In this paper we consider a provider that offers a SOA application implemented as a composite service to several users with different Qos requirements. For such a system, we present a scalable framework to the QoS-aware self-adaptation based on a two layer reference architecture. The first layer addresses the adaptation at the provisioning level: operating at a slower time scale, its role is to identify the set of candidate services to implement the system functionality at the required user QoS. The second layer addresses the adaptation at the service selection level: operating on a faster time scale, its role is to determine at running time the actual services which are bound to each user request while meeting both provider and user QoS. We formulate the adaptation strategy of both layers as suitable optimization problems which can be efficiently solved using standard techniques. Numerical experiments show the effectiveness of the proposed approach.