| The application areas of real-time embedded systems range these days from industrial controls to consumer gadgets, communication and medical devices. The systems are equipped with complex application features, a significant amount of software, and net working interconnection despite heterogeneity. The development of these complex software centric real-time embedded applications can benefit from a high level abstraction and programming language model, provided by virtual machine (VM) that features portability, productivity, and safety.; In order for VM to be adequate for real-time embedded applications, it has to satisfy the constraints and demands of the application. In this dissertation, we focus on developing software and analysis techniques essential to make VM infrastructure suitable for real-time embedded applications. The goal of our study is to retune VM systems for predictability, reliability, and availability.; The dissertation describes an extended VM infrastructure which includes three subsystems. The Schedulable Garbage Collection makes behavior of garbage collection predictable by controlling the pause time due to garbage collection operations. Then, with a joint scheduling algorithm of garbage collection and real-time application tasks, the subsystem can guarantee to meet task deadlines and to make a heap memory available. The Schedulable Persistence System aims to ensure fault-tolerance that is demanded in mission- and safety-critical systems. It demonstrates that a persistence service can be conducted non-intrusively and cost-efficiently with concurrent and preemptible object serialization. In the Schedulable Online Testing Framework, software testing is piggybacked on a running system that operates to serve an actual mission. The framework ensures the timely correctness of the regular service with a flexible scheduling mechanism. By employing a persistence service, the testing operation can be initiated with a consistent system state and run in an isolated partition. Thus, the testing operation, while having a full access to external interference and unexpected behavior of the target application environment, cannot cause unbounded pause time or spoil the regular operation.; The subsystems are prototyped on a CLI (Common Language Infrastructure) runtime environment. Execution data is measured and used to analyze performance characteristics and to establish scheduling models. With execution traces and performance analysis, the feasibility of the VM infrastructure is demonstrated. |
