De nos jours les données ainsi que les applications dans le nuage (cloud) connaissent une forte croissance, ce qui pousse les fournisseurs à chercher des solutions garantissant un lien réseau stable et résilient à leurs utilisateurs. Dans cette thèse on étudie les protocoles réseaux et les stratégies de communication dans un environnement de centre de données distribués. On propose une architecture cloud distribuée, centrée sur l’utilisateur et qui a pour but de: (i) migrer des machines virtuelles entre les centres de données avec un temps d’indisponibilité faible; (ii) fournir un accès résilient aux machines virtuelles; (iii) minimiser le délai d'accès au cloud. On a identifié deux problèmes de décision: le problème d'orchestration de machines virtuelles, prenant en compte la mobilité des utilisateurs, et le problème de basculement et de configuration des localisateurs, prenant en compte les états des liens inter- et intra-centre de données. On évalue notre architecture en utilisant une plate-forme de test avec des centres de données distribués géographiquement et en simulant des scenarios basés sur des traces de mobilités réelles. On montre que, grâce à quelques modifications apportées aux protocoles d'overlay, on peut avoir des temps d'indisponibilité très faibles pendant la migration de machines virtuelles entre deux centres de données. Puis on montre qu’en reliant la mobilité des machines virtuelles aux déplacement géographiques des utilisateurs, on peut augmenter le débit de la connexion. De plus, quand l’objectif est de maximiser le débit entre l’utilisateur et sa ressource, on démontre par des simulations que la décision de l'emplacement des machines virtuelles est plus importante que la décision de basculement de point d'entrée du centre de données. Enfin, grâce à un protocole de transport multi-chemins, on montre comment optimiser les performances de notre architecture et comment à partir des solutions de routage intra-centre de données on peut piloter le basculement des localisateurs. / While many business and personal applications are being pushed to the cloud, offering a reliable and a stable network connectivity to cloud-hosted services becomes an important challenge to face in future networks. In this dissertation, we design advanced network protocols, algorithms and communication strategies to cope with this evolution in distributed data center architectures. We propose a user-centric distributed cloud network architecture that is able to: (i) migrate virtual resources between data centers with an optimized service downtime; (ii) offer resilient access to virtual resources; (iii) minimize the cloud access latency. We identify two main decision making problems: the virtual machine orchestration problem, also taking care of user mobility, and the routing locator switching configuration problem, taking care of both extra and intra data center link states. We evaluate our architecture using real test beds of geographically distributed data centers, and we also simulate realistic scenarios based on real mobility traces. We show that migrating virtual machines between data centers at negligible downtime is possible by enhancing overlay protocols. We then demonstrate that by linking cloud virtual resource mobility to user mobility we can get a considerable gain in the transfer rates. We prove by simulations using real traces that the virtual machine placement decision is more important than the routing locator switching decision problem when the goal is to increase the connection throughput: the cloud access performance is primarily affected by the former decision, while the latter decision can be left to intra data center traffic engineering solutions. Finally, we propose solutions to take profit from multipath transport protocols for accelerating cloud access performance in our architecture, and to let link-state intra data center routing fabrics piloting the cloud access routing locator switching.

Mobile Ad hoc Networks (MANETs) are re-emerging as a popular networking facility for wireless device users. A growing number of diversified applications are now accessible via wireless devices. The different applications may have different Quality of Service (QoS) requirements, which may better be satisfied by using different routing methods or metric types. Existing ad hoc network routing solutions do not consider various application-level requirements when making a routing decision. They typically make routing decisions based upon limited information acquired at the network layer. Most of the existing routing protocols make use of a single routing metric. Using a single metric type and/or information, only acquired at the network layer may not be able to accommodate different QoS requirements, imposed by diversified user-level applications or application-level data types.The aim of this thesis is to design an efficient routing function for ad hoc networks while at the same time satisfying users‟ and/or applications‟ QoS and security requirements. To achieve this, the thesis investigates and specifies routing requirements that could best support application-level QoS and security requirements in MANETs. It also investigates and critically analyses the state of the art in MANET routing, and the mechanisms used for protecting the routing functions. To overcome the weaknesses and advance the state of the art in MANET routing, this thesis proposes two major solutions. The first solution is the Secure ETX (SETX) routing protocol. It is a secure routing solution that can provide routing functions efficiently in malicious MANET environment. The SETX protocol provides a security mechanism to counter black hole attacks in MANETs on the ETX metric acquisition process. Simulation studies have been carried out and discussed in the thesis. Simulation results show that the SETX protocol can provide a marked improvement in network performances in the presence of black hole attacks, and it can do so with a negligible level of additional overhead.The second solution is a novel routing decision making called the Flexible Routing Decision (FRD) framework. The FRD framework supports routing decision making by using multiple metric types (i.e. multi-criteria routing decision making) and uses a cross-layer approach to support application-level QoS requirements. This allows users to use different routing metrics types, making the most appropriate routing decision for a given application. To accommodate the diversified application-level QoS requirements, multiple routing metric types have been identified and interpreted in the FRD framework design. The FRD framework has overcome some weaknesses exhibited by existing single metric routing decision making, used in MANETs. The performance of a routing decision making of FRD is also evaluated using NS2 simulation package. Simulation results demonstrate that the FRD framework outperforms the existing routing decision making methods.