A Study of Access Point Allocation and ChannelAssignment for Dependable Wireless Mesh Network

September 2010

Walaa A. A. Hassan

Graduate School ofNatural Science and Technology

(Doctor Course)Okayama University

Dissertation submitted toGraduate School of Natural Science and Technology

ofOkayama University

forpartial fulfillment of the requirements

for the degree ofDoctor of Philosophy.

Written under the supervision of

Professor Nobuo Funabiki

and co-supervised byProfessor Yoshitaka Morikawa

andProfessor Masaharu Hata

Okayama University, September 2010.

To Whom It May Concern

We hereby certify that this is a typical copy of the original doctor thesis ofMrs. Walaa A. A. Hassan

Seal of Seal of

the Supervisor Graduate School of

Prof. Nobuo Funabiki Natural Science and Technology

Abstract

The rapid progress of wireless communications and the availability of lightweight, small-size, and portable computing devices have given great impacts on our communities. Thesetechnologies have made the dream of communications at anytime and at anywhere cometrue. A common solution for such ubiquitous communications is the use of the wireless localarea network (WLAN). WLAN has been widely studied and deployed as an access networkto the Internet in many places.

With the progression of WLAN, the wireless mesh network has been developed as a flexi-ble, adaptive, and reconfigured network architecture while offering cost-effective solutions toservice providers. A wireless mesh network is composed of multiple wireless routers that areconnected wirelessly, in addition to wireless communications between user hosts and routers,in order to expand the coverage area that is usually limited into a small space by a single APin a conventional WLAN. Thus, data communications between routers are usually offeredby multi-hop wireless communications.

Among several variations of under-studying wireless mesh networks, our research grouphas focused on the one that targets the Internet access service, uses only access points(APs) as wireless routers, and realizes wireless communications between APs mainly on theMAC layer using the wireless distribution system (WDS). Henceforth, we call this Wire-less Internet-access Mesh NETwork as WIMNET for convenience. WIMNET consists of anumber of APs and their wireless links that can cause failures easily. As a result, the de-pendability enhancement of WIMNET becomes a very important task for a secure accessnetwork to the Internet, where even some failures of APs and/or links may not cause themalfunction or the disconnection of WIMNET.

In this thesis, we first formulate the link-fault and AP-fault dependable AP allocationproblems for WIMNET, after giving the background about WIMNET and the overview ofthe network model used in our study, and present their algorithms by extending the existingAP allocation algorithm. We verify the effectiveness of our approach through extensive sim-ulations using the WIMNET simulator. Three network topologies are adopted as simulatedinstances for evaluations. The simulation results show that a small number of additional APsare required to satisfy the dependability constraint, and the degradation of the throughputis not serious regardless of the increasing interference by increasing APs.

Then, we introduce the route availability (RA) index for WIMNET to represent howmany hosts can be connected with the GW on average under a gaiven set of probabilities.Four network topologies are adopted as simulated instances for evaluations. The simulationresults show that the RA index is actually improved in our fault dependable AP allocationsfound by our algorithms, and the difference between RA indices and the correspondingsimulation results is small.

i

After that, we present the dynamic channel assignment (DCA) technique for WIMNET.The DCA technique is composed of the initial stage and the dynamic stage, and is giventhrough modifications from the existing study for the static channel assignment, with anewly defined decision function. Two network topologies are adopted as simulated instancesfor evaluations. The application of our DCA technique gives the significant performanceimprovement.

Finally, we survey several secure multi-hop routing techniques for WIMNET for our nextworks, where we will adopt the secure multi-hop routing technique to further enhance thedependability of WIMNET.

In our future works for the dependable AP allocation problem, we will evaluate theperformance of our proposal in more practical network fields for WIMNET, such as stations,shopping malls, and large buildings. At this time, we will consider the effect of indoorenvironments more precisely in our model. Besides, we will extend our approach to enduremultiple failures of links and/or APs. For the RA index, we will investigate the relationshipbetween the RA index and the AP-fault dependability through simulations, and will to ablethe AP allocation algorithm extension to maximize the RA index. For the DCA technique,we will consider the application of this technique to a real AP.

ii

List of Publications

Journal

1. W. Hassan, N. Funabiki, and T. Nakanishi, ”Extensions of the Access Point AllocationAlgorithm for Wireless Mesh Networks”, IEICE Trans. Commun., vol. E93-B, no. 6,pp.1555-1565, June 2010.

International Conference Proceeding

2. W. Hassan, N. Funabiki, and T. Nakanishi, ”A dynamic channel assignment tech-nique for wireless Internet-access mesh networks,” The 14th Asia-Pacific Conferenceon Communications (APCC2008), Oct. 14-16, 2008.

Others

3. W. Hassan, T. Farag, N. Funabiki, and T. Nakanishi, ”Dependability extensions ofthe access point allocation algorithm for wireless mesh networks,” IEICE TechnicalReport, NS2009-23, pp. 37-42, May 2009.

4. W. Hassan, T. Farag, N. Funabiki, and T. Nakanishi, ”Improvement of DependableAccess Point Allocation Algorithm and Introduction of Route Availability for WirelessMesh Networks, ” IEICE Technical Report, NS2009-122, pp. 11-16, December 2009.

iii

List of Figures

1.1 A wireless mesh network architecture. . . . . . . . . . . . . . . . . . . . . . . 2

2.1 WIMNET elements. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62.2 Network model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

3.1 AP allocation result for dependability extensions in 16 square-room field. . . 193.2 AP allocation result for dependability extensions in central library field. . . . 203.3 AP allocation result for dependability extensions in office building floor field. 21

4.1 Serial path and parallel path. . . . . . . . . . . . . . . . . . . . . . . . . . . 254.2 Sample WIMNET for RA calculation. . . . . . . . . . . . . . . . . . . . . . . 274.3 RA index changes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 284.4 RA index changes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 294.5 Validation of RA index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

5.1 Seven patterns of numbers of associated hosts with one AP. . . . . . . . . . . 365.2 Average throughputs and numbers of channel changes for 3 × 3 network with

max. two NICs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375.3 Average throughputs and numbers of channel changes for 3 × 3 network with

max. three NICs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385.4 Average throughputs and numbers of channel changes for 5 × 5 network with

max. two NICs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395.5 Average throughputs and numbers of channel changes for 5 × 5 network with

max. three NICs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39

6.1 Classification of ad hoc network routing protocols. . . . . . . . . . . . . . . . 436.2 Wormhole attack by colluding nodes A and B. . . . . . . . . . . . . . . . . . 456.3 SAODV operations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 496.4 Security aware routing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

iv

List of Tables

3.1 Numbers of allocated APs. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213.2 Total throughputs with no fault (Mbps). . . . . . . . . . . . . . . . . . . . . 223.3 Total throughputs for link-fault extension with one link fault (Mbps). . . . . 223.4 Total throughputs for AP-fault extension with one AP fault (Mbps). . . . . . 223.5 CPU time (minutes). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

6.1 Attacks on Routing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 466.2 Operation example and message formats in ARAN. . . . . . . . . . . . . . . 476.3 Operation example of SRP and format of the SRP messages. . . . . . . . . . 48

v

Contents

Abstract ii

List of Publications iii

List of Figures iv

List of Tables v

1 Introduction 1

2 Background and Network Model 52.1 WIMNET Elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2 Design Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.3 WIMNET Simulator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72.4 Network Model for WIMNET Design . . . . . . . . . . . . . . . . . . . . . . 82.5 Signal Propagation Model in Indoor Environment . . . . . . . . . . . . . . . 9

3 Dependability Extensions of AP Allocation Algorithm 103.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103.2 Previous Work of AP Allocation Algorithm . . . . . . . . . . . . . . . . . . . 113.3 Link-fault Dependable AP Allocation extension Algorithm . . . . . . . . . . 143.4 AP-fault Dependable AP Allocation extension Algorithm . . . . . . . . . . . 163.5 Evaluation by Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183.6 Related Works . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

4 Route Availability Index 244.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244.2 Route Availability Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 244.3 Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

5 Dynamic Channel Assignment Technique 305.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305.2 Formulation of Dynamic Channel Assignment Problem . . . . . . . . . . . . 315.3 FCA Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

vi

5.4 DCA Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355.5 Performance Evaluation by Simulator . . . . . . . . . . . . . . . . . . . . . . 365.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38

6 Secure Multi-hop Routing 406.1 Security Technology Overview . . . . . . . . . . . . . . . . . . . . . . . . . . 406.2 Security Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 416.3 Multi-hop Routing Protocols and Their Attacks . . . . . . . . . . . . . . . . 426.4 Securing Multi-hop Routing Protocols . . . . . . . . . . . . . . . . . . . . . 466.5 Summary and Open Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

7 Conclusion and Future Works 52

Acknowledgment 54

References 55

vii

Chapter 1

Introduction

Recently, an increasing number of people are using wireless local area networks (WLANs)for ubiquitous communications with the Internet. WLANs have been widely studied andactually deployed as access networks to the Internet in many places. WLANs provide un-precedented freedom and mobility for a growing number of users using laptop PCs andPDAs. They no longer need wires to stay connected with their workplaces and the Internet.Besides, they have the low cost and the relative ease of deployments. From these reasons,WLANs are considered as an attractive communication paradigm.

Along this context, the wireless mesh network (WMN) has been studied as a flexible,adaptive, and reconfigurable network architecture in offering cost-effective solutions to net-work service providers.WMNs are multi-hop wireless networks formed by mesh routers andmesh clients. Mesh routers are typically stationary and do not have energy constrains.On the other hand, mesh clients are usually mobile and energy constrained. Some meshrouters are designed as gateways (GWs) which are connected to the Internet through wiredconnections [1]. An architecture of WMN is shown in Figure 1.1. In this WMN, data commu-nications between mesh routers are offered by multihop wireless communications, in additionto communications between mesh routers and mesh clients. The WMN can be applied to awide variety of applications including:

• community networking,

• building automation,

• intelligent transport system network,

• defense system, and

• citywide surveillance system.

Among several variations of under-studying wireless mesh networks, we have focused onthe one that targets the Internet access service, uses only access points (APs) as wirelessrouters, and realizes communications between APs mainly on the MAC layer using thewireless distribution system (WDS). Henceforth, we call it Wireless Internet-access MeshNETwork (WIMNET) for convenience. WDS allows a wireless network to be expanded usingmultiple APs without a wired backbone. The advantage of WDS over other solutions is to

1

Figure 1.1: A wireless mesh network architecture.

preserve the MAC addresses of client packets across links between APs [2]. The limitationof WDS is that all the APs in the same WDS cluster must be configured to use the sameradio channel, and the same encryption method with the same encryption key. WDS is aninexpensive and easy way to extend the wireless Internet-access network coverage.

For the optimal design of WIMNET, we have studied several combinatorial optimizationproblems related to the design issue, which requires the optimization of multiple objec-tives while satisfying complex constraints. The objectives include the maximization of thethroughput, and the minimization of the cost and the delay. The constraints include thecoverage of any client host in the network field through wireless communications from APs,the wireless connectivity between APs and GWs, the WDS cluster size limitation, and thelimited locations of APs and GWs. Considering these conditions, we have formulated thefollowing combinatorial optimization problems, and proposed their algorithms [3][4][5]:

• AP allocation problem,

• WDS clustering problem,

• GW selection problem,

• Routing tree generation problem, and

2

• Channel configuration problem.

The dependability of WIMNET includes ensuring the survivability of network servicesdespite failures of AP and/or links, or attacks from insides/outsides [6]. As the dependableInternet access network, the availability of WIMNET should be maximized. Therefore, thedesign of AP allocation of dependable WIMNET that endures the link/AP failure with theinvestigation of the route availability index is an essential target. To increase the availabilityof WIMNET, the dynamic channel configuration is another important issue to increase theoverall performance of the network. Finally, it becomes another important issue to securethe multi-hop routing against both dynamically changing topology and malicious attacks.

In the AP allocation, at least one AP must act as a gateway (GW) to the Internet. Thepackets to/from user hosts, if associated with APs other than GWs, need to reach one ofGWs through multi-hop communications between APs to access the Internet. As a result,communications around GWs can become bottlenecks. Besides, the link quality in indoorenvironments may be degraded by obstacles such as walls.

Moreover, WIMNET may suffer from failures of links and/or APs due to hardware faultsin the large scale system and/or environmental changes in the network field. A link/APfailure often causes the disconnection of APs that is crucial as a network infrastructure.Thus, the AP allocation of enduring the link/AP failure becomes the important design issueof WIMNET to improve the dependability. Under these circumstances, the installation costand the delay should be minimized, while any host in the service area must be covered byat least one AP in spite of link/AP failure. For example, the delay of WIMNET can beminimized by reducing the maximum hop count between APs [7].

In order to improve the dependability of WIMNET, the route availability index (RA) isproposed in this thesis. When each AP is assigned the probability of functioning properly andeach host is assigned the probability of accessing to the Internet, this index can be calculated.Actually, it represents how many hosts can be connected with the GW on average. Whenthe number of APs is fixed, this index should be maximized for the better dependability,where more hosts can be covered under failures of APs and/or links on average.

The dynamic channel assignment (DCA) strategy changes channel assignments dynami-cally, to meet changing traffic demands. The secure routing is an important issue of WIM-NET availability, because by attacking the routing mechanism, an adversary can modify thenetwork topology and therefore, affect the good functioning of the network.

In this thesis, we present a novel method of increasing the dependability of WIMNET.First, we formulate the dependable AP allocation problem in indoor environments for WIM-NET. Then, we present the link-fault dependable AP allocation algorithm and the AP-faultdependable AP allocation algorithm by extending the existing AP allocation algorithm so asto endure one link fault and one AP fault separately. In these algorithms, the connectivityof APs and the host coverage are always checked when one link or AP is removed fromthe network. If some APs are disconnected due to a link/AP fault or some hosts are notcovered by an AP fault, additional redundant APs are allocated to the places such that themaximum number of faults can be covered while maximizing the throughput. After that,we investigate the RA index by calculating how many hosts can be connected with the GWon average under some probabilities. Through simulations using the WIMNET simulator[3] that has been developed by our group, we verify the effectiveness of our proposal by

3

comparing RA indexes between original AP allocations and their dependable allocations.Then, we present a formulation for the DCA problem and a DCA technique through

modifications of our existing study for the fixed channel assignment (FCA) problem [4].We newly define a decision function for the channel reassignment. The effectiveness ofour approaches is evaluated through simulations using the WIMNET simulator. Finally, wesurvey the secure multi-hop routing in WIMNET, and summarize the its aspects to maximizethe network availability with the future studies.

The remaining of this thesis is organized as follows. Chapter 2 introduces an overviewof the characteristics and challenges in the WIMNET design, especially for the availabilityand survivability of WIMNET. Then, the difficulty in achieving the dependable networkarchitecture is discussed. Besides, the outline of the WIMNET simulator is presented, whichhas been used for performance evaluations in our studies. Finally, the network model in thisthesis is presented.

Chapter 3 formulates the AP allocation problem for the dependable WIMNET to endurelink/AP failure, and presents its two-stage heuristic algorithm composed of the initial APallocation (greedy stage) and the AP allocation optimization with the association host opti-mization (improvement stage). The algorithm finds the AP allocation where the connectivitybetween the GW and the hosts are maintained, even if one link/AP is failed. The effective-ness of this approach is verified through network simulations using the WIMNET simulator.In simulation results, the dependability is realized with a small number of additional APsand the network performance is not degraded for one link or one AP fault.

Chapter 4 formulates the RA index to estimate the certainty of connections betweenhosts and the Internet gateway under some probabilities. The effectiveness of our proposalis verified through simulations in three network instances.

Chapter 5 presents a formulation for the DCA problem and a DCA technique throughmodifications of our existing study for the FCA problem [4]. We newly define a decisionfunction for the channel reassignment. Our technique is divided into two stages, the initialstage and the dynamic stage. The significant performance improvement is observed byapplying our DCA technique over two network instances.

Chapter 6 surveys secure multi-hop routing protocols to increase the availability and sur-vivability of the network. This chapter covers relevant security goals for WIMNET in generaland secure routing. Then, secure routing objectives of WIMNET are summarized, and thepossible attacks are illustrated. Some research proposals and open issues are introduced forour future studies.

Finally, Chapter 7 concludes this thesis with some future works.

4

Chapter 2

Background and Network Model

In this chapter, we summarize the WIMNET architecture as the first target of this thesisand its design challenges. Then, we describe the details of the WIMNET simulator and thenetwork model that have been used through the thesis.

2.1 WIMNET Elements

WIMNET is composed of the following three distinct network elements shown in Figure 2.1:

Network Gateway (GW): at least one GW must be deployed to allow access to the In-ternet.

Access Point (AP): APs form the wireless network infrastructure for providing the con-nectivity between mobile hosts distributed in the network field.

Mobile Host: mobile hosts may include any device embedding wireless capabilities such asPDAs and laptop PCs. They can access to the Internet through a GW with direct ormulti-hop communications.

WIMNET has a relatively stable topology except for the occasional failures of nodesand/or additions of new nodes. The traffics from end users may be changed frequently.Most traffics in WIMNET are either forwarded to or from a GW. Traffics between a pair ofmobile hosts inside the same WIMNET are usually smaller than them. The APs in WIM-NET have functionalities of enabling the mesh networking. The APs may have single ormultiple network interface cards (NICs) using the same or different communication tech-nologies depending on the requirements. They can provide the wide coverage area using thesame small transmission power by using multi-hop communications where intermediate APsrelay the packets. The APs can be built on general-purpose computers, or can be built ondedicated hardware platforms. Normally, a mobile host has a single NIC.

2.2 Design Challenges

The optimal design of WIMNET is actually a very difficult task, because it must consider sev-eral factors at the same time, including the wireless multi-hopping routing, the performance

5

Figure 2.1: WIMNET elements.

(throughput, packet loss, and delay), and the security. The challenges in the WIMNETdesign are summarized as follows.

Channel assignmentFor each wireless link, a set of non-interfered channels is available. A link is assigneda channel by selecting the transmitting channel of the NIC at the source node. Thislink channel should be optimally selected so that the interference between adjacentlinks can be minimized to improve the performance. A channel can be overloadedwhen too many links are assigned the same channel. There are several methods thatmay be able to overcome this overloaded channel in wireless networks. One of themis the use of multiple NICs and the assignment of multiple channels. In our previouswork in [4], we proposed a technique to avoid the bottleneck communication amonglinks around GWs by properly choosing their channels and the channels at NICs ofthe corresponding APs. Besides, in WIMNET, hosts may move around in the networkfield, which can raise issues such as the quality of service, the channel interference,and the network load management. As a result, the dynamic channel assignment tothe links between APs becomes essential to improve the performance, which will bediscussed in Chapter 5.

6

ScalabilityThe scalability is one of the major issues for the optimal design of WIMNET [8]. Forthe successful deployment and revenue generation, WIMNET should be able to expandthe system capacity, in terms of the number of supporting users, the density of users,the size of the geographical coverage area, and the data transmission rate. To ensurethe scalability of WIMNET, the architecture and the communication protocols fromthe MAC layer to the application layer need to be scalable.

Connectivity through multi-hop communicationAdvantages of WIMNET come from the mesh connectivity between APs through multi-hop communications. To ensure the reliable mesh connectivity, the self-organizationand topology control mechanisms are necessary. The high-quality AP allocation androuting can significantly improve the performance of WIMNET. In WIMNET, themulti-hopping has a significant influence on the network utilization and performance.If the wireless mesh network is not well-designed, an AP with several hops apart fromthe GW would obtain a much lower bandwidth share than an AP that is close tothe GW. This directs to the critical unfairness problem, and even potentially to thestarvation problem. An optimal AP allocation may help to solve the problem. Inthis thesis, we extend the AP allocation algorithm in [9] to ensure the connectivity ofWIMNET by enduring one AP/link failure, which will be discussed in Chapter 3.

SecurityOne week point of WIMNET for the wide-scale deployment is the lack of securityguarantees. For example, the multi-hop routing mechanism in WIMNET needs to besecured. The attacker may affect the routing and the functionality of WIMNET byinserting false routing messages. For the security in WIMNET, especially the protocolsof multi-hop routing are important, which will be discussed in Chapter 6.

2.3 WIMNET Simulator

We use the WIMNET simulator to verify the effectiveness of our approaches in this thesis.The WIMNET simulator simulates least functions for wireless communications of hosts andAPs that are required to calculate the throughput and delay, where it has been developedto evaluate a large-scale WIMNET with reasonable CPU time on a conventional PC. Asequence of functions such as host movements, communication request arrivals, and wirelesslink activations are synchronized by a single global clock called a time slot. Within anintegral multiple of time slots, a host or an AP can complete the one-frame transmission andthe acknowledgement reception.

From our past experiments [10] and some references [11][12], we set 30Mbps for themaximum transmission rate for IEEE 802.11a and 20Mbps for IEEE 802.11g. Note thatthis transmission rate can cover about 26 hosts [13][14]. Then, if the duration time of onetime slot is set 0.2ms and each frame size is 1, 500bytes, two time slots can complete the30Mbps link activation because (1, 500byte × 8bit × 10−6M)/(0.2ms × 2slot × 10−3s) =30Mbps, and three slots can complete the 20Mbps link activation because (1, 500byte ×8bit× 10−6M)/(0.2ms× 3slot× 10−3s) = 20Mbps. We note that the different transmission

7

rate can be set by manipulating the time slot length and the number of time slots for onelink activation. When two or more links within their wireless ranges may be activated atthe same time slot, randomly selected one link among them is successfully activated, andthe others are inserted into waiting queues to avoid collisions, supposing the distributedcoordination function (DCF) and the request-to-send/clear-to-send (RTS/CTS) functions.

In order to evaluate the throughput shortly, every host has 1, 000 packets to be trans-mitted to the GW, and the GW has 125 packets to every host before starting a simulation.Then, when every packet reaches the destination or is lost, the simulation is finished. Here,no packet is actually lost by assuming the queue with the infinite size at any AP in oursimulations. The packets for each request are transmitted along the shortest path that iscalculated. Only the connection-less communication is implemented this time, where theretransmission between end hosts is not considered.

The throughput comparison using this simple WIMNET simulator is actually sufficientto show the effectiveness of our approach, because it simulates the basic behaviors affectingthe throughput of the wireless mesh network, such as the contention resolution among theinterfered links and the packet relay action for the multi-hop communication. Note thatour experimental results in a simple topology confirmed the correspondence of throughputbetween the simulator and the measurement. The packet retransmissions of the interferedlinks, if implemented, can worse the throughput by the poor AP allocation in comparisons,because they cause more interference between links.

2.4 Network Model for WIMNET Design

A closed area such as one floor in an office/school building, a conference hall, or a library,is considered as the network field for WIMNET in this thesis. Like [15][16], we adopt thediscrete formulation for the AP allocation problem. On this field, as shown in Figure 2.2,discrete points called host points are considered as locations where hosts and/or APs mayexist. Every host point is associated with the number of possibly located hosts there. Besides,a subset of host points are given as battery points where the electricity can be supplied tooperate APs. Thus, any AP location must be selected from battery points. Here, we notethat some host points are allowed to be associated with zero hosts, so that some batterypoints can exist without any host association. A subset of battery points can be candidatesfor GWs to the Internet. This GW selection is also the important mission of the AP allocationproblem.

In WIMNET, each AP must take two different roles for wireless communications, specifi-cally a wireless hub for its associated hosts and a wireless bridge for relaying packets betweenAPs. To reduce the radio interference between these two communications, we use differentprotocols with different channels (radio frequencies) for them. Actually, we assign the IEEE802.11b/g protocol with 2:4GHz for the former role, and the 802.11a protocol with 5GHzfor the latter. Note that each protocol has several non-interfered frequency channels.

8

Figure 2.2: Network model.

2.5 Signal Propagation Model in Indoor Environment

In an indoor environment, the estimation of the signal strength received at a point is essentialto determine the availability of the wireless link from its source node (host or AP) to thispoint, because it is strongly affected by obstacles between them. To estimate it properly,our study employs the following log-distance path loss model that has been used successfullyfor both indoor and outdoor environments [17][18][19]:

Pd = P1 − 10 · α · log10 d−∑

k

nk ·Wk + Xσ (2.1)

where Pd represents the received signal strength (dBm) at a point with the distance d (m)from the source, P1 does the received signal strength (dBm) at a point with 1 m distancefrom it when no obstacle exists, α does the path loss exponent, nk does the number oftype-k obstacles along the path between the source and the destination, Wk does the signalattenuation factor (dB) for the type-k obstacle, and Xσ does the Gaussian random variablewith the zero mean and the standard deviation of σ (dB). Thus, the model determines thereceived signal strength not only by the distance between the source and the destination,but also by the effect from obstacles along the path between them.

The proper value for the parameter pair (α, σ) depends on the network environment.Measurements in literatures reported that α may exist in the range of 1.8 (lightly obstructedenvironment with corridors) to 5 (multi-floored buildings), and σ does in the range of 4 to 12dB [18]. After calculating the received signal strength at a point, we regard that the wirelesslink from its source can exist to this point if the strength is larger than the threshold.

9

Chapter 3

Dependability Extensions of APAllocation Algorithm

In this chapter, we present two dependability extensions of the AP allocation algorithm inindoor environments for WIMNET in [20].

3.1 Introduction

In WIMNET, all the packets to/from hosts must pass through GWs to access the Internet.If a host is associated with an AP other than GW, the packets must reach GW throughmultihop wireless communications between APs. The routing path for connecting APs andthe GW in WIMNET becomes a tree rooted at the GW, where two or more APs mayhave the same parent AP, and thus, multiple links may come out from the same AP inthe path. Since the bandwidth of one link is usually small in wireless networks, the trafficconcentration into the limited links around the GW increases the communication delay anddecreases the performance of WIMNET. Besides, in indoor environments, where WIMNETis mainly deployed, the link quality can easily be degraded by obstacles such as walls, doors,and furniture, which can reduce the transmission speed of the link. Then, because theAP allocation determines the routing between APs and the GW and the link quality, wehave proposed the AP allocation algorithm for WIMNET using one Internet GW in indoorenvironments [9]. We have adopted the long-distance path loss model [17] to estimate thelink quality. This algorithm can minimize the number of APs to reduce the installation andrunning costs, while maximizing the network performance and minimizing the transmissionpowers of APs.

As the number of APs increases in a large-scale WIMNET, one major concern mustbe resolved as best as possible. This concern is the link/AP fault occurrence. WIMNETmay suffer from malfunctions of links and/or APs due to hardware faults in the large-scalesystem and/or to environmental changes in the network field. Even one link/AP fault cancause the disconnection of APs, which is crucial as the network infrastructure [21]. Thus,the dependable AP allocation of enduring one link/AP fault is an important design issueof WIMNET. To realize this one link/AP fault dependability, redundant APs should beallocated properly, while the number of such APs should be minimized to sustain the cost

10

increase.In this chapter, we present extensions of the AP allocation algorithm to solve the de-

pendable AP allocation problem in a large-scale WIMNET. In the dependable AP allocationto one link/AP fault, we present the link-fault dependable AP allocation algorithm and theAP-fault dependable AP allocation algorithm. Both algorithms use the AP allocation so-lution of our previous algorithm as the initial state. Then, the connectivity of APs andthe host coverage are checked when one link or AP is removed from the network. If someAPs are disconnected due to a link/AP fault or some hosts are not covered by an AP fault,additional redundant APs are allocated to the places such that the maximum number offaults can be covered while maximizing the throughput. The effectiveness of our proposal isverified through simulations using the WIMNET simulator which is illustrated in sec. 2.3.

The rest of this chapter is organized as follows: Section 3.2 introduces our previouswork of the AP allocation algorithm. Sections 3.3 and 3.4 present the link-fault and AP-fault dependable AP allocation algorithms respectively. Section 3.5 discusses performanceevaluations using the WIMNET simulator. Section 3.6 refers some related works to thischapter. Section 5.6 provides the conclusion of this work.

3.2 Previous Work of AP Allocation Algorithm

In this section, we briefly introduce our previous work of the AP allocation problem forWIMNET in [9].

3.2.1 Objectives

The proper AP allocation in WIMNET needs to consider several conflicting factors at thesame time. First, the resulting WIMNET must be feasible as the Internet access network.That is, any AP must be connected to at least one gateway to the Internet through multi-hopcommunications, and any host in the service area be covered by at least one AP. Then, theperformance of WIMNET should be maximized [22], while the AP installation/operationcost be minimized [23]. The performance is usually improved by reducing the maximum hopcount (the number of hops) between an AP and the gateway [7]. Besides, the maximum loadlimit for any AP should be satisfied to enforce the load balance between APs, where theirproper load balance also improves the performance [24]. Furthermore, the signal transmissionpower of an AP should be minimized to reduce the operation cost and the interference oflinks using the same channel. Hence, the objectives of the AP allocation can be summarizedas follows:

• to minimize the number of installed APs,

• to minimize the maximum propagation delay,

• to minimize the transmission delay,

• to minimize the total interference,

• to minimize the maximum hop count to reach a gateway from any AP along the shortestpath, and

11

• to minimize the transmission power of each AP.

3.2.2 Problem Formulation

Now, we define the AP allocation problem for WIMNET.

• Input: A set of host points HP = {hi} with the number of possibly located hosts hni

for the host point hi, a set of battery points BP = {bj} ⊆ HP with the AP installationcost bcj for the battery point bj, a GW g ∈ BP , the number of hosts that each AP cancover as the load limit L, and a set of discrete AP transmission powers TP for P1.

• Output: A set of AP allocations S with the selected transmission power pj for bj ∈ S.

• Constraint: To satisfy the following five constraints:

A1) to cover every host point that has possibly located hosts by an AP,

A2) to connect every AP directly or indirectly,

A3) to allocate APs at battery points (S ⊆ BP ),

A4) to select one transmission power from TP for each AP, and

A5) to associate L or less hosts for any AP.

• Objective: To minimize the following cost function:

F = A∑

bj∈S

bcj + B maxbj∈S

∑

(p,q)∈Pj

1

spq

+C max(p,q)∈Pall

{tpq + tqp

spq

}

+D∑

(i,j)∈Pall

∑

(p,q)∈Pall

(tij + tpq)f(i, j, p, q)

+E

∑bj∈S

pwj

|S| (3.1)

where A-E represent constant coefficients, spq does the transmission speed of the linkfrom APp to APq that depends on the received signal strength at APq, Pj does therouting path from GW to the AP allocated at the battery point bj, tpq does the trafficof the link from APp to APq that is proportional to the number of hosts using this linkfor communications with GW in the paper, and Pall does the union of Pj for ∀bj ∈ S.f(i, j, p, q) represents the interference to the link from APp to APq from the link fromAPi to APj, where if the received strength at APp or APq of the signal transmitted fromAPi is larger than or equal to the threshold (−90 dB in this paper), f(i, j, p, q) = 1,and otherwise, it is 0.

The A-term represents the total installation cost of APs, the B-term does the prop-agation delay between GW and the furthest AP, the C-term does the transmission

12

delay at the most congested link, the D-term does the total interference between ad-jacent links, and the E-term does the average transmission power. This cost functionhas slightly been modified from the previous one to improve the AP allocation resultby considering transmission speeds of links, congestions of links, and the interferencebetween adjacent links. We choose the following values for A-E coefficients. A = 100,B = 0.05, C = 0.05, D = 0.05, and E = 0.05.

3.2.3 AP Allocation Algorithm

Our heuristic algorithm for the AP allocation problem is reviewed here. This algorithm iscomposed of the initial stage and the improvement stage to find an AP allocation satisfyingthe complex constraints while optimizing the objectives. The routing path is selected by theshortest path in both stages, every time a new AP is added or an existing AP is removed.If multiple shortest paths exist, it selects the firstly found one.

Initial Stage

The initial stage consists of the host coverage process and the load balance process to findAP allocations satisfying the constraints. Here, the maximum transmission power is alwaysassigned to any AP to minimize the number of APs.

• Host coverage process: Starting from the selection of GW g, the host coverageprocess repeats the selection of one battery point that is connected with at least oneselected battery point and can cover the largest number of uncovered hosts withoutconsidering the load limit constraint, until every host point is covered by at least oneAP.

• Load balance process: The host coverage process usually does not satisfy the loadlimit constraint for host associations, where some APs may be associated with morethan L hosts. If so, the following load balance process selects new battery points foradditional APs to reduce their loads.

1. Associate each host point to the AP such that the received power is maximumamong the APs.

2. For each AP that does not satisfy the load limit constraint, select the closestbattery point from it into S.

3. Repeat 1. and 2. until all APs satisfy the load limit constraint.

Improvement Stage

In the initial stage, the AP allocation may be far from the best one in terms of the costfunction F due to the greedy nature of the procedure and to the additional APs by the loadbalance process. The improvement stage improves it by modifying the location, the powertransmission, and the host association jointly by using a local search method. The followingprocedure is repeated for a constant number of iterations T :

13

1. Randomly select a battery point bj /∈ S that is connected to an AP in S, and add itinto S with the maximum transmission power.

2. Apply the AP association refinement (latter described in this subsection).

3. Remove from S any AP that satisfies the following three conditions:

1) it is different from bj and GW,

2) all the host points associated with the AP can be associated with the remainingAPs, where for the new association of the host point, the load limit constraintis checked from the AP whose signal power is largest if two or more APs can beassociated with it, and

3) all the APs are connected if removed.

4. If removed, re-associate all the host points associated with this AP to the APs foundin b.

5. Change the transmission power of any possible AP to the smallest one in TP such thatthis AP can still cover any associated host and maintain the links necessary to connectall the APs.

During the iterative search process, the best solution in terms of the cost function F isalways kept for the final output.

AP Association Refinement

After AP locations are modified, some host points may have better APs for associations interms of the received power than their current ones. Thus, the following procedure is appliedto improve them:

1. Find the better AP for association to every host point in terms of the received powerin Eq. (2.1).

2. Apply the following procedure for every host point that is associated with a differentAP from the best:

a. Change the association of this host point to the best AP, if its load is smallerthan the load limit.

b. Otherwise, swap the associated APs between such two host points, if this swappingbecomes better.

3.3 Link-fault Dependable AP Allocation extension Al-

gorithm

3.3.1 Fault Dependability in WIMNET

WIMNET can be disconnected even for one link fault or one AP fault occurrence, whenthe AP allocation by the algorithm in Sect. 3.2.3 is used. To improve the dependability

14

of WIMNET, the AP allocation algorithm should be extended to find an AP allocationsuch that the APs are connected even for one link fault or one AP fault occurrence. Thisdependability can be achieved by allocating redundant APs to the original allocation toprovide backup routes [25]. At the same time, the number of such APs and the maximum hopcount should be minimized for the cost reduction and the performance improvement. Thus,we summarize the design goal in dependability extensions of the AP allocation algorithm asfollows:

1. to endure one link fault or one AP fault,

2. to minimize the number of additional APs, and

3. to minimize the maximum hop count.

3.3.2 Constraint for Link-fault Dependability

To achieve the link-fault dependability, the network must be connected if any link is removedfrom there. Then, another constraint must be satisfied in the AP allocation in addition tothe original five constraints (A1) to A5) ):

B1) to provide the connectivity among the APs if any link is removed.

3.3.3 Algorithm Extension for Link-fault Dependability

Now, we present the algorithm extension for the link-fault dependability. The idea hereis that after maximizing the transmission power from any AP to increase the connectivityamong them, we find any link whose removal disconnects the network, which is called thebridge. If bridges exist, we sequentially allocate an additional AP at the battery point thatcan resolve the maximum number of bridges until all of them are resolved. Then, we findthe minimum-delay routing tree to this link-fault dependable AP allocation by applyingthe algorithm in [5]. Finally, we minimize the transmission powers of APs such that theconstraints of the problem are satisfied. The following procedure describes the link-faultdependability extension:

1. Input the AP allocation from the algorithm in [9].

2. Maximize the transmission power for any AP and find the links between two APs.

3. Find the set of bridges BR.

4. Apply the following procedure if BR = ∅:a. Apply the AP association refinement in 3.2.3.

b. Apply the routing tree algorithm in [5].

c. Minimize the transmission power of the APs such that all the constraints aresatisfied.

d. Terminate the procedure.

15

5. For every bridge in BR, find the set of battery points that can resolve this bridge if anew AP is allocated there. Let this set of the battery points found here be BS.

6. Calculate the number of bridges in BR for each battery point in BS that the APallocated there can resolve.

7. Find the battery point in BS that can resolve the largest number of bridges in BR,and allocate an AP there.

8. Update BR.

9. Go to 4.

3.4 AP-fault Dependable AP Allocation extension Al-

gorithm

3.4.1 Constraint for AP-fault Dependability

To achieve the AP-Fault dependability, the network must be connected and every host mustbe covered by a remaining AP, if any AP is removed from there. Here, any GW is notremoved because we assume no fault at GW. Then, the following two constraints must besatisfied in the AP allocation in addition to the five constraints (A1) to A5) ):

C1) to cover any host by an existing AP if any AP is removed, and

C2) to provide the connectivity among the APs if any AP is removed.

3.4.2 Algorithm Extension for AP-fault Dependability

We present the algorithm extension to the AP-fault dependability. For the AP-fault depend-ability, at least the link-fault dependability must be satisfied, because if one AP is removedfrom the network, its incident links are also removed. Thus, in this extension, we use thelink-fault dependable AP allocation and maximize the transmission power of any AP as theinitial state.

First, we find any host point that cannot be covered if one AP is removed from thenetwork due to the fault, called the critical point, in the initial state. The critical pointsatisfies the following either condition:

1) only this fault AP covers it, or

2) all the backup APs reach association load limits, including the re-associated hosts bythis AP fault.

If critical points exist, we sequentially allocate an additional AP to the battery point thatcan cover the maximum number of critical points until all of them are resolved. Then, wefind any AP whose removal disconnects the network, called the cut AP. If cut APs exist,

16

we sequentially allocate an additional AP to the battery point that can cover the maximumnumber of cut APs until all of them are resolved.

After these steps, we apply the improvement stage in Section 3.4.3 for finding the betterAP allocation. Then, we apply the algorithm in [5] to find the routing tree to the AP-fault de-pendable allocation. Finally, we minimize the transmission powers such that the constraintsare satisfied. The following procedure describes the AP-fault dependability extension:

1. Input the link-fault dependable AP allocation.

2. Maximize the transmission power for any AP and find the links between APs.

3. Find the set of critical host points CR.

4. Apply the following critical host resolution procedure until CR = ∅:(a) For every host point in CR, find the set of battery points that can cover this

critical point if a new AP is allocated there. Let this set of the battery pointsfound here be CS.

(b) Calculate the number of critical points in CR for each battery point in CS thatthe AP allocated there can cover.

(c) Find the battery point in CS that can cover the largest number of critical pointsin CR, and allocate an AP there.

(d) Update CR.

5. Find the set of cut APs CA.

6. Apply the following cut AP resolution procedure until CA = ∅:(a) For every cut AP in CA, find the set of battery points that can cover this cut AP

if a new AP is allocated there. Let this set of the battery points found here beCB.

(b) Calculate the number of cut APs in CA for each battery point in CB that theAP allocated there can cover.

(c) Find the battery point in CB that can cover the largest number of cut APs inCA, and allocate an AP there.

(d) Update CA.

7. Apply the improvement stage in 3.4.3.

8. Apply the AP association refinement in 3.2.3.

9. Apply the routing tree algorithm in [5].

10. Minimize the transmission power of the APs such that all the constraints are satisfied.

11. Terminate the procedure.

17

3.4.3 Improvement Stage

The improvement stage for the AP-fault dependable extension has been slightly modifiedfrom the corresponding one for the original AP allocation problem, such that any AP mustbe connected with at least two APs in order to preserve the link/AP fault dependability.The following procedure is repeated for a given constant number of iterations AT , where thebest solution in terms of the cost function F is always kept for the final solution during theiterative search process:

1. Randomly select a battery point bj /∈ S that is connected to at least two APs in S,and add it into S with the maximum transmission power.

2. Apply the AP association refinement in 3.2.3.

3. Remove from S any AP that satisfies the following four conditions:

1) it is different from bj and GW,

2) all the host points associated with the AP can be re-associated with the remainingAPs, where for the new association of each host point, the load limit constraintis checked from the AP whose signal power is largest if two or more APs can beassociated,

3) no cut AP appears if removed, and

4) no critical host point appears if removed.

4. If removed, re-associate all the host points associated with this AP to the APs foundin 2).

5. Change the transmission power of any possible AP to the smallest one in TP such thatthis AP can still cover any associated host and maintain the links necessary to connectall the APs.

3.5 Evaluation by Simulations

In this section, we evaluate our proposal of the dependable AP allocation algorithm exten-sions through simulations using the WIMNET simulator described in 2.3.

3.5.1 Simulated Instances

In our simulations, three network instances are examined, where the same set of networkparameters in [9] are used.

Figures 3.1, 3.2, and 3.3 illustrate the field and AP allocation results with the routingtree for square room, central library, and office building floor instances respectively. Notethat the white circle represents an AP by the original AP allocation algorithm, the graycircle does an additional AP by the link-fault dependability extension, and the black circledoes an additional AP by the AP-fault dependability extension.

18

Figure 3.1: AP allocation result for dependability extensions in 16 square-room field.

Square room has been used to investigate the optimality of the original AP allocationalgorithm in terms of the number of allocated APs. The network field is composed of 16square rooms with 60m sides. In each room, 25(= 5× 5) host points are allocated with the10m interval, where each host is associated with one host. The 16 host points along the wallsare selected as battery points, assuming that electrical outlets are installed on walls. Thus,the total of 400 host points are distributed regularly in the network field. The maximumload constraint L is set 25 so that the lower bound on the number of allocated APs is 16 tocover the host points by exactly one AP by 400/25 = 16.

Central library is a field similar to the first floor in the central library at Cairo universityas a practical simulated instance. Like the previous instance, each host point is associatedwith one host, and the maximum load limit L is set 25. In the this field, the total size is64m × 32m, and 411 host points are allocated, where the host points along the walls areselected as battery points. Note that the size of the largest room at the top right, calledTaha Hussin Hall, is 18m × 12m with 74 host points. The lower bound on the number ofAPs to satisfy the load constraint is 17(=

⌈41125

⌉).

Office building floor is another practical field similar to one floor in our office building.This field is composed of two rows of different-sized rectangular rooms and one corridor. Onerow has 12 small square rooms with 5m× 5m size with 4 host points, and another row has 5large rectangular rooms with 10m× 12.5m size with 20 host points. The host points along

19

Figure 3.2: AP allocation result for dependability extensions in central library field.

the walls parallel to the corridor are selected as battery points. Besides, 29 battery pointsare allocated with the same interval in the corridor with no host association. The batterypoint in front of the center of the fifth small room in the corridor is given as GW. The totalnumber of expected hosts is 148(= 4 × 12 + 20 × 5). The maximum load limit L is set 25.

Thus, the lower bound on the number of APs to satisfy the load constraint is 6(=⌈

14825

⌉)

3.5.2 AP Allocation Results by Dependability Extensions

First, we discuss the solution quality in terms of the number of APs in AP allocation resultsfor dependability extensions. Table 3.1 compares the numbers of APs in the original APallocation algorithm, the link-fault extension, and the AP-fault extension. For the artificialnetwork field of 16 square rooms (Square field), our dependability extensions can providethe link-fault dependability with additional three APs, and the AP-fault dependability withadditional seven APs. The latter result is much better than the trivial solution for the AP-fault dependability using 15 additional APs where two APs are allocated in each room. Forthe practical fields in the central library (Library field) and office building floor (Office field),no additional AP is necessary for the link-fault dependability and only three additional APsfor the AP-fault dependability. Because most APs can communicate with GW in one hop,any link can easily be backed up by other links. These results verify the effectiveness of ourproposal for dependability extensions in WIMNET in terms of the AP allocation cost.

Then, we investigate throughput changes with or without link/AP faults among AP al-

20

Figure 3.3: AP allocation result for dependability extensions in office building floor field.

Table 3.1: Numbers of allocated APs.Instance Original Link-fault AP-fault

Square field 16 19 26Library field 17 17 20Office field 6 6 9

location results for dependability extensions. Table 3.2 compares total throughputs amongAP allocations for the three cases when no link/AP has fault. The result indicates that thetotal throughput is slightly degraded as the number of APs increases for the fault depend-ability extensions because of the increase of the interference among wireless links betweenAPs using the single channel.

Tables 3.3 and 3.4 show the average, maximum, and minimum throughputs in the link-fault dependable and AP-fault dependable allocations when one link or AP is removed fromthe network to assume the occurrence of a fault. By comparing these results, we concludethat our proposal can provide sufficient throughputs, even if one link fault or one AP faultoccurs in WIMNET.

Here, we note that in the fault dependable AP allocation, some APs may become redun-dant. Thus, the routing without using such APs can be able to improve the performanceby reducing the interference. Besides, if multiple NICs are used at APs for multiple channelcommunications, the results can be changed by reducing the interference. The performanceevaluation in such cases will be in our future studies.

21

Table 3.2: Total throughputs with no fault (Mbps).Instance Original Link AP

Square field 13.0 12.9 12.6Library field 23.9 23.9 23Office field 32.7 32.7 31.27

Table 3.3: Total throughputs for link-fault extension with one link fault (Mbps).Instance Ave. Max. Min.

Square field 12.4 12.9 10.9Library field 23.37 23.74 23Office field 27.28 28.4 27

For reference, we measure the CPU time of the algorithm to find the AP-fault dependableAP allocation for each instance on a Core 2 Duo processor with the 4GB memory and Ubuntu8.04.2 Linux-OS. Table 3.5 shows the CPU time.

3.6 Related Works

Several studies have been reported for the dependability in multihop wireless networks in-cluding wireless mesh networks. This section briefly introduces some of them.

In [26], Gupta et al. presented efficient detection and recovery mechanisms of one failedgateway or its link in a clustered wireless sensor network. The detection is based on theconsensus of healthy gateways. The recovery re-associates the sensors that are managed bythe failed gateway to other clusters based on the range information. The effectiveness isverified through simulations.

In [27], Varshney et al. presented the multilevel fault tolerance design of wireless net-works using adaptable building blocks (ABBs). The ABB has several levels of componentssuch as base stations, base station controllers, databases, and links, similar to cellular net-works, where the reliability such as MTBF/MTTR can differ significantly by using differentnumber of components. The fault tolerance design is achieved at the three levels of thecomponent and link, the building block, and the interconnection. If the computed depend-ability attributes are not acceptable, the process of adding the incremental redundancy atthe three levels is repeated. They present an analytical model of measuring the dependabil-ity enhancement, and evaluate the network survivability and the network availability withdifferent interconnection architectures, block-level redundancy, mobility, and fault tolerance

Table 3.4: Total throughputs for AP-fault extension with one AP fault (Mbps).Instance Ave. Max. Min.

Square field 12.31 12.6 11.1Library field 21.75 22.65 21Office field 30.57 31.27 29.03

22

Table 3.5: CPU time (minutes).Instance Time (m)

Square field 114Library field 38Office field 11

at the three levels in ring, star, and SONET dual ring topologies.In [28], Pan et al. studied detection and repair methods of faulty APs for large-scale

wireless networks. For the detection, they presented three algorithms. The first one is thatif an AP gives reports to the network operation center, it is regarded as no fault. Thesecond one modifies the first one such that the no-fault probability of an AP is exponentiallydecreased as the time interval of no report increases. The third one further improves it byconsidering the path of APs that the host is moving along, where if an AP along the pathdoes not report, it can be regarded as a fault. For the repair, they presented the ellipseheuristic algorithm to find the best schedule of repairing faulty APs by minimizing the totalmoving length and the downtime of popular APs. They evaluate their proposal using the freedata set available from Dartmouth College that includes log messages from client association,authentication, and others in their wireless networks for nearly four years.

3.7 Conclusion

This chapter has presented dependability extensions of the AP allocation algorithm for wire-less mesh networks such that one link fault or one AP fault does not impair the connectivityof the network. Our algorithm extensions find the link-fault or AP-fault dependable APallocations separately, where it seeks the minimization of the number of additional APs andthe maximization of the throughput. The effectiveness of our proposal is verified throughnetwork simulations, where the dependability is realized with a small number of additionalAPs and the network performance is not degraded for one link or one AP fault.

23

Chapter 4

Route Availability Index

In this chapter, we present the route availability index to evaluate the dependability of anAP allocation for WIMNET by estimating the certainty of connections between hosts andthe Internet gateway.

4.1 Introduction

To evaluate the dependability of WIMNET quantitatively, it is better to introduce an indexrepresenting how many hosts can be connected with the GW on average, after assigningevery AP the probability of functioning properly and every host the probability of accessingto the Internet. When the number of APs is fixed, this index should be maximized forthe better topology where more hosts can be covered on average. Therefore, we proposethis route availability (RA) index in WIMNET in this chapter. Using the four instances,we verify the effectiveness of the RA index by comparing the values between the originalAP allocations and their dependable extensions by the algorithm in the previous section.Besides, we verify the validity of the RA index through simulations.

The rest of this chapter is organized as follows: Section 4.2 introduces the RA index.Section 4.3 discusses performance evaluations. Section 5.6 provides the conclusion.

4.2 Route Availability Index

In this section, we introduce the route availability (RA) index for WIMNET to representhow many hosts can be connected with the GW on average under a given set of probabilitiesfor the proper functioning of APs and the Internet accessing of hosts. If the number of APsis fixed, the better AP allocation in terms of the dependability should have the larger value.Thus, the RA index can be used to evaluate the dependability of WIMNET.

4.2.1 Definition of Route Availability Index

To calculate the RA index, we introduce several input parameters and rules so that it can becalculated in a short time. For the inputs, the network topology is described by the graphG = (V1, V2, E). The i-th node vi ∈ V1 represents the i-th AP (APi), the i-th node vi ∈ V2

24

Figure 4.1: Serial path and parallel path.

represents the i-th host (HSi), and the edge between two nodes represents the wireless linkbetween two APs or between an AP and a host. The i-th AP is assigned ai (0 ≤ ai ≤ 1) torepresent its proper functioning probability, and the i-th host is assigned pi (0 ≤ pj ≤ 1) torepresent its Internet accessing probability.

To calculate the RA index, the route availability probability ri (0 ≤ ri ≤ 1) must becalculated for APi. The route availability probability for an AP represents the probabilitysuch that at least one of all the shortest paths from the GW is available under a given setof proper functioning probabilities of APs. Here, to enumerate all the shortest paths, wemodify the Dijkstra algorithm.

Then, we define how to calculate the route availability probability of the serial path thatis composed of a series of APs, and that of the parallel path where two or more paths aremerged into one as shown in Figure 4.1. The route availability probability of the serial pathof two APs, APi and APj, is given by the multiplication of their probabilities ai · aj. Theroute availability probability of the parallel path from APi and APj merging into APk isgiven by (1 − (1 − ri · ak)(1 − rj · ak)). If a host can be associated with multiple APs, theroute availability probability for this host follows the same calculation as the parallel path.

Then, the RA index is given by the following equation:

25

RA =M∑

i=1

(1− ∏

j∈Si

(1− rj · pi)) (4.1)

where M represents the number of hosts, and Sj represents the set of APs that the j − thhost HSj can be associated with.

4.2.2 Procedure for RA Index Calculation

The RA index calculation procedure is described as follows.

1. Find all the shortest paths from the GW to every AP.

2. Calculate the route availability probability for every AP by following the procedure in4.2.1.

3. Calculate the RA index in Eq. (4.1).

4.2.3 Example of RA Calculation

By using the sample WIMNET in Figure 4.2, we explain how to calculate the RA index.There is only one shortest path from the GW to AP1, (AP1, AP2, AP3, GW). Thus, r1 isgiven by (a1 · a2 · a3 · aGW ). For AP4, there are two shortest paths from the GW, (AP4, AP2,AP3, GW) and (AP4, AP5, AP6, GW). Thus, r4 is given by (1− (1− a4 · a2 · a3 · aGW ) · (1−a4 · a5 · a6 · aGW )).

4.3 Evaluation

In this section, the RA index is evaluated through simulations in four instances.

4.3.1 Simulated Instances

The three network instances used in Chapter 3 and an additional instance with 6× 6 roomsare adopted in this chapter. Later, we mention the last topology as the 6 × 6 field. Thisfield is composed of 36 square rooms with 60m sides. In each room, 25(= 5× 5) host pointsare allocated with the 10m interval, where each host is associated with one host. The 16host points along the walls are selected as battery points, assuming that electrical outletsare installed on walls. The battery point near the center is selected as the GW. A total of900 host points are distributed regularly in the field. The maximum load constraint L is set25 so that the lower bound on the number of allocated APs is 36 to cover the host pointsby exactly one AP by 900/25 = 36. After applying the AP-fault dependability extension inSection 3.4, the number of installed APs becomes 57.

26

.

Figure 4.2: Sample WIMNET for RA calculation.

4.3.2 Route Availability Index Results

Then, we investigate changes of RA indices for the AP allocations found by the AP allocationalgorithm in [9] and by the dependable AP allocation algorithm with the improvementstage in this paper for the two instances. We vary the proper functioning probability ofeach AP from 0.5 to 1, whereas we fix the Internet accessing probability of each host to0.85. Figure 4.3(a) and (b) illustrate the RA index results for Square filed and Office fieldrespectively. The RA index is increased by about 14% for Square filed and by about 12.5%for Office field on average.

Figures 4.4 (a) and (b) illustrate the RA index results for Central Library filed and 6× 6field respectively. As noticed in Figure 4.4, the average increase of the RA index in AP-faultdependable over AP-allocation is 0.9%. The reasons for these results, first the added numberof APs in AP-fault dependable is only three APs. Second, due to the open space, each APcan reach the GW through maximum two hop counts. For 6 × 6 field, the RA index isincreased about 13% on average.

4.3.3 Route Availability Index Validation

In this subsection, we verify the validity of the RA index in Eq. (4.1) through simulations.We generate 100 random network instances for each of the above fields. In each instance,

27

(a)Square field (b) Office field

Figure 4.3: RA index changes.

we randomly select failed APs with the probability 0.9. In addition, we randomly select theInternet connected host with the probability 0.85. After that, the routing tree of the workingAPs and connecting hosts is created using the shortest path method. Then, we calculate thenumber of hosts that can connect to the Internet through WIMNET.

In the following results, we compare the RA index in Eq. (4.1) with the average numberof connecting hosts among the generated 100 instance. Figure 4.5(a) shows the validationof the RA index for Square filed. The difference between the RA index in Eq. (4.1) and thecorresponding value from simulations is 3.8% for the conventional AP allocation and 4.3%for the AP fault-dependability allocation. Figure 4.5(b) shows the RA index validation forOffice field. As noticed, the difference between the RA index and the simulation result is2.9% for the AP allocation and 3.1% for the AP fault-dependability allocation. The sameobservation can be seen in Figure 4.5(c).

However, for 6 × 6 field in Figure 4.5(d), the difference between the RA index and thesimulation result is almost 11% for the AP allocation and 4.3% for the AP fault-dependabilityallocation. One reason of this degradation comes from the number of hop count. In thisfield, most of the APs can reach the GW through 7 or 8 hops. As a result, the number ofshortest paths is increased, which also increases the calculation error of the RA index.

4.4 Conclusion

This chapter introduced the route availability (RA) index to evaluate the dependability ofthe AP allocation by estimating the certainty of connections between hosts and the Internetgateway. The calculation of the RA index is performed in four instances. The effectiveness ofour proposal is verified through simulations of generating network instances with randomlyselected failed APs.

28

(a)Central Library filed (b) 6× 6 field

Figure 4.4: RA index changes.

(a)Square field (b) Office field

(c)Central Library filed (d) 6× 6 field

Figure 4.5: Validation of RA index.

29

Chapter 5

Dynamic Channel AssignmentTechnique

In this chapter, we present the dynamic channel assignment technique for WIMNET [29].

5.1 Introduction

In WIMNET, each AP takes two different roles in wireless communications. One is for com-munications between APs and hosts, and another is for communications between two APs.To reduce the radio interference between these communications, we use different protocolswith different channels (radio frequencies) for them. Actually, the IEEE 802.11b/g protocolis assigned to the first role with the 2.4 GHz unlicensed frequency band. The IEEE 802.11aprotocol is assigned to the second role with the 5.4 GHz unlicensed frequency band. How-ever, this protocol separation by the roles of APs is not sufficient to allow the scale-up ofWIMNET while maintaining the performance, because more and more traffics concentrateinto GWs as the increase of associated hosts.

In order to afford the increase of traffic capacity, the efficient use of limited channelsis very important by assigning proper channels. This is one of the fundamental problemsin WIMNET. There are two strategies for assigning channels to APs: the Fixed ChannelAssignment (FCA) and the Dynamic Channel Assignment (DCA) [30]. The FCA strategystatically allocates channels to APs in advance, according to estimated traffic intensities inthem, as in [4]. The DCA strategy may change channel assignments dynamically in realtime, to meet changing traffic intensities. In DCA, however, channel reassignments shouldbe avoided as best as possible to minimize the stopping time of network operations duringchannel reassignment procedures. In general, DCA exhibits better capacity, and betterhandoff performance in terms of lower forced termination. This is the motivation of ourstudy of the DCA problem for WIMNET in this paper.

Our contribution in this chapter is the proposal of the DCA problem formulation andthe DCA technique through modifications of our existing study for the FCA problem in[4]. Here, we newly define a decision function for the channel reassignment. Our techniqueis divided into two stages: the initial stage and the dynamic stage. In the initial stage, aproper number of network interface cards (NICs) with proper channels are assigned to APs,

30

using the information of the maximum number of hosts associated with each AP, such thatthe network performance of WIMNET is maximized, while the NIC configuration cost isminimized according to [4].

In the dynamic stage, the decision function decides whether the channel reassignmentshould be done or not, according to the traffic balance between the links adjacent to GW,because the most traffic to/from the Internet passes through one of them. This decisionfunction is designed to minimize the number of channel reassignments to reduce the stoppingtime of network operations during channel reassignment procedures. If the function returnsyes, the channel is reassigned to the links using the FCA algorithm so that the performance ofthe network is maximized. The effectiveness of our approach is evaluated through simulationsusing the WIMNET simulator [3].

5.2 Formulation of Dynamic Channel Assignment Prob-

lem

In this section, we formulate the DCA problem for WIMNET as a combinatorial optimizationproblem. This problem aims both of the maximization of the channel utilization and theminimization of the number of channel reassignments to satisfy the demanded traffic.

5.2.1 Input

First, we describe the inputs to the DCA problem for WIMNET.

• The AP network topology graph, G = (V, E):

– A node vi in V represents the i-th AP (APi), where the number of nodes isrepresented by N .

– An edge eij ∈ E represents the wireless link between APi and APj.

– One node is designated as the GW to the Internet, APg.

– The communication route to the GW is selected for APi, Pi.

• The AP routing subgraph, R = (V, ER):

– An edge in ER(⊆ E) is used in at least one communication route.

• The link interference matrix, D:

– The radio interference between two neighbor links is represented by the link in-terference matrix D.

– The element dijpq in D is 1 if the link from APi to APj is interfered with the linkfrom APp to APq, and 0 otherwise.

• The channel interference matrix C:

– The interference between two channels is described by the channel interferencematrix C.

31

– The element c(x, y) in C is 1 if x = y, and 0 otherwise for the IEEE 802.11aprotocol with orthogonal channels [31].

• The maximum number of hosts associated with APs, H:

– The expected maximum number of hosts associated with each AP is given asthe input, because it is usually necessary to design the Internet access networkappropriately.

– The element hi in H represents the number for APi.

– The expected traffic, tij, through the link from APi to APj for i = 1, ..., N andj = 1, ..., N , is calculated from the number of hosts associated with each AP,assuming that every host has the same amount of traffic to/from the Internet.

• The design constraints of WIMNET:

– The total number of NICs, B.

– The maximum number of NICs at APi, bi,

– The number of channels, M .

5.2.2 Output

• The number of NICs assigned to APi, xi.

• The channel assigned to the link between APiand APj, yij.

5.2.3 Constraint

This problem must satisfy the following constraints:

• The total number of NICs must be B or smaller:N∑

i=1xi ≤ B.

• The number of NICs at APi must be positive, and bi or degi or smaller, where degi isthe number of incident edges to APi in R: 1 ≤ xi ≤ bi and 1 ≤ xi ≤ degi.

• The channel must be feasible: 1 ≤ yij ≤ M .

• The channel at the both directions of any link must be identical: yij = yji.

• The number of different channels assigned to the links incident to APi must be xi orsmaller.

32

5.2.4 Objective

This problem aims to minimize the following objective functions ENIC , Elink and Estop:

ENIC = maxi

{∑j∈Ni

(tij + tji)

xi

}(5.1)

where Ni represents the set of APs adjacent to APi.

Elink =N∑

i=1

N∑

j=i+1

N∑

p=1

N∑

q=p+1

tij · tpq · dijpq · c (yij, ypq) (5.2)

Estop = (# of channel reassignments) (5.3)

ENIC seeks the minimization of the maximum load per one NIC, Elink seeks the min-imization of interfered traffics among neighbor links, and Estop seeks the minimization ofchannel reassignments.

5.3 FCA Algorithm

In this section, we review the two-stage algorithm for the channel configuration problem forAP communications in WIMNET in [4], which is composed of the NIC assignment stage andthe channel assignment stage.

5.3.1 NIC Assignment Stage

The NIC assignment stage repeats one NIC assignment to one AP where the traffic per NIC(NIC traffic, tNIC) is maximum while the constraints are satisfied, so as to minimize ENIC .

Initialization

(1) Calculate the traffic per AP from the link traffic tij:

tAPi =

∑

j∈Ni

(tij + tji). (5.4)

(2) Assign one default NIC to every AP: xi = 1.

(3) Calculate the NIC traffic:

tNICi =

tAPi

xi

. (5.5)

(4) Initialize the number of assigned NICs: X=N.

33

Sequential NIC Assignment

(1) Assign one NIC to an AP (let APk) such that tNICk is maximum with xk < bk. Then,

xk + + and X + +.

(2) Terminate this stage if X = B or every AP reaches the upper limit (xk = bk).

(3) Update the NIC traffic:

tNICk =

tAPk

xk

. (5.6)

(4) Go to step (1).

5.3.2 Channel Assignment Stage

The channel assignment stage basically repeats one channel assignment to one link so as tominimize Elink.

1. Calculate the traffic collision:

colij = tij ·N∑

p=1

N∑

q=p+1

tpq · dijpq. (5.7)

2. Sort all the links in descending order of traffic collision.

3. Assign channel 1 to the first link (the most congested link) and NICs at its both endAPs.

4. Assign channels to links by the following procedure:

(a) Channel assignment to links without choice:

(i) When the same channel is assigned to NICs at both end APs of a link, thischannel is assigned to the link. Here, if two or more such channels exist, thechannel to minimize Elink is selected among them, where unassigned links arenot considered for its calculation.

(ii) When one channel is assigned to the sole NIC at one end AP and no channelis assigned to the NICs at another end AP of the link, this channel is assignedto the link and the NIC.

(b) Channel assignment to links with choice:

(i) When all the NICs at either end AP of the link are assigned channels, thechannel to minimize Elink among them is assigned to the link and the NIC.Note that unassigned links are not considered for the calculation of Elink.

(ii) When both end APs of the link have at least one unassigned NIC, the channelto minimize Elink among all of M channels is assigned to the link and theNICs.

34

(c) Assignment priority change for impossible links :

When different channels are assigned to the NICs at both end APs of an unas-signed link, no channel can be assigned there. At this case, the assignment priorityof such a link is increased by multiplying its traffic collision by a constant (> 1),and repeat this stage from 2).

(d) NIC assignment movement for unused NIC :

When some NICs are not assigned any channel after the completion of the channelassignment to every link, such NICs are moved to different APs satisfying theconstraints, and repeat this stage from 2).

5.4 DCA Technique

In this section, we propose a dynamic channel assignment technique for the DCA problem,which is composed of the initial stage with the NIC assignment and the initial channelassignment, and the dynamic stage with the decision function for changing the channelassignment.

In the initial stage, we use the algorithm in the previous section to configure the numberof NICs per AP. Here, the traffic is computed by using the expected maximum number ofhosts associated to each AP that is given as inputs.

In the dynamic stage, we present the decision function for the channel reassignment. Theidea of this decision function comes from the fact that the number of links incident to GWis usually limited even in a large-scale WIMNET, and the traffic of each such link shouldevenly distributed among them, so as to avoid that they become bottlenecks of the wholecommunications in WIMNET, where most traffics must pass through one of these limitedlinks to access to the Internet. When the evenness of them is not satisfied by the currentchannel assignments, they should be changed. The decision function is actually defined asfollows.

First, we calculate the total traffic of the links incident to GW using the same channelp, GTp:

GTp =∑

egi∈ GL,ygi=p

tgi (5.8)

where GL represents the set of the links incident to GW.Then, we calculate the decision factor:

maxp,q

|GTp

GTq

− 1| ≥ δ (5.9)

where p and q represent any pair of the channels assigned to the links in GL. Each channelhas a traffic capacity assigned to it. δ represents the unbalance threshold of the bandwidthconsumption for each channel. δ should be given empirically by considering the networktopology, the traffic pattern, and the number of NICs assigned to GW. If the decision functionreturns ”Y es”, the channel reassignment is applied by using the FCA algorithm in theprevious section. If the decision function returns ”No”, the last channel assignment iscontinued.

35

Figure 5.1: Seven patterns of numbers of associated hosts with one AP.

5.5 Performance Evaluation by Simulator

5.5.1 Simulated Instances

Simulations for the DCA technique are executed with three different schemes. In the staticchannel assignment scheme (Static), we obtain the static channel assignment by using themaximum number of hosts associated to each AP, and fix this channel assignment to anytraffic change. In the always changing channel assignment scheme (Always Change), when-ever the traffic is changed, we change the channel assignment. In the dynamic channelassignment scheme (Dynamic), the decision function decides either to change the channelassignment or to keep the last channel assignment.

In our simulations, the expected maximum number of hosts associated to each AP is givenas the input. The host distribution among 24 discrete times for each AP is randomly selectedfrom the seven patterns of the number of associated hosts with each AP shown in Figure5.1. n represents the maximum number of associated hosts. The maximum number of NICsat one AP bi is set 2 or 3. Because the parameter δ depends on the network topology and themaximum number of NICs in/around the GW, we simulate two different network topologies,and summarize their simulation results with various values for δ. The throughput and thenumber of channel assignment changes are observed under different traffic load patterns tothe network.

36

Figure 5.2: Average throughputs and numbers of channel changes for 3 × 3 network withmax. two NICs.

5.5.2 Simulation Results in 3× 3 (N=9) Network Topology

In our first simulations, 3 × 3 (N = 9) APs are regularly placed with the 100m interval ona 300m × 300m field. The coverage distance from a host/AP is set 100m. Thus, only theadjacent APs on the left, right, top and bottom can be communicated with each other. Thecenter AP on the field is selected as GW to the Internet.

In Figure 5.2, where the maximum number of NICs per AP is 2, we can notice that there isa proportional relationship between the average throughput and the number of reassignmentchanges in our technique. For example, in the case with δ = 0.1, the number of changesover 24 times is 11.29, and the throughput is almost the same as that of Always changing.In the last case with δ = 0.8, we found that the number of changes is 2.14 over 24 times,and the throughput of our scheme is still better than that of Static. δ for a specific networkshould be carefully selected in order to decrease the number of changes and to enhance thethroughput at the same time.

Figure 5.2 shows that, for the 3 × 3 network with maximum two NICs, the best case ofδ that makes a balance between the number of changes and the throughput is 0.5.

Figure 5.3 shows the results for the 3 × 3 network with maximum three NICs. Here, wenotice that the best value for δ is 0.9 to decrease the number of changes and to enhance theperformance of the network.

5.5.3 Simulation Results in 5× 5 (N=25) Network Topology

In our second simulations, 5 × 5 (N = 25) APs are regularly placed with the 100m intervalon a 500m × 500m field. Figure 5.4 illustrates the simulation results with maximum two

37

Figure 5.3: Average throughputs and numbers of channel changes for 3 × 3 network withmax. three NICs.

NICs. As noticed, the proper range of δ varies from the 3 × 3 network. The best choice ofδ that improves the performance of the whole network and decreases the number of changesis 0.25, where 5.8 changes are taken place over 24 times on average, and the performance ismuch better than that of the static scheme.

Figure 5.5 shows the results for the 5 × 5 network with maximum three NICs. We cansee that the best value of δ is 0.5. In this figure, when we notice the last case with δ, =0.8, we need to say that some value of δ may lead to worse the performance of the network.Thus, the mathematical formula of calculating the proper value for δ is in our future studies.

5.6 Conclusion

This chapter has presented the formulation of the dynamic channel assignment (DCA) prob-lem for WIMNET, and a DCA technique composed of the initial stage and the dynamic stagethrough modifications from the existing study with a newly defined decision function. Theeffectiveness of our approach is verified through network simulations using the WIMNETsimulator. The significant performance improvement is observed by choosing an appropriateδ with the decreasing number of changes. In future works, we will present a mathematicalformula for δ, and implement APs with multiple NICs for AP communications so that weapply our DCA technique to justify our approach in the real world.

38

Figure 5.4: Average throughputs and numbers of channel changes for 5 × 5 network withmax. two NICs.

Figure 5.5: Average throughputs and numbers of channel changes for 5 × 5 network withmax. three NICs.

39

Chapter 6

Secure Multi-hop Routing

In this chapter, we discuss security challenges in multi-hop wireless networks that are relevantto the wireless mesh network including WIMNET. Especially, we focus on secure multi-hoprouting protocols as the most fatal security challenge in the wireless mesh network.

6.1 Security Technology Overview

The wireless mesh network including WIMNET is basically exposed to the same fundamentalcommon threats for both the wired and the wireless networks. Messages can be intercepted,modified, delayed, and replayed. Even new messages can be inserted. The network resourcescan be made unavailable by denial of service (DoS) attacks.

Therefore, the following security requirements [32] [33] must be considered in the wirelessmesh network.

AvailabilityThe authorized actions can take place at any time. The availability also ensures thesurvivability of the network despite of attacks and failures. The availability can becompromised by following ways.

• Signal jamming is the (usually deliberate) transmission of radio signals thatdisrupt communications by decreasing the signal to noise (SN) ratio.

• Denial of Service (DoS) attack can be launched at any layer of a wireless meshnetwork. Firewall rules could be adjusted to stop any request from certain ad-dresses or networks. However, modern attacks use ”zombies” systems in all overthe world. This attack is called distributed DoS (DDoS), and it is nearly impos-sible to counter it. An intrusion detection system and/or an intrusion preventionsystem have been deployed to monitor these attacks.

• Battery exhaustion attack, also known as ” sleep deprivation attack”, is areal threat and is more hazardous than simple DoS attacks. Attacks on CPUcomputation may deny the availability of the service while battery exhaustioncan disable the victims.

40

AuthenticationAuthenticity enables a node to ensure the identity of the peer node that it is communi-cating with. Without the authenticity, unauthorized access to resources and sensitiveinformation, and interfering with the operation of other nodes are increased.

IntegrityIntegrity ensures that the contents of data or correspondences are preserved intactthrough the transfer from the sender to the receiver. Attacks on Integrity are usuallymade in two ways:

1. by the intentional alteration of the data for vandalism or revenge, or

2. by the unintentional alteration of the data caused by either the operator inputerror, the computer system error, or the faulty application error.

Therefore, the integrity ensures that data cannot be modified without being detected.

ConfidentialityConfidentiality is the assurance that any sensitive data is being accessed and viewedonly by those who are authorized to see it. For the confidentiality, the authenticityneeds to be implemented first. It is pointless to protect the secrecy of a communicationwithout first ensuring that one is talking to the right principal.

Non-repudiationThe sender of a message cannot later deny the transmission of any information, or thereceiver cannot deny the reception of it. It means preventing the denial of previouscommitments or actions.

Further security services include the user privacy (anonymity, pseudonymity, user profiling,and tracing), and the traffic flow confidentiality.

The protection of the communication traffics involves the confidentiality by the encryp-tion, the authentication of the communication partners, the protection of the integrity, andthe authenticity of the exchanged messages. The integrity protection refers not only to theintegrity of a single message, but also to the correct order of the related messages. Thismeans the avoidance of the replay, the reordering, or the deletion of messages. They shouldalso be used within the wireless mesh network in authenticating the mesh nodes and es-tablishing the session keys that protect the confidentiality and the integrity of the trafficsexchanged between them.

6.2 Security Challenges

The wireless mesh network is extremely vulnerable to attacks due to their dynamicallychanging topology, the absence of conventional security infrastructures, and the open mediaof communication. Unlike their wired counterparts, they cannot be secured. In this section,a brief overview of the security challenges in a secure wireless mesh network is presented.Theses major challenges demand a set of cross-layer, self-adapted security mechanisms.

41

Like a wireless network, the shared wireless medium makes it easy for attackers to launchjamming attacks, to eavesdrop the communication between the mesh router, and to injectmalicious information into the shared medium as discussed in sec. 6.1. Given the fact that thecorrectness of the routing messages is fatal to achieve the wireless multi-hop routing in thewireless mesh network, the most harmful kind of the malicious information is the fabricatedrouting message. Accordingly, the first security challenge that should be considered is thesecure multi-hop routing.

Second, the wireless mesh network may be dynamic because of changes in both its topol-ogy and its membership. The lack of association is an important security challenge. In otherwords, the mutual authentication of the network nodes has to be made. In the wireless meshnetwork, APs and hosts hold significantly different characteristics such as the mobility andthe power constrains. As a result, there are two authentication types, the authentication ofthe APs at the initialization (or re-initialization) phase, and the authentication of the hostsduring the session established by the hosts.

Third, the wireless mesh network should consider the physical vulnerability. The meshrouters are located outdoor, usually on roof tops or traffic light poles. They are not physicallyprotected like the wired routers and wireless LAN access points. Therefore, it is much easierfor attackers to capture the AP and get full control of the device.

6.3 Multi-hop Routing Protocols and Their Attacks

In this section, a brief overview of the routing protocols in ad hoc networks is presented.Then, the attacks on these protocols are described below.

6.3.1 Multi-hop Routing Protocols

The routing protocols for a wireless ad hoc network have similar requirements such as therouting through multi-hop links, the self-configuration, and the self-adaptation. Althoughvery few routing protocols have been proposed specifically for the wireless mesh network,the similarities between the wireless mesh network and the wireless ad hoc network make itfeasible for the wireless mesh network to borrow the ideas from the domain of wireless adhoc networks, which have been extensively studied in the literature.

The routing is a fundamental function in packet networks that is responsible for trans-ferring packets through the network from the source to the destination. A router takes tworoles. First is to acquire and maintain the routing information (routing tables), called the(control plane). Second is to forwards data packets, called (data plane). One way to classifythe ad hoc network routing protocols is illustrated in Figure 6.1 [34].

Topology-based protocols are based on traditional routing concepts, such as maintainingrouting tables or distributing the link-state information. Topology-based protocols canbe proactive or reactive.

• Proactive protocols try to maintain consistent, up-to-date routing informationwithin the system. Popular examples for such protocols include Destination-Sequenced Distance-Vector Routing (DSDV) and Optimized Link State Routingprotocol (OLSR).

42

Figure 6.1: Classification of ad hoc network routing protocols.

• Reactive protocols establish a route between a source and a destination onlywhen it is needed. Popular examples for such protocols include Dynamic SourceRouting (DSR) and Ad-hoc On-demand Distance Vector routing (AODV).

Position-based protocols use information about the physical locations of the nodes toroute data packets to their destinations. Examples for such protocols include GreedyPerimeter Stateless Routing (GPSR), Greedy Other Adaptive Face Routing (GOAFR).GOAFR uses greedy forwarding packets. Distance Routing Effect Algorithm for Mo-bility (DREAM) and Location-Aided Routing (LAR) are restricted directional floodingalgorithms.

Hybrid approaches try to combine the advantages of both proactive and reactive pro-tocols Typically, hybrid protocols use the proactive approach to update the routinginformation at each node regarding the node’s local neighborhood, and use the reactiveapproach when routes to far away destinations are needed.

6.3.2 Attacks on Multi-hop Routing Protocols

Some aspects of the wireless mesh network have interesting security problems as mentionedin 6.2. The routing is one such aspect. The routing protocols require the nodes to exchangetheir routing information within the neighborhood to make efficient routing decisions. As aresult, the routing protocols for the wireless mesh network are designed to achieve:

• self- configuration of the routing tables,

• self-adaptation to changes in the wireless link quality, and

• maximization of the performance metrics such as end-to-end delay, throughput andpacket loss rate.

43

In this subsection, we discuss the attacks and threats on multi-hop routing protocols.There are two types of attacks, passive attacks, and active attacks.

Passive AttacksIn a passive attack, the attacker does not disrupt the operation of a routing proto-col. It only attempts to discover the valuable information by listening to the routingtraffic. The major advantage for the attacker in a passive attack is that in a wirelessenvironment, this attack is usually impossible to detect. This also makes defendingagainst such attacks difficult. An attacker tries to discover information by listeningto network traffic. The routing data can reveal information about the relation andlocation of nodes, and about the network topology in general.

• Eavesdropping:In Eavesdropping, an attacker tries to discover information by listening to networktraffic. Routing data can reveal information about the relation and location ofnodes, and about the network topology in general.

Active AttacksTo perform an active attack, the attacker must be able to inject arbitrary packets intothe network. The goal of this attack may be to attract packets destined to other nodesto the attacker for analysis or just to disable the network. A major difference from thepassive attacks is that an active attack can sometimes be detected

Next, we present several types of active attacks that can usually be easily performedagainst an ad hoc network. [35] [36]:

• Sinkhole, Wormhole:If an attacker gets control of two nodes with a wired communication link (tunnel)between them, the wormhole attacks could be launched by sending all the packetsreceived from one node through the tunnel and replaying these packets at the otherend of the tunnel. Then, the attacker can choose to drop the packets to performthe DoS attack (black hole), or selectively forward the packets (grey hole), oralternatively use its place on the route as the first step in a man-in-the-middleattack. Figure 6.2 shows an example of the wormhole attack. Because the packetsthrough the tunneled link (A → B) arrive sooner than the packets through anyother multi-hop wireless links, any other intermediate nodes are excluded from thenetwork, and the traffic between the source and destination nodes is completelyunder the control of the attacker.

• Routing Table Overflow:The attacker attempts to create routes to nonexistent nodes. The goal of thisattack is to create enough routes to prevent new routes from being created or tooverwhelm the protocol implementation.

• Rushing Attack:The rushing attack is a malicious attack that is targeted against on-demand rout-ing protocols such as AODV or DSR [34]. To limit the overhead of ROUTEREQUEST (RREQ) flooding, each node typically forwards only the first RREQ

44

Figure 6.2: Wormhole attack by colluding nodes A and B.

originating from any route discovery. An attacker disseminates RREQs quicklythroughout the network, suppressing any later legitimate RREQs when nodesdrop them due to the duplicate suppression. Then, the attacker can increase theprobability that routes including him will be discovered rather than other validroutes. This type of attack becomes an effective DoS attack against on-demandrouting protocols.

• Byzantine Attack:Attacks where adversaries have full control of a number of authenticated devicesand behave arbitrarily to disrupt the network are referred to as Byzantine attacks.Traditional secure routing protocols are vulnerable to this class of attacks, sincethey usually assume that once authenticated, a node can be trusted to executethe protocol correctly [37].

• Sleep Deprivation:The sleep depravation torture is briefly introduced in [38]. A malicious usermay interact with a node in a legitimate way, but for no other purpose thanto consume its battery energy. The battery life is usually a critical parameter.Battery powered devices try to conserve the energy by transmitting only whennecessary. An attacker can attempt to consume batteries by requesting routes, orby forwarding unnecessary packets to the node using, for example, a black hole

45

attack.

• Location Disclosure:The location disclosure attack can reveal something about the locations of nodesor the structure of the network. The information gained might reveal which othernodes are adjacent to the target, or the physical location of a node.

Table 6.1 summarizes some attacks on routing protocols.

Table 6.1: Attacks on Routing.Routing Phase Security Attack

Routing Discovery Phase Routing table overflow attack,Location disclosure attack

Route Maintenance Phase False Route Control Message (Rushing Attack)

Data forwarding phase Route Data Dropping

Advanced / sophisticated attacks Blackhole/sinkhole attack,Byzantine attack, Rushing attack,

Resource consumption (Sleep deprivation attack),Location disclosure attack

6.4 Securing Multi-hop Routing Protocols

In the previous section, we illustrated types of attacks against ad hoc network routing pro-tocols. In this section, we discuss some techniques that can be used to defend them fromthose attacks. To protect routing messages from attacks, the routing protocols need effectivemechanisms to:

• to authenticate the received routing message to validate that it is sent by a legitimatenode, and

• to check the integrity of the received routing message to validate that it has not beenaltered by the attacker.

6.4.1 Cryptographic-based Solutions

Security of the routing can be enhanced using cryptographic measures of protecting the in-tegrity and the authenticity, and the confidentiality of routing messages (routing packets).Cryptographic measures prohibit the adversary from manipulating or even intercepting rout-ing messages, but they require some real-world trust relations to distinguish ”trustworthy”and ”malicious” nodes [39]. In this subsection, we illustrate some examples of secure routingprotocols.

46

Authenticated routing for ad hoc networks (ARAN)The ARAN, proposed in [40], detects and protects the network against malicious actions. Itutilizes cryptographic certificates to achieve the authentication, the message integrity, andthe non-repudiation in an ad-hoc environment. Every node that forwards RREQ or RREPmust sign it. Table 6.2 illustrates the operation and message format of ARAN. S and Dare identifiers of the source and the destination, respectively. F1 and F2 are identifiers ofthe intermediate nodes. n is a nonce generated by S, and t is the current timestamp whengenerating the request. certX , and sigX are the public key certificate and the digital signatureof X, respectively. All signatures are generated on message fields that precede the signature.The source begins the route discovery by broadcasting a RREQ. n, t, and sigS togetheruniquely identify the message, and they are used to detect and discard duplicates of thesame RREQ. When D receives the first RREQ, it performs the verification, and updates itsrouting table. Then, it sends a RREP to S. ARAN is indeed proven to be secure. However,this method requires heavy weight computations, and the size of the message increases ateach hop.

Table 6.2: Operation example and message formats in ARAN.S −→ ∗ : (RREQ, D, certS, n, t,sigS )F1 −→ ∗ : (RREQ, D, certS, n, t,sigS,sigF1 , certF1)F2 −→ ∗ : (RREQ, D, certS, n, t,sigS,sigF2 , certF2)D −→ F2 : (RREP, S, certD, n, t,sigD )F2 −→ F1 : (RREP, S, certD, n, t,sigD,sigF2 , certF2)F1 −→ S : (RREP, S, certD, n, t,sigD,sigF1 , certF1)

Secure Routing Protocol (SRP)SRP [34] is a secure variant of DSR. It is based on the symmetric-key authentication (MAC-Message Authentication Code). SRP requires that for each route discovery, only the sourceand the distention share a key. Therefore, only the end-to-end authentication is possible.SRP introduces another principle: no optimization. This means that, the intermediatenodes do not send replies to route discovery messages and they do not cache informationfrom overhead control packets. Table 6.3 illustrates the operation and message format ofSRP. S and D are identifiers of the source and destination nodes, respectively. F1 and F2

are identifiers of the intermediate nodes. id is a randomly generated query identifier, snis a query sequence number maintained by S and D. macX is the MAC generated by X.S generates a RREQ and broadcasts it to its neighbors. The integrity of this RREQ isprotected by macS that is computed with a key shared between S and D. Each intermediatenode that receives the RREQ for the first time appends its identifier to the request, andre-broadcasts it. When the RREQ reaches D, it contains the list of the identifiers of theintermediate nodes. This list is considered as a route found between S and D. SRP is simple,but it does not prevent the manipulation of the mutable information added by intermediatenodes. This drawback opens the door for some attacks.

Secure Efficient Ad hoc Distance vector routing (SEAD)SEAD is a secure routing protocol based on the DSDV algorithm where the distentionsequence number and the hop count values are protected using one-way hash chain [41].SEAD tries to ensure that the sequence number cannot be increased and the hop count

47

Table 6.3: Operation example of SRP and format of the SRP messages.S −→ ∗ : (RREQ, S ,D, id, sn, macS, [] )F1 −→ ∗ : (RREQ, S ,D, id, sn, macS, [F1] )F2 −→ ∗ : (RREQ, S ,D, id, sn, macS, [F1, F2] )D −→ F2 : (RREP, S ,D, id, sn, macS, [F1, F2],macD )F2 −→ F1 : (RREP, S ,D, id, sn, macS, [F1, F2],macD )F1 −→ S : (RREP, S ,D, id, sn, macS, [F1, F2],macD )

values cannot be decreased. A one-way hash chain is built on the one-way hash function.Like a normal hash function, in a one-way hash function, H maps an input of any lengthto a fixed-length bit string. Thus, H :{0, 1}∗ → {0, 1}ρ, where ρ is the length in bits ofthe output of the hash function. The function H should be simple to compute yet must becomputationally infeasible in general to invert. Each node generates a one-way hash chainof length n + 1 using H, where n is a multiple of the maximum diameter m of the network(i.e. n = k · m). A node chooses a random initial value x ∈ {0, 1}ρ, and computes thelist of values h0, h1, h2, h3, ..., hn where h0=x, hi = H(hi − 1) for 0 < i ≤ n. It is assumedthat hn is securely distributed to all other nodes in the network. When a node S sendsout a route update message about itself with the sequence number i and the hop count 0,it reveals h(k−i)m in the same message. A neighboring node will update its routing tableentry that belongs to S by recording the sequence number i, the hop count value 1, andthe hash value H(h(k−i)m) = h(k−i)m+1 . Then, it sends out a route update message, andits neighboring node will record the sequence number i, hop count 2, and the hash valueH(h(k−i)m+1) = h(k−i)m+2 for S, and so on. Each route update concerning S is verified by thenodes using a previously known hash value from the hash chain of S. For example, node Dknows the sequence number i, the hop count c, and the hash value h = h(k−i)m+c for node S.D receives a route update for S with the sequence number j, the hop count c, and the hashvalue h. Then, D accepts this update only if H(i−j)m+c−c(h) = h, where Hz means that weinvoke H iteratively z times.

Secure ah hoc on-demand distance vector routing (SAODV)SAODV was proposed as a security extension to the AOVD [32]. SAOVD uses two mecha-nisms to secure routing messages: the digital signature to authenticate the non-mutable(thenode sequence numbers, the addresses of source and destination, request identifier) fields ofmessages, and the hash chain to secure the hop count information (the only mutable infor-mation). Hence, every node that receives the message (either an intermediate node or thefinal destination) can verify that the hop count has not been decremented by an attacker.There are four specific fields in the routing message that are used by the hop count protec-tion mechanism [35]: HopCount, MaxHopCount, Hash, and TopHash. Every time a nodeoriginates a RREQ or a RREP message, it performs the operations show in Figure 6.3.

6.4.2 Reputation-based Solutions

Watchdog and Pathrater [34] are two mechanisms that together attempt to detect andmitigate a gray hole (misbehaved nodes that do not forward data packets that they shouldforward). Watchdog is responsible for listening in the promiscuous mode and trying to catch

48

Figure 6.3: SAODV operations.

the transmission of data packets by the next node in the path. Pathrater assesses the resultsof the watchdog and selects the most reliable path for the packet delivery. In other words,Pathrater is used to select routes that likely avoid the gray holes.

6.4.3 Security Aware Routing

Security Aware routing (SAR) utilizes security metrics for the route discovery andmaintenance functions [39]. Figure 6.4 illustrates the battlefield communication scenariofor SAR. In this network, two commanders have a security path to themselves. During themission, the commanders detect that some of the civilians have defected or have becomecompromised. The commanders decide that they can only trust nodes owned by the marinesto route their packets. It turns out that the route through marines may be more secure butit may not be the shortest path. A discovered path that is passing through the marines iscalled a security aware route.

Technique for intrusion resistant ad hoc routing algorithm (TIRARA) [33] im-plements network layer survivability mechanisms for detecting and recovering from intruderinduced malicious faults that work in concert with existing ad hoc routing algorithms andaugment their capabilities. It works mainly against DoS attacks. Each node has a policythat defines the list of authorized flows that can be forwarded by the node.

49

Figure 6.4: Security aware routing.

6.5 Summary and Open Issues

This chapter has discussed several security issues, attacks, and challenge designs of thewireless mesh network in general. Especially, the problem of securing routing in multi-hopwireless mesh network is illustrated, notable in an ad hoc network. First, we gave an overviewof the security technology in the wireless mesh network. Then, we described the securitychallenges concerned with the wireless mesh network. After that, we described why and howad hoc routing protocols can be attacked. Following this, we presented general descriptionsof routing protocols for the wireless mesh network. Finally, we presented general methodsof how the multi-hop routing protocols can be secured via existing secure routing protocolsexamples.

Despite many research works that have been spent on wireless mesh networks and multi-hop networks, there are still challenging open issues as follows:

• There is no single efficient and reliable security solution suitable for the wireless meshnetwork since many of those solutions may be compromised due to vulnerabilities ofchannels and nodes in the shared media, the absence of reliable links to infrastructure,and the dynamic topology change.

• Attackers may launch man-in-the-middle and modification attacks against routing pro-tocols.

• Without strong authorizations, attackers may enter into the network, and impersonatelegitimate nodes, and may not follow the protocol rules.

• Mechanisms to authenticate using peer-based mutual authentication schemes need thesecurity analysis for the wireless mesh network.

50

• The group key management remains in a challenge by the absence of the central au-thority, the trusted third party, or the server to manage the keys. Some distributed andself-organizing key management schemes may be needed for the wireless mesh network.

51

Chapter 7

Conclusion and Future Works

In this thesis, we studied the access point (AP) allocation problems and their related issues torealize the dependable wireless Internet-access mesh network (WIMNET). For this purpose,we gave the background about WIMNET and the overview of the network model used inour study.

First, we formulated the link-fault and AP-fault dependable AP allocation problems forWIMNET, and presented their algorithms by extending the existing AP allocation algo-rithm. We verified the effectiveness of our approach through extensive simulations usingthe WIMNET simulator. Three network topologies were adopted as simulated instances forevaluations. The simulation results showed that a small number of additional APs are re-quired to satisfy the dependability, and the degradation of the throughput is not significantregardless of the increasing interference by increasing APs.

Then, we introduced the route availability (RA) index for WIMNET to represent howmany hosts can be connected with the GW on average under a set of probabilities. Fournetwork topologies were adopted as simulated instances for evaluations. The simulationresults showed that the RA index is actually improved in our fault dependable AP allocationsfound by our algorithms, and the difference between RA indices and the correspondingsimulation results is small.

After that, we presented the dynamic channel assignment (DCA) technique for WIMNET.The DCA technique is composed of the initial stage and the dynamic stage, and was giventhrough modifications from the existing study for the static channel assignment, with a newlydefined decision function. Two network topologies were adopted as simulated instancesfor evaluations. The application of our DCA technique gave the significant performanceimprovement.

Finally, we surveyed several secure multi-hop routing techniques for WIMNET for ournext works, where we will adopt the secure multi-hop routing technique to further enhancethe dependability of WIMNET.

In our future works for the dependable AP allocation problem, we will evaluate theperformance of our proposal in more practical network fields for WIMNET, such as stations,shopping malls, and large buildings. At this time, we will consider the effect of indoorenvironments more precisely in our model. Besides, we will extend our approach to enduremultiple failures of links and/or APs. For the RA index, we will investigate the relationshipbetween the RA index and the AP-fault dependability through simulations, and will to able

52

the AP allocation algorithm extension to maximize the RA index. For the DCA technique,we will consider the application of this technique to a real AP.

53

Acknowledgment

It is my great pleasure to thank those who made this thesis possible.I owe my deepest gratitude to my supervisor, Prof. Nobuo Funabiki, who has supported

me throughout my thesis with his patience and knowledge.I would like to thank my co-supervisors, Prof. Yoshitaka Morikawa and Prof. Masaharu

Hata, for their continuous help, guidance, and their proofreading of this work.I would like to express my gratefulness to Prof. Toru Nakanishi for his help and support

during my research.I would like to acknowledge Ministry of Higher Education of Egypt and Cairo University

for financially supporting my doctoral course.I am indebted to all of my laboratory colleagues for assisting and helping me over the

years, and I gave a special acknowledge to Ms. Kanako Uemura.I wish to express my sincere gratitude to Mr. Shigeto Tajima in Osaka University for

helping me during my research.I am heartily thankful to my colleague and my husband, Tamer, for his great support

along my life.I am eternally grateful for the unconditional love, guidance, and support from my parents

and my sisters during my years in the graduate school, and all over my life.Lastly, I offer my regards and blessings to all of those who supported me in any respect

during the completion of my work.

54

Bibliography

[1] I. F. Akyildiz, X. Wang, and W. Wang, ”Wireless mesh networks: a survey,” Comput.Network. ISDN Syst., vol.47, no.4, pp.445-487, Mar. 2005.

[2] ”Wireless distribution system linked router network,” DD-WRT Wiki. Retrieved Dec.31, 2006.

[3] S. Tajima, T. Higashino, and N. Funabiki, ”An Internet gateway access-point selectionproblem for wireless infrastructure mesh networks,” Proc. Int. Work. Future Mob. Ubi.Inform. Tech. (FMUIT), pp. 133-137, Mar. 2006.

[4] N. Funabiki, T. Nakanishi, W. Hassan, and K. Uemura, ”A channel configuration prob-lem for access-point communications in wireless mesh networks,” Proc. IEEE Int. Conf.Networks (ICON), Nov. 2007.

[5] N. Funabiki, K. Uemura, T. Nakanishi, and W. Hassan, ”A minimum-delay routing treealgorithm for access-point communications in wireless mesh networks,” Proc. Int. Conf.Research Innovation Vision for the Future (RIVF-2008), pp. 161-166, Jul. 2008.

[6] M. Siddiqui, and C. Hong, ”Security issues in wireless mesh networks,” Int. Conf. onMultimedia and Ubiquitous Engineering(MUE07), April 26th-28th 2007.

[7] J. Li, J. Jannotti, D. S. J. De Couto, D. R. Karger, and R. Morris, ”A scalable locationservice for geographic ad hoc routing,” Proc. Int. Conf. MobiCom, pp. 120-130, 2000.

[8] S. Misra, S. C. Misra, and I. Woungang, ”Guide to wireless mesh networks,” SpringerLondon, December 12, 2008.

[9] T. Farag, N. Funabiki, and T. Nakanishi, ”An access point allocation algorithm forindoor environments in wireless mesh networks”, IEICE Trans. Commun., vol. E92-B,no. 3, pp.784-739, March 2009.

[10] H. Kato, N. Funabiki, and T. Nakanishi, ”Throughput measurements under variouscontention window size in wireless mesh networks,” IEICE Tech. Report, NS2007-115,pp. 55-60, Dec. 2007.

[11] Proxim Co., ”A detailed examination of the environmental and protocol parameters thataffect 802.11g network performance,” http://www.proxim.com/learn/library/, 2003.

55

[12] A. Sharma, R. Raghavenda, K. Puttaswamy, H. Lundgren, K. Almeroth, and E. Belding-Ro, ”Experimental characterization of interference in a 802.11g wireless mesh network,”Tech. Paper, Univ. California Santa Barbara, Dec. 2005.

[13] M.S. Gast, ”802.11 wireless networks - the definitive guide,” O’Reilly, 2002.

[14] A. Bahri and S. Chamberland, ”On the wireless local area network design problem withperformance guarantees,” Comput. Networks, vol. 48, pp. 856-866, 2005.

[15] Y. Lee, K. Kim, and Y. Choi, ”Optimization of AP placement and channel assignmentin wireless LANs,” Proc. Work. Wireless Local Networks, Nov. 2002.

[16] F. Li, Y. Wang, X.-Y. Li, and A. Nusairat, ”Gateway placement for throughput op-timization in wireless mesh networks,” Proc. IEEE Int. Conf. Commun. (ICC’07), pp.24-28, Jun. 2007.

[17] T. S. Rappaport, ”Wireless communications - principles and practice”, Prentice Hall,Jan. 1996.

[18] D. B. Faria, ”Modeling signal attenuation in IEEE 802.11 wireless LANs - vol. 1,”Tech.Report TR-KP06-0118, Kiwi Project, Stanford Univ., July 2005.

[19] S. Kouhbor, J. Ugon, A. Rubinov, A. Kruger, and M. Mammadov, ”Coverage in WLANwith minimum number of access points,” Proc. IEEE Vehi. Tech. Conf. (VTC) , pp.1166-1170, May 2006.

[20] W. Hassan, N. Funabiki, and T. Nakanishi, ”Extensions of the access point allocationalgorithm for wireless mesh networks”, IEICE Trans. Commun., vol. E93-B, no. 6,pp.1555-1565, June 2010.

[21] Y. Zhang, J. Luo, and H. Hu, ”Wireless mesh networking: architectures, protocols andstandards,” Auerbach Pub., 2006.

[22] G. de la Roche, R. Rebeyrotte, K. JaffrRunser, and J-M. Gorce, ”A QoS-based FAPcriterion for indoor 802.11 wireless LAN optimization,” Proc. IEEE Int. Conf. Commun.(ICC2006), pp. 5676-5681, Jun. 2006.

[23] L. Nagy and L. Farkas, ”Indoor base station location optimization using genetic algo-rithms,” IEEE Int. Symp. Personal, Indoor, and Mobile Radio Commun. (PIMRC), vol.2, pp. 843-846, 2000.

[24] P.-H. Hsiao, A. Hwang, H. T. Kung, and D. Vlah, ”Load-balancing routing for wirelessaccess networks,” Proc. IEEE INFOCOM, pp. 986-995, 2001.

[25] R. Ramamurthy, Z. Bogdanowicz, S. Samieian, D. Saha, B. Rajagopalan, S. Sengupta,S. Chauduri, and K. Bala, ”Capacity performance of dynamic provisioning in opticalnetworks,” J. Lightwave Technol., vol. 19, pp. 40-48, Jan. 2001.

56

[26] G. Gupta and M. Younis, ”Fault-tolerant clustering of wireless sensor networks”, Proc.IEEE Wireless Commun. Networking Conf. (WCNC), vol. 3, pp. 1579-1584, March2003.

[27] U. Varshney and A. D. Malloy, ”Multilevel fault tolerance in infrastructure-orientedwireless networks: framework and performance evaluation,” Int. J. Network Manage-ment, vol. 16, pp. 351-374, Feb. 2006.

[28] H. J. Pan and S. Keshav, ”Detection and repair of faulty access points,” Proc. IEEEWireless Commun. Networking Conf. (WCNC), pp. 532-538, April 2006.

[29] W. Hassan, N. Funabiki, and T. Nakanishi, ”A dynamic channel assignment techniquefor wireless internet-access mesh networks,” Proc. the 14th Asia-Pacific Conf. on Comm.(APCC2008), Oct. 14-16, 2008.

[30] A. Hac and C. Mo, ”Dynamic channel assignment in wireless communication networks,”Int. J. Network Mgmt., vol. 9, pp. 44-66, 1999.

[31] A. Mishra, E. Rozner, S. Banerjee, and W. Arbaugh, ”Exploiting partially overlap-ping channels in wireless networks; turning a peril into an advantage,” Proc. InternetMeasurement Conf., 2005.

[32] W. Zhang, Z. Wang, S. K. Das, and M. Hassan, ”Security issues in wireless meshnetworks,” In: E. Houssain, K. Leung, ”Wireless mesh networks: architecture andprotocols,” Springer, Heidelberg, 2008.

[33] Y. Zhang, J. Luo, and H. Hu, ”Wireless mesh networks: architecture, protocols andstandards,” Auerbach Publ., 2007.

[34] L. Buttyan and J.-P. Hubaux, ”Security and cooperation in wireless networks: thwart-ing malicious and selfish behavior in the age of ubiquitous computing,” CambridgeUniversity Press, 2007.

[35] M. Zapata and N. Asokan, ”Securing ad hoc routing protocols,” ACM Workshop onWireless Secuirty (WiSe), pp. 110, Sep. 2002.

[36] Y.-C. Hu and A. Perrig, ”A survey of secure wireless ad hoc routing,” IEEE Securityand Privacy Magazine, vol. 2, no. 3, pp. 28-39, May/June 2004.

[37] B. Awerbuch, D. Holmer, C. NitaRotaru and H. Rubens, ”An on-demand secure routingprotocol resilient to byzantine fFailures,” Proc. ACM Workshop on Wireless Security(WiSe), pp. 21-30, 2002.

[38] P. G. Argyroudis and D. O’Mahony, ”Secure routing for mobile ad hoc network,” IEEEComm. Surveys and Tutorials, Vol. 7, no. 3, pp. 2-21, 2005.

[39] A. Mishra, ”Security and quality of service in ad hoc wireless networks,” CambridgeUniversity Press, 2008.

57

[40] K. Sanzgiri, B. Dahill, B. N. Levine, C. Shields, and E. M. Belding-Roye, ”A securerouting protocol for ad hoc networks,” Proc. 10th IEEE Int. Conf. on Network Protocols(ICNP02), p. 78-89, 2002

[41] Y.-C. Hu, D. B. Johnson, and A. Perrig, ”SEAD: secure efficient distance vector routingfor mobile wireless ad hoc networks,” Proc. 4th IEEE Workshop on Mobile ComputingSystems and Applications (WMCSA2002), pp. 3-13, 2002.

58