A preliminary analysis of self-adaptive systems according to different issues

Abstract

Self-adaptive systems dynamically change their structure and behavior in response to changes in their execution environment to ensure the quality of the services they provide. Self-adaptive systems are usually built of a managed part, which implements their functionality, and a managing part, which implements the self-adaptive mechanisms. Hence, the complexity of self-adaptive systems results also from the existence of the managing part and the interaction between the managed and the managing parts. The available evaluation approaches of self-adaptive systems focus on their performances, i.e., on the benefits (e.g., degree of autonomy, support for detecting anomalous behavior, adaptivity time, quality of response) achieved through the self-adaptive mechanisms of the managing part. In this paper, we evaluate the quality of the design of self-adaptive systems (including the managed and the managing parts) as it is done in traditional software engineering. We are interested in the internal software quality of self-adaptive systems, as the existence of the managing part and its interaction with the managed part leads to a tightly coupled system. We analyze the self-adaptive systems through the detection of different issues such as architectural and code smells and the detection of design patterns. The smells provide some hints on possible design and implementation problems, and help software engineers to improve the quality of the systems. While, design patterns are usually indicators of the application of good practices in the software development and allow to capture part of the design rationale. In this way, they can help software engineers to understand, reuse, and extend self-adaptive systems. In this paper, we have considered the detection of 3 architectural smells, 18 code smells, and 15 design patterns in 11 self-adaptive systems written in the Java programming language. The results indicate that the 3 architectural smells, 9 out of the 18 code smells, and the 15 design patterns have been detected in all the analyzed self-adaptive systems. We also discuss the possible reasons behind the presence of these quality issues, and provide our lessons learned.