Evolving Models at Run Time to Address Functional and Non-Functional Adaptation Requirements

Increasingly, applications need to dynamically self-reconfigure as new environmental conditions arise at run time. In order to selfreconfigure, an adaptive system must determine which target system configuration will yield the desired behavior based on current execution conditions. However, it may be impractical to evaluate all potential system configurations in a reasonable time frame. This paper presents a model-based approach that leverages evolutionary computation to automatically generate, at run time, target system models that balance tradeoffs between functional and non-functional requirements in response to run-time monitoring of environmental conditions. Specifically, this approach generates graph-based representations of architectural models for potential target system configurations. The current run-time system models serve to constrain the degree of change and novelty in the newly generated models. This approach is applied to the dynamic reconfiguration of a set of remote data mirrors, where operational and reconfiguration costs are minimized, while maximizing data reliability and network performance.