Quantitative analysis and systematic parametrization of a two-level real-time scheduler

The computational power of embedded systems have increased steadily during the recent years. In contrast to former approaches which allowed at least one application per computational node the memory size and computational power of today allows to host more than one application per node. Often applications are delivered by suppliers as a whole including the operating systems where the application tasks run on top. In this case virtualization is a common software approach to maintain isolation between different applications on the same computation system. Virtual machine monitors are able to divide the resources of a physical system into several logical subsystems. However, those monitors which are available today do not focus on the preservation of real-time properties. Consequently, our working group develops and investigates a two-level hierarchy of real-time schedulers, where a global scheduler assigns temporal resources to guest systems, while each subsystem has its own local scheduler for its application tasks. In this contribution, we focus on a formal investigation of the real-time properties of the two-level scheduling hierarchy. The starting points are independent applications building subsystems, each containing a set of tasks and a local scheduler, which have to be integrated and configured at the global scheduling level. Utilization bounds are derived unfolding the overhead of such an approach. Furthermore we propose systematic process for the computation of the task parameters for both levels of scheduling. Representatively the whole approach is applied to the rate monotonic assignment of priorities to tasks at the low scheduling level. For reasons of abstraction all these tasks are mapped into a single task proxy. This enables the global scheduler to treat all of its subsystems as periodic tasks allowing again for the application of the rate monotonic assignment of priorities to tasks.