For cyber-physical systems, real-time virtualization optimizes the hardware utilization by consolidating multiple systems into the same platform, while satisfying the timing constraints of their real-time tasks. This paper considers virtualization based on unikernels, i.e., single address space kernels usually constructed by using library operating systems. Each unikernel is a guest operating system in the virtualization and hosts a single real-time task.
We consider deferrable servers in the virtualization platform to schedule the unikernel-based guest operating systems and analyze the worst-case response time of a sporadic real-time task under such a virtualization architecture. Throughout synthesized tasksets, we empirically show that our analysis outperforms the restated analysis derived from the state-of-the-art, which is based on Real-Time Calculus. Furthermore, we provide insights on implementation-specific issues and offer evidence that the proposed scheduling architecture can be effectively implemented on top of the Xen hypervisor while incurring acceptable overhead.