Evolutionary planning of virtualized cyber-physical compute and control clusters

Virtualization technology has the potential to notably advance the automation process in the domain of cyber-physical systems (CPS). It can improve both dependability and availability as well as significantly reduce the procurement, operation and maintenance costs of such systems. However, in the context of virtualization, research has put the most emphasis on topics of hardware utilization and fault-tolerance. There is little literature on how to model, integrate and consolidate a CPS by means of virtualization. In this paper we present a methodology for planning safe and efficient virtualized cyber-physical compute and control clusters - execution platforms for time-constrained virtual machines (VMs) that encapsulate CPS applications. We discuss the used methods, describe the corresponding models and the required system architecture. In contrast to typical resource allocation problems from other domains (e.g. cloud computing), in this case, the planning process must take real-time requirements of applications into account. In order to achieve this, we combine evolutionary algorithms with formal system performance analysis - in particular algorithms considered in classical scheduling theory. Such an approach allows not only to optimally dimension the compute and control clusters, but also provides strict guarantees regarding the timing predictability of the integrated CPS. Further, the embedment of a formal performance analysis technique notably eases the modeling of a system. As a consequence, the modeling process is fast, flexible and accessible not only to experts but also to system designers as they do not have to struggle with complex and time consuming mathematical formulations. Finally, our approach also provides answers to several practical questions that arise when integrating a CPS by means of virtualization.