Preventing timing failures in mixed-criticality clouds with dynamic real-time containers

Real-time containers can be seen as the convergence of hard real-time operating systems and container-based virtualization technologies that can be used to build mixed-criticality cloud systems. In this paper we investigate the implementation of this concept based on earliest deadline first and server-based scheduling on a Linux kernel with the RTAI real-time patch. With respect to earlier proposals, our solution allows more flexibility with simplified admission checks and better utilization of CPU resources. We compare the solution with the SCHED DEADLINE scheduling class, already available in the Linux kernel, and we show, with experiments on real hardware, that the proposal allows to prevent timing failures to tasks running within real-time containers, despite the presence of disturbances.