A Q-Leaning-Based On-Line Planning Approach to Autonomous Architecture Discovery for Self-managed Software

초록

Two key concepts for architecture-based self-managed software are flexibility and autonomy. Recent discussion have focused on fexibility in self-management, but the software engineering community has not been paying attention to autonomy as much as flexibility in self-management. In this paper, we Focus oil achieving the autonomy of software systems by on-line planning in which a software system can decide all appropriate plan in the presence of change, evaluate the result of the plan, and learn the result. Our approach applies Q-leaning, which is one of the reinforcement learning techniques, to self-managed systems. The paper presents a case study to illustrate the approach. The result of the case study shows that our approach is effective for self-management.