This work adresses the service level agreements (SLA) life cycle in service- oriented architectures. This life cycle starts with a specification and a modeling phase of the required quality levels. The second phase of this life cycle consists in negotiating these levels. This phase is accomplished by signing a service level agreement by the client and the provider. The last phase consists in monitoring the obligations defined in the SLA until the termination of its validity period. Concerning the first phase, we noticed that the existing SLA models are incomplete especially if the required service levels have to be indirectly computed according to other elementary quality metrics. To overcome this problem, we defined a new SLA model named SLAOnt. This model is characterized by its semantic richness and completeness. Regarding the second phase, we noticed that the existing SLA negotiation approaches consist in choosing a subset of clauses and values among choices predefined by the provider. However, the client may not understand these choices if he is not an expert of the IT domain. In this case, he has not the possibility of expressing his needs with his own knowledge and language. For this reason, we propose a novel negotiation approach based on a semantic matching of quality of service objectives between clients and providers. This approach is based on semantic inferences in all its stages to detect indirect correspondences that are not detectable by the classic negotiation and matching techniques. Our approach provides higher precision when compared to the existing ones. This phase ends with a generation of a complete service level agreement in case of compatibility. Concerning the monitoring phase, we noticed a lack of an automatic and a generic approach to facilitate the task of supervision, analysis and detection of the SLA violations. To address these shortcomings, we relied on the same semantic capacities as the negotiation phase to establish a complete monitoring approach of the signed service level agreements between service providers and consumers. Indeed, our QoS monitoring approach is based on inferential reasoning allowing us to detect violations and even undetectable d egradations without using semantics.
