A self-adaptation framework for dealing with the complexities of software changes

Abstract

Software Self-adaption (SA) is a promising technology to reduce the cost of software maintenance. However, the complexities of software changes such as various and producing different effects, interrelated and occurring in an unpredictable context challenge the SA. The current methods may be insufficient to provide the required self-adaptation abilities to handle all the existent complexities of changes. Thus, this paper presents a self-adaptation framework which can provide a multi-agent system for self-adaptation control to equip software system with the required adaptation abilities. we employ the hybrid control mode and construct a two-layer MAPE control structure to deal with changes hierarchically. Multi-Objective Evolutionary Algorithm and Reinforcement Learning are applied to plan an adequate strategy for these changes. Finally, in order to validate the framework, we exemplify these ideas with a meta-Search system and confirm the required self-adaptive ability.