Designing VM schedulers for embedded real-time applications

Virtual Machines (VMs) allow for platform-independent software development and their use in embedded systems is increasing. In particular, VMs are rewarding in the context of mixed-criticality applications to provide isolation between critical and non-critical tasks running on the same processor. In this paper, we study the design of a real-time system based on a VM monitor/hypervisor that supports multiple VMs/domains. Since each VM in the system runs several real-time tasks, scheduling the VMs leads to a hierarchical scheduling problem. So far, most published techniques for analyzing hierarchical scheduling deal with the schedulabil-ity problem, i.e., for a given hierarchical scheduler, testing whether a set of real-time tasks meet their deadlines. In this paper, we are rather concerned with the synthesis of hier-archical/VM schedulers; that is, how to design a scheduler such that all real-time tasks running on the different VMs meet their deadlines. We consider a setup where the tasks are scheduled on multiple VMs under fixed priorities according to the Deadline Monotonic (DM) policy. The VMs are scheduled under fixed priorities on a Rate Monotonic (RM) basis using one or more processors. A partitioned scheduling of VMs is considered, i.e., VMs are not allowed to migrate from one processor to the other. In this context, we propose a method for selecting optimum time slices and periods for each VM in the system. Our goal is to configure the VM scheduler such that not only all tasks are schedulable but also the minimum possible resources are used. Finally, to illustrate the proposed design technique, we present a case study based on automotive control applications.