Routage Cross-Layer et Gestion de la Mobilité dans les Réseaux Maillés Sans fil

Thèse de Doctorat de l Université Paris VI Pierre et Marie Curie Spécialité Systèmes Informatiques présentée par M. Luigi Iannone pour obtenir le grade de Docteur de l Université Pierre et Marie Curie Routage Cross-Layer et Gestion de la Mobilité dans les Réseaux Maillés Sans fil Soutenance prévue le 3 Avril 2006 devant le jury composé de Andrzej DUDA Rapporteur Prof. à l INPG-ENSIMAG de Grenoble Catherine ROSENBERG Rapporteur Prof. à l Université de Waterloo Isabelle GUERIN-LASSOUS Examinateur Chargé de Recherche INRIA Luigi RIZZO Examinateur Prof. à l Université de Pisa Bertil FOLLIOT Examinateur Prof. à l Université Pierre et Marie Curie Serge FDIDA Directeur Prof. à l Université Pierre et Marie Curie

Thèse de Doctorat de l Université Paris VI Pierre et Marie Curie Spécialité Systèmes Informatiques présentée par M. Luigi Iannone pour obtenir le grade de Docteur de l Université Pierre et Marie Curie Routage Cross-Layer et Gestion de la Mobilité dans les Réseaux Maillés Sans fil Soutenance prévue le 3 Avril 2006 devant le jury composé de Andrzej DUDA Rapporteur Prof. à l INPG-ENSIMAG de Grenoble Catherine ROSENBERG Rapporteur Prof. à l Université de Waterloo Isabelle GUERIN-LASSOUS Examinateur Chargé de Recherche INRIA Luigi RIZZO Examinateur Prof. à l Université de Pisa Bertil FOLLIOT Examinateur Prof. à l Université Pierre et Marie Curie Serge FDIDA Directeur Prof. à l Université Pierre et Marie Curie Numéro bibliothèque:

Doctor of Science Thesis Pierre and Marie Curie University (Paris VI) Specialization Computer Science presented by M. Luigi Iannone Submitted in partial satisfaction of the requirements for the degree of Doctor of Science of the Pierre and Marie Curie University Cross-Layer Routing and Mobility Management in Wireless Mesh Networks Commitee in charge: Andrzej DUDA Reviewer Prof. at INPG-ENSIMAG of Grenoble Catherine ROSENBERG Reviewer Prof. at University of Waterloo Isabelle GUERIN-LASSOUS Examinator INRIA Research Officer Luigi RIZZO Examinator Prof. at University of Pisa Bertil FOLLIOT Examinator Prof. at Pierre and Marie Curie University Serge FDIDA Advisor Prof. at Pierre and Marie Curie University

À ma précieuse famille. The man who follows the crowd will usually get no further than the crowd. The man who walks alone is likely to find himself in places no one has ever been. (Alan Ashley-Pitt)

Remerciements Ces années de thèse passée au sein du LIP6 représentent une expérience inoubliable. La période passé ici en France ne m a pas seulement apporté un diplôme de Docteur de l Université Paris VI, elle m a aussi permis de grandir tant au niveau humain que professionnel. Il est difficile de trouver les mots pour remercier toutes les personnes qui j ai rencontré dans cette période. Il est aussi impossible de faire une liste exhaustive, car elle serait trop longe et il y aura toujours quelqu un qui j oublier. Je tiens à remercier Andrzej DUDA et Catherine ROSENBERG pour avoir consacré un peu de leur précieux temps à rapporter ce travail. Je tiens à exprimer ma gratitude à Luigi RIZZO, Isabelle GUERIN-LASSOUS et Bertil FOLLIOT pour avoir accepté de faire partie de mon jury en leur qualité d examinateurs. Un remerciement particulier a Serge FDIDA qui m a donne la possibilité de faire une thèse en m accueillant au LIP6, qui a dirigé mes travaux et qui m a toujours soutenu. Je voudrais remercier Marcelo DIAS DE AMORIM pour les conseilles et l aide qui m a donné tout au long de ma thèse, à partir de la période où on a partagé le bureau. Mes remerciements vont aussi à Konstantin KABASSANOV qui m a permis de réaliser toute la partie pratique de mon travail. Finalement, j aimerais exprimer ma gratitude toute particulière à Ramin KHALILI avec qui j ai passé beaucoup de soirée a discuter de recherche et de vie durant notre promenades sur les quai de la seine. Mes derniers remerciements vont à toutes les personnes que j ai rencontré pendant toute la période de ma thèse et que m ont permis de apprendre toujours quelque chose de nouveau et qui m ont aussi aidé. Merci à tous, de tout mon cœur.

Résumé La révolution sans fil déclenchée par le succès de la norme IEEE 802.11 a poussé la communauté de recherche dans la conception, l analyse, le développement, et le déploiement de nouvelles solutions sans fil. En particulier, les Réseaux Maillés Sans fil (Wireless Mesh Networks - WMN), dont l architecture à deux niveaux est basée sur les Routeur Sans fil Maillés (Wireless Mesh Routers - WMR), ont capturé l intérêt de la recherche universitaire et de l industrie, en raison de leur capacité à satisfaire à la fois les exigences des Fournisseurs d Accès à Internet (FAI) sans fil et les utilisateurs. En dépit de la grande quantité de recherche effectuée dans ce contexte, plusieurs problèmes restent encore à résoudre. Les points suivants sont devenus cruciaux : (i) l utilisation traditionnelle du plus court chemin en termes de nombre de sauts n est pas représentative de la qualité des liens sans fil, ce qui entraîne une réduction du débit ; (ii) il manque une solution définitive capable de gérer de façon efficace la mobilité des utilisateurs ; (iii) la conception des WMRs est encore incomplète, les réalisations existantes étant basées sur des solutions expérimentales, souvent propriétaires. Dans cette thèse, nous abordons les points mentionnés ci-dessus et proposons une solution efficace pour les réseaux sans fil, basée sur les propriétés réelles des liens sans fil. Nos contributions sont les suivantes. D abord, nous concevons et formalisons MRS (Mesh Routing Strategy), une stratégie de routage cross-layer. MRS ne dépend d aucune spécificité de la couche MAC et se base sur des heuristiques simples et capables d optimiser la puissance de transmission et de calcul des chemins. En second lieu, nous concevons et implémentons MeshDV, une solution de routage adaptée aux WMNs, qui inclut MRS et peut également gérer efficacement la mobilité des utilisateurs. MeshDV combine le calcul proactif des chemins entre les routeurs et une mise en place réactive des communications entre les clients. Troisièmement, nous mettons au point MeshDVbox, un WMR basé sur IPv6 et construit en utilisant des composants disponibles sur le marché public. Des MeshDVbox ont été déployés dans notre laboratoire pour réaliser une plateforme de tests expérimentale. Cette thèse adresse des aspects théoriques, à travers la formalisation de MRS, mais aussi des aspects pratiques, à travers le développement de MeshDV. Le travail d évaluation s étend de la simulation de MRS à l implémentation de MeshDV (qui intègre MRS) et au déploiement d une plateforme de test en utilisant MeshDVbox. Mots-clés : Réseaux sans-fil multi-sauts, réseaux maillés sans fil, protocoles de routage, conception cross-layer, contrôle de puissance, capacité de transfert, gestion de la mobilité. i

Abstract The wireless revolution triggered by the success of IEEE 802.11 standard has pushed the research community to address the design, analysis, development, and deployment of new wireless solutions. In particular, Wireless Mesh Networks (WMN), a two-tier architecture based on Wireless Mesh Routers (WMR), has captured the interest from both academia and industry, because of their ability to accommodate both Wireless ISPs and user requirements. Despite the large amount of research developed in this area, several issues still need to be solved. The following points have become a main concern: (i) the traditional use of the hop-count metric does not capture the very nature of wireless links, resulting in poor transport capacity of wireless systems; (ii) the lack of a definitive solution able to effectively manage mobile users; (iii) the design of WMR is still weak, with custom implementations often based on proprietary solutions. In this thesis we tackle the abovementioned issues and design an efficient solution for multi-hop wireless networks that uses the real properties of wireless links. We provide the following contributions. First, we design and formalize Mesh Routing Strategy (MRS), a cross-layer routing approach. MRS does not depend on any specificity of the MAC layer and relies on simple heuristics able to optimize transmission power and route selection, resulting in important increase of the transport capacity of the mesh backbone. Second, we design and implement MeshDV, a routing framework adapted to WMNs, which includes MRS and is also able to easily manage user mobility. MeshDV combines proactive route computation for routers and on-demand path setup for clients. Proactive route computation is achieved by MRS, while on-demand path setup is performed by new mechanisms that take benefit from the two-tier architecture of WMNs. Third, we develop MeshDVbox, a custom IPv6 based Wireless Mesh Router built using off-the-shelf components and deployed in our laboratory to realize a WMN test-bed. The tests performed on the test-bed demonstrate good results, for both throughput and latency in mobility management. This thesis addresses both theoretic aspects, in the formalization of MRS, as long as practical issues, in the development of MeshDV. The evaluation work ranges from simulation of MRS to real implementation of MeshDV (which integrates MRS) and to the deployment of a real test-bed using MeshDVbox. Key Words: Wireless multi-hop networks, wireless mesh networks, routing protocols, cross-layer design, power control, throughput capacity, mobility management. iii

Table of contents 1 Introduction 1 1.1 Wireless Mesh Networks........................... 4 1.2 The challenge of Routing in Wireless Multi-hop Networks........ 8 1.3 Contribution of the thesis.......................... 10 1.4 Outline.................................... 14 2 Routing and Mobility in Wireless Multi-hop Networks 15 2.1 The problem of finding good paths..................... 16 2.2 The problem of managing mobility..................... 19 2.3 Shortest-Path (Hop-count) routing approaches.............. 20 2.3.1 Destination Sequenced Distance Vector Routing Protocol.... 21 2.3.2 Ad Hoc On-Demand Distance Vector Routing Protocol..... 23 2.3.3 Dynamic Source Routing Protocol................. 24 2.3.4 Optimized Link State Routing Protocol.............. 26 2.4 Mesh Routing Protocols........................... 27 2.5 Standardization Activities.......................... 29 2.6 Cross-Layer based approaches........................ 30 2.7 Routing and mobility............................. 33 2.8 Conclusion.................................. 34 3 MeshDV Architecture and Mobility Management 37 3.1 MeshDV Architecture............................ 38 3.2 MeshDV communication setup/holding................... 41 3.2.1 Communication setup........................ 42 3.2.2 Holding a communication...................... 44 3.3 MeshDV implementation architecture................... 46 3.3.1 Enhanced DV Module........................ 46 3.3.2 Client Manager Module....................... 47 3.3.3 NDP-Proxy Module......................... 50 3.3.4 Forwarder Module.......................... 50 v

vi Table of contents 3.4 MeshDV Analysis............................... 53 3.4.1 MeshDV Hardware platform and implementation issues..... 53 3.4.2 MeshDV Performance........................ 56 3.5 Conclusion.................................. 59 4 Cross-Layer Routing to improve capacity 61 4.1 Capacity Law and Cross-Layer Approach................. 62 4.1.1 Capacity Law and Cross-Layer Approach............. 63 4.1.2 Capacity Law and Physical Layer Modeling............ 63 4.2 Routing Metrics............................... 65 4.2.1 Transmission Rate Metric...................... 65 4.2.2 Interference Metric.......................... 65 4.2.3 Packet Error Rate Metric...................... 68 4.3 Proof of NP-completeness.......................... 69 4.4 Routing Strategy............................... 70 4.4.1 Local Power Optimization Strategy................. 71 4.4.2 Path Cost and Route Computation................. 75 4.5 Conclusion.................................. 76 5 Performance Analysis of Mesh Routing Strategy 79 5.1 The Capacity Index............................. 81 5.2 NS-2 enhancement.............................. 82 5.3 Simulation Setup............................... 84 5.4 Simulation Results.............................. 85 5.5 Conclusion.................................. 92 6 Conclusion and Further Research 93 6.1 Conclusion.................................. 94 6.1.1 MRS.................................. 95 6.1.2 MeshDV................................ 96 6.1.3 MeshDVbox.............................. 97 6.2 Further Research............................... 98 Appendix - Thesis French Version 101 A La Problématique 105 A.1 Les réseaux maillés sans fil.......................... 106 A.2 Le Routage dans les réseaux multi sauts sans fil.............. 107 A.3 Protocoles de routage pour les réseaux maillés sans fil.......... 108 A.4 La gestion de la mobilité........................... 109 B Contributions de cette thèse 111 B.1 L architecture de MeshDV.......................... 112 B.2 Le WMR MeshDVbox............................. 114 B.3 Performance de MeshDV........................... 116

Table of contents vii B.4 Mesh Routing Strategy............................ 118 B.4.1 Les Métriques de Routage...................... 118 B.4.2 Stratégie de routage......................... 120 B.5 Performance de MRS............................. 122 B.5.1 Résultats des simulations...................... 123 C Conclusion 127 C.1 Conclusion.................................. 128 C.2 Perspectives.................................. 130 References 131 List of Acronyms 143 List of Figures 147

Chapter1 Introduction Last years have witnessed a still increasing interest in wireless networking by the research community, but also industries and Wireless Internet Service Providers (ISP). Initially triggered by the unexpected success of the Institute of Electrical and Electronics Engineers (IEEE) 802.11 standard [1], this wireless revolution is supported by the proliferation of wireless computing and communication devices, driving a deep change in our information society, causing modifications in the way networks and applications should be designed. Compared to the wired Internet, which is facing its middle age, wireless networking is relatively young. The great amount of research done insofar has already leaded to new architectures and protocols paradigms. Born as a natural extension of the wired Internet, by deploying Wireless Local Area Networks (WLAN) based on wired Access Points (AP), ever since wireless networking has evolved toward new architectures. The concept of ad hoc networking, which has introduced the idea of multi-hop routing in mobile wireless networks, has been the first step through more sophisticated and large scale approaches like Sensors Networks and, more generally, Self-Organizing Networks

2 (SON). Aiming to offer seamless connectivity, Wireless Mesh Networks (WMN) have gained more and more interest from both academic and industry, as a mean to speed-up the deployment of large scale networks, reducing costs while guaranteeing a minimum level of service. Recently, all this amount of innovation has pushed the research community to think about wireless networking in a more general way, leading to the very young idea of Autonomic Communication, a new paradigm for next generation networks. The above cited advancements in networking, with mobile devices becoming cheaper and cheaper, and the explosion of Internet users made the ubiquitous computing ([2], [3]) reality. The wireless technology fits well the requirements to be the last mile interconnecting technology, in order to build stub networks in a fast, cheap, and simple way, offering broadband seamless connectivity. In this context, however, not all of the actual wireless architectures are suitable for this purpose. In particular, ad hoc networks are not able to give a minimum level of service guarantee, since each user, while part of the routing/forwarding infrastructure, behaves independently, leading to issues like network merge and split. Thus, they cannot guarantee the very basic need of each user: connectivity. At the same time, since the network is completely composed of users equipment, Wireless Internet Service Providers (ISP) do not have any control on the network. Users and operators have different viewpoints, needs, and priorities. In particular users are usually interested in: Ubiquitous Connectivity. Users demand seamless connectivity, whenever they want, wherever they move. No strange setups. The average user usually has no interest to a service that requires performing more than few steps, with strange or complicate setups, to obtain relative benefits. Furthermore, the average user is used buying a service, not paying to be part of a service (e.g. ad hoc). Control their resources. Users are selfish; they do not share resources so easily. Finding good incentives to motivate user to collaborate is still an open issue. Average user may agree in collaborating with neighbors or other user she/he knows, but has hesitation on sharing his resources with strangers (e.g. freeriders in file sharing applications). The open question remains: Why should I (user) forward packets on behalf of others?

Chapter 1. Introduction 3 No third part software. Like the previous point, users prefer to have the feeling that they have the total control of their equipments (laptop/pc/communication devices). They prefer not having to install software like connection managers, which may contain tracking features collecting information about each connected user. Quality of Service (QoS). For the average user, beside the ubiquitous connectivity, QoS means broadband. Independently from the particular use of the network, which may range from web surfing to heavy file transfers, the only quality usually perceived is the downloading time. On the other hand, Wireless ISPs are interested in: Offering Broadband Connectivity. This is the basic service operators can sell. Ease and low cost deployment. Using wireless links to build a stub network in order to offer connectivity, thus avoiding heavy and expansive cable deployment. Control of the network. If traffic goes only through proprietary devices, operators can control the network, perform traffic engineering, and guarantee a certain QoS. This would be hardly possible if they rely on users devices to form the basic infrastructure. Capacity. The concept of capacity in the context of wireless multi-hop networks goes over the concept of bandwidth of single links. Since wireless links are not independent one from another, the key parameter becomes the total amount of traffic the network is able to relay. The consequences of a higher capacity of the network for the operator are two-fold. More capacity means to have a more cost/effective solution. At the same time, more capacity means the possibility of managing the traffic of an increased number of users, thus increasing the benefits. In order to accommodate, sometimes contrasting, requirements of both users and Wireless ISPs, simple flat architecture like ad hoc is not sufficient. This is the reason why there is an increasing interest for WMNs, whose architecture is two-tier and seems able to satisfy the requirements of both parties.

4 1.1. Wireless Mesh Networks 1.1 Wireless Mesh Networks Wireless Mesh Networks are an emerging two-tier architecture based on multi-hop transmission, with two fundamental objectives: (a) to offer connectivity to end-users; (b) to form a self-organized wireless backbone. These two characteristics make WMNs different from ad hoc and other infrastructure-based networks. In a WMN, nodes can be distinguished in Wireless Mesh Routers (WMR) and mesh clients. 1 Wireless Mesh Routers interconnect and self-organize themselves in order to form a meshed wireless backbone. This backbone can be a self-standing network, simply offering inter-user connectivity. Otherwise, if connection is available through one or more WMRs acting as gateways, it might be considered as a local wireless extension of the Internet. Each WMR covers a region where it offers connectivity by acting as an AP, thus building a WLAN sub-network offering access to any allowed client. End-user terminals associate to a WMR without the need to run any routing feature or particular software module, and communicate by means of the WMRs backbone. Figure 1.1 shows an example of this type of mesh network. The two-tier architecture of WMNs differentiates them from ad hoc networks. The latter is a flat architecture where each node is an end-user terminal enhanced with routing features in order to forward traffic on behalf of others. Each node in ad hoc networks can move freely in the environment. Thus, routing has to face this mobility issue by assuming that the topology is highly dynamic, links are fragile, and no dedicated infrastructure components are present. In this context, a winning approach is the on-demand path setup coupled with high reactive approaches to maintain up to date information and path breakage recovery. Routing in wireless mesh networks is only performed between WMRs, i.e. the wireless backbone. In a WMN, end-users still move freely in the environment although WMRs are fixed. In this context, proactive path setup may be a rewarding choice, since the backbone has a low level dynamics. On a large time-scale, however, WMRs can still be added (e.g., to increase capacity) or reduced (e.g., in case of failure). Communication between end-users is provided by the WMR managing the region where they are located. Thus, in the case of WMNs, mobility has no impact on building paths between WMRs, but the issue becomes to locate which client is associated to which 1 For simplicity of presentation, in the remaining of this thesis mesh clients will be called merely clients, and we will use the terms clients and users as synonyms. Furthermore, the terms WMN, wireless mesh networks, and mesh network will be all used as synonyms.

Chapter 1. Introduction 5 Mesh Backbone Subnetwork Client Access Subnetwork Wireless Mesh Router Figure 1.1: An example of Wireless Mesh Network. WMR and to correctly setup and maintain a communication between users associated to different WMRs. Solutions like Mobile IP [4] or Host Identity Protocol (HIP) [5] can be used to manage end-users migrating from one WMR s region to another. Nevertheless, as we will see in the next chapter, they are not appropriate to WMNs. An alternative approach is to let the routing protocol include mechanisms to manage the mobility of users. Wireless Mesh Networks fit well all requirements of both users and Wireless ISPs, mentioned in the previous section. As long as users move in an area covered by good radio signal, they have seamless connectivity, with good QoS (compatible with the radio standard used) without the need of complicate setups. Indeed, WMNs offer to users common WLAN access, which is a sufficiently consolidated technology that is correctly supported by any Operating System (OS), and connection is correctly managed without the necessity to install third party connection manager software modules. Resources of the user are controlled by herself/himself, who can decide at which level she/he wishes to share them. From an ISP s viewpoint there is the possibility to sell connectivity. As an owner of the mesh backbone, an ISP may reduce costs related to its deployment. At the same

6 1.1. Wireless Mesh Networks time, it can manage the network the way it prefers, update it without bothering users (e.g., to deploy a new solution to increase the capacity of the backbone). Furthermore, since usually WMRs have at least two wireless interfaces, one or more for the mesh backbone and one or more for the client access, it plays a key role for technology integration (e.g. Wireless Fidelity (WiFi) [6], WiMAX [7], ZigBee [8]). D. Beyer well describes the benefits provided by mesh networks ([9]), namely: easy and fast network deployment, low cost of installation and maintenance, flexibility, scalability in both size and density. Yet if promising, the mesh technology has still open issues that need to be solved. Akyildiz et al. give a complete overview of WMNs [10], with an analysis of the challenges and the open issues of this kind of networks. These open issues range from the physical layer, where high-speed physical techniques are under study, up to the application layer, in order to make the existing Internet applications to work under the architecture of WMNs. At Medium Access Control (MAC) level research is focusing on improving existing protocols, but also in finding new ones with an improved synchronization in order to exploit the multi-hop nature of communications [11]. While the Internet Protocol (IP) is accepted as a standard protocol at network layer, several other protocols present at this level need to be improved. Routing protocols that have enhanced metrics able to better model the wireless link cost are an active research field. Broadcast is explored in order to avoid the excessive retransmission of the same message in the medium. Address auto-configuration still suffer of the lack of a general solution. Mobility, which has important consequences in the correct behavior of all protocols, is under heavy research. Transport layer and related protocols have showed poor and unpredicted behavior in the wireless environment, since hypothesis valid in the wired Internet do not hold anymore in the wireless context. Besides the above issues, WMNs throughput capacity, i.e. the global traffic the network is able to relay, is among the key factors affecting the scalability, in terms of users able to effectively take advantage of the network. Increasing the effective throughput capacity, pushing it toward the theoretical bound, would let WMNs become

Chapter 1. Introduction 7 even more a cost-effective solution for wireless ISPs. Moreover, the capacity of WMNs is affected by many factors such as the architecture, network topology, traffic pattern, WMRs density, transmission power level, and mobility. A clear understanding of the relationship between network capacity and the above factors provides a guideline for protocol development, architecture design, deployment and operation in the network. Traditionally, different protocol layers are required to be transparent from each other. This makes the protocol development and implementation a simple and scalable process. However, the methodology of layered protocol design does not necessarily lead to the optimum solution. The physical channel in a wireless environment is variable in terms of capacity, bit error rate, signal propagation, and interference. Despite the great advances achieved in last years, there is no way to guarantee fixed capacity, zero packet loss rate, or reliable connectivity as expected by higher layers. Therefore, higher layer protocols will be inevitably affected by the unreliable physical channel. Thus, in order to improve protocol efficiency, cross-layer design becomes indispensable. Cross-layer design can significantly improve network performance. However, certain issues must be considered when carrying out cross-layer protocol design. Cross-layer design presents some risks due to: Loss of protocol layer abstraction. Incompatibility with existing protocols. Unforeseen impact on the future design of the network. Difficulty in maintenance and management. Thus, certain guidelines, like the following ones, need to be respected: Well define all dependencies between different parts of the protocol stack to avoid unpredicted behaviors. Avoiding to introduce several control loops on the same parameter at different levels of the protocol stack, risking to introduce instability. Have a structured implementation, avoiding to introduce fragmented functionalities resulting in a spaghetti-like code difficult to maintain. More details can be found in the work of Kawadia et al. [12].

8 1.2. The challenge of Routing in Wireless Multi-hop Networks To overcome all the above deficiencies, this thesis focuses on a cross-layer approach in the design of routing protocols targeting both the increase of the transport capacity of the mesh backbone, as well as an efficient users mobility management. We show that well designed metrics may drastically increase the capacity of the mesh backbone. Moreover, we show that a well design modular routing architecture, leveraging on the two-tier of WMNs, can efficiently manage users mobility. 1.2 The challenge of Routing in Wireless Multi-hop Networks Protocols and algorithms paradigms thought as consolidated solutions in the context of the wired world, have showed heavy and unpredictable limits in the context of the young wireless world. Classical routing approaches are not sufficient anymore in the wireless context, since the definition of good path has now a different and more complex meaning. Furthermore, mobility management needs smarter solutions than the simple Mobile IP proposition, since in the multi-hop context, even the addressing do not follow anymore the hierarchy of the Internet. This lack of hierarchy and sub-networking in the addressing space of wireless networks has led to the concept of multi-hop forwarding. In the wired Internet, routing intrinsically means to change sub-network, i.e. interface, and the hierarchical architecture allows to aggregate paths. In multi-hop communications, this is not true anymore, since the ingoing and outgoing interfaces are often the same, and no route aggregation is possible. What changes in each transmission is the intended next hop at the data link level. In a certain way, this means to get back to the very early and basic definition of routing. In this context, the first attempts of multi-hop routing protocols have failed. They were an adaptation of existing approaches for the wired environment, which have proved not to be well performing in the wireless context. Since route aggregation is no longer possible, the average routing table sizes per number of nodes is increased, leading to a more complex management, overhead in update transmissions, and lower performance due to the increased computational cost. The above mentioned issues are only partially responsible for the poor behavior of traditional routing approaches. A major reason consists in the inaptness of the hopcount shortest-path metric usually used in the wired Internet. In the wired Internet

Chapter 1. Introduction 9 there are three main routing protocols: Routing Information Protocol (RIP) ([13]), Open Shortest Path First (OSPF) ([14]), and Border Gateway Protocol (BGP) ([15], [16]). RIP is the simplest protocol and it uses the hop-count metric. OSPF and BGP use instead an integer link cost that is set when the demon is configured, in order to perform traffic engineering by operators. This integer cost is totally unrelated from the quality of the link, remaining basically very close to the hop-count concept. This approach cannot be used in mobile wireless networks, due to the continuous changes in the topology. For this reason, the hop-count metric of RIP has been widely adopted, however, this metric do not capture the very nature of wireless links. Today s devices may offer several different transmission rates at physical layer. The selection of the more appropriate depends on the distance of the next-hop, on the interferences on the link, and on the previously experienced losses. Packet losses and achievable throughput on each link may vary heavily in time and space. Even if all radio devices use the same technology, which is a basic requirement for any communication, each link may experience a very different behavior. This factor is totally ignored by routing protocols using the hop-count shortest path metric. The result is that traditional routing set up very poorly behaving paths. Thus, research has moved to new approaches since the wireless environment present some peculiarities that cannot be overcome by traditional (layered) paradigms. In the quest of finding routing metrics able to correctly model the wireless link behavior, a myriad of routing protocols propositions have been published. Most of them are based on cross-layer approaches including power control features, but often relying on strict theoretical assumption. Although several works, including the present thesis demonstrate that routing joint with power control and with well designed metrics may drastically improve capacity of multi-hop networks, differently from our proposition, other cross-layer solutions have seldom achieved real implementation. This is partially due to: (i) lacks in current technology, which is not able to support some advanced features; (ii) the complexity in implementing a cross-layer solution into today s protocol stacks present in current OSs; (iii) the unrealistic assumption of proposed solutions. Another issue that arises in wireless multi-hop networks is mobility. Nevertheless, it is only perceived at routing level as a path fragility and network topology discovery issue and never really integrated with the routing algorithm. We argue that, in the context of mesh networks, approaches that take full advantage of the two-tier architecture may lead the routing protocol to effectively manage user mobility.

10 1.3. Contribution of the thesis The above discussion lead to recognize the following points as the key requirements in the design of routing solutions for wireless multi-hop networks: Give to the routing layer the awareness of the underlying issues, by means of metrics that well model the physical layer. Define the objective of route optimization when performing cross-layer optimization. Accommodate on-demand and proactive mechanisms in order to have low path setup delays without exploding routing tables sizes or introducing excessive overhead. Avoid unrealistic or theoretic assumptions that do not match reality and limit the implementation of the proposed solution. Integrate mobility in the routing, not just considering only its consequences (i.e. route fragility). 1.3 Contribution of the thesis WMNs are a promising technology that is already being commercialized ([17], [18], [19]). Nevertheless, existing WMRs are based either on adaptation of protocols designed for the wired Internet or on the adaptation of solution originally designed for ad hoc networks. Moreover, WMRs all use the Internet Protocol Version 4 (IPv4) as a network layer protocol and only few are using Internet Protocol Version 6 (IPv6), the new standard network layer protocol. We believe that in order to have good, robust, and well performing protocols, they have to be designed taking full advantage of the two tier architecture of WMNs and the enhanced mechanisms of IPv6. The common practice to integrate existing blocks, usually designed for IPv4, leads to non optimal solutions that under-exploit both the network architecture and the enhanced features of IPv6. Real wireless multi-hop networks still perform poorly when compared to the theoretical upper bound they can achieve, despite the fact that their capacity has been thoroughly analyzed and formalized [20]. On the one hand, this gap can be partially justified by the fact that some theoretical hypothesis do not hold in reality. On the other hand, much can still be done to increase actual performance of multi-hop networks. In particular, we argue that cross-layer routing based on power control and on

Chapter 1. Introduction 11 well designed metrics can drastically increase the throughput capacity experienced by real WMNs. This thesis starts investigating the main issues of routing and mobility in wireless multi-hop networks. In particular, by analyzing the routing problem, we show the limits of traditional hop-count shortest-path approaches in wireless environments. We show the main reasons of this failure and the deep impact that routing has on the experienced transport capacity of real networks. We survey cross-layer approaches proposed in the literature to overcome these issues, aiming to increase the transport capacity. These solutions are usually based on a specific Medium Access Control (MAC) layer or rely on tight theoretic hypothesis that do not hold in reality. The result is a proliferation of interesting solutions that, however, can hardly be implemented. In this thesis, we do not propose any solution that relies on specific MAC protocol or on hypothesis that limits its application in a real test-bed. We also survey the way mobility impacts performance and the way it is usually managed at the routing layer. Even if mobility management and routing in WMNs are difficult problems and are usually solved separately, we claim that an integrated approach can lead to better solutions with higher performance. In this context, the contributions of this thesis are three-fold: MeshDV ([21], [22]) a simple routing framework adapted to WMNs that is also able to easily manage user mobility; MeshDVbox ([22], [23]) a custom mesh router realization based on offthe-shelf components; Mesh Routing Strategy (MRS) ([24], [25], [26], [27]) a general MAC-independent cross-layer routing approach that increases the transport capacity of the mesh backbone. While the first two contributions are complementary specific solution for WMNs, Mesh Routing Strategy (MRS) is quite general and may be applied to any multi-hop wireless network. In more details: MeshDV: MeshDV is a routing protocol framework expressly designed for Wireless Mesh Networks, able to effectively manage clients mobility. MeshDV is composed of a fully modular architecture that eases its development. It combines proactive route computation for routers and on-demand path setup for clients. In particular, proactive route computation is performed using a Distance Vector (DV) approach, which sets up meshed paths between each WMR present on the mesh backbone. Since the WMRs are fixed, a proactive DV approach fits well in order to reduce overhead introduced by network wide broadcasts, which are particularly

12 1.3. Contribution of the thesis expensive in wireless environments in terms of medium resources. Path setup between clients is obtained in an on-demand fashion through new mechanisms that take advantage of the particular characteristics of mesh networks. Furthermore, these mechanisms fully leverage on existing standard IPv6 mechanisms, allowing MeshDV to gracefully integrate them and use them in the most appropriate manner for WMNs. This design choice greatly eases the management of clients mobility. Indeed, client-to-client path setup signaling also takes in charge their mobility, advertising when necessary clients wandering from a WMR to another. Another advantage is that it scales well. It also behaves correctly in small scale WMNs, where all services are placed outside the mesh and reachable through one or more gateways. MeshDV, however, does not rely on the presence of gateways, and works well also in standalone WMNs. In large scale WMNs, where services are spread into the network on servers associated to WMRs, MeshDV adapts automatically. Indeed, it is a totally distributed approach that does not make any assumption on the traffic pattern. MeshDVbox: MeshDVbox routers are the physical platform on which MeshDV has been implemented and tested. These custom WMRs have a simple architecture based on commercial components and open source code. Inside MeshDVbox, only the IPv6 protocol stack is enabled, in order to avoid the common practice of developing complete IPv4 solutions with only some IPv6 features. In these approaches, indeed, not all issues solved in the IPv4 case are then tackled in IPv6, often some key enhanced services are implemented only for IPv4, making the IPv6 version of the proposal not able to work without IPv4. On our mesh router, two interfaces are present: one working as an access point for client connections, using IEEE 802.11b [28] technology; the other working in ad hoc mode forming the mesh backbone with the others WMRs peer interfaces, using IEEE 802.11a [29]. In this way, we have the nice property that the mesh backbone sub-network and the client access sub-network are physically independent. This contribution, like the MeshDV prototype, is part of the RNRT-Infradio Project [30] whose main objective, among others, is to deploy a WMN test bed on the University Campus of Jussieu in Paris (University of Pierre et Marie Curie Paris VI). MRS: MRS is the final results on the research we performed in cross-layer routing. We first formalized cross-layer routing as a power optimization problem [27],

Chapter 1. Introduction 13 Interference Interference Power = Growing Tx Power PER Ideal Optimum PER Power Figure 1.2: Relation between Interference and Packet Error Rate when transmission power is changed. concluding that this leads to an NP-Complete problem. We proposed a simple heuristic, able to find good transmission power level trade-offs, high throughput paths, reduced interference, and increased reliability. The main task of the proposed heuristic is to optimize, through power control, the communication toward each neighbor, in terms of rate, interference, and Packet Error Rate (PER). In particular, interference is estimated by using an innovative approach based on what we called Interference Trend Index Estimator, an index able to forecast the interference generated on each transmission of a packet. Figure 1.2 shows the relation between Interference, PER and transmission power. Concerning the interference, we observe that the more (less) power used, the higher (lower) the interference (cf. top left side of figure 1.2). Concerning the PER, we observe that the more (less) power used, the lower (higher) the PER (cf. bottom left side of figure 1.2). Composing this two functions, operation indicated by the symbol in figure 1.2, leads to find the relation between interference and PER when the transmission power is changed (cf. right side of figure 1.2). MRS searches the optimal trade-off on the Interference-PER curve of each local hop, by setting an optimal transmission power level that minimizes the distance from the ideal optimum. The ideal optimum is the origin of the axis, where a transmission does not

14 1.4. Outline produce interference and there are no packet losses. When a packet is sent, the transmission power is dynamically adapted, since different next-hops may have different optimal power levels. Thus, by a distance vector approach, our routing solution selects optimal paths to reach any other WMR of the network. Even if MRS is designed using lower layer parameters, it does not rely on any particular MAC approach; it uses general parameters of any wireless communication device. We thoroughly analyze our proposal by simulations and provide a comparison to other wireless multi-hop routing protocols. While demonstrating the impact and the importance of choosing good paths in wireless multi-hop networks, in order to increase the overall performance, MRS shows itself able to offer drastically improved throughput capacity while reducing the costs in terms of transmission power. MRS fits well in the modular architecture of MeshDV, improving its performance. The work described in this thesis, has both a theoretic aspect, in particular on the formalization of MRS, but also a practical one, in the design and implementation work of MeshDV and MeshDVbox. The work done in studying the theoretical laws of wireless multi-hop networks, in particular the transport capacity scaling law, helped in finding appropriate routing metrics for the MRS proposition, therefore increasing the throughput capacity. The implementation work performed in the framework of the RNRT-Infradio Project has enabled us to deeply look in real devices, test-bed issues, and users mobility problems. The realization of MeshDV and the MeshDVbox is the final result of this work. 1.4 Outline The remainder of this thesis contains five chapters. In chapter 2, a deeper insight is given on the routing problem in wireless multi-hop networks and the mobility management issue. Chapter 3 proposes a general routing framework able to effectively increase the mesh backbone capacity while managing user mobility, presenting also the main results for the mobility management part. In chapter 4, there is the detailed proposal of cross-layer routing which is thoroughly evaluated in chapter 5. The last chapter concludes this thesis, resuming the main results and contributions and proposing future works and research directions.

Chapter2 Routing and Mobility in Wireless Multi-hop Networks Routing (and forwarding) in such a complex and dynamic context like wireless multi-hop networks, exhibits different and still unsolved technical challenges. Differently from the wired Internet, wireless networks have specific limitations, properties, and issues, such as limited bandwidth, dynamic topology, link interference, link range limitations, and inherent broadcast nature. Moreover, the common practice of using a flat addressing structure in wireless multi-hop networks does not allow route aggregation like in the hierarchical wired Internet, introducing several drawbacks. 1 The impossibility of performing route aggregation leads to an increase in routing table size and also in the number and the size of routing messages. The result is a greater overhead in terms of computation, memory usage, and bandwidth. Link quality assessment becomes a key feature of well performing routing protocols, 1 The use of a flat addressing structure in wireless multi-hop networks is not an intrinsic property of these networks, but the way addresses are usually assigned to each node in real deployments.

16 2.1. The problem of finding good paths in order to overcome numbers of issues rose by the wireless environment and the mobility nature of wireless equipments. In architectures like ad hoc, routing is performed on each terminal that joins the network, which may also follow a high mobility behavior. In a WMN, routing is performed only on WMRs, which can only join or leave the network, building a full set of meshed paths between each pairs of WMRs. The mobility issue here consists of users moving in the area covered by WMRs. Even if ad hoc and WMRs are different, they have similar issues in both routing and mobility management. Moreover, literature present numbers of examples of solution for WMRs derived from previous works in ad hoc. For this reason, in this chapter we present milestones works for both WMNs and ad hoc without making any particular differentiation. However, before focusing on this issue, hereafter we detail the two major concerns to be considered when routing in the wireless multi-hop context: the selection of good paths and user mobility management. 2.1 The problem of finding good paths Routing in multi-hop wireless networks using the traditional Hop-Count Shortest-Path metric is not a sufficient condition to construct good paths, as showed by De Couto et al. [31]. By good paths, we mean paths that are able to effectively transport data with reasonable delay, throughput, and reliability. By means of experimental results from real test-beds, De Couto et al. show that usually there are multiple minimum hop-count paths, many of which have poor throughput. The main reason of this is the fact that the wireless link has a variable behavior characteristic which is not captured by the hop-count metric. For example, radio links between nodes may have loss rates low enough that the routing protocol is willing to use them, but high enough that much of the capacity is consumed by retransmissions. This condition, affected also by the unstable behavior of the wireless channel, has unpredictable bottom-up side effects through the whole protocol stack. The above issue is emphasized by traditional route discovery mechanisms, which are not able to discover the characteristics of the wireless links, especially the multi-rate capabilities of modern radio devices. The usual approach consists in HELLO packets broadcasted on the wireless medium. Broadcast packets are always sent at the lowest transmission rate, since for them there is no Acknowledgement (ACK) and it allows to reach the largest number of neighbors. Indeed, different transmission range have

11 Mbps Chapter 2. Routing and Mobility in Wireless Multi-hop Networks 17 Source Source N2 Destination N1 N3 5.5 Mbps 2 Mbps 1 Mbps Figure 2.1: The gray-zone phenomenon in the case of IEEE 802.11b wireless cards. different transmission rate: the higher the transmission rate, the smaller the transmission range. Lundgren et al. [32] have demonstrated that this approach has a severe drawback that they call gray-zones. A node in a gray-zone can be sensed by broadcast HELLO packets, but it cannot relay traffic at rates higher than the rate of the broadcast. Figure 2.1 shows an example of this phenomenon in the case of Institute of Electrical and Electronics Engineers (IEEE) 802.11b wireless cards. If node N1 wishes to communicate to node N3 the usual approach is to do it directly, since N1 and N3 sense each other through broadcast sent at 1 Mbps. The result is that data exchange will never use a transmission rate higher than 1 Mbps. If instead of direct communication, multi-hop through N2 would be used, data could be sent using two transmissions at 11 Mbps, which means roughly an end-to-end rate of 5.5 Mbps, five times greater than the previous approach. Routing protocols selecting shortest hop-count paths have the natural attitude of using long hops, i.e. relay positioned in a gray-zone. Furthermore, the higher is the density of the network, the more likely hop-count routing protocols choose next-hops in a gray-zone, as we will show in chapter 5, resulting in a strong reduction in throughput performance. The fact that the average transmission rate lowers when density increases has also severe effects in the overall transport capacity of the entire network, as we will show later. Gray-zones is not the only phenomenon that can reduce the transport capacity of the network, in particular in the case of IEEE 802.11 technology. Indeed, Heusse et al. [33]