Service-oriented computing promotes the construction of applications by composing distributed services that are advertised in an open service market. In such an environment, individual services may change and evolve dynamically, requiring composite services to adapt to such changes. The prevailing strategy is to react on failures and replace the defective component of the composite service. However, this reactive approach does not fully exploit the opportunities of a dynamic market where older services may be replaced by better ones.

In this paper we promote a novel architecture for automated, dynamic, pro-active, and transparent maintenance and improvement of composite services. We leverage fine-grained client-side monitoring techniques to generate information regarding functional and non-functional properties of service behavior. A reputation manager is responsible for collecting and aggregating this information, and provides economical incentives for honest sharing of feedback. Composite services can thus use reliable reputation information to pro-actively improve their aggregate performance.