Rethinking MAPE: Next Generation Adaptation Control for Learning Adaptive Systems

Over the last three decades, computer systems completely changed from large-scale and mainly isolated systems to interactive, often mobile devices. Such devices are equipped with various sensors to gather information about their environment. Further, pervasive communication technology is required, i.e., changes to the dynamics of the environment, such as changing the quality or availability of communication channels. Self-adaptive and self-organizing (SASO) systems can adjust their behavior to environmental changes to maintain their functionality and performance. Those systems provide adaptive behavior by integrating an adaptation control system, often relying on the so-called MAPE functionality, standing for the basic functionalities of monitoring the system and the environment, analyzing if adaptation is required, and planning and executing the adaptation. However, being a well-known concept that emerged around 20 years ago, the MAPE functionality misses an inherent integration of learning. Further, it focuses on reactive, central, coordinated adaptation reasoning, which is not always beneficial and contradicts decentralized, local decision-making. This article analyzes those shortcomings and presents a new system model for integrating learning into the MAPE cycle, focusing on proactive adaptation in distributed systems. The paper concludes with a discussion of research challenges using the running example of a communication system.