Ali Abbas Jaffari Syed

• The complexity of modern real-time systems is increasing day by day. This inevitable rise in complexity predominantly stems from two contradicting requirements, i.e., ever increasing demand for functionality, and required low cost for the final product. The development of modern multi-processors and variety of network protocols and architectures have enabled such a leap in complexity and functionality possible. Albeit, efficient use of these multi-processors and network architectures is still a major problem. Moreover, the software design and its development process needs improvements in order to support rapid-prototyping for ever changing system designs. Therefore, in this dissertation, we provide solutions for different problems faced in the development and deployment process of real-time systems. The contributions presented in this thesis enable efficient utilization of system resources, rapid design & development and component modularity & portability. In order to ease the certification process, time-triggered computation model is often used in distributed systems. However, time-triggered scheduling is NP-hard, due to which the process of schedule generation for complex large systems becomes convoluted. Large scheduler run-times and low scalability are two major problems with time-triggered scheduling. To solve these problems, we present a modular real-time scheduler based on a novel search-tree pruning technique, which consumes less time (compared to the state-of-the-art) in order to schedule tasks on large distributed time-triggered systems. In order to provide end-to-end guarantees, we also extend our modular scheduler to quickly generate schedules for time-triggered network traffic in large TTEthernet based networks. We evaluate our schedulers on synthetic but practical task-sets and demonstrate that our pruning technique efficiently reduces scheduler run-times and exhibits adequate scalability for future time-triggered distributed systems. In safety critical systems, the certification process also requires strict isolation between independent components. This isolation is enforced by utilizing resource partitioning approach, where different criticality components execute in different partitions (each temporally and spatially isolated from each other). However, existing partitioning approaches use periodic servers or tasks to service aperiodic activities. This approach leads to utilization loss and potentially leads to large latencies. On the contrary to the periodic approaches, state-of-the-art aperiodic task admission algorithms do not suffer from problems like utilization loss. However, these approaches do not support partitioned scheduling or mixed-criticality execution environment. To solve this problem, we propose an algorithm for online admission of aperiodic tasks which provides job execution flexibility, jitter control and leads to lower latencies of aperiodic tasks. For safety critical systems, fault-tolerance is one of the most important requirements. In time-triggered systems, modes are often used to ensure survivability against faults, i.e., when a fault is detected, current system configuration (or mode) is changed such that the overall system performance is either unaffected or degrades gracefully. In literature, it has been asserted that a task-set might be schedulable in individual modes but unschedulable during a mode-change. Moreover, conventional mode-change execution strategies might cause significant delays until the next mode is established. In order to address these issues, in this dissertation, we present an approach for schedulability analysis of mode-changes and propose mode-change delay reduction techniques in distributed system architecture defined by the DREAMS project. We evaluate our approach on an avionics use case and demonstrate that our approach can drastically reduce mode-change delays. In order to manage increasing system complexity, real-time applications also require new design and development technologies. Other than fulfilling the technical requirements, the main features required from such technologies include modularity and re-usability. AUTOSAR is one of these technologies in automotive industry, which defines an open standard for software architecture of a real-time operating system. However, being an industrial standard, the available proprietary tools do not support model extensions and/or new developments by third-parties and, therefore, hinder the software evolution. To solve this problem, we developed an open-source AUTOSAR toolchain which supports application development and code generation for several modules. In order to exhibit the capabilities of our toolchain, we developed two case studies. These case studies demonstrate that our toolchain generates valid artifacts, avoids dirty workarounds and supports application development. In order to cope with evolving system designs and hardware platforms, rapid-development of scheduling and analysis algorithms is required. In order to ease the process of algorithm development, a number of scheduling and analysis frameworks are proposed in literature. However, these frameworks focus on a specific class of applications and are limited in functionality. In this dissertation, we provide the skeleton of a scheduling and analysis framework for real-time systems. In order to support rapid-development, we also highlight different development components which promote code reuse and component modularity.