Model-based Explainable Coordination of Complex Reconfigurations

点击翻译按钮获取中文摘要

Today, software engineers develop systems, which are able to make decisions about themselves at runtime. Such changes could for example be provisioning and deprovisioning nodes in elastic cloud systems to satisfy service level objectives or changing flight trajectories by individual drones as part of a drone swarm to achieve certain swarm objectives. Those systems have goals as well as a set of possible actions. They continuously aim at realizing the given goals by performing the actions as needed. This project proposal specifically considers architectural reconfigurations as actions, i.e., actions where the system structure changes by means of adding or removing software and hardware components or changing their parameters.Unfortunately, it is still difficult to verify at design time whether systems will use their reconfigurations appropriately at run time. There exist already approaches to analyze single reconfigurations. However, in practice usually multiple reconfigurations are independently developed and applied. This results in, potentially negative, influences and unwanted behavior of the complete system. Such effects of interactions between reconfigurations cannot be properly modelled nor analyzed at run-time at the moment. This also results in a lack of adequate comprehensibility of the system’s behavior since the applications of and the interactions between reconfigurations are autonomous decisions of the system and the reasons are not transparent.Hence, the main research question of the project proposal is:"How to support software engineers in developing validated and explainable systems which apply complex and coordinated reconfigurations?"MENTOR will close this gap. We will develop a modeling approach and analysis method for explainable, complex, coordinated, reactive and proactive, reconfigurations. This includes extensions of previous work with respect to modeling uncertainty, expected effects of reconfigurations as well as automated planning of coordinated reconfigurations. Furthermore, we will improve the efficiency of our tool which forecasts the behavior of reconfigurable systems. Finally, we specifically focus on providing explanations of the reconfiguration behavior. This includes explanations on why reconfigurations are executed – and why not, reasons for decisions during the coordination of reconfiguration as well as reasons for emergent, uncoordinated reconfigurations. We consider in the project two complementary domains and corresponding demonstrators: elastic cloud systems and mechatronic systems of systems. Focusing on these two domains enables us to distinguish generalizable aspects from domain specific results.