Cloud Computing is becoming the new paradigm for information technology to provide resources as Internet-based services. These services are basically categorized according to three layers also called SPI model (Software, Platform, Infrastructure). The same way as ``non-Cloud'' applications, Cloud services must be able to adapt themselves according to the evolution of their environment. There are many works on dynamic adaptation such as server consolidation and green computing but these works are generally specifics to one layer and do not take the others into account. However Kephart et al. have explain in 2000 that even if adaptations are, in theory, independant, they have an impact on the overall system. Consequently, an adaptation at the infrastructure layer can have an impact at the platform or at the application layers.This thesis provides an abstraction to manage adaptation as an orthogonal concern overs Cloud layers. Based on Model atRuntime (M@R) techniques which offer to use design tools to build and validate new configuration of the system at the runtime, this abstraction is able to modelize all the Cloud layers. To show the usability of this abstraction, we provide three experimentations showing the extensibility and genericity of our approach, showing that performance overhead on the system (infrastructure or platform) is weak and showing that the abstraction allows to build multi-layers adaptations.
