Continuous verification of system of systems with collaborative MAPE-K pattern and probability model slicing

Abstract

The phenomenon of cooperation among independent systems to achieve common goals has been growing. In this regard, the concept of system of systems (SoS), wherein numerous independent systems cooperate with each other, has been proposed. The key characteristic of an SoS is the operational and managerial (O/M) independence of each constituent system (CS). Each CS of a collaborative SoS with high O/M independence provides different levels of internal-knowledge sharing and is entitled to voluntary participation in the SoS (i.e., dynamic reconfiguration). To increase goal-achievement rate, we need to verify SoS considering the knowledge-sharing and dynamic reconfiguration constraints. The dynamic reconfiguration of SoSs can be managed using continuous verification, which involves environment monitoring, modeling systems for operation in changing environments, and verifying the model runtimes. However, O/M independence introduces the following challenges: (1) the low knowledge sharing level causes inaccurate modeling, which leads to inaccurate verification results, and (2) dynamic reconfiguration requires frequent re-verification at runtime, which incurs high verification costs. In this paper, we propose a continuous-verification-of-SoS (CVSoS) approach to solve these two challenges. To address the low knowledge-sharing level, we propose the collaborative MAPE-K pattern. The key to collaborative MAPE-K is the retrieval of knowledge from the other collaborating CSs. To address dynamic reconfiguration, we propose a new slicing algorithm for SoS models. This algorithm promotes synchronization dependence, which is essential for representing interactions between CSs. Furthermore, we demonstrate the accuracy of this algorithm. We evaluated CVSoS across multiple SoS domains, which revealed that the SoS goal-achievement rate increases by up to 64% using the collaborative MAPE-K pattern and that slicing the benchmark and SoS models improved the verification time by an average of 67%.