Real-time virtualization with Xvisor

Abstract

Embedded virtualization has gained attention in recent years due to increasing usage of embedded systems in cyber-physical systems and the Industry 4.0 revolution. Especially in combination with multi-core embedded systems, virtualization reduces the number of embedded systems and simultaneously delivers a secure and separated environment in each virtualized system. Applications in such cyber-physical systems often require real-time guarantees with hard deadlines. To guarantee those real-time constraints in virtualization, both hypervisor and guest operating system must support real-time scheduling. Selecting the optimal scheduling algorithm on both scheduling levels is hard and is only optimal for the analysed application. Due to the multiple scheduling levels, a set of scheduling algorithm combinations must be analysed which is too costly without analysis on higher abstraction levels. By using an analysis methodology to find this optimal combination using higher abstraction levels analysis, we reduce the set at every abstraction level. In this paper, we present a real-time hypervisor, based on Xvisor, for multi-core embedded systems. We modified the hypervisor to support real-time scheduling and the compositional schedulability analysis and validated the analysis methodology using this embedded hypervisor.