Enhancing P2P Systems over Wireless Mesh Networks

(1)

Faculty of Economic Sciences, Communication and IT Computer Science

DISSERTATION Karlstad University Studies

2011:60

Marcel Cavalcanti de Castro

Enhancing P2P

Systems over Wireless

Mesh Networks

(2)

Marcel Cavalcanti de Castro

Enhancing P2P

Systems over Wireless Mesh Networks

Karlstad University Studies

2011:60

(3)

Marcel Cavalcanti de Castro. Enhancing P2P Systems over Wireless Mesh Networks

DISSERTATION

Karlstad University Studies 2011:60 ISSN 1403-8099

ISBN 978-91-7063-398-0

Distribution:

Karlstad University

Faculty of Economic Sciences, Communication and IT Computer Science

S-651 88 Karlstad Sweden

+46 54 700 10 00

www.kau.se

Print: Universitetstryckeriet, Karlstad 2011

(4)

A minha esposa Roberta e aos meus filhos Sofia e Matteus

(5)

(6)

Enhancing P2P Systems over Wireless Mesh Networks

MARCEL CAVALCANTI DE CASTRO

Department of Computer Science, Karlstad University, Sweden

Abstract

Due to its ability to deliver scalable and fault-tolerant solutions, applications based on the peer-to-peer (P2P) paradigm are used by millions of users on the internet. Recently, wireless mesh networks (WMNs) have attracted a lot of interest from both academia and industry, because of their potential to provide flexible and alternative broadband wireless internet connectivity. However, due to various reasons such as unstable wireless link characteristics and multi-hop forwarding operation, the performance of current P2P systems is rather low in WMNs.

This dissertation studies the technological challenges involved while deploying P2P systems over WMNs. We study the benefits of location-awareness and resource replication to the P2P overlay while targeting efficient resource lookup in WMNs. We further propose a cross-layer information exchange between the P2P overlay and the WMN in order to reduce resource lookup delay by augmenting the overlay routing table with physical neighborhood and resource lookup history information.

Aiming to achieve throughput maximization and fairness in P2P systems, we model the peer selection problem as a mathematical optimization problem by using a set of mixed integer linear equations. A study of the model reveals the relationship between peer selection, resource replication and channel assignment on the performance of P2P systems over WMNs. We extend the model by formulating the P2P download problem as chunk scheduling problem. As a novelty, we introduce constraints to model the capacity limitations of the network due to the given routing and channel assignment strategy.

Based on the analysis of the model, we propose a new peer selection algorithm which incorporates network load information and multi-path routing capability.

By conducting testbed experiments, we evaluate the achievable throughput in multi- channel multi-radio WMNs. We show that the adjacent channel interference (ACI) problem in multi-radio systems can be mitigated, making better use of the available spectrum.

Important lessons learned are also outlined in order to design practical channel and channel bandwidth assignment algorithms in multi-channel multi-radio WMNs.

Keywords: peer-to-peer overlay, wireless mesh networks, peer selection, channel assignment, routing, optimization, adjacent channel interference, channel bandwidth adaptation.

(7)

(8)

Acknowledgments

The five years and a half spent working on my Ph.D. thesis at the Computer Science department is an experience that I will never forget. This thesis would have never been completed without the help and support of many persons who contributed to it directly or indirectly.

First, I would like to thank Prof. Andreas Kassler for giving me the opportunity to pursue my Ph.D. studies under his supervision, and for providing me with excellent support and dedication during all these years. I am privileged for having him as my supervi- sor.

I would like to thank Prof. Mario Gerla, Prof. Yevgeni Koucheryavy, and Prof. Evgeny Osipov for their willingness to be on the examination committee, and Prof. Edmundo Monteiro for accepting the role of opponent and its comments in a preliminary version of this thesis.

I would like to thank the European Regional Development Fund through the In- terreg IVB project E-CLIC (European Collaborative Innovation Centres for Broadband Media Services), NEWCOM++ (Network of Excellence in Wireless Communications), Knowledge Foundation of Sweden, and STINT (stiftelsen för interntionalisering av högre utbildning och forskning) for the financial support.

I am very grateful for having the opportunity to work with so many wonderful peo- ple at the Computer Science department, whom deserve credit for providing such a nice and friendly work environment. A special thanks to the members of the Distributed System and Communications Research Group (DISCO) and the co-authors of the publications included in this thesis. It has been a privilege to work with all of you.

Lastly, a huge thank to all of those who supported me in any respect during the completion of this work. This thesis is dedicated to my wife, Roberta M. Agostini, and my parents Jazon & Adelcy Castro for supporting and encouraging me to pursue this degree. Without my wife’s encouragement, I would not have finished the degree.

Karlstad, November 2011 Marcel C. Castro

iii

(9)

(10)

List of Figures

2.1 Wireless Mesh Network Architecture[20] . . . 14

2.2 Design choices of P2P and WMN integration . . . 23

3.1 Architecture: components and interconnections . . . 29

3.2 Example of a store and find resource action using DHT functionalities. . 32

3.3 Message sequence diagram for a store and find resource action using put/get DHT functionalities. . . 33

3.4 Impact of Bamboo management traffic . . . 41

3.5 Percentage of management overhead compared to user traffic . . . 41

3.6 Percentage of lookups completed . . . 42

3.7 CDF of lookup delay in the 36 nodes topology . . . 42

3.8 Comparison between the number of logical hops in Georoy and Bamboo in a grid topology. . . 47

3.9 Comparison between the number of physical hops in Georoy and Bam- boo in a grid topology. . . 47

3.10 Comparison between the average lookup delay in Georoy and Bamboo in a grid topology. . . 48

3.11 Comparison between the percentage of lookups completed in Georoy and Bamboo in a grid topology. . . 49

3.12 Comparison between the routing stretch factor in Georoy and Bamboo in a grid topology. . . 50

3.13 Comparison between the number of logical hops in Georoy and Bamboo for random topologies. . . 51

3.14 Comparison between the number of physical hops in Georoy and Bam- boo for random topologies. . . 51

3.15 Comparison between the average lookup delay in Georoy and Bamboo for random topologies. . . 52

3.16 Number of logical and physical hops in Bamboo in a grid topology with 225 nodes. . . 53

3.17 Number of logical and physical hops in Georoy in a grid topology with 225 nodes. . . 53

ix

(15)

LIST OF FIGURES

3.18 Average lookup delay for Georoy and Bamboo in a grid topology with

225 nodes. . . 54

3.19 Example of the one-hop DHT neighborhood and lookup request. Node 000 is requesting a resource located at node 101. The overlay neighborhood, network topology and key lookup process are represented in the left, middle and right part of the figure respectively. . . 55

3.20 Average cache size versus node density for standard and cross-layer cache schemes using 500 and 1000 nodes . . . 57

3.21 Routing stretch versus node density for 1000 nodes . . . 57

4.1 Collision domain for linkl(1,2) using single channel WMN . . . 67

4.2 K-Partition channel assignment . . . 69

4.3 BFS channel assignment . . . 69

4.4 Impact of replication for files located at the gateways, two NICs and six channels. . . 74

4.5 Impact of replication for files located at the random nodes, two NICs and six channels. . . 75

4.6 Throughput versus number of channels . . . 77

4.7 Impact of number of radios for BFS and random replication . . . 77

4.8 Tradeoff between maximum throughput and fairness for the degree of replication of two at gateways, two NICs and six channels . . . 78

4.9 Theoretical and simulation results of the normalized throughput for different degrees of replication at the gateways and single channel assignment 79 5.1 Chaska topology . . . 89

5.2 Optimum makespan for the P2P download problem: chunk-based versus non-chunk-based, assuming single channel WMN deployment. . . 90

5.3 Optimum makespan for different channel assignment strategies and increasing number of seeds . . . 91

5.4 Optimum makespan for K-Partition and BFS channel assignment strategies, 1 seed, 6 channels, 2 NICs . . . 92

5.5 Optimum makespan versus number of channels for grid topology, 1 seed, 6 leechers, and 2 NICs . . . 92

5.6 Interaction between peer selection, routing and channel assignment . . . . 94

5.7 Total download time versus number of seeds for different peer selection schemes using 10 leechers . . . 100

5.8 Total download time versus number of seeds for different peer and path selection schemes . . . 101

5.9 CDF of total download time for Chaska topology, 15 seeds and 30 leechers and different peer and path selection schemes . . . 102

5.10 Total download time versus number of seeds for channel reassignment . . 103

6.1 Testbed experimental setup (a) and development board used (b) . . . 109 x

(16)

LIST OF FIGURES

6.2 IEEE 802.11a transmit spectrum mask[21] . . . 111 6.3 Critical ACI scenarios in multi-radio mesh networks . . . 112 6.4 ACI experimental setup: basic setup (upper part) and 2-boxes setup (lower

part) . . . 114 6.5 Impact of board crosstalk, radiation leakage, and antenna engineering

on 802.11a multi-radio performance in terms of normalized aggregated throughput and PHY rate of 6 Mbit/s . . . 116 6.6 Impact of ACI under different link qualities in terms of normalized ag-

gregated throughput and PHY rate of 6 Mbit/s . . . 118 6.7 Throughput A-C for different channel combinations (20 cm antenna dis-

tance . . . 120 6.8 PHY rate with highest throughputA − C for different channel combina-

tions (20 cm antenna distance) . . . 123 6.9 Throughput versus channel bandwidth in 802.11 networks for different

modulation schemes . . . 125 6.10 Example of channel overlapping for a 20/40 MHz channel combination

and different channel separation . . . 126 6.11 Normalized throughput for various interferer and receiver combination

at 20 and 40 MHz . . . 129 6.12 Normalized throughput of linkA− B1for various interferer and receiver

combination . . . 131

xi

(17)

(18)

List of Tables

2.1 Overview of IEEE 802.11 physical layer[22] . . . 21 3.1 Bamboo management timers (secs) . . . 40 3.2 Average routing traffic received per node in the 36 nodes topology . . . 42 3.3 Number of dropped packets and correspondent reason in the 36 nodes

topology . . . 42 6.1 Measurements parameters . . . 110 6.2 Percentage of packets sent at a given PHY rate for the SampleRate algo-

rithm, with and without ACI . . . 122

xiii

(19)

(20)

Chapter 1 Introduction

WMNs have attracted the interest of the research community and industry over the last decade. WMNs have matured to a point where the wireless mesh technology is been used as an attractive means to provide connectivity in complement to access as offered by wireless local area networks (WLANs) or cellular networks. Various qualities of this paradigm include the combination of low cost, high speed, large coverage and robustness.

WMNs consist of a backbone of quasi-stationary mesh routers forming a multi-hop wireless network, in which mesh routers wirelessly relay traffic on behalf of others. By using gateway functionality, mesh routers allow the integration of the wireless mesh network with the internet or existing wireless networks. The WMNs can be visualized as an integration of two planes where the access plane provides connectivity to the clients while the forwarding plane relays traffic between the mesh routers[20, 23]. Currently, mesh routers have been equipped with multiple radios, which allow them to send and receive on multiple channels in parallel and consequently increase network capacity.

Due to their flexible structure, WMNs possess a wide range of application scenarios. The most common application has been the deployment of wireless community networks, where users own the mesh routers and form the wireless mesh backbone, en- abling access to other users for mutual benefit[24]. WMNs have also been used at control systems in order to deploy public area surveillance[25] and temporary infrastructure in disaster and emergency situations[26]. Other applications also considered to WMNs include industrial automation [27], traffic control [28], and sensor monitoring systems [29].

Allied to the rapid proliferation of wireless technologies, the increasing development of P2P communication systems in the internet has also gained a lot of attention. Using the definition by[30], P2P communication refers to technology that enables two or more peers to collaborate spontaneously in a network of equals (peers) by using appropriate information and communication systems without the necessity for central coordination.

P2P systems and WMNs share many key characteristics such as self-organization and de- centralization due to the common nature of their distributed components designed to

(21)

2 Introduction

operate in dynamic network environments and hop-by-hop connection establishment.

The common characteristics shared by both technologies also dictate that P2P systems and WMNs are faced with the same fundamental challenge, that is to provide connectivity in decentralized environments.

P2P systems represent the next frontier to wireless communication, as wireless users might expect to use the same services which are already available on the internet, like popular file-sharing systems such as BitTorrent[31], voice over P2P applications such as Skype[32], and emerging live or on-demand streaming applications such as PPLive [33]. In addition to that, P2P systems also represent an important alternative to scale WMN services, such as mobility and control management [34, 35], network routing [36], and anonymity [37]. Therefore, deploying P2P systems over WMNs represents exciting possibilities, but at the same time several challenges need to be investigated.

To leverage distributed resource lookup and peer discovery, P2P systems implement an abstract overlay network on top of the physical network topology. Generally, the P2P systems are designed for the wired internet, and rely on the IP routing infrastructure which is resource rich especially in terms of bandwidth availability. As a result, such overlay networks use a membership management in order to keep the connectivity among neighboring peers and consequently maintain the P2P overlay network. If deployed over wireless mesh networks, the maintenance of such overlay may require the exchange of frequent control traffic with overlay neighbors that are physically located several wireless hops away. This is because no direct mapping exists between the overlay network and the physical network topology. This brings down the network performance in WMNs as delay and packet loss increase with the number of wireless hops traversed.

Thus, solutions that adapt P2P overlays to current WMNs conditions are important to scale the aforementioned services.

Using multiple radios increases the network capacity in WMNs, as multiple orthogonal channels are available and assigned such that the interference in the network is min- imized. In addition, finding paths with better channel diversity can leads to increase in the overall network capacity. As a result, the channel assignment and routing protocol are quite dependent on each other, as the channel assignment determines the physical network topology on which the routing protocol works.

Given that P2P systems are deployed over WMNs, and the resource lookup process is completed by finding the list of peers holding the requested resource, the peer selection phase takes place. In multi-channel multi-radio WMNs, the peer selection, which considers how to select neighbors from a set of peers, may also be influenced by the channel assignment and routing as they dictate the network capacity and topology. Thus, an important issue to solve is how to model the achievable performance of the P2P download process in a WMN given the capacity constraints imposed by the channel assignment and routing. To address this issue, it is important to study the interactions between the peer selection, channel assignment and routing layers. By using mathematical optimization models, it is possible to derive such interactions as important network optimization tasks which combines channel assignment, routing, peer selection strategy, resource replication, and the impact of number of peers, channels and radios can be easily performed

(22)

1.1. Problem Definition and Research Questions 3

in order to derive achievable P2P download performance. Despite the use of limited assumptions, the solution of such models are an important asset as it gives more insights in the theoretical understanding of P2P systems in WMNs, and helps to define novel peer selection strategies in multi-channel multi-radio WMNs, taking into account their limitations in terms of resource availability.

Apart from the mathematical models, researchers have also been allowed to undertake testbed evaluations given the increasingly cheaper and more accessible WMNs technology. The use of testbed evaluations has allowed hands-on experience on implementing theoretical ideas and also worked as a tool to improve mathematical and simulation models. For example, the increasing deployment of multi-radio WMN testbeds has shown that in practice the amount of orthogonal channels is reduced as the use of close-by radios operating on adjacent channels causes considerable network interference, known as the ACI. In addition, the need for spectrum flexibility without further increase in hardware costs has allowed the possibility for adapting the channel bandwidth according to changing environment conditions such as interference and network traffic demands[22].

However, the consequences of using cheap off-the-shelf hardwares in WMN testbeds also needs to be considered while analyzing the results derived from the measurements, as imperfect radio shielding, low quality board design and antenna engineering contribute to the overall interference in the network, and consequently impact on the achieved network capacity.

Given those considerations, we describe in Section 1.1 the problem definition and research questions that are investigated in this dissertation. In Section 1.2 we present the thesis outline and the contributions of this dissertation, together with the clarifications about the work done in cooperation with other researchers. Other related papers to this dissertation are presented in Section 1.3.

1.1 Problem Definition and Research Questions

In this dissertation we study:

How to enhance P2P system performance over wireless mesh network environments.

To address this problem, we sub-divide it into three important research questions:

I. How to organize P2P overlay membership in wireless mesh networks in order to provide efficient P2P resource lookup ?

Due to the characteristics of the wireless links and multi-hop forwarding paradigm, the performance of traditional distributed services, such as P2P systems, is rather low in wireless mesh networks. To address this question requires first a detailed analysis of the challenges involved in the deployment of P2P systems in wireless mesh networks described in Chapter 3. Normally, P2P systems implement an abstract overlay network on top of the physical network topology in order to leverage resource indexing and peer discovery. Since the maintenance of the P2P overlay involves a membership management overhead to the wireless networks, an ac-

(23)

4 Introduction

ceptable trade-off is required between the P2P resource lookup performance and overlay membership management. Targeting efficient resource lookup in wireless mesh networks, we analyze in Chapter 3 different cross-layer information exchange strategies among the P2P overlay and the wireless mesh network.

II. How to model the achievable performance of a P2P system in wireless mesh networks given the capacity constraints imposed by the channel assignment and routing layers ? Aiming to achieve high performance and fairness in P2P systems over wireless mesh networks, the peer selection problem is the one that identifies the set of best peers to be chosen during the resource exchange phase. To address this question, four different peer selection schemes are formulated in Chapter 4 using a mathematical model that describes a set of linear equations. Numerical results are derived in order to estimate the achievable capacity of the P2P system given the constraints imposed by the network capacity. As outlined in Chapter 5, an extension of the peer selection problem which allows peers to upload resource’s segments among themselves marks an important step towards the decrease of the amount of time required to disseminate resources in the network. Moreover, the underlying routing and channel assignment layers in WMNs have a big impact on the P2P download problem, as the selection of peers having suboptimal paths in the network will impact on resource consumption in all intermediate nodes in WMN scenarios. There- fore, the study of the interactions of those lower layers with the peer selection is carried out through packet-level simulations. Given the insights gained through the models and simulations, a novel peer selection algorithm is proposed which accounts for the path load information and multi-path routing capability.

III. What performance improvements can be achieved using common off-the-shelf multi- radio devices and what is the impact of interference on performance ?

The availability of cheap off-the-shelf multi-radio devices have enabled the wide deployment of wireless mesh testbeds. Those testbed deployments have shown that certain assumptions used in mathematical and simulation models are not always true in practice. One important example is related to the amount of orthogonal channels available in multi-channel multi-radio WMNs. Through testbed experiments, we have shown that in practice the amount of orthogonal channels in WMNs is reduced as the use of close-by radios operating on adjacent frequency bands causes considerable network interference, known as the ACI. The impact of ACI on achievable performance is studied in Chapter 6. Moreover, the use of cheap hardwares implies an additional performance degradation in WMN deployments, as imperfect radio shielding, bad board design and antenna engineering contribute to the ACI problem. In addition, the need for spectrum flexibility and high network performance have motivated the use of channel bandwidth adaptation. Thus, an evaluation of the channel bandwidth adaptation and ACI while targeting performance improvement in multi-channel multi-radio WMN scenarios is carried out in our studies. Given the knowledge acquired through such testbed experiments,

(24)

1.2. Thesis Outline and Contributions 5

important lessons learned can be used to develop better channel and channel bandwidth assignment algorithms.

1.2 Thesis Outline and Contributions

The thesis is organized as follows. In Chapter 2, we introduce the background work necessary to understand the research presented in this thesis. Thereafter, four chapters are presented. Each chapter presents a short introduction to the problem discussed along with the contributions and the detailed description of the performed studies. Related works and conclusions of the achieved results in the chapter are outlined, followed by a discussion of the validity of the results and their possible limitations. The concluding remarks and future directions are then outlined in Chapter 7.

We now outline the individual contributions of each chapter along with the work done in collaboration.

Chapter 3: Adapting the P2P Overlay to WMNs

Chapter 3 starts by introducing our proposed architecture for P2P systems over wireless mesh networks, describing the required architectural components and their interactions.

To better describe the challenges involved while deploying P2P systems over wireless mesh networks, we conduct simulation studies showing the trade-off between P2P resource lookup efficiency and the overlay management overhead. Given those insights, we analyze the benefits of location-aware distributed hash tables (DHTs) and resource replication in order to increase the overlay lookup efficiency in WMN scenarios. Through- out the simulation results, we have shown that by constructing the overlay DHT using physical location information at larger topologies contributes to lower lookup delays as longer physical paths in the network are avoided while routing resource lookup through the overlay. Moreover, the use of multiple replicas increases the probability of finding the requested resource within a shorter distance, benefiting both location-aware and location- unaware DHTs. Through interactions with the routing layer, we replace long-range overlay neighbors by physical neighbors, and exploit the resource lookup history through information caching. By using limited caching allied to neighbor-of-neighbor information provided by the routing layer at mesh routers, we show that the routes selected by the P2P overlay are almost as efficient as the shortest path routes. These investigations have previously been published in the following papers.

[1]. Marcel C. Castro, Laura Galluccio, Andreas Kassler and Corrado Rametta. Oppor- tunistic P2P Communications in Delay Tolerant Rural Scenarios. In the EURASIP Journal on Wireless Communications and Networking, 2011:1–14, 2011.

[2]. Marcel C. Castro, Andreas Kassler, Carla-F. Chiasserini, Claudio Casetti and Ibra- him Korpeoglu. Peer-to-Peer Overlay in Mobile Ad-hoc Networks. In X. Shen, H.

Yu, J. Buford, M. Akon, editors,Handbook of Peer-to-Peer Networking, Springer, pages 1045–1080, 2010.

(25)

6 Introduction

[3]. Marcel C. Castro, Andreas Kassler, Gabriel Kliot, Roy Friedman, Raphäel Kum- mer, Peter Kropf, Pascal Felber. Minimizing DHT Routing Stretch in MANETs.

In the 9^{t h}Scandinavian Workshop on Wireless Adhoc Networks (Adhoc’09), Uppsala, Sweden, 4–5 May, 2009.

Concerning Paper[1], the author of this dissertation shares the development of the original idea, evaluation and analysis of the results with Corrado Rametta. The problem formulation was refined in collaboration with Laura Galluccio and Andreas Kassler, who provided valuable insights in the evaluation and writing phases. Concerning Paper[2] , the author of this dissertation developed all the original ideas. In Paper[3], the author of this dissertation shares the development of the original idea with Raphäel Kummer and Gabriel Kliot. The DHT implementation was carried out by Raphäel Kummer, while the location-awareness and caching modifications were done in conjunction with the author of this dissertation. The evaluation of the implementation, the experiments, and paper writing were carried out by the author of this dissertation.

Chapter 4: Peer Selection Problem Formulation in WMNs

Chapter 4 starts the study of the interdependency between the P2P overlay, channel assignment and routing layers in wireless mesh networks. The collision domain model [38, 39] is used to estimate achievable wireless link capacity in multi-hop communication scenarios. We first present a novel mathematical model which allows to calculate the achievable performance of a P2P system in wireless mesh networks given the capacity constraints imposed by the channel assignment and routing layers. While previous models have applied the collision domain model to estimate the achievable throughput in wireless mesh networks, we are the first who extends this model to include peer selection, channel assignment, routing and resource replication on the problem formulation for multi-channel multi-radio wireless mesh networks. We devise four peer selection strategies targeting network throughput maximization and fairness among peers during the P2P download process. Numerical results on the achievable throughput and fairness are presented, showing the relationship between peer selection, replica placement and channel assignment. An analysis of the results shows that with a lower degree of resource replication and enough channels, it is better to replicate randomly the resource in the network compared to replication at the gateways. This is due to the fact that the download traffic can be spread out more effectively in the network as more links operating on orthogonal channels can be used simultaneously. The optimum number of channels can be identified for a given number of radios, beyond which the throughput can not be increased by adding more channels. According to the results, the optimum number of channels depends on the channel assignment, peer selection and number of radios available in the network. Part of these results were included in:

[4]. Marcel C. Castro, Durga M. Prasad, Andreas Kassler, Stefano Avallone. Peer-to- Peer Selection and Channel Assignment for Wireless Mesh Networks. Inthe Inter-

(26)

1.2. Thesis Outline and Contributions 7

national Workshop on Network Modeling and Analysis (IWNMA), pages 1–8, Banga- lore, India, 2^nd January 2011.

The work published in Paper[4] was coauthored with Durga M. Prasad during his internship at Karlstad University. The author of this dissertation proposed the original problem formulation and acted as leading author of the paper. The development of the model in Octave was mostly done by Durga M. Prasad. The numerical results, the val- idation of the model with the simulator, and the paper writing were carried out by the author of this dissertation.

Chapter 5: Extending the Peer Selection Problem for Multiple Chunks

Chapter 5 presents an extension of the mathematical model proposed in Chapter 4 by incorporating the availability of multiple chunks in the peer selection problem formulation. We formulate the P2P download problem as a chunk scheduling problem given the constraints imposed by the channel assignment and routing protocol on the achievable capacity. While such chunk-based peer selection scheduling exists, we are the first who considers channel assignment and routing protocol in the network constraints formulation targeted to multi-channel multi-radio wireless mesh networks. The new chunk-based peer selection problem, formulated as a mixed integer linear program, allow us to devise the optimum makespan, that is the minimal time required to disseminate a file among all peers involved in the P2P download process. Using a numerical analysis, we compare the chunk based scheduling with the non-chunk based scheduling described in Chapter 4. In addition to that, a packet-level simulation is carried out in order to further analyze the interdependency between P2P overlay, channel assignment and routing layers for larger set of peers, chunks and topologies. A new peer selection algorithm which incorporates path load information, multi-path routing capability and channel reassignment is proposed while targeting improvements on the P2P download process. An analysis of the results shows that the combination of the breadth first search (BFS) channel assignment together with the possibility of mutual resource exchange among peers contributes to a faster resource dissemination in the context of wireless mesh networks. Moreover, we show that by using the new peer selection metric which accounts for network path load together with the availability of multiple paths among peers, we can reduce by up to forty percent the total download time to disseminate a resource, compared to BitTorrent[31].

Part of these results were included in:

[5]. Marcel C. Castro, Andreas Kassler. On the Interaction Between Peer Selection, Routing and Channel Assignment. In the 10^{t h} Scandinavian Workshop on Wireless Adhoc Networks (Adhoc’11), Stockholm, Sweden, 10–11 May, 2011.

The work presented in Section 5.2 of Chapter 5 was a joint work with Nadia Ulrich.

The author or this dissertation developed the original idea by proposing the extended model presented. The development of the model in Octave was jointly performed by both authors. The porting of the model and evaluation carried out in SCIP and CPLEX

(27)

8 Introduction

were all carried out by the author of this dissertation. The work presented in Section 5.3 represents the work published in Paper[5]. The author of this dissertation developed all the original ideas and implementations contained in the aforementioned publication.

Andreas Kassler provided valuable insights during the evaluation and writing phases.

Chapter 6: Practical Considerations for Channel and Channel Bandwidth Assign- ment in WMNs

In Chapter 6 we focus our attention to the practical considerations for channel and channel bandwidth assignment in multi-channel multi-radio wireless mesh networks. By conducting testbed experiments using off-the-shelf multi-radio devices, we show that in practice the amount of orthogonal channels used by the channel assignment algorithms in WMNs is reduced as the use of close-by radios operating on adjacent channels causes considerable network interference, known as the ACI. Allied to that, we study the hardware design problems caused by imperfect radio shielding, low quality board design and antenna engineering, which added to the performance degradation caused by ACI. In order to increase network capacity and maximize the number of available channels, channel bandwidth adaptation is used in our wireless mesh testbed. Furthermore, we list important points to be considered in the design of practical channel and channel bandwidth assignment algorithms. An analysis of the results shows that by using a good hardware engineering for the node enclosure, board, radio cards, sufficient antenna separation, and orthogonal polarization it is indeed possible to achieve orthogonal channels using the IEEE 802.11a standard in multi-radio mesh nodes. Moreover, the results indicate that joint effects of ACI together with channel heterogeneity need to be considered while targeting network throughput prediction at higher PHY rates. The results also show that it is better to use wider channel bandwidths if the radio interfaces are operating on adjacent channels as it forces them to share the medium via clear channel assessment (CCA) mechanism avoiding than the ACI while transmitting on adjacent channels. These investigations are included in:

[6]. Peter Dely, Marcel C. Castro, Sina Soukhakian, Arild Moldsvor, Andreas Kassler.

Practical Considerations for Channel Assignment in Wireless Mesh Networks. In the IEEE Broadband Wireless Access Workshop, held in conjunction with Globecom 2010, Miami, USA, 6–10 December, 2010.

[7]. Marcel C. Castro, Andreas Kassler, Stefano Avallone. Measuring the Impact of ACI in Cognitive Multi-Radio Mesh Networks. Inthe IEEE 72^nd Vehicular Technology Conference (VTC Fall 2010), Ottawa, Canada, 6–9 September, 2010.

The work presented in Sections 6.4.2 and 6.4.3 of Chapter 6 represents the work published in Papers[6], coauthored with Peter Dely and Sina Soukhakian. The author of this dissertation proposed the original idea contained in Paper[6]. The experiments were jointly performed by the authors, while the paper editing was done by Peter Dely in conjunction with the author of this dissertation. The work presented in Section 6.5

(28)

1.3. Other Related Papers 9

represents the work published in Paper[7]. The author of this dissertation developed all the ideas contained in Paper[7]. Andreas Kassler, Arild Moldsvor and Stefano Avallone provided valuable insights in the experimental evaluation and writing phases.

1.3 Other Related Papers

The following publications, although not specifically included in the dissertation, contain material that is related to the contributions of this dissertation:

• Marcel C. Castro, Laura Galluccio, Andreas Kassler, Sergio Palazzo, Corrado Ram- etta. On the comparison between performance of DHT-based protocols for oppor- tunistic networks. In Proceedings of Future Network and MobileSummit 2010, pages 1–8, Florence, Italy, 16–18 June, 2010.

• Marcel C. Castro, Peter Dely, Andreas Kassler, Francesco Paolo D’Elia, Stefano Avallone. OLSR and Net-X as a Framework for Channel Assignment Experi- ments (POSTER). In the 4^{t h} ACM International Workshop on Wireless Network Testbeds, Experimental Evaluation and Characterization (WiNTECH), held in conjunction with MobiCom, Beijing, China, 20–25 September, 2009.

• Marcel C. Castro, Peter Dely, Andreas Kassler, Nitin Vaidya. QoS-Aware Channel Scheduling for Multi-Radio/Multi-Channel Wireless Mesh Networks. In the 4^{t h} ACM International Workshop on Wireless Network Testbeds, Experimental Evalua- tion and Characterization (WiNTECH), held in conjunction with MobiCom, Beijing, China, 20–25 September, 2009.

• Marcel C. Castro and Andreas J. Kassler. Packet Aggregation for VoIP in Wireless Meshed Networks. In Y. Koucheryavy, G. Giambene, and D. Staehle, editors,The Traffic and QoS Management in Wireless Multimedia Networks, chapter Multihop Wireless Networks, 2008.

• Marcel C. Castro, Eva Villanueva, Iraide Ruiz, Susana Sargento, and Andreas Kassler.

Performance Evaluation of Structured P2P over Wireless Multi-hop Networks. In the International Conference on Advances in Mesh Networks (MESH 2008), Cap Es- terel, France, 25–31 August, 2008.

• Nico Bayer, Marcel C. Castro, Peter Dely, Andreas Kassler, Yevgeni Koucheryavy, Piotr Mitoraj, and Dirk Staehle. VoIP Service Performance Optimization in Pre- IEEE 802.11s Wireless Mesh Networks. In the IEEE International Conference on Circuits & Systems for Communications (ICCSC2008), Shanghai, China, 26–28 May 2008.

• Marcel C. Castro, Peter Dely, Jonas Karlsson, Andreas Kassler. Capacity Increase for Voice over IP Traffic through Packet Aggregation in Wireless Multihop Mesh Networks. In the International Workshop on Wireless Ad Hoc, Mesh and Sensor

(29)

10 Introduction

Networks (WAMSNET07) , Jeju-Island, Korea, 10–12 December 2007. Best Paper Award.

• Andreas J. Kassler, Marcel C. Castro, Peter Dely. VoIP Packet Aggregation based on Link Quality Metric for Multihop Wireless Mesh Networks. In Proceedings of Future Telecommunications Conference (FTC2007), Beijing, China, 11–12 October 2007.

• Marcel C. Castro, Andreas Kassler. SIP based Service Provisioning for hybrid MANETs. In Proceedings of International Workshop on Telecommunications (IWT 2007), Santa Rita do Sapucaí, Brazil, 12–15 February, 2007.

• Marcel C. Castro, Andreas Kassler. Challenges of SIP in Internet Connected MA- NETs. In Proceedings of International Symposium of Wireless Pervasive Computing (ISWPC 2007) , San Juan, Puerto Rico, 5–7 February, 2007.

• Marcel C. Castro, Andreas Kassler. SIP in hybrid MANETs - A gateway based approach. In the 4^{t h} Swedish National Computer Networking Workshop (SNCNW 2006), Luleå, Sweden, 26–27 October, 2006.

• Marcel C. Castro, Andreas Kassler. Optimizing SIP service provisioning in internet connected MANETs - Invited Paper. In the 14^{t h} International Conference on Software, Telecommunications and Computer Networks (SoftCOM 2006)- Symposium on QoS in Wireless Multimedia Networks, Split, Croatia, 29 September – 1 October, 2006.

(30)

Chapter 2 Background

The objective of this chapter is to present the background information required to assist the reader on the understanding of the research presented in this thesis. We first introduce the P2P system concept and its classification through a set of well-known examples.

Having in mind that P2P communication paradigm is very important in wireless multi-hop networks, as centralized servers might not be available or located in the internet, we introduce the wireless mesh network architecture. We also describe the benefit of using multi-channel multi-radio WMNs and important challenges such as routing, channel assignment, wireless capacity modeling, and the IEEE 802.11 physical (PHY) and medium access control (MAC) layers. An overview on different principles that guide the various integration and interaction possibilities for peer-to-peer systems in wireless mesh networks is presented through a set of proposed solutions in the literature.

2.1 Peer-to-Peer Systems

Recently applications based on the P2P communication paradigm are increasing in pop- ularity. There are numerous P2P systems proposed with very different architectures and protocols. Examples are popular file-sharing applications (e.g., Gnutella[40], and Bit- Torrent[31]), voice over IP solutions (e.g. Skype [32]) , and P2P video streaming (e.g.

PPLive, SopCast, CoolStreaming[33]).

A P2P system is a collection of autonomous end-system devices calledpeers that form a set of interconnections called anoverlay to share the peer’s content. Commonly, P2P architectures can be organized into three main classes: centralized, distributed, and hybrid.

(31)

12 Background

2.1.1 Centralized

The centralized architecture was the first class of P2P systems, popularized by Napster [41] which paved the way for file-sharing distribution. Napster made use of centralized index server responsible for maintaining the list of connected peers and the files they provide. The central index could be queried by a peer to lookup for the address information (e.g. IP and port) of all peers sharing the requested file. After obtaining the list of peers that contained the desired file, the peer established direct connection with the peers, and downloaded the file without the involvement of the central index server.

The main disadvantage of the centralized architecture is related to its scalability, as the central index directory represents a single point of failure. Despite its huge number of users, Napster went out of service not because of technical failure, but because the single administration was vulnerable to the legal challenges of record companies[42].

2.1.2 Distributed

The main idea of distributed architectures is to establish and maintain the peer index without using central entities. Common distributed architectures are classified as unstructured and structured.

The unstructured P2P overlays proposes that each peer in the system would be responsible for indexing its own resource. Thus, search operations along the peers are necessary in order to find the desired resource in the overlay which may take long time and consume network resource extensively as there is no relation between the overlay topology and resource location. In Gnutella[40], the search operation is broadcasted in the network and no guarantee can be given that the resource can be found. Attempts to decrease the search overhead has been proposed by using expanded time to live (TTL) rings[43] or random walk searches directed to peers with higher degree of connectivity [44]. In unstructured approaches, the load on each peer grows linearly with the number of search operation and the system size.

Since flooding request in the network is costly in terms of scalability, structure P2P overlays made use of an organized overlay in order to achieve higher flexibility. As a consequence, DHT based solutions have been proposed. DHTs offer an indexing service by mapping each resource and each peer storing the resource on a certain key assigned to a common identifier space via secure hash algorithm (SHA)-1 hash functions[45].

The one-way hash function leads to every peer being responsible for a range of keys and having a virtual link with a subset of peers. When someone requests a key to a peer, it compares its own identifier (ID) with the key and, if it falls in its range, it replies to the requester. Otherwise the request is forwarded to the neighbor whose ID is the closest with respect to the searched key.

Chord[46], Pastry [47], Tapestry [48], and Viceroy [49] are all solutions that exploit a DHT approach. They differ in the way they build and maintain the structure of the logical overlay. For example, Chord uses a logical ring where every node has an assigned ID and is responsible for all the keys between its ID and its predecessor ID. Moreover, in

(32)

2.1. Peer-to-Peer Systems 13

order to speed up the search process, a finger table is used to connect the node to other nodes in the network.

After Chord, other robust algorithms were proposed like Pastry[47] and Tapestry [48]. These protocols follow basically the same methodology for the next-hop choice, i.e.

the node with the longest common prefix with respect to the searched key is selected, but exploit different routing mechanisms in the overlay. Common features of these schemes are that the size of the routing tables typically increases logarithmically with the size of the network.

Several enhancements have been proposed to DHTs in order to improve performance over resource scarce networks. For example, in Bamboo[50], a proactive overlay management approach intended to react faster to unexpected changes in the underlay network.

While targeting a strict relationship between the overlay and underlay, Georoy[51] have proposed a location-aware variant of the Viceroy algorithm, where a geographical hash function is proposed.

2.1.3 Hybrid

Compared to the previous architectures, the hybrid architecture proposes auxiliary mechanisms in order to facilitate the resource location in P2P systems. For example, in hybrid solutions such as FastTrack[52], a two-tier architecture is proposed in which high capacity peers act as superpeers and forms an upper level which provides resource location services to the other peers at the lower level. Despite superpeers sharing some similari- ties with the centralized architectures, two important factors distinguish them from the central index server. First, a superpeer is not as powerful as the central server as it is in charge of a subset of peers in the network. Second, a superpeer does not only coordi- nate the resource location, but it also acts as a normal peer contributing with its own resources.

BitTorrent[31] introduced the idea of distributing larger files into pieces. By using a mutual distribution of the pieces between the set of peers, called a swarm, BitTorrent pro- portionated further scalability during the download phase. However, as in the centralized architecture, the resource index still relies on a centralized entity known as tracker.

Although file download in BitTorrent is decentralized, the central tracker is recognized as a scalability bottleneck been also susceptible for single control availability[53]. Thus in case the tracker is inoperable, new users can not be bootstrapped into the network.

To overcome this issue, BitTorrent allows the possibility to have multiple trackers per file. In case the primary tracker is down or have a long response time, next trackers are selected.

The use of multiple tracker can increase the reliability, but still depends on the scalability of the centralized trackers. Alternatively, several BitTorrent clients implement a distributed tracker, also known as trackerless, which serve the same purpose as the central tracker but implemented through DHT.

(33)

14 Background

Internet

X X

Access point

Base station Base station

Sink node Sensor

Mesh router Mesh router

with gateway Mesh router

with gateway

Wired clients

Wireless clients Mesh router

with gateway/bridge

Mesh router with gateway/bridge

Wi-Fi networks

Cellular networks

WiMAX networks

Sensor networks Wireless Mesh

backbone

X X

Figure 2.1: Wireless Mesh Network Architecture [20]

2.2 Wireless Mesh Networks

WMNs provide a low cost and easy to deploy solution to build broadband wireless access networks, thereby ensuring connectivity to remote locations. WMNs consist of a backbone of quasi-stationary mesh routers forming a multi-hop wireless networks, in which mesh routers wirelessly relay traffic on behalf of others.

Figure 2.1 shows a WMN architecture, composed of mesh routers and mesh clients.

Commonly, mesh routers form an infrastructure/backbone in order to provide connectivity to mesh clients. With gateway functionality, mesh routers can also provide internet access among themselves and to their clients. The mesh clients are static or mobile devices which are connected to the mesh routers. Different from mesh routers that are connected to the electricity network, wireless mesh clients are more sensitive to power consumption as they can rely on battery. Typical application scenarios for WMNs are community networks, city networks, public safety or the extension of WLAN hotspots.

WMNs can concurrently support a variety of wireless radio and access technologies such as IEEE 802.16 (WiMAX), IEEE 802.11 (WLAN) and IEEE 802.15 (Bluetooth and Zigbee), thus providing the flexibility to integrate different radio access networks[54].

(34)

2.2. Wireless Mesh Networks 15

With the tremendous growth of WLANs and the proliferation of IEEE 802.11 devices, the IEEE 802.11 wireless technology is the most common technology used on the deployment of WMNs. In this thesis we focus on the use of IEEE 802.11 for deploying WMNs.

2.2.1 Multi-channel Multi-radio WMNs

Traditionally, WMNs were based on single channel single radio networks. Thus, due the multi-hop communication characteristic, the throughput capacity of WMNs were highly limited. It has been shown by[55], that on a string topology of n nodes and using carrier sense multiple access with collision avoidance (CSMA/CA) based MAC protocol, the network throughput degraded approximately to 1/n of the channel bandwidth.

Due to the availability of inexpensive off-the-shelf wireless radios, mesh routers have been recently equipped with multiple radios, allowing them to send and receive on multiple frequency bands in parallel and consequently increase the throughput capacity of WMNs. By using all available orthogonal channels and maximizing spatial channel reuse, network traffic can be increased as a larger number of simultaneous transmissions is possible.

While multi-channel multi-radio provides several benefits such as increased throughput capacity, they also face several challenges as routing and channel assignment.

2.2.2 Routing

The main task of routing protocols is to select path(s) between the source node and the destination node. Due to its common characteristics, several mobile ad-hoc networks (MANETs) routing protocols have been used in deployments of WMNs. In order to select the best path(s) among all available, the routing protocol make use of routing metrics.

Routing Protocols

The routing protocols are normally categorized into reactive and proactive schemes. In the reactive scheme, routes are established on-demand when data transfer is requested from a given source to the destination. By operating on-demand, such scheme reduce the amount of routing overhead, however the routing establishment phase may incur an additional delay to the communication. Well-known representatives of such scheme are dynamic source routing (DSR)[56] and ad-hoc on-demand distance vector (AODV)[57].

In AODV, if a source node wants to transfer data, it first need to check if a valid route to the destination already exist. In case no valid route exist, the node initiates a route request (RREQ) message to the destination node broadcasted over the network [57].

The destination, or intermediate node holding already a route to the destination, replies with a route reply (RREP) message. When link fails, a route error (RERR) message is generated in order to trigger route repair.

(35)

16 Background

The proactive routing scheme is characterized by the periodically routing updates exchanged among the nodes, independent of the data transfer. The proactive scheme incurs negligible delays to the communication as routes are always available. In order to avoid excessive routing overhead, the concept of multi point relay (MPR) is introduced by optimized link state routing (OLSR)[58]. The MPRs are selected nodes which forward broadcast messages during the flooding process. Thus, each node maintains the topology information about the network through link state information acquired via topology control (TC) messages generated by the MPRs. Additionally, host and network associa- tion (HNA) messages are generated in order to disseminate external routing information into the network.

In IEEE 802.11s[59], the hybrid wireless mesh protocol (HWMP) is the default routing protocol adopted, and it represents an adaptation of AODV and tree-based routing at layer 2. The IEEE 802.11s standard also proposes the radio aware optimized link state routing (RA-OLSR) as an optional routing protocol. The RA-OLSR is an adaptation of OLSR at layer 2 working with arbitrary routing metrics.

Several benefits arises in case multiple paths are used. The option to allow the use of multiple paths between source and destination have been proposed as an extension to the routing protocols described so far. Examples of multi-path routing protocols are the ad-hoc on-demand multipath distance vector (AOMDV)[60] and the Horizon [61].

As shown in[60], the multi-path routing achieve faster and efficient recovery from route failures and therefore can guarantee better network resilience if compared to AODV.

Multi-path routing also allows network nodes to perform load balancing among the multiple paths selected, and therefore increase throughput and fairness among network flows [61].

Routing Metrics

What is common to all routing protocol is the use of a routing metric in order to select the best route(s) among all available paths in the network. In single channel WMNs, the minimum hop count is the most common metric as it aims at reducing the amount of concurrent link access in order to minimize the network bandwidth utilization in the network. However such simple strategy does not consider important factor inherent to wireless scenarios, such as interference level, link errors and criticality. Moreover, in multi-channel multi-radio WMNs the channel diversity is also an important factor that needs to be taken into account during the path selection.

A variety of routing metrics have been proposed for WMNs, providing a large range of routing algorithms with different objectives. The expected transmission count (ETX) is one of the first routing metrics that accounts for wireless link quality. According to [62], the ETX of a link is the predicted number of data transmission required to send a packet over that link, including retransmission. ETX has been implemented in various routing protocols such as AODV, DSR and OLSR.

Since the size of the probe packets is fixed and the loss ratio in wireless networks varies with packet size, inaccurate prediction are implicit to ETX. Moreover, ETX ignores the

(36)

fact that different links can operate over different data rate and therefore consumes different amount of channel time. The expected transmission time (ETT) extends ETX by applying a packet pair probing technique in order to estimate the link rates.

Given the availability of multiple orthogonal channels, the weighted cumulative ETT (WCETT) metric proposes an adaptation of ETT accounting for the channel diversity [62]. The WCETT metric gives a trade-off between throughput and delay as it measures path latency and bottleneck link along available paths. One limitation of WCETT is its lack on accounting for inter-flow interference, which may cause path selection with higher number of interfering neighbors and therefore worst performance.

Metrics such as the metric of interference and channel-switching (MIC)[63], the interference aware (iAWARE)[64] and the metric for interference and channel diversity (MIND)[65], are an evolution of WCETT as they make use of number of interfering neighbors, SNR/SINR correlation, and traffic load, respectively, in order to account for interference.

2.2.3 Channel Assignment

The use of multi-channel multi-radio WMNs opens up the possibility for different channel assignment schemes with the common objective of assignment channels to radio interfaces in order to achieve efficient channel utilization and minimize interference.

Depending on the criteria selected, different classifications of channel assignment arises. In[66, 67], the channel assignment is classified according to the frequency that channels are modified, and can be described as static, dynamic, semi-dynamic and hybrid.

Other classifications, e.g. according to the level of centralization and implementation as- pects, are also available[68].

Static Channel Assignment

In the static channel assignment, channels are permanently fixed to the radio interfaces.

The most simple example of such assignment is the use of a common channel assignment, where the radio interfaces of each node are all assigned to the same set of channels[62].

Such approach guarantees similar network connectivity if compared to the single channel approach.

Dynamic and Semi-dynamic Channel Assignment

On the other hand, the dynamic channel assignment proposes channel changes on-demand, e.g. on a per-packet basis, and therefore requires frequent channel switching within each node. According to[69], the channel switching time on current off-the-shelf hardwares are between 200µs and 20 ms, causing a very high overhead for per-package switched channel. Moreover, a high level of coordination between the nodes is necessary in order to assign the channels at the right time and maximize the overall channel utilization.

In the semi-dynamic scheme, channels are re-assigned at larger time intervals, e.g.

several minutes or hours, depending on specific metrics such as network demand and/or

(37)

18 Background

interference. A channel assignment scheme that takes into account the network load demand in order to re-assign channels to radio interfaces is proposed by[70]. Thus, depending on the node position and traffic load,[70] re-assigns links to channels in order to accommodate the link expected demands.

By assuming that traffic is commonly target to the internet, several channel assignment strategies make use of a BFS fashion assignment where nodes close to the gateways have higher priority while reassigning channels[70, 71, 72]. By incorporating such parent-child relationship in the channel reassignment, those strategies avoid non con- vergence states during the channel reassignment period. However, ripple effects might occurs as the channel reassignment at a given node might change due to the variations in one of its neighbor’s channel assignment[67].

While capturing packets from the medium, each node in[71] can measure the number of radio interfaces and the bandwidth consumed by those radios in order to estimate interference on a given channel. Based on those periodical channel interference information gathered by a central server,[71] proposes a channel re-assignment scheme that minimize the interference within the wireless mesh network.

Hybrid Channel Assignment

The hybrid channel assignment strategy combines both semi-dynamic and dynamic assignment properties. Net-X[73], a known hybrid approach, applies a semi-dynamic assignment to the its fixed interface, used primarily for receiving data from neighbors, and a dynamic assignment to the switchable interfaces, used to transmit data to its neighbors.

A local balancing algorithm that looks for the less utilized channel of a given node’s neighborhood is applied to the fixed interface by assuming similar traffic distribution over the channels[73]. If a node notices that the number of nodes using the same fixed channel as itself is large, it can reassign its interface to a less used channel and inform its neighbors. Based on the observation that traffic demands are mostly not uniformly distributed in a WMNs,[9] proposes a demand-aware channel assignment which tries to assure that nodes with high traffic demands are assigned to less loaded channels.

The switchable interface can be used to transmit to neighbors whose fixed interfaces may potentially be on different channels[73]. Thus, the channel in the switchable interface can be changed frequently without having to inform all neighbors, as it depends only on the next hop destination of the packet been transmitted.

Since the channel switching time incurs a non-negligible delay, a queuing algorithm to buffer packets is required, as well as a round robin scheduling policy to transmit buffered packets in order to reduce frequent switching. In order to consider the requirements of real time traffic,[10] proposed to replace the round robin scheduler by a quality of service (QoS) aware scheduler that takes into account packet priorities. By applying the QoS-aware channel scheduler,[10] reduces the end-to-end delay and network jitter of voice over IP (VoIP) traffic while still guaranteeing reasonable throughput of non-priority TCP traffic.

(38)

2.2.4 Joint Routing and Channel Assignment

In multi-channel multi-radio WMNs, the problem of routing and channel assignment are interdependent. The channel assignment determines the set of links sharing the same channel and consequently the channel capacity and network topology. On the other hand, the routing impacts the link bandwidth along the selected path, changing also the interference level among links that share a common channel. Thus, a joint routing and channel assignment mechanism is desirable in order to maximize network capacity.

Several works in the literature propose optimization frameworks in order to jointly solve the routing and channel assignment problem, by using linear programming formu- lations[74, 75] and also heuristic approaches [66, 70, 76, 77].

In[70], the routing and channel assignment problem are solved using an interactive approach, where an initial link load estimation is used to identify the capacity of a link based on the set of communication nodes and interference range links. Given the expected load estimation, channel assignment and routing are performed interactively in order to guarantee the expect traffic demands.

Following a distributed heuristic approach,[77] proposed a joint coordination between channel assignment and routing based on the traffic information measured and exchanged among two hop neighbors. Since the network global view is not assumed, every node performs locally channel assignment and routing decisions based on the channel cost metric (CCM) which represents the cost of channel time weighted by channel utilization. In case the measured metric on a given channel is higher than its predefined threshold, representing a channel overload situation, the node start to check for each neighbor node and channel, if a channel re-assignment, re-routing, or a combination both is necessary. Channel re-assignment is negotiated among a node’s two hop neighborhood, whereas the search for new routes is restricted by only changing the interfaces between adjacent nodes, while the node sequence of the entire path remains the same.

Moreover, joint optimizations involving other layers have also been proposed, such as the joint optimization of channel bandwidth adaptation, topology control and routing by[78] and the channel assignment, routing and rate assignment by [79].

2.2.5 Modeling Capacity in WMNs

In order to model the capacity in wireless networks, two popular wireless interference models formalized by[80] are used; the physical interference model and the protocol interference model.

The physical interference model, also known as signal to interference plus noise ratio (SINR) model, is based on the power capture model which states that a transmission is correctly received if the SINR at the receiver is greater than a threshold so that the transmitted signal can be decoded with an acceptable bit error probability. Given that, [80] derives lower and upper bounds on the wireless capacity.

The known accuracy of the physical interference model comes with its high com- plexity cost when it involves cross-layer optimizations in a multi-hop network[81]. To

Enhancing P2P Systems over Wireless Mesh Networks

Marcel Cavalcanti de Castro

Enhancing P2P

Systems over Wireless

Mesh Networks

Marcel Cavalcanti de Castro

Enhancing P2P

Systems over Wireless Mesh Networks

Enhancing P2P Systems over Wireless Mesh Networks

Abstract

Acknowledgments

Contents

List of Figures

List of Tables

Chapter 1

Introduction

1.1 Problem Definition and Research Questions

1.2 Thesis Outline and Contributions

1.3 Other Related Papers

Chapter 2

Background

2.1 Peer-to-Peer Systems

2.2 Wireless Mesh Networks