Adaptation impact and environment models for architecture-based self-adaptive systems

Abstract

Self-adaptive systems have the ability to adapt their behavior to dynamic operating conditions. In reaction to changes in the environment, these systems determine the appropriate corrective actions based in part on information about which action will have the best impact on the system. Existing models used to describe the impact of adaptations are either unable to capture the underlying uncertainty and variability of such dynamic environments, or are not compositional and described at a level of abstraction too low to scale in terms of specification effort required for non-trivial systems. In this paper, we address these shortcomings by describing an approach to the specification of impact models based on architectural system descriptions, which at the same time allows us to represent both variability and uncertainty in the outcome of adaptations, hence improving the selection of the best corrective action. The core of our approach is a language equipped with a formal semantics defined in terms of Discrete Time Markov Chains that enables us to describe both the impact of adaptation tactics, as well as the assumptions about the environment. To validate our approach, we show how employing our language can improve the accuracy of predictions used for decision-making in the Rainbow framework for architecture-based self-adaptation.

Towards priority-awareness in autonomous intelligent systems

“…In order to support decision-making under uncertainty exhibited by an AIS, a number of runtime modelling techniques exist that use prioritisation of NFRs.The techniques in [1,4,15,17] make use of Markov-based approaches such as Markov Decision Process (MDPs), Partially Observable Markov Decision Processes (POMDPs) and Discrete Time Markov Chains (DTMCs) along with probablistic model checking. As these techniques are Markov-based they also support the quantification of uncertainty by the use of probabilities over the variables of the states of the environment.…”

“…As such, a number of runtime modelling techniques have been developed that take into account NFRs' priorities [1,4,15,17]. These modelling techniques are based on optimisation methods, including decision analysis and utility theory [25,26].…”

“…[Problems] A feature of such AIS is that they typically employ single objective optimization techniques. These techniques use a uni-scalar cumulative utility value to represent a combined priority for all NFRs [1,2,4,15,17]. However, the adaptive actions taken by the AIS can have different impacts, either positive or negative, on the satisfaction levels of individual NFRs.…”

Reasoning about Human Participation in Self-Adaptive Systems

“…Note that, to obtain both the latency and impact on the different quality dimensions 1 define boolean PNA=exists n:ProcessorNodeT in M.components | !n.isActive; 2 define set ActivePNs={select n:ProcessorNodeT in M.components | !n.isActive}; 3 define set of tactics in practice, the approach relies on expert knowledge or field data about similar existing systems, although nothing prevents the use of machine learning techniques to obtain that information. Although in this paper we consider fixed cost/benefit impacts for illustration purposes, Stitch also supports the specification of sophisticated impact models that are context-sensitive, and can capture probabilistic aspects in the outcome of tactic executions [6]. …”

Machine Learning Meets Quantitative Planning: Enabling Self-Adaptation in Autonomous Robots

“…Examples include those about software components associated with hardware elements like sensors that degrade over time, becoming less accurate or consuming more energy than expected. To characterize expected system behavior, existing approaches in self-adaptive systems almost always rely on manual modeling by domain experts [12], which is expensive and potentially unreliable, or they use small and artificial models that need to be adjusted as the system evolves [36]. Regardless of the approach employed, the vast space of configurations and environmental conditions of such systems almost invariably lead to strong simplifying assumptions and models that do not Overview of our approach: We use machine learning to identify optimal configurations that will be used for run-time adaptations.…”