A Framework Based on Learning Techniques for Decision-making in Self-adaptive Software

Abstract

Abstract-The development of Self-adaptive Software (SaS) presents specific innovative features compared to traditional ones since this type of software constantly deals with structural and/or behavioral changes at runtime. Capabilities of human administration are showing a decrease in relative effectiveness, since some tasks have been difficult to manage introducing potential problems, such as change management and simple human error. Self-healing systems, a system class of SaS, have emerged as a feasible solution in contrast to management complexity, since such system often combines machine learning techniques with control loops to reduce the number of situations requiring human intervention. This paper presents a framework based on learning techniques and the control loop (MAPE-K) to support the decision-making activity for SaS. In addition, it is noteworthy that this framework is part of a wider project developed by the authors of this paper in previous work (i.e., reference architecture for SaS [1]). Aiming to present the viability of our framework, we have conducted a case study using a flight plan module for Unmanned Aerial Vehicles. The results have shown an environment accuracy of about 80%, enabling us to project good perspectives of contribution to the SaS area and other domains of software systems, and enabling knowledge sharing and technology transfer from academia to industry.

DmS-Modeler: A Tool for Modeling Decision-making Systems for Self-adaptive Software Domain

“…Moreover, a selection of current and past selfhealing approaches were addressed, as well as explanations for the origins, principles, and theories of self-healing for such approaches. These two studies provided the theoretical basis for the design of our framework [9].…”

“…The Action Plan Module aims at assisting in the adaptation of such entities, providing controls for: (i) dynamic behavior, (ii) individual reasons, and (iii) execution state in relation to the environment. Thereby, a framework based on learning techniques [20] and the MAPE-K loop [21], [22] was developed in previous work as part of this module to support the decision-making [9]. Section IV presents details on this framework as well as the design and implementation of the DmS-Modeler tool.…”

“…In short, this plan must be elaborated based on new requirements (e.g., structural and/or behavioral) and adaptation level of each software entity. Based on this scenario, a framework for decision-making was developed by Affonso et al [9] to support the generation of such plan identifying anomalies and proposing one or more solutions to solve them. Concerning the design, MAPE-K loop [22], learning techniques [23], [24], and an external approach [4] of adaptation were the resources used in this framework.…”

“…Therefore, our framework [9] acts as a non-intrusive supervision modality, i.e., a supervisor system (engine) can be coupled to a software entity (adaptable software) to monitor its internal state of operation or the execution environment in which it is inserted.…”

“…From this point onwards, SaS may be also referred to as software entities or simply entities. Moreover, as part of this RA, a framework for decision-making was developed [9], whose purpose is to identify anomalies and propose one or more solutions to solve them. The RA and this framework have improved the SaS development; however, the design of Decision-making Systems (DmS), i.e., instances of such framework, was still manually conducted.…”