Adaptive Exchange of Distributed Partial Models@run.time for Highly Dynamic Systems

Abstract

Abstract-Future software systems will be highly dynamic. We are already experiencing, for example, a world where CyberPhysical Systems (CPSs) play a more and more crucial role. CPSs integrate computational, physical, and networking elements; they comprise a number of subsystems, or entities, that are connected and work together. The open and highly distributed nature of the resulting system gives rise to unanticipated runtime management issues such as the organization of subsystems and resource optimization.In this paper, we focus on the problem of knowledge sharing among cooperating entities of a highly distributed and selfadaptive CPS. Specifically, the research question we address is how to minimize the knowledge that needs to be shared among the entities of a CPS. If all entities share all their knowledge with each other, the performance, energy and memory consumption as well as privacy are unnecessarily negatively impacted. To reduce the amount of knowledge to share between CPS entities, we envision a role-based adaptive knowledge exchange technique working on partial runtime models, i.e., models reflecting only part of the state of the CPS. Our approach supports two adaptation dimensions: the runtime type of knowledge and conditions over the knowledge. We illustrate the feasibility of our technique by discussing its realization based on two state-of-the-art approaches.

Modeling Collaborations in Adaptive Systems of Systems

“…Partial Knowledge: The inherent complex and distributed nature of adaptive SoS leads to multiple local views in the systems and different versions of data over time [59,86]. Therefore, the modeling language should consider partial knowledge and conflict resolution mechanisms for different versions or even outdated data.…”

“…The semantic (R-10) and partial runtime models (R-11) are not considered. Götz et al [86] target the point of partial available knowledge and illustrate the definition of views via a model query language, but lack in a clear semantic and integration of the runtime model information into an overall control architecture. Furthermore, Rajhans et al [151] focus on the integration of multiple, different, potentially partial domain models during the development of a CPS, but lack in an integration concept of this information at runtime.…”

“…Götz et al [86] present preliminary ideas, where distributed feedback loops formulate queries that are applied on the knowledge base on interaction participants. Therefore, the authors emphasize the importance of considering different views over the complete available knowledge base, which is related to R-11.…”

“…Therefore, the authors emphasize the importance of considering different views over the complete available knowledge base, which is related to R-11. Unfortunately, in [86], they present only first ideas and pinpoint to possible solutions rather than describe a working modeling approach.…”

“…Thus, runtime models may exists in different versions distributed in systems throughout the adaptive SoS. As a consequence, the adaptive SoS has to deal with partial available knowledge as well as outdated or versioned runtime models [86], which leads to the following thesis goal.…”