University of Houston

The Virtual Machine (VM) technology partitions a host physical machine into a group of VMs. Typically, a VM machine only preempts a part of a dedicated physical resource temporally or spatially. This fact greatly impacts the real-time task scheduling in a VM because most traditional real-time scheduling theories are based on dedicated resources. The real-time community has introduced some hierarchical real-time resource model to address this issue. Among them, the Regularity-based Resource Partitioning (RRP) Model could provide maximal transparency for task scheduling. However, the current theoretical results on the RRP Model are still far from the complete theoretical fundamentals required by a real-time VM monitor. First, only a naive algorithm has been found for resource partitioning. Second, only the Periodic Task Model is investigated for the task scheduling, and even for this task model, only one simple case is investigated. Last but not least, fault tolerance has not been taken into account in the design of the Regularity-based Model. We shall enhance the RRP Model in these three aspects: resource partitioning, task scheduling and fault tolerance. With these theory enhancement, a 2-layer real-time resource model is completely presented and the theoretical fundamentals of a real-time VM monitor are ready from the resource management perspective.