With the advent of cloud computing and Software-as-a-Service, Service-Based Application (SBA) represents a new paradigm to build rapid, low-cost, interoperable and evolvable distributed applications. A new application is created by defining a workflow that coordinates a set of third-party Web services accessible over the Internet. In such distributed and loose coupling environment, the execution of SBA requires a high degree of flexibility. For example, suitable constituent services can be selected and integrated at runtime based on their Quality of Service (QoS); furthermore, the composition of service is required to be dynamically modified in response to unexpected runtime failures. In this context, the main objective of this dissertation is to design, to develop and to evaluate a service middleware for flexible execution of SBA by using chemical programming model. Using chemical metaphor, the service-based systems are modeled as distributed, selforganized and self-adaptive biochemical systems. Service discovery, selection, coordination and adaptation are expressed as a series of pervasive chemical reactions in the middleware, which are performed in a distributed, concurrent and autonomous way. Additionally, on the way to build flexible service based systems, we do not restrict our research only in investigating chemical-based solutions. In this context, the second objective of this thesis is to find out generic solutions, such as models and algorithms, to respond to some of the most challenging problems in flexible execution of SBAs. I have proposed a two-phase online prediction approach that is able to accurately make decisions to proactively execute adaptation plan before the failures actually occur.

Les Architectures Orientées Services (SOA) sont adoptées aujourd'hui par de nombreuses entreprises car elles représentent une solution flexible pour la construction d'applications distribuées. Une Application Basée sur des Services (SBA) peut se définir comme un workflow qui coordonne de manière dynamique l'exécution distribuée d'un ensemble de services. Les services peuvent être sélectionnés et intégrés en temps réel en fonction de leur Qualité de Service (QoS), et la composition de services peut être dynamiquement modifiée pour réagir à des défaillances imprévues pendant l'exécution. Les besoins des architectures orientées services présentent des similarités avec la nature: dynamicité, évolutivité, auto-adaptabilité, etc. Ainsi, il n'est pas surprenant que les métaphores inspirées par la nature soient considérées comme des approches appropriées pour la modélisation de tels systèmes. Nous allons plus loin en utilisant le paradigme de programmation chimique comme base de construction d'un middleware. Dans cette thèse, nous présentons un middleware "chimique'' pour l'exécution dynamique et adaptative de SBA. La sélection, l'intégration, la coordination et l'adaptation de services sont modélisées comme une série de réactions chimiques. Tout d'abord, l'instantiation de workflow est exprimée par une série de réactions qui peuvent être effectuées de manière parallèle, distribuée et autonome. Ensuite, nous avons mis en oeuvre trois modèles de coordination pour exécuter une composition de service. Nous montrons que les trois modèles peuvent réagir aux défaillances de type panne franche. Enfin, nous avons évalué et comparé ces modèles au niveau d'efficacité et complexité sur deux workflows. Nous montrons ainsi dans cette thèse que le paradigme chimique possède les qualités nécessaires à l'introduction de la dynamicité et de l'adaptabilité dans la programmation basée sur les services. / With the advent of cloud computing and Software-as-a-Service, Service-Based Application (SBA) represents a new paradigm to build rapid, low-cost, interoperable and evolvable distributed applications. A new application is created by defining a workflow that coordinates a set of third-party Web services accessible over the Internet. In such distributed and loose coupling environment, the execution of SBA requires a high degree of flexibility. For example, suitable constituent services can be selected and integrated at runtime based on their Quality of Service (QoS); furthermore, the composition of service is required to be dynamically modified in response to unexpected runtime failures. In this context, the main objective of this dissertation is to design, to develop and to evaluate a service middleware for flexible execution of SBA by using chemical programming model. Using chemical metaphor, the service-based systems are modeled as distributed, selforganized and self-adaptive biochemical systems. Service discovery, selection, coordination and adaptation are expressed as a series of pervasive chemical reactions in the middleware, which are performed in a distributed, concurrent and autonomous way. Additionally, on the way to build flexible service based systems, we do not restrict our research only in investigating chemical-based solutions. In this context, the second objective of this thesis is to find out generic solutions, such as models and algorithms, to respond to some of the most challenging problems in flexible execution of SBAs. I have proposed a two-phase online prediction approach that is able to accurately make decisions to proactively execute adaptation plan before the failures actually occur.