There is a large scheduling theory developed for, let's say, "standard" systems; those systems that are entirely controlled by a single operating system. The system can be mono or muli-processor, but there is one central scheduling algorithm which administers the processing units.

The hypervisor technology allows to run several complete RTOS on a single system. Spatial isolation among partitions can be easily understood and managed by the system designer; but temporal isolation deserves more attention.

The scheduling problem introduced by the hypervisor is known as hierarchical scheduling. Each activity (task, thread, process, etc.) is scheduled first by its own local operating system scheduler, which is then scheduled by the hypervisor. The performance of the application depends now on two scheduling levels.

The solution proposed in the ARINC 653-1 specification is to use a strong cyclic policy (configuration table) to schedule hypervisor partitions, and preemptive fixed priority on the RTOS. This solution raises a number of scheduling issues.

We are developing a set of tools to analyse hierarchical scheduled systems. An early prototype is available at our scheduling tools site.