Towards a General Supporting Framework for Self-Adaptive Software Systems

When confronted with their internal and environmental dynamics, various software systems increasingly require self-adaptation capabilities. The vision above requires the self-adaptation approach to tackle these challenges and some software systems have their own specific implementations. However, in this paper a general supporting framework is proposed for software systems to self-adapt with running environmental dynamics and meanwhile fulfill various user requirements. Three key issues are covered in the framework: 1) the overall control architecture, which adopts the double closed-loop style and respectively includes the self-adaptation loop and the self-learning loop; 2) a general descriptive language, which is a application-independent and unified language to represent self-adaptation knowledge about target systems; 3) three implementation mechanisms, including forward reasoning, planning and reinforcement learning using feedback, which are supported by the above descriptive language and executed at runtime in different modules. Finally, one scenario of on-demand services of massive data mining tasks is selected and the case study demonstrates how the framework is customized as required and how the approach works.