excessive number ofisolated nodes). The nodes move according to a randomwalk 2D mobility model inside the rectangular bounds.Each node moves with a speed chosen randomly between2 and 4 m/s. The direction and speed of the nodes are



303

304

305

306

307

308

309

310

311

312

313

314

315

316

317

318

319

320

321

322

323

324

325

326

327

328



8 August 2014

Q1

updated after they have moved 100 m. These values areselected with the assumption that the mobile nodes aretransported by people moving around a segment of a city.The initial node positions and their trajectories are shownin Fig. 1. For the wireless channel, the log-distance propa-gation loss model has been considered.

According to [32], before the transmission of dataframes, mesh stations must create and maintain a logicaltopology using the mesh peering management protocol.Every mesh station discovers its mesh neighbors (peers)by means of beacon frames which are periodically sent.When a new neighbor has been discovered, the meshstation starts a peer link open handshake. It begins with

Fig. 1. Initial node positions

Fig. 2. Centrality metrics for on


a Peering Open management frame that contains meshconfiguration parameters. If the target station agrees withsuch parameters, it responds with a Peering Confirm frame.The same procedure is executed in the opposite directionto ensure bidirectional links. Only if the complete hand-shake procedure is successfully executed, a peer link isestablished between two mesh stations. Standard doesnot specify why or when a peer link must be closed. Thens-3 mesh networking implementation [34] triggers aclose peer link procedure when the number of successivelost beacons achieves a maximum configurable value. Bydefault this threshold is set to 5. If there is an active dataflow, the ns-3 mesh model also executes a close procedure

and their trajectories.

e of the WMN snapshots.



329

330

331

332

333

334

335

336



8 August 2014

Q1

when a station is unable to transmit to a peer a number ofsuccessive frames. This value is also set by default to 5.

337

338

339

340

341

342

4.2. Centralized implementation

Based on the presented scenario, first of all we haveanalyzed several snapshots of the mesh network datasettaken every 5 s. For each snapshot and for each node inside

Fig. 3. Resulting topology with the 40

5 10 15 20 25 300

0.1

0.2

0.3

0.4

0.5

0.6

0.7

t

Net

wor

k Fr

agm

enta

tion

Degree 40%RDegree 50%RCloseness 40%RCloseness 50%RBetweenness 40%RBetweenness 50%R

Fig. 4. Time evolution of the network fragmentation for the thre


the network we compute its degree, closeness andbetweenness centrality metrics. ORA [35] and UCINET[36] software have been used to centrality calculations,while network graphics are done with ORA. With the cen-trality values, we are able to rank and identify the mostcentral nodes from those three points of view. Fig. 2 showsthis ranking for one specific snapshot (at time = 5 s) and forthe three metrics. Blue nodes represent the most central

% most central nodes as routers.

35 40 45 50 55 60

[s]

e centrality metrics and for 40% and 50% selected routers.



343

344

345

346

347

348

349

350

351

352

353

354

355

356

357

358

359

360

361

362

363

364

365

366

367

368

369

370

371

372

373

375375

376

377

378

379

380

381

382

383

384

385

386

387

388

389

390

391

392

393

394

395

396

397

398

399

400

401

402

403

404

405

406

407

408

Table 1Summary of the resulting network fragmentation and the number of links per connected node values.

Centrality %R Network fragmentation Links per connected node

Mean SD Max Mean SD Max

Degree 40 0.547 0.090 0.639 7.405 0.839 8.63550 0.356 0.148 0.579 7.226 0.424 8.158

Closeness 40 0.481 0.053 0.580 6.511 0.371 7.20550 0.352 0.030 0.393 6.796 0.362 7.395

Betweenness 40 0.113 0.042 0.173 5.430 0.212 5.69150 0.051 0.021 0.079 6.123 0.299 6.573

Fig. 5. Egocentric network structure from the perspective of the node N18.



8 August 2014

Q1

stations and the red ones the least. As expected, Fig. 2confirms that different nodes are selected as most centralones for each different metric. This is because each metricevaluates a very different parameter and there is no rela-tionship among them.

As previously said, to reduce the complexity, communi-cation overhead and energy consumption of the network,we want to find a subset of mesh stations that forms a con-nected dominating set. So, instead of allowing each node toperform routing tasks, we choose only the most centralstations to do it. In a roughly first approach, we limit thenumber of routers to the forty (40%) most central nodesfor each network dataset and for each centrality metric.We virtually remove all the links between non-routerstations and keep all the other. If a non-router stationhas a link with more than one router, we keep all theselinks in order to preserve network resilience and to enablefuture load balancing and fair energy consumptionimprovements. Fig. 3 shows the resulting virtual topologyfor each centrality metric and for the same snapshot con-sidered in Fig. 2. Blue nodes represent the selected routers,the green are the mesh stations that have a link with atleast one router and the red represents the isolated nodes.

To determine which of the centrality metrics is the mostsuitable for our topology control application, we compare(for all the network datasets and for the three centralitymetrics) the resulting network fragmentation F, that is,the proportion of nodes in a network that are disconnectedfrom each other. It can be computed in an efficient waythrough the following equation [37]:

F ¼ 1�P

k sk sk � 1ð ÞN N � 1ð Þ ð4Þ

where sk is the number of nodes of the kth component ofthe graph that represents the network.

Fig. 4 shows the time-evolution of the network frag-mentation for the three centralities and for two differentpercentages of selected routers. As can be seen, the valuesfor the betweenness centrality are considerably lesser thanfor the other two centrality metrics. And as expected, forall the centralities the fragmentation decreases when agreater percentage of nodes are selected as routers.Besides, Table 1 summarizes the average, the standarddeviation and the maximum values for this measure. Ascan be observed, the mean and maximum fragmentationusing the betweenness centrality are much lesser thanwith the other two options. Furthermore, the between-


ness-based selection exhibits a more stable (confident)behavior since its standard deviation is the least amongthe three options. Another metric that we use to comparethe three different centralities is the average number oflinks which remain active after router selection perconnected node. Again, betweenness centrality exhibitsbetter performance since on average it requires lessernumber of links for each node that remains connected.The maximum and the standard deviation of this metricare also the least among the analyzed options. Other cen-trality metrics like hub, authority, and bridging have alsobeen evaluated and for all the cases the betweennesscentrality exhibits better outcomes. Based on this, we canconclude that by far the betweenness centrality is the met-ric that better identifies the subset of nodes that should beconsidered as routers.

4.3. Distributed implementation

In this work, we are dealing with a totally dynamicnetwork, in which the links among mesh stations continu-



409

410

411

412

413

414

415

416

417

418

419

420

421

422

423

424

425

426

427428

430430

431

432

433

434435

437437

438

439

440

441

442

443



8 August 2014

Q1

ously change due to their movements and channelconditions. This is done in an autonomous and distributedway through the mesh peering management protocol.Accordingly, our topology control algorithm must be imple-mented in a fully distributed manner. Mesh stations mustbe able to decide which nodes act as routers based onlyon their local information, and without the need of anycentral controller. In the previous section we determinedthat betweenness centrality is the best choice for our pur-pose. Therefore, we must use the egocentric version ofbetweenness centrality [38]. This version requires onlythe first order (one-hop) neighborhood information of eachnode instead of the complete network topology. Moreover,this information is available in each mesh station since theykeep an updated record of their peer links.

An efficient algorithm to compute the egocentricbetweenness is presented in [39]. Each node representsits ego network (one-hop neighbors) by means of the adja-cency matrix A. The elements of this matrix are given by:

Aij ¼1 if there is a peer link between nodes i and j

0 otherwise

�

ð5Þ

Fig. 6. Routers selection using 1-h

Fig. 7. Routers selection using 2-h


As an example, Fig. 5 shows the egocentric network struc-ture from the perspective of the node N18 at a given timeinstant.

The adjacency matrix for this ego network is given by:

N18 N16 N27 N44 N45 N61 N64 N73 N87

N18 0 1 1 1 1 1 1 1 1N16 1 0 0 0 1 0 0 0 1N27 1 0 0 1 0 1 1 0 0N44 1 0 1 0 0 1 1 0 0N45 1 1 0 0 0 0 0 1 1N61 1 0 1 1 0 0 1 0 0N64 1 0 1 1 0 1 0 0 0N73 1 0 0 0 1 0 0 0 1N87 1 1 0 0 1 0 0 1 0

Observe in Fig. 5 that all the shortest paths between non-adjacent stations that pass through the ego (the stationwho is calculating its egocentric betweenness) have alength of two hops. Therefore, when i – j, and being 1 thematrix with all its elements equal to 1, the expressionA2½1� A�i;j gives the number of shortest paths with length

op egocentric betweenness.

op egocentric betweenness.



444

445

446

447

448

449

Fig. 8. Routers selection using the sociocentric betweenness.

(a) Complete topology (b) Selected routers

Fig. 9. Connectivity achieved with the probe packets approach.

Fig. 10. Resulting backbone selecting two routers per neighborhood.



8 August 2014

Q1

2 that links stations i and j. Thus, the egocentric between-ness is the sum of the reciprocal of the resulting non-ceroelements [39].


Note that the peering management protocol guaranteesthat all the links are bidirectional, and then the matricesare symmetric and only the elements above the main diag-



Fram

eCo

ntro

l

Dur

a�on

Des

�n.

Add

ress

Sour

ceA

ddre

ss

BSS

Iden

�f.

Sequ

en.

Cont

rol

Fram

eCo

ntro

l

Dur

a�on

Des

�n.

Add

ress

Sour

ceA

ddre

ss

BSS

Iden

�f.

Sequ

en.

Cont

rol

Fram

eCo

ntro

l

Dur

a�on

Des

�n.

Add

ress

Sour

ceA

ddre

ss

BSS

Iden

�f.

Sequ

en.

Cont

rol

2 2 6 6 6 2 4

4

FCSRSF

Fram

eCo

ntro

l

Dur

a�on

Des

�n.

Add

ress

Sour

ceA

ddre

ss

BSS

Iden

�f.

Sequ

en.

Cont

rol

bytes2 2 6 6 6 2 6 6 6 4

FCS

Sour

ceA

ddre

ss

Nei

g. #

1A

ddre

ss

Nei

g. #

2A

ddre

ss …NAF

Fram

eCo

ntro

l

Dur

a�on

Des

�n.

Add

ress

Sour

ceA

ddre

ss

BSS

Iden

�f.

Sequ

en.

Cont

rol

2 2 6 6 6 2 6 4 4

FCS

Sour

ceA

ddre

ss

Egoc

ent.

Rem

ain.

Ener

gy

EAF

Fram

eCo

ntro

l

Dur

a�on

Des

�n.

Add

ress

Sour

ceA

ddre

ss

BSS

Iden

�f.

Sequ

en.

Cont

rol

2 2 6 6 6 2 4

4

FCSRSF

Fig. 11. Frames format.

05

1015

12

34

50

2

4

6

8

10

12

x 104

Neighbors

NT=100 nodes

Routers

Topo

logy

Con

trol R

ate

(bps

)

TUT = 5.12 s.

TUT = 2.56 s.

(a) 100-nodes mesh network0

510

15

12

34

50

0.5

1

1.5

2

2.5

x 105

Neighbors

NT=200 nodes

Routers

Topo

logy

Con

trol R

ate

(bps

)

TUT = 5.12 s.

TUT = 2.56 s.

(b) 200-nodes mesh network

Fig. 12. Topology control protocol bit rate for different NT and TUT values.

5 10 15 20 25 30 35 400

1

2

3

4

5

6x 10

5

Number of Data Flows

Rat

e of

Rou

ting

Man

agem

ent M

essa

ges

[bps

]

Top 40RAll R

Fig. 13. Rate of routing management messages as a function of the number of active flows.



8 August 2014

Please cite this article in press as: A. Vázquez-RodasQ1 , L.J. de la Cruz Llopis, A centrality-based topology control protocol for wireless meshnetworks, Ad Hoc Netw. (2014), http://dx.doi.org/10.1016/j.adhoc.2014.07.026


450

451452

454454

455

456

457

458

459

460

461

462

463

464

465

466

467

468



8 August 2014

Q1

onal must be taken into account. For the node N18 thisresults in:

N18 N16 N27 N44 N45 N61 N64 N73 N87

N18 � 0 0 0 0 0 0 0 0N16 � � 1 1 0 1 1 3 0N27 � � � 0 1 0 0 1 1N44 � � � � 1 0 0 1 1N45 � � � � � 1 1 0 0N61 � � � � � � 0 1 1N64 � � � � � � � 1 1N73 � � � � � � � � 0N87 � � � � � � � � �

469

470

5 10 15 200

1

2

3

4

5

6x 105

Number of

Tota

l dat

a fo

rwar

ding

s [p

kts]

Fig. 14. Total data forwarding as a funct

5 10 15 201

2

3

4

5

6

7

8 x 104

Number of

Succ

essf

ully

RX

pack

ets

[pkt

s]

Fig. 15. Total number of successf


Therefore, the egocentric betweenness for the node N18 isequal to 16.333. This way, whit just local information,every node is able to calculate its egocentric betweennessvalue. This value should be updated every time a stationestablishes a new peer link or it closes a previously estab-lished one. In a practical implementation, the nodes willrefresh this value at predetermined time intervals.

Once selected a distributed way to obtain the centralityvalues, we also need a distributed way to select the nodesthat will act as routers. Note that in this case there is not acentral control with all the information needed to selectthe nodes with higher centralities. Our proposal here isthat every node selects as router the node (or nodes) withhigher egocentric betweenness in its neighborhood. Notethat with this way of proceeding, the same node can beselected as the router for different neighborhoods. This

25 30 35 40

Data Flows

Top 40RAll R

ion of the number of active flows.

25 30 35 40

Data Flows

Top 40RAll R

ully received data packets.



471

472

473

474

475

476

477

478

479

480

481

482

483

484

485

486

487

488

489

490

491

492

493

494

495

496

497

498

499

500

501

502

503

504



8 August 2014

Q1

way, the total number of routers is reduced, and this num-ber is determined by the procedure itself. Of course, thissimple method guaranties that every station is connectedto a router, but it does not assure that all the routers areinterconnected between them. Therefore, some furtheractions will be needed. On the other hand, the main advan-tage is that it is a very simple and implementable method.

With this background, the procedure to select the meshstations that will develop routing tasks can be summarizedas follows: first, each station calculates and broadcasts itsegocentric betweenness centrality value, and second, inits respective ego network, each station selects the neigh-bor (or neighbors) with the highest betweenness to beused as routers. The details and practical implementationof the protocol are exposed in Section 4.4.

In the following we present the evaluation of this topol-ogy control algorithm for the network dataset previously

5 10 15 203

4

5

6

7

8

9

10

11

12

13

Number of

Tota

l For

war

ding

s pe

r RX

Pack

et

Fig. 16. Total data forwardings per

5 10 15 200.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

Number o

Net

wor

k Ef

ficie

ncy

Fig. 17. Network efficiency in terms of packet de


described. We analyze different alternatives to create aconnected dominating set in a distributed way. In firstplace, we consider that each station computes its egocen-tric betweenness with the first-order (one-hop) neighbor-hood information and selects as router just the one withthe highest egocentric betweenness. As expected due tothe distributed router selection, instead of a connectedbackbone, this procedure generates a set of isolatedclusters that have no common routers among them.Fig. 6 presents the worst case obtained in the simulations,that is, the time instant with the higher number of isolatedclusters. For this specific case, 25% of stations have beenselected as routers. Although this is the worst case, it isworth to mention that the 8.33% of the network snapshotsanalyzed construct a fully connected backbone with thismost simple procedure, and with 30% of stations selectedas routers on average.

25 30 35 40

Data Flows

Top 40RAll R

successfully received packet.

25 30 35 40

f Data Flows

Top 40RAll R

livery ratio as a function of the traffic load.



505

506

507

508

509

510

511

512

513

514

515

516

517

518

519

520

521

522

523

524

525

526

527

528

529

530

531

532

0 2 6 10 14 18 22 26 30 34 38 42 46 50 54 58 62 66 700

5

10

15

20

25

30

35

40

45

50

Energy Consumption [J]

(a) Histogram

No.

of n

odes

0 10 20 30 40 50 60 700

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1


(b) Cdf

Cum

mul

ativ

e pr

obab

ility

All RTop 40R

Top 40RAll R

Fig. 18. Energy consumption statistics for the 15-flows reactive mode case.



8 August 2014

Q1

To improve the connectivity among the selected rou-ters, in the following some different approaches are pro-posed and evaluated.

4.3.1. K-hops egocentric betweenness centralityThe first possibility is to consider a higher number of

hops when computing the egocentric centrality. Forinstance, we can assess the case in which the stations com-pute it taking into consideration the two-hop neighbor-hood information. Again, only the most central station isselected as router in every neighborhood. Fig. 7 shows asignificant connectivity improvement for the same net-work snapshot shown in Fig. 6. Nevertheless, there are stilltwo unconnected clusters. In this case, 31% of stations havebeen selected as routers.


The theoretical limit for this approximation is the casein which the stations compute the sociocentric instead ofthe egocentric betweenness centrality. Namely, eachstation needs to know the complete network topologyinformation, which is not a practical consideration fordynamic WMNs. Fig. 8 shows that this procedure createsa complete backbone with 34% of stations selected as rou-ters. However this total connectivity among all the selectedrouters is not achieved for all the network snapshots thathave been evaluated. Then, this procedure does not guar-antee the creation of a CDS.

4.3.2. Probe packetsIn the following, we describe another possible solution

which guaranties that all the selected routers create a



533

534

535

536

537

538

539

540

541

542

543

544

545

546

547

548

549

550

551

552

553

554

555

556

557

558

559

560

561

562

563

564

565

566

567

568

569

570

571

572

573

574

575

576

577

578

579

580

581

582

583

584

585

586

587

588

589

590



8 August 2014

Q1

single connected backbone at the expense of increasing thesignaling overhead. After a first router selection round,based on egocentric betweenness, the non-router stationsvirtually linked to at least two routers verify the connectiv-ity between them. If there is no path between any tworouters connected to a non-router station, that stationmust be marked as router. To this end, the station trans-mits a probe message through one (and only one) of itslinks. This special message must be retransmitted by therouters and, if connectivity exists, it will be received bythe sending station through the other links. If the messageis not received before a predetermined time, the stationmarks itself as router and communicates this decision toits neighborhood. To illustrate this, Fig. 9(a) shows a differ-ent network snapshot. The cyan bigger nodes represent thestations that have been selected as routers based on one-hop egocentric betweenness and a selection with just onerouter per station. As it can be seen, selected routers formstwo components (Fig. 9(b)) that can be linked througheither of stations N88; N33; N54; N52; N51; N91, N82 andN11. These nodes will be marked as routers according tothe proposed procedure.

4.3.3. Higher number of routers per neighborhoodWe have evaluated another procedure in which the sta-

tions, instead of selecting only the most central node, markas routers the two nodes with the higher one-hop egobetweenness values in each neighborhood. Fig. 10 showsthe resultant backbone for the same mesh network dataset.In this case, 43% of stations have been selected as routers.This process generates a fully connected backbone for the

0 2 4 6 8 100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Delay [s]

(a)

Cum

ulat

ive

prob

abilit

y

0 2 40

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Delay

(b

Cum

ulat

ive

prob

abilit

y

0 2 4 6 8 100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Delay [s]

(d)

Cum

ulat

ive

prob

abilit

y

0 2 40

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Delay

(e

Cum

ulat

ive

prob

abilit

y

Top40R-5FlowsAllR-5FlowsTop40R-10FlowsAllR-10FlowsTop40-15FlowsAllR-15Flows

Top40R-30FlowsAllR-30Flows

Fig. 19. Cumulative distribution function


83.33% of the network snapshots tested. The increase inthe number of redundant routers can be exploited by loadbalancing techniques, and it is also valuable from a net-work resilience perspective. Due to the fact that thisapproach is the one with a lower amount of overload, wehave chosen it to propose a practical implementation inthe next section. Obviously, incrementing the number ofrouters selected by each station will ensure a fully com-plete backbone for all the scenarios at the expense of ahigher router redundancy. As a future line of work we willstudy the minimum number of routers per neighborhoodneeded to guaranty a desired connectivity (expressed asa percentage of the total connectivity).

4.4. Practical implementation

Energy efficiency and fair energy consumption are con-sider for the practical implementation of the protocol. It iscarried out in three steps:

– Step 1: Every node broadcast its own neighbors list bymeans of a Neighborhood Advertising Frame (NAF). Thisis a special management frame where every neighbor isidentified by its MAC address.– Step 2: Every node receives the NAFs from its neigh-bors. With this information, they are able to computetheir own egocentric value. Once computed, theybroadcast that value in the Egocentric AdvertisingFrame (EAF). In order to allow fair energy consumptionin the network, the nodes remaining energy is alsoincluded in the frame.

6 8 10 [s]

)

0 2 4 6 8 100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Delay [s]

(c)

Cum

ulat

ive

prob

abilit

y

6 8 10 [s]

)

0 2 4 6 8 10Delay [s]

(f)

Cum

ulat

ive

prob

abilit

y





0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

of data packets end-to-end delay.



591

592

593

594

595

596

597

598

599

600

601

602

603

604

605

606

607

608

609

610

611

612

613

614

615

616

617

618

619

620

621

622

623

624

625

626



8 August 2014

Q1

– Step 3: Every node receives the egocentric value of allof its neighbors and selects the one with the highervalue as its router. If different options are available(i.e., different neighbors with the same centralityvalue), the node with the higher value of remainingenergy is chosen. A multi-metric protocol, assigning dif-ferent weights to the centrality and energy remainingvalues, is one of the targets of a future work. The deci-sion is communicated to the new router with the Rou-ter Selection Frame (RSF). When the new routerreceives the RSF, it switches to the router mode (or rou-ter state) and activates its routing functionality. As saidin the previous section, in order to guarantee obtaininga connected topology, a bigger number of routers can bechosen by every node.

Due to the network variability, this three steps proce-dure must be repeated every certain interval time, called

5 4 30

1

2

3

4

5

6

7x 105

Lifetime of reactive ro

Rat

e of

Rou

ting

Man

agem

ent M

essa

ges

[bps

]

Fig. 20. Rate of routing management messages

5 44

4.5

5

5.5

6

6.5

7 x 104

Lifetime of reactive r

Succ

essf

ully

RX

pack

ets

[pkt

s]

Fig. 21. Total number of successfully received pac


the Topology Updating Time (TUT). Besides, the routerstate must be seen as a soft state. That is, to keep the rout-ing functionality activated, a node must receive at least aRSF every TUT seconds. Otherwise the node understandsthat its routing functionality is not needed anymore andswitches it off.

Fig. 11 shows the specific frame format for the newthree management frames needed by this protocol. TheNAF and EAF frames can be included into the mesh-specificmanagement frames. They are broadcast frames and there-fore they do not need to be acknowledged by the receiver.The RSF frame is also a mesh-specific management framebut in this case it is a unicast frame and therefore it mustbe acknowledged by the receiver. The three frames canbe identified by the sub-type code in the Frame Controlfield.

Calling NT the total number of nodes in the networkand NR the number of routers that every node selects,

2 1

uting information [s]

Top 40RAll R

as a function of the reactive path lifetime.

3 2 1outing information [s]

Top 40RAll R

kets as a function of reactive path lifetime.



627

628629

631631

632

633

634

635

636

637

638

639

641641

642

643

644

645

647647

648

649

650

651

652

653

654

655

656



8 August 2014

Q1

the total number of bytes transmitted by this protocolprocedure is:

NT size NAFð Þ þ size EAFð Þ½þNR size RSFð Þ þ size ACKð Þ½ �� bytes ð6Þ

where sizeðXÞ stands for the size in bytes of the frame X.As it can be seen in Fig. 11, the size of the NAF depends

on the number of neighbors. Calling NN to the averagevalue of this number, the average size of these frames is34þ 6NN bytes. The size of EAF and RSF frames are 42and 28 bytes respectively. Substituting these values in Eq.(6) we obtain:

NT 34þ 6NN þ 42þ 42NRð Þ¼ NT 76þ 6NN þ 42NRð Þ bytes ð7Þ

5 4 33

4

5

6

7

8

9

10

Lifetime of reactive r

Tota

l For

war

ding

s pe

r RX

Pack

et

Fig. 22. Total data forwardings per successfully receive

5 40.2

0.22

0.24

0.26

0.28

0.3

0.32

0.34

Lifetime of reactive

Net

wor

k Ef

ficie

ncy

Fig. 23. Network efficiency as a func


This amount of bytes is sent by the protocol every TUTseconds. Therefore, the bit rate added by the topologycontrol protocol to the network, RTC , is:

657

RTC ¼8NT 76þ 6NN þ 42NRð Þ

TUTbps ð8Þ

Obviously, RTC grows linearly with NT ; NN; NR, andð1=TUTÞ. These dependencies are shown in Fig. 12 for sev-eral values of the involved parameters. It is important toremark that the lifetime of a network topology must begreater than or at least equal to the lifetime of a routingpath. In the figure two typical values have been considered(one of them, 5.12 s, is the standard default value). Theresulting protocol bit rate will affect the node energy con-sumption and the network load, and will be taken intoaccount later in the performance evaluation.

2 1

outing information [s]

Top 40RAll R

d packet as a function of reactive path lifetime.

3 2 1

routing information [s]

Top 40RAll R

tion of reactive path lifetime.



658

659

660

661

662

663

664

665

666

667

668

669

670

671

672

673

674

675

676

677

678

679

680

681

682

683

0 10 20 30 40 50 60 70 800

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1


(b) Cdf

Cum

mul

ativ

e pr

obab

ility

Top R pathTAll R pathT

2.5 7.5 12.5 17.5 22.5 27.5 32.5 37.5 42.5 47.5 52.5 57.5 62.5 67.5 72.5 77.50

5

10

15

20

25

30

35

40

45


(a) Histogram

No.

of n

odes

Top 40RAll R

Fig. 24. Energy consumption statistics for the nodes working with the reactive mode, 20 peer-to-peer data flows and reactive routing information lifetimeof 3 s.



8 August 2014

Q1

5. Performance evaluation

In this section, extensive simulations have been carriedout to evaluate the performance of WMNs with the pro-posed centrality based topology control mechanism. Forthe evaluation, we consider the following metrics andparameters:

– The rate of routing management messages.– The total number of forwardings per successfully

received packet.– The network efficiency in terms of packet delivery

ratio.– The energy consumed on a node by node basis.


– The end-to-end packet delay.– The transmission bit-rate savings taking into account

the overload added by the topology control protocol.

We take into account the two routing modes of thehybrid wireless mesh protocol (HWMP). In the reactiveon-demand mode, a source mesh station starts a pathdiscovery procedure only when it has a data packet to sendto any destination. This operation mode is more suitablefor peer to peer communication inside the mesh network.On the other hand, proactive mode is more appropriatewhen a single or a few mesh stations are configured asgateways for external communications and therefore theyare the targets of most of data connections. In this mode, a



684

685

686

687

688

689

690

691

692

693

694

695

696

697

698

699

700

701

702

703

704

705

706

707

708

709

710

711

712

713



8 August 2014

Q1

single or few mesh stations are configured to be path treeroots, and they periodically initiate path selectionprocedures.

All simulations are done with the ns-3 network simula-tor. To model the mesh network, we have used one of thenetwork snapshots described in Section 4.2. It consists of100 IEEE 802.11s-based mesh stations randomly distrib-uted in a 1040 � 520 m rectangular area. For all the exper-iments, we compare the performance of the mesh networkwhen all the station are configured as routers with the casethat only 40 most central nodes are selected as routers(from now they will be called the All R and Top 40R topol-ogies for convenience). All the simulations are 500 s longand we have carried out 100 statistically independent runsfor every experiment. The average of these runs with 95%confidence interval is shown for all the results. In the

0 1 2 3 4 50

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Dela

(a

Cum

ulat

ive

prob

abilit

y

0 1 2 3 4 50

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Dela

(b

Cum

ulat

ive

prob

abilit

y

Fig. 25. Cumulative distribution function of data packets e


following, we present and discuss the simulation resultsfor different mesh network conditions and configurations.

5.1. Reactive mode. Impact of traffic load

The aim of this part is to evaluate the impact of thetraffic load on the network performance when peer-to-peer communications are more relevant and then, meshstations are operating in the reactive mode. We vary thenumber of simultaneously active UDP data flows from 5to 40. The source and the destination nodes are randomlyselected for each independent run and for each flow. Allthe UDP flows are configured with an exponential distribu-tion for the interarrival time with mean value of 20 ms. Forthe packet length we use an exponential distribution withmean of 512 bytes that has been truncated to a minimum

6 7 8 9 10y [s]

)

6 7 8 9 10y [s]

)

Top40R-pathT=1sTop40R-pathT=2sAllR-pathT=1sAllR-pathT=2s

Top40R-pathT=3sTop40R-pathT=4sTop40R-pathT=5sAllR-pathT=3sAllR-pathT=4sAllR-pathT=5s

nd-to-end delay for different reactive path lifetimes.



714

715

716

717

718

719

720

721

722

723

724

725

726

727

728

729

730

731

732

733

734

735

736

737

738

739

740

741

742

743

744

745

746

747

748

749

750

751

752

753

754

755

756

757

758

759

760

761

762

763

764

765

766

767

768

769

770

771

772

773

774

775

776

777

778

779

780

781

782

783

784

785

786

787

788

789



8 August 2014

Q1

of 12 bytes (header size), and a maximum of 1500 bytes.Mesh stations are configured with the default value of5.12 s for the lifetime of reactive routing information.

First of all, Fig. 13 shows the evolution of the mean rateof routing management messages as a function of the num-ber of data flows. Evidently, since mesh stations are usingthe on-demand path selection mode, a greater number ofsources demand a greater number of path selection proce-dures. Therefore, the rate of routing management mes-sages increases when the number of data flows increases.As expected because of the lower number of mesh stationswith path selection functionality, a significant reductionoccurs (76.14% on average) when the mesh network oper-ates with the topology control mechanism. Observe thatthis reduction directly impacts on the total energy con-sumed by the network.

The next parameter that we consider is the total num-ber of unicast data packets forwarded by all the stationsin the network (Fig. 14)). We can clearly identify two dif-ferent trends of this parameter. For low traffic loads, upto 15 concurrent flows for the network without topologycontrol and up to 20 flows when topology control isapplied, the data forwardings growth linearly with thenumber of flows. Beyond these values the network startsto exhibit a saturated behavior and then the growthbecomes less significant. This is also verified when westudy the total number of successfully received packets(Fig. 15)). This figure confirms that the network is satu-rated beyond 15 data flows for the All R topology andbeyond 20 flows for the Top 40R case. Over these values,although there are more sources generating data packets,there is practically no increase in the number of success-fully received data packets. The topology control mecha-nism produces a savings of the total data forwardings of32.77% on the average. This is another important reductionof the total amount of energy consumed by the network. Itis important to remark that this is not in detriment of thenumber of received packets. In fact, when the topology

20.48/8.192 10.24/4.096 5.12/2.0480

0.5

1

1.5

2

2.5

3

3.5

4 x 106

Active path to root time

Rat

e of

Rou

ting

Man

agem

ent M

essa

ges

[bps

]

Fig. 26. Rate of proactive routing managemen


control is applied, the stations get on average 2.3% moredata packets than without topology control. Precisely, tobetter analyze this relation we compute the total numberof data forwardings per successfully received packet.Fig. 16 shows the evolution of this measure and it can beobserved that for all the cases the network with topologycontrol is more efficient. Namely, the number of forwar-dings is reduced a 33.43% on average. We also verify thatan increase of traffic sources causes more collisions andtherefore a greater number of retransmissions and forwar-dings are required to deliver each data packet. This grow-ing trend continues until the network reaches thesaturated state.

The next parameter that has been evaluated is the net-work efficiency in terms of packet delivery ratio. Fig. 17shows that the network with the topology control becomesmore efficient when the number of sources is equal orgreater than 20. This is because the impact of the signalingrequired for path selection is not so important when a fewnumber of stations are generating data traffic. Neverthe-less, in more practical situations, the routing overhead ismore relevant and then the advantage of reducing thenumber of stations involved in routing is evident. Withinthe interval of this experiment, although this is not ourmain objective, the average improvement of the networkefficiency is about 3.7% when the topology control isapplied.

This small improvement in the efficiency comestogether with a significant reduction of the global energyconsumption inside the network, which is in fact the maingoal of this proposal. To evaluate this, we consider the totalamount of energy dissipated by each node for all the trans-missions and receptions during the entire simulation time.The energy model presented in [40] has been consideredfor all the reported cases. Fig. 18(a) shows the histogramof the energy consumed by the nodes for the 15-activeflows case (which corresponds to a non-saturated situa-tion). As can be observed, the maximum energy dissipated

2.56/1.024 1.28/0.512 0.64/0.256

out / Root Interval [s]

Top 40RAll R

t messages for different update times.



790

791

792

793

794

795

796

797

798

799

800

801

802

803

804

805

806

807

808

809

810

811

812

813

814

815



8 August 2014

Q1

by a node has been reduced from 72 to 63.6 J when topol-ogy control is applied. The total energy consumed by thenetwork has been reduced around a 41% (from 2247.4 to1324.4 J). Besides, the cumulative distribution function ofthe energy consumption (Fig. 18(b)) shows also the betterbehavior achieved with the topology control. For instance,it can be observed that around 27% of nodes dissipate lessthan 10 J for the All R case, and that this figure is increasedto 65% for the Top 40R case.

Finally, Fig. 19 shows the cumulative distribution func-tion of data packets end-to-end delay. As expected, agreater number of data sources increases the contentionfor channel access and also the path selection messagesoverhead. This implies that path selection requests require

20.48/8.192 10.24/4.096 5.12/2.0485

10

15

20

25

30

35


Tota

l For

war

ding

s pe

r RX

Pac

ket

Fig. 27. Total data forwardings per successfully receiv

20.48/8.192 10.24/4.096 5.12/2.0480

0.05

0.1

0.15

0.2

0.25

0.3


Net

wor

k E

ffici

ency

Fig. 28. Network efficiency for diffe


more time to be resolved and then the end-to-end delay ofdata packets grows with the number of traffic sources. Forall the cases the delay whit the Top 40R is lower than withthe All R topology. The difference is more significant forlow traffic loads (Fig. 19 and is reduced when the networkis saturated (Fig. 19c–f).

5.2. Reactive mode. Impact of reactive routing informationlifetime

The aim of this set of experiments is to analyze howdynamic a network could be. Mesh stations continueworking in reactive mode and 20 simultaneous peer-to-peer data flows are established between randomly

2.56/1.024 1.28/0.512 0.64/0.256


Top 40RAll R

ed packet for different proactive path lifetimes.

2.56/1.024 1.28/0.512 0.64/0.256


Top 40RAll R

rent proactive path lifetimes.



816

817

818

819

820

821

822

823

824

825

826

827

828

829

830

831

832

833

834

835

836

837

838

839

840

841

842

843

844

845

846

847

0 100 200 300 400 500 600 7000

10

20

30

40

50

60

70


(a) Histogram

No.

of n

odes

0 100 200 300 400 500 600 700 8000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Energy Consumption [J](b) Cdf

Cum

mul

ativ

e pr

obab

ility

Top 40RAll R

Top 40RAll R

Fig. 29. Energy consumption statistics for the network working in the proactive mode with one root station. The root node is included and the proactiverouting information lifetime is set to the default value of 5.12 s.



8 August 2014

Q1

selected sources and destinations. A more dynamic andvariable network requires a more frequent activation ofthe mesh path selection mechanism in order to keepthe paths updated. As stated in the IEEE 802.11-2012standard [32], we can control the time for which meshstations consider the forwarding information to be validby means of the dot11MeshHWMPactivePathTimeout con-trol variable. Therefore, in this section we evaluate theperformance of the network when this lifetime variesfrom 5.12 to 1 s. The evolution of the mean rate of rout-ing management messages is shown in Fig. 20. Of coursewe can observe that, although the traffic load is equiva-lent for all the cases, the rate of path selection relatedmessages grows as lifetime of reactive routing informa-tion decreases. As expected, topology control producesan important reduction of around 83.43% because of


the reduced number of mesh stations involved in rout-ing. This directly corresponds to the same significantlyenergy consumption savings due to the lower amountof routing messages exchanged by the mesh stations.We verify that the limitation in the number of routershas no negative impact in the total number of success-fully received packets. In fact, as can be seen in Fig. 21,topology control produces an average increase of around8.75% in the number of successfully received packetswith regard to the All R configuration. The decreasingbehavior is explained as follows: lower duration ofestablished paths implies a greater number of path dis-covery messages which in turn increases the collisionsand the contention inside the network. Thus, data pack-ets have less transmissions opportunities and then thereare less successfully received packets.



848

849

850

851

852

853

854

855

856

857

858

859

860

861

862

863

864

865

866

867

868

869

870

871

872

873

874

875

876

877

878

879

0 20 40 60 80 100 120 140 1600

10

20

30

40

50

60

70


(a) Histogram

No.

of n

odes

0 20 40 60 80 100 120 140 1600

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1


(b) Cdf

Cum

mul

ativ

e pr

obab

ility

Top 40RAll R

Top 40RAll R

Fig. 30. Energy consumption statistics for the network working in the proactive mode with one root station. Only common mesh nodes are included andthe proactive routing information lifetime is set to the default value of 5.12 s.



8 August 2014

Q1

This behavior is also confirmed when we analyze thetotal number of data forwardings per successfully receivedpacket (Fig. 22)). We can observe that, although the trafficload remains equivalent for all the cases and even theincreasing number of collisions requires more data packetretransmissions, the high network contention caused bythe routing messages overhead reduces the transmissionopportunities of data packets and then the number of for-wardings decreases instead of remain constant or growdue to collisions. Nevertheless, it is important to remarkthat the topology control contributes with an averagereduction in the number of data forwardings of about42.19%. This represents again a considerably energyconsumption savings.

The network efficiency in terms of packet delivery ratioalso exhibits a decreasing tendency when the lifetime or


reactive routing information decreases (Fig. 23). Aspreviously said, this is due to the big amount of routingmanagement messages overhead required for a fasterupdate of the established paths. An important observationis that, for a given network efficiency value, the topologycontrol allows almost twice routing information updatescompared with the network operating with all routers.For example, the efficiency reached by the All R networkwith the default active path timeout value of 5.12 s couldbe achieved with the Top 40R topology even if the pathupdates are done approximately every 2.7 s. In summary,topology control also improves the scalability when moredynamism and variability is required.

With the same criteria of the previous section, Fig. 24shows the energy consumption statistics for the networkconfigured with a reactive path lifetime of 3 s. The



880

881

882

883

884

885

886

887

888

889

890

891

892

893

894

895

896

897

898

899

900

901

902

903

904

905

906

907

0 2 4 6 8 100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Delay [s]

(b)C

umul

ativ

e pr

obab

ility

0 2 4 6 8 100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Delay [s]

(c)

Cum

ulat

ive

prob

abilit

y

0 2 4 6 8 100

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Delay [s]

(a)

Cum

ulat

ive

prob

abilit

y

Top40R-ActivePath=20.48s-Interv=8.192sAllR-ActivePath=20.48s-Interv=8.192s

AllR-ActivePath=10.24s-Interv=4.096sTop40R-ActivePath=10.24s-Interv=4.096s

Top40R-ActivePath=5.12s-Interv=2.048sAllR-ActivePath=5.12s-Interv=2.048s

Fig. 31. Cumulative distribution function of data packets end-to-end delay for different proactive path lifetimes.



8 August 2014

Q1

histogram (Fig. 24(a)) shows that the maximum energydissipated by a node in the network reduces from 79.3 to71.3 J when topology control is applied. The total energydissipated in the network reduces around a 38.44% (from2655.8 to 1635 J). The cdf (Fig. 24(b)) shows that whileonly 23% of nodes consumes less than 10 J in the All R con-figuration, this percentage increases to 67 in the Top 40Rnetwork.

The data packets end-to-end delay analysis (Fig. 25))reveals that the delay decreases when the path updatesare more frequents. This is due to the fact that the networkreacts earlier to link failures and to paths that should nolonger be valid for data forwarding. But it is also becausethere is a lesser number of data packets forwarded by


intermediate stations and a lesser number of successfullyreceived packets. When the level of routing overheadincreases, most of the packets that finally arrives to thedestination are those that traverse a few number of inter-mediate nodes. Therefore, those packets arrives in shortertime. On the other hand, it is important to remark thatthe application of topology control improves the perfor-mance of the network regarding end-to-end delay for allthe simulation cases. Even the worst case results for thenetwork with topology control (Lifetime = 5.12 s) are bet-ter than the results for network with all routers and life-time greater than or equal to 2 s. The reason for this isthat the paths are established faster when a lesser numberof stations are involved in the path discovery procedure.



908

909

910

911

912

913

914

915

916

917

918

919

920

921

922

923

924

925

926

927

928

1 2 30

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

x 106

Number of Roots

(a)

Rat

e of

Rou

ting

Man

agem

ent M

essa

ges

[bps

]

1 2 30.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

0.55

0.6

Number of Rooots

(b)

Net

wor

k Ef

ficie

ncy

Top 40RAll R

All RTop 40R

Fig. 32. Proactive mode, variable number of root stations and constant traffic load: (a) rate of routing management messages and (b) network efficiency.



8 August 2014

Q1

5.3. Proactive mode. Impact of proactive routing informationlifetime

Proactive mode of HWMP protocol is more suitablewhen the target of data communications is mainly outsideof the mesh basic service set. The most common use casesare mesh stations accessing to Internet or the internet-working with other network technologies. In these cases,one or more mesh stations are configured to be the rootmesh station(s) and they play the role of gateways. Eachroot periodically sends path request (PREQ) messages in


order to build a proactive tree, and thus, create paths fromall the mesh stations to the root.

As in the previous section, we begin evaluating how fastproactive data forwarding paths could be updated. A fasterroute update is required by more dynamic networks inwhich stations continuously vary their established peerlinks. Obviously, this implies a greater path selectionmanagement messages overhead. As stated in the IEEE802.11-2012 standard [32], the lifetime of the proactiverouting information is configured by the dot11Mesh-HWMPactivePathToRootTimeout control variable. This



929

930

931

932

933

934

935

936

937

938

939

940

941

942

943

944

945

946

947

948

949

950

951

952

953

954

1GW - 25Flows 2GW - 50Flows 3GW -75Flows0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1.8

2

x 106

Num. Roots - Num. Flows

(a)

Rat

e of

Rou

ting

Man

agem

ent M

essa

ges

[bps

]

1GW - 25Flows 2GW - 50Flows 3GW - 75Flows

0.16

0.18

0.2

0.22

0.24

0.26

0.28

Num. Roots - Num. Flows

(b)

Net

wor

k Ef

ficie

ncy

All RTop 40R

Top 40RAll R

Fig. 33. Proactive mode, variable number of root stations and traffic load: (a) rate of routing management messages and (b) network efficiency.



8 August 2014

Q1

variable should be greater than dot11MeshHWMProotInter-val which in turn controls the frequency of proactive PREQframes sent by the root. For the evaluation we configureone station as the root and then we vary the values of thesepair of variables from 20.48 to 0.64 s for the active path toroot timeout and from 8.192 to 0.256 s for the root inter-val. As it can be seen, we maintain constant the defaultratio between these variables in all the reported cases.For each independent run, the traffic load consists of 25simultaneously active data flows, five uploads fromrandomly selected mesh stations to the root and 20 down-loads from the root to randomly selected stations. Theexpected growth of routing management messages rate


as the lifetime of routing information decreases is verifiedin Fig. 26. We also confirm that proactive routing overheadis greater than the reactive one. For instance, when thedefault routing lifetime values are used and for the 25flows case, the routing management rate increases fromaround 432 Kbps (Fig. 13)) to 1.1 Mbps for the All R config-uration, and from around 97 Kbps to 190 Kbps for thenetwork with topology control. The application of topologycontrol produces an average reduction of around 78.87% inthe routing management messages overhead.

The increase of routing overhead due to faster pathupdates causes more collisions and then data packets needto be retransmitted a greater number of times. This is



955

956

957

958

959

960

961

962

963

964

965

966

967

968

969

970

971

972

973

974

975

976

977

978

0 100 200 300 400 500 600 700 800 9000

10

20

30

40

50

60


(a) Histogram

No.

of n

odes

0 100 200 300 400 500 600 700 800 9000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1


(b) Cdf

Cum

mul

ativ

e pr

obab

ility

Top 40RAll R

Top 40RAll R

Fig. 34. Energy consumption statistics for the network working in the proactive mode with three root stations and 75 active flows. The root nodes areincluded and the proactive routing information lifetime is set to the default value of 5.12 s.



8 August 2014

Q1

verified in Fig. 27 which shows the evolution of total dataforwardings per successfully received packet as a functionof routing information lifetime. It can be seen that,although the traffic load remains equivalent for all thecases, the average number of forwandings required to suc-cessfully transmit each data packet increases as the life-time decreases. The average reduction achieved with theapplication of topology control for this parameter isaround 51.75%.

The growing routing overhead makes increasingly diffi-cult to successfully transmit data packets and then the net-work efficiency gradually reduces (Fig. 28). As previously


said, although this is not the main objective of the topologycontrol mechanism, the network efficiency significantlyimproves when only the most central stations are involvedin the path selection algorithm. The average improvementachieved in the presented range is about 58.17%. Thisimprovement is much more significant than in the reactivemode case in which the path selection overhead level islower. Fig. 28 also shows that, for a given network effi-ciency, the mesh network with topology control supportsalmost four times faster path selection updates.

The energy consumption statistics for all the nodes inthe network, including the root mesh station configured



979

980

981

982

983

984

985

986

987

988

989

990

991

992

993

994

995

996

997

998

0 50 100 150 200 2500

10

20

30

40

50

60


(b) Histogram

No.

of n

odes

0 50 100 150 200 250 3000

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1


(b) Cdf

Cum

mul

ativ

e pr

obab

ility

Top 40RAll R

Top 40RAll R

Fig. 35. Energy consumption statistics for the network working in the proactive mode with three roots and 75 active flows. Only the non-root stations areincluded and the proactive routing information lifetime is set to the default value of 5.12 s.



8 August 2014

Q1

with the default values for the routing information lifetime(i.e. the active path to root is 5.12 s and the root interval is2.048 s) are shown in Fig. 29. As expected, the station con-figured as root dissipates significantly more energy thannon-root stations since all the traffic (data and routing)starts and ends in that node. For the root station, theenergy dissipation reduction achieved with the topologycontrol is around 14.79% (from 775.1 to 660.5 J). The totalenergy dissipated in the network reduces around 30.4%(from 3233.1 to 2250.2 J). These higher energy dissipation


values also confirm the greater overhead of the proactivemode regarding to the reactive one. The cdf of Fig. 29(b)shows that 52% of nodes dissipates less than 20 J for theAll R configuration and this value increases to 73% in theTop 40R network. On the other hand, since the root is aspecial type of station, Fig. 30 shows the energy consump-tion statistics for only non-root stations. The histogramand the cdf in Fig. 30 identifies a drawback of the proposedtopology control mechanism: three nodes selected asrouters are overloaded regarding the All R configuration.



999

1000

1001

1002

1003

1004

1005

1006

1007

1008

1009

1010

1011

1012

1013

1014

1015

1016

1017

1018

1019

1020

1021

1022

1023

1024

1025

1026

1027

1028

1029

1030

1031

1032

1033

1034

1035

1036

1037

1038

1039

1040

1041

1042

1043

1044

1045

1046

1047

1048

1049

1050

1051

1052

1053

1054

1055

1056

1057

1058

1059

1060

1061

1062

1063

1064

1065

1066

1067

1068

1069

1070

1071

1072

1073

1074



8 August 2014

Q1

This fact also confirms that the remaining energy of nodesmust be included in the router selection process as statedin Section 4.4.

The end-to-end delay cdf (Fig. 31) also confirms thateach successfully received data packet had to be retrans-mitted a greater number of times. It can be observed thatfaster proactive path updates increases the data packetdelays. This behavior holds for both network configura-tions, with or without topology control. Nevertheless,delays are smaller when topology control is applied dueto the lower channel access contention.

5.4. Proactive mode with more than one root station

A way to improve the network efficiency is to distributethe root or gateway functionality to a greater number ofmesh stations. This section evaluates the effect of addingone or two roots to the previous configuration. For thereported simulations, the proactive path lifetime remainsconstant in its default values (activePathToRootTime-out = 5.12 s and rootInterval = 2.048 s). In the first set ofexperiments, the traffic load also remains constant (foreach independent run there are 25 continuously activedata flows: five uploads from randomly selected mesh sta-tions to the roots and 20 downloads from the roots to ran-domly selected stations). Fig. 32(a) shows the rate ofrouting management messages as a function of the numberof root stations. Obviously, since proactive mode is used,this rate increases when there are a greater number ofroots independently of the traffic load. It is important toobserve that the growth is not linear because of, while eachroot sends periodic PREQs, the non-root stations justresponse with a PREP to the root with the best path metric.The average path selection overhead reduction achievedwith the application of the topology control in this caseis around 69.14%.

The evident improvement of the network efficiency interms of packet delivery ratio due to the distribution ofthe load in a greater number of roots can be verified in

5 10 15 200

200

400

600

800

1000

1200

1400

1600

Number of

Tota

l Sav

ing

[Kbp

s]

Fig. 36. Global bit-rate savings a


Fig. 32(b). On the other hand, the reduction of path selec-tion overhead decreases the collisions probability and thusa greater number of data packets successfully arrives to thetarget stations when topology control is applied. The aver-age value of the improvement due to topology control isaround 19.71% for the reported experiments. The improve-ment is more significant when the level of congestionaround the root(s) is greater (i.e. the one root case).

Finally, we evaluate the situation in which the numberof roots and the traffic load increases at the same time.Namely, for the single root network data traffic consist of25 flows, for the network with two roots there are 50 flows,and for three roots there are 75 active flows. For all thecases, downloads represents the 80% traffic and uploadsthe remaining 20%. The results for this set of simulationsare shown in Fig. 33. As expected, since proactive modeis used, the rate of routing management messages(Fig. 33(a)) is pretty similar to the constant traffic results.There is a slight difference, a bit more noticeable for thenetwork with tree roots and all stations as routers (around145 Kbps), due to the higher level of channel contentionproduced by the 75 data flows. This higher contentionreduces the transmission opportunities of routing mes-sages and therefore its rate is lower than in the 25 flowscase. This does not happen when the topology control isapplied since the path selection overhead has beenreduced a 68.2% on average.

Although the traffic load is increased, the distribution ofroot functionality to a greater number of roots stillimproves the network efficiency (Fig. 33(b)) of the networkconfigured with all stations as routers. The reduction of thenumber of mesh stations involved in routing implies theavailability of fewer alternative paths, and causes adecrease by 2% of the network efficiency with topologycontrol and two roots with respect to the one root case.This does not hold for the three roots case, which achievesaround a 6% of improvement in comparison with the oneroot network because of a better distribution of the trafficload. Nevertheless, for all the cases, the efficiency of the

25 30 35 40

Data Flows

s a function of traffic load.



1075

1076

1077

1078

1079

1080

1081

1082

1083

1084

1085

1086

1087

1088

1089

1090

1091

1092

1093

1094

1095

1096

1097

1098

1099

1100

1101

1102

1103

1104

1105

1106

1107

1108



8 August 2014

Q1

network with topology control is better than the all routersnetwork even though consuming less energy. The averageimprovement is about 19.54%.

The energy consumption statistics for the three rootsand 75 active flows configuration are shown in Fig. 34 forall the nodes in the network and in Fig. 35 for only non-root stations. Again, the three root stations can be easilyidentified. Although the global energy dissipation isreduced around a 10.25% (from 5782.9 to 5190.3 J) withthe topology control, we can recognize, as in the previoussection, four overloaded non-root stations regarding theAll R network. However, the nodes that dissipate less than20 J increases from 30% to 64% in the Top 40R case.

5.5. Total bit rate savings for the distributed implementation

This section includes a first approximation to the eval-uation of the proposed distributed implementation. We

5 40

200

400

600

800

1000

1200

1400

1600

Lifetime of reactive

Tota

l Sav

ing

[Kbp

s]

Fig. 37. Global bit-rate savings as a fu

20.48/8.192 10.24/4.096 5.12/2.0481200

1400

1600

1800

2000

2200

2400

2600

2800


Tota

l Sav

ing

[Kbp

s]

Fig. 38. Global bit-rate savings as a fun


analyze the total bit-rate savings achieved with the appli-cation of the topology control mechanism. The computa-tions include the reduction of data forwardings, thereduction of path selection related messages and also thecost of the router selection protocol (Eq. (8)). For all thereported cases, in order to compute the bit rate added bythe topology control we make the following consider-ations: NT ¼ 100 nodes, NN ¼ 8 (i.e. NN equal to the averagedegree of all the nodes in the network), NR ¼ 2 whichresults on average in around 40% of nodes selected as rou-ters, and the periodicity (TUT) of the required topologycontrol related messages is the same that the routing infor-mation lifetime. Remember from Section 4.4 that TUT Pdot11MeshHWMPactivePathTimeout.

For the first set of experiments, reported in Section5.1, Fig. 36 shows the total bit-rate savings achievedwith the topology control application as a function ofthe number of data flows. In this case TUT = 5.12 s and

3 2 1

routing information [s]

nction of reactive path lifetime.

2.56/1.024 1.28/0.512 0.64/0.256


ction of proactive path lifetimes.



1109

1110

1111

1112

1113

1114

1115

1116

1117

1118

1119

1120

1121

1122

1123

1124

1125

1126

1127

1128

1129

1130

1131

1132

1133

1134

1135

1136

1137

1138

1139

1140

1141

1142

1143

1144

1145

1146

1147

1148

1149

1150

1151

1152

1153

1154

1155

1156

1157

1158

1159

1160

1161

1162

1163

1164

1165

1166

1167

1168

1169

1170

1171

1172

1173

1174

1175

1176

1177

1178

1179

1180

1181

1182

1183

1184

1185

1186

1187

1188

1189

1190

1191

1192

1193

1194

1195

1196

1197

1198

119912001201120212031204120512061207120812091210121112121213121412151216121712181219122012211222122312241225122612271228122912301231



8 August 2014

Q1

remains constant for all the different traffic loads. Anaverage savings of 1.24 Mbps is reached with thetopology control and the previously mentioned networksaturation is also verified in this figure. It is importantto remark that the transmission/reception energy savingsof the mesh stations is directly proportional to thisreported quantity.

On the other hand, Fig. 37 shows the evolution of thisglobal traffic reduction as a function of the lifetime of thereactive routing information (experiments reported in Sec-tion 5.2). As previously justified, in these experiments theperiodicity of topology control updates is again the samethan the path lifetime. The average reduction for thereported cases is around 1.46 Mbps.

Finally, the global bit-rate reduction for the proactiveone-root case (experiments reported in Section 5.3) isshown in Fig. 38. The average value of the savings is around1.86 Mbps and as it was expected this is greater than in thereactive mode configuration.

6. Conclusions

In this paper we have evaluated the feasibility of usingthe centrality metrics from social network analysis to cre-ate a topology control mechanism based on a connecteddominating set. Specifically, for a dynamic wireless meshnetwork formed by user hand-held devices, we have pro-posed and evaluated some mechanisms to select whichof those devices must act as routers, forwarding the pack-ets received from other hand-held to their destination. Themechanism implementation has been proposed in twoways, centralized and distributed. For the centralizedapproach, we have evaluated the three most common cen-trality measures: degree, closeness and betweenness cen-trality. In all those cases, the more central nodes havebeen selected to form the backbone. The experimentalanalysis has shown that the network fragmentation usingthe betweenness centrality is considerably lesser thanthe resultant with degree or closeness centralities. Thus,it is possible to conclude that by far the betweenness cen-trality is the metric that better identifies the stations thatshould be considered for routing tasks. In the second place,taking into consideration that a centralized implementa-tion could be unfeasible in most scenarios, a distributedimplementation has been proposed. In this case only thebetweenness centrality has been considered. To distributethe responsibility, the concept of the egocentric networkperspective of each station is needed. That is, every stationcomputes its own centrality knowing only its own estab-lished links (neighbors). Besides, due to the dynamic andtime evolving nature of the network with which we aredealing, we also discuss and evaluate different distributedmechanisms to select the central stations that will act asrouters. First of all, every station selects the node in itsneighborhood with the highest centrality. This mechanismworks well in most cases if a minimum density of nodesexists. Then again, the complete connectivity is notassured. To guarantee that connectivity, some approacheshave been proposed and compared in terms of overloadand effectiveness. As a future line of work, the minimumnumber of routers per neighborhood needed to guaranty


a desired connectivity in the distributed implementationwill be studied.

The effectiveness of this centrality-aware topology con-trol protocol has been verified through extensive simula-tions. The evaluation has been carried out for bothreactive and proactive path selection modes. For all thestudied cases the efficiency of the network was improvedor at least remain equal but always with a considerablylower amount of transmission/reception energy consump-tion. Another common factor in all the simulation results isthe reduction of the packet end-to-end delays. This is adirect consequence of the global bit-rate savings whichimplies lower channel contention. The reduction of thenumber of mesh stations involved in the path selectionprocedure also improves the scalability when the variabil-ity of the network demands faster path updates. Finally, weconfirm that greater energy savings are obtained in theproactive path selection case because of the greater over-head present with this operation mode.

As current and future lines of work we are carrying outperformance evaluations considering human mobilitymodels. Besides, we are designing and evaluating amulti-metric protocol which assigns different weights tothe egocentric betweenness centrality and the noderemaining energy values.

Acknowledgments

This work was supported by the Spanish ResearchCouncil under project COPPI (TEC2011-26491), and bythe Ecuadorian SENESCYT (2010) and Salesian PolytechnicUniversity grant programs.

References

[1] I.F. Akyildiz, X. Wang, W. Wang, Wireless mesh networks: a survey,Comput. Netw. 47 (4) (2005) 445–487. http://dx.doi.org/10.1016/j.comnet.2004.12.001. <http://www.sciencedirect.com/science/article/pii/S1389128604003457>.

[2] W. Khan, Y. Xiang, M. Aalsalem, Q. Arshad, Mobile phone sensingsystems: a survey, Commun. Surv. Tutorials, IEEE 15 (1) (2013) 402–427, http://dx.doi.org/10.1109/SURV.2012.031412.00077.

[3] Y. Jiang, A. Jaiantilal, X. Pan, M. Al-Mutawa, S. Mishra, L. Shi,Personalized energy consumption modeling on smartphones, in: D.Uhler, K. Mehta, J. Wong (Eds.), Mobile Computing, Applications, andServices, Lecture Notes of the Institute for Computer Sciences, SocialInformatics and Telecommunications Engineering, vol. 110, Springer,Berlin Heidelberg, 2013, pp. 343–354. http://dx.doi.org/10.1007/978-3-642-36632-1_20.

[4] A. Aziz, Y. Sekercioglu, P. Fitzpatrick, M. Ivanovich, A survey ondistributed topology control techniques for extending the lifetime ofbattery powered wireless sensor networks, Commun. Surv. Tutorials,IEEE 15 (1) (2013) 121–144, http://dx.doi.org/10.1109/SURV.2012.031612.00124.

[5] S. Narayanaswamy, V. Kawadia, R.S. Sreenivas, P.R. Kumar, Powercontrol in ad-hoc networks: theory, architecture, algorithm andimplementation of the COMPOW protocol, in: European WirelessConference, 2002, pp. 156–162.

[6] N. Li, J. Hou, L. Sha, Design and analysis of an MST-based topologycontrol algorithm, IEEE Trans. Wirel. Commun. 4 (3) (2005) 1195–1206, http://dx.doi.org/10.1109/TWC.2005.846971.

[7] Y. Xu, J. Heidemann, D. Estrin, Geography-informed energyconservation for ad hoc routing, in: Proceedings of the 7th AnnualInternational Conference on Mobile Computing and Networking,MobiCom ’01, ACM, New York, NY, USA, 2001, pp. 70–84. http://dx.doi.org/10.1145/381677.381685.

[8] F. Ye, G. Zhong, J. Cheng, S. Lu, L. Zhang, Peas: a robust energyconserving protocol for long-lived sensor networks, in: Proceedings.


http://dx.doi.org/10.1016/j.comnet.2004.12.001

http://dx.doi.org/10.1016/j.comnet.2004.12.001

http://www.sciencedirect.com/science/article/pii/S1389128604003457


http://dx.doi.org/10.1109/SURV.2012.031412.00077

http://dx.doi.org/10.1007/978-3-642-36632-1_20

http://dx.doi.org/10.1007/978-3-642-36632-1_20



http://dx.doi.org/10.1109/TWC.2005.846971

http://dx.doi.org/10.1145/381677.381685

http://dx.doi.org/10.1145/381677.381685


12321233123412351236123712381239124012411242124312441245124612471248124912501251125212531254125512561257125812591260126112621263126412651266126712681269127012711272 Q312731274127512761277127812791280128112821283128412851286128712881289129012911292129312941295129612971298129913001301130213031304130513061307130813091310

13111312131313141315131613171318131913201321132213231324132513261327132813291330133113321333133413351336Q4133713381339134013411342134313441345134613471348134913501351135213531354

1355

13571357

13581359136013611362136313641365136613671368136913701371

13731373

13741375137613771378137913801381138213831384



8 August 2014

Q1

23rd International Conference on Distributed Computing Systems,2003, pp. 28–37. http://dx.doi.org/10.1109/ICDCS.2003.1203449.

[9] J. Wu, F. Dai, M. Gao, I. Stojmenovic, On calculating power-awareconnected dominating sets for efficient routing in ad hoc wirelessnetworks, J. Commun. Netw. 4 (1) (2002) 59–70, http://dx.doi.org/10.1109/JCN.2002.6596934.

[10] R. Misra, C. Mandal, Minimum connected dominating set using acollaborative cover heuristic for ad hoc sensor networks, IEEE Trans.Parall. Distrib. Syst. 21 (3) (2010) 292–302, http://dx.doi.org/10.1109/TPDS.2009.78.

[11] Z. Yuanyuan, X. Jia, H. Yanxiang, Energy efficient distributedconnected dominating sets construction in wireless sensornetworks, in: Proceedings of the 2006 International Conference onWireless Communications and Mobile Computing, IWCMC ’06, ACM,New York, NY, USA, 2006, pp. 797–802. http://dx.doi.org/10.1145/1143549.1143709.

[12] A.-L. Barabási, Network Science, 2012. <http://barabasilab.neu.edu/networksciencebook/>.

[13] T.G. Lewis, Network Science: Theory and Applications, WileyPublishing, 2009.

[14] M. Newman, Networks: An Introduction, Oxford University Press,Inc., New York, NY, USA, 2010.

[15] D. Katsaros, N. Dimokas, L. Tassiulas, Social network analysisconcepts in the design of wireless ad hoc network protocols,Netw., IEEE 24 (6) (2010) 23–29, http://dx.doi.org/10.1109/MNET.2010.5634439.

[16] E. Stai, V. Karyotis, S. Papavassiliou, Topology enhancements inwireless multihop networks: a top-down approach, IEEE Trans.Parall. Distrib. Syst. 23 (7) (2012) 1344–1357, http://dx.doi.org/10.1109/TPDS.2011.250.

[17] C. Verma, B. Tamma, B.S. Manoj, R. Rao, A realistic small-world modelfor wireless mesh networks, Commun. Lett., IEEE 15 (4) (2011) 455–457, http://dx.doi.org/10.1109/LCOMM.2011.020111. 100266.

[18] A. Banerjee, R. Agarwal, V. Gauthier, C.K. Yeo, H. Afifi, F. Lee, A self-organization framework for wireless ad hoc networks as smallworlds, IEEE Trans. Veh. Technol. 61 (6) (2012) 2659–2673, http://dx.doi.org/10.1109/TVT.2012.2197768.

[19] D.J. Watts, S.H. Strogatz, Collective dynamics of ‘small-world’networks, Nature 393 (6684) (1998) 409–410.

[20] R. Agarwal, A. Banerjee, V. Gauthier, M. Becker, C.K. Yeo, B.S. Lee,Achieving small-world properties using bio-inspired techniques inwireless networks, Comput. J. http://dx.doi.org/10.1093/comjnl/bxs024. <http://comjnl.oxfordjournals.org/content/early/2012/03/28/comjnl.bxs024.abstract>.

[21] E.M. Daly, M. Haahr, Social network analysis for routing indisconnected delay-tolerant manets, in: Proceedings of the 8thACM International Symposium on Mobile Ad Hoc Networking andComputing, MobiHoc ’07, ACM, New York, NY, USA, 2007, pp. 32–40.http://dx.doi.org/10.1145/1288107.1288113.

[22] M. Kas, S. Appala, C. Wang, K. Carley, L. Carley, O. Tonguz, What ifwireless routers were social? Approaching wireless mesh networksfrom a social networks perspective, Wirel. Commun., IEEE 19 (6)(2012) 36–43, http://dx.doi.org/10.1109/MWC.2012.6393516.

[23] G. Pallis, D. Katsaros, M. Dikaiakos, N. Loulloudes, L. Tassiulas, On thestructure and evolution of vehicular networks, in: IEEE InternationalSymposium on Modeling, Analysis Simulation of Computer andTelecommunication Systems, MASCOTS ’09, 2009, pp. 1–10. http://dx.doi.org/10.1109/MASCOT.2009.5366230.

[24] M. Girvan, M.E.J. Newman, Community structure in social andbiological networks, Proc. Nat. Acad. Sci. 99 (12) (2002) 7821–7826,http://dx.doi.org/10.1073/pnas.122653799. <http://www.pnas.org/content/99/12/7821.abstract>.

[25] A. Cuzzocrea, A. Papadimitriou, D. Katsaros, Y. Manolopoulos, Edgebetweenness centrality: a novel algorithm for QoS-based topologycontrol over wireless sensor networks, J. Netw. Comput. Appl. 35 (4)(2012) 1210–1217. http://dx.doi.org/10.1016/j.jnca.2011.06.001.<http://www.sciencedirect.com/science/article/pii/S1084804511001135>.

[26] J. Wu, H. Li, On calculating connected dominating set for efficientrouting in ad hoc wireless networks, in: Proceedings of the 3rdInternational Workshop on Discrete Algorithms and Methods forMobile Computing and Communications, DIALM ’99, ACM, New York,NY, USA, 1999, pp. 7–14. http://dx.doi.org/10.1145/313239.313261.

[27] D. Kim, Y. Wu, Y. Li, F. Zou, D.-Z. Du, Constructing minimumconnected dominating sets with bounded diameters in wirelessnetworks, IEEE Trans. Parall. Distrib. Syst. 20 (2) (2009) 147–157,http://dx.doi.org/10.1109/TPDS.2008.74.

[28] M. Thai, F. Wang, D. Liu, S. Zhu, D.-Z. Du, Connected dominating setsin wireless networks with different transmission ranges, IEEE Trans.


Mobile Comput. 6 (7) (2007) 721–730, http://dx.doi.org/10.1109/TMC.2007.1034.

[29] L.C. Freeman, Centrality in social networks conceptual clarification,Soc. Netw. 1 (3) (1978–1979) 215–239. http://dx.doi.org/10.1016/0378-8733(78)90021-7. <http://www.sciencedirect.com/science/article/pii/0378873378900217>.

[30] L.C. Freeman, A set of measures of centrality based on betweenness,Sociometry 40 (1) (1977) 35–41. <http://links.jstor.org/sici?sici= 0038-0431%28197703%2940%3A1%3C35%3AASOMOC%3E2.0.CO%3B2-H>.

[31] ns-3 Network Simulator. <http://www.nsnam.org/>.[32] IEEE Standard for Information Technology–Telecommunications and

Information Exchange between Systems Local and Metropolitan AreaNetworks–Specific Requirements Part 11: Wireless LAN MediumAccess Control (MAC) and Physical Layer (PHY) Specifications, IEEE Std802.11-2012 (Revision of IEEE Std 802.11-2007) (2012) 1–2793 http://dx.doi.org/10.1109/IEEESTD.2012. 6178212.

[33] S. Kurkowski, W. Navidi, T. Camp, Constructing manet simulationscenarios that meet standards, in: IEEE Internatonal Conference onMobile Adhoc and Sensor Systems, MASS, 2007, pp. 1–9. http://dx.doi.org/10.1109/MOBHOC.2007.4428640.

[34] K. Andreev, P. Boyko, IEEE 802.11s mesh networking NS-3 model, in:Workshop on ns3, 2010, p. 43.

[35] K.M. Carley, J. Pfeffer, J. Reminga, J. Storrick, D. Columbus, ORA User’sGuide 2012, Tech. Rep., DTIC Document, 2012.

[36] S.P. Borgatti, M.G. Everett, L.C. Freeman, UCINET for windows:software for social network analysis (2002).

[37] S.P. Borgatti, Identifying sets of key players in a social network,Comput. Math. Org. Theory 12 (1) (2006) 21–34, http://dx.doi.org/10.1007/s10588-006-7084-x. <http://dx.doi.org/10.1007/s10588-006-7084-x>.

[38] P.V. Marsden, Egocentric and sociocentric measures of networkcentrality, Soc. Netw. 24 (4) (2002) 407–422. doi:http://dx.doi.org/10.1016/S0378-8733(02)00016-3. URL http://www.sciencedirect.com/science/article/pii/S0378873302000163.

[39] M. Everett, S.P. Borgatti, Ego network betweenness, Social Networks27 (1) (2005) 31–38. http://dx.doi.org/10.1016/j.socnet.2004.11.007.<http://www.sciencedirect.com/science/article/pii/S037887330400067X>.

[40] W. Heinzelman, A. Chandrakasan, H. Balakrishnan, Energy-efficientcommunication protocol for wireless microsensor networks, in:Proceedings of the 33rd Annual Hawaii International Conference onSystem Sciences, vol. 2, 2000, p. 10. http://dx.doi.org/10.1109/HICSS.2000.926982.

Andrés Vázquez Rodas received the Elec-tronics Engineering degree (Honors) in 2004from the Salesian Polytechnic University,Cuenca, Ecuador and the M.Sc. (Honors) fromthe University of Cuenca, Ecuador in 2010. Hewas also an assistant professor at the SalesianPolytechnic University until 2010. He is cur-rently pursuing his Ph.D. at the Department ofTelematic Engineering at the UniversitatPolitècnica de Catalunya BarcelonaTech (UPC),Barcelona, Spain. His research interestsinclude wireless mesh networks, wireless

sensor networks, multimedia services and complex networks.

Luis J. de la Cruz Llopis received the tele-communication engineering degree in 1994and the Ph.D. in telecommunications engi-neering in 1999, both from the UniversitatPolitècnica de Catalunya BarcelonaTech (UPC),Barcelona, Spain. He is currently an AssociateProfessor at the Department of TelematicEngineering of the UPC. His current researchinterests include multimedia service trans-mission with QoS and network security.


http://dx.doi.org/10.1109/ICDCS.2003.1203449

http://dx.doi.org/10.1109/JCN.2002.6596934

http://dx.doi.org/10.1109/JCN.2002.6596934

http://dx.doi.org/10.1109/TPDS.2009.78


http://dx.doi.org/10.1145/1143549.1143709

http://dx.doi.org/10.1145/1143549.1143709

http://barabasilab.neu.edu/networksciencebook/

http://barabasilab.neu.edu/networksciencebook/

http://dx.doi.org/10.1109/MNET.2010.5634439

http://dx.doi.org/10.1109/MNET.2010.5634439



http://dx.doi.org/10.1109/LCOMM.2011.020111.100266

http://dx.doi.org/10.1109/TVT.2012.2197768

http://dx.doi.org/10.1109/TVT.2012.2197768

http://dx.doi.org/10.1093/comjnl/bxs024

http://dx.doi.org/10.1093/comjnl/bxs024

http://comjnl.oxfordjournals.org/content/early/2012/03/28/comjnl.bxs024.abstract

http://comjnl.oxfordjournals.org/content/early/2012/03/28/comjnl.bxs024.abstract

http://dx.doi.org/10.1145/1288107.1288113

http://dx.doi.org/10.1109/MWC.2012.6393516

http://dx.doi.org/10.1109/MASCOT.2009.5366230

http://dx.doi.org/10.1109/MASCOT.2009.5366230

http://dx.doi.org/10.1073/pnas.122653799

http://www.pnas.org/content/99/12/7821.abstract

http://www.pnas.org/content/99/12/7821.abstract

http://dx.doi.org/10.1016/j.jnca.2011.06.001



http://dx.doi.org/10.1145/313239.313261


http://dx.doi.org/10.1109/TMC.2007.1034

http://dx.doi.org/10.1109/TMC.2007.1034

http://dx.doi.org/10.1016/0378-8733(78)90021-7

http://dx.doi.org/10.1016/0378-8733(78)90021-7

http://www.sciencedirect.com/science/article/pii/0378873378900217

http://www.sciencedirect.com/science/article/pii/0378873378900217

http://links.jstor.org/sici?sici=0038-0431%28197703%2940%3A1%3C35%3AASOMOC%3E2.0.CO%3B2-H



http://www.nsnam.org/

http://dx.doi.org/10.1109/IEEESTD.2012.6178212

http://dx.doi.org/10.1109/IEEESTD.2012.6178212

http://dx.doi.org/10.1109/MOBHOC.2007.4428640

http://dx.doi.org/10.1109/MOBHOC.2007.4428640

http://dx.doi.org/10.1007/s10588-006-7084-x

http://dx.doi.org/10.1007/s10588-006-7084-x

http://dx.doi.org/10.1007/s10588-006-7084-x

http://dx.doi.org/10.1007/s10588-006-7084-x

http://dx.doi.org/10.1016/S0378-8733(02)00016-3

http://dx.doi.org/10.1016/S0378-8733(02)00016-3



http://dx.doi.org/10.1016/j.socnet.2004.11.007

http://www.sciencedirect.com/science/article/pii/S037887330400067X

http://www.sciencedirect.com/science/article/pii/S037887330400067X

http://dx.doi.org/10.1109/HICSS.2000.926982

http://dx.doi.org/10.1109/HICSS.2000.926982


Documents

A centrality-based topology control protocol for wireless mesh networks