A portable worst-case execution time analysis framework for real-time Java architectures

Real-time and embedded systems are systems that react continuously to their environment within time constraints and at a speed imposed by the environment. The success of such systems relies upon their capability of producing functionally correct results within defined timing constraints. In such systems, the role of Worst-Case Execution Time (WCET) analysis is fundamental since the WCET estimations of hard real-time threads have to be known prior to performing the schedulability analysis. This thesis is mainly concerned with describing a portable WCET analysis for realtime Java programs, where dynamic dispatching issues are addressed to be able to offer greater flexibility but without resulting in unpredictable timing behaviour. The dynamic dispatching feature is addressed by the means of introducing minimum annotations in the platform-independent analysis phase to allow the use of dynamic dispatching in hard realtime applications and this does not necessarily result in unpredictable timing analysis. Two measurement-based analysis techniques that can perform platform-dependent analysis are introduced to accommodate a diverse set of implementations on the underlying platforms and virtual machines for embedded systems. To improve the utilisation of resources of the overall system, a framework is also proposed to reclaim gain time, unused processor resources allocated for hard real-time threads. Gain time is produced when the hard real-time threads execute in less than their WCET estimations. This framework is integrated with the portable WCET analysis so that real-time Java applications can be developed without loss of predictability and performance. Hence, the Real-Time Specification for Java (RTSJ) augmented with the portable WCET analysis taking account of dynamic dispatching and a gain time reclaiming framework provides a flexible real-time Java environment to develop mission-critical real-time and embedded systems.