An automated semantic-based approach for creating task structures

For the design of safety-critical embedded systems, many different aspects have to be considered to guarantee the correctness of the system. Besides functional correctness, these systems also have to meet real-time constraints expressed in terms of end-to-end deadlines. End-to-end deadlines claim that certain system events must occur within a given time interval. This means that all involved computation and communication processes have to be finished within this time bound. To verify such time bounds, different techniques may be used, such as scheduling-analysis as an analytical method, and model-checking as a computational method. For this step, the software parts of the model are typically represented as a task network. A task network is a directed graph of task nodes representing application processes, and edges indicating dependencies between processes. In practice, there is a gap between the specification model a designer creates in high level modeling tools like MATLAB Simulink, and the task network representation used for timing analysis. On the one hand, the designer has to decide manually which parts of the model should form a task, and, on the other hand, semantics of the specification model has to be preserved when creating the task network. Otherwise, it cannot be assured that the timing analysis really verifies the specified system. Additionally, tools such as Simulink abstract from any concrete target hardware architecture and how the different software parts are mapped to hardware resources for execution. The approach proposed in this PHD-thesis is called task creation and forms the first part of a design space exploration framework for safety-critical embedded systems. This framework addresses the complete design flow from specification models of new automotive features captured in Simulink to their distributed execution on hierarchical bus-based electronic architectures. During task creation, in a first step, a task network is derived automatically from a given Simulink model by preserving the partial order of block executions. To be able to represent Simulink models, the formalism of task networks is extended to so-called function networks offering more expressiveness. As the obtained network is typically unbalanced in the sense of computational node weights, in a second step, nodes are merged to form application tasks following an optimization metric called cohesion. This metric is defined such that nodes are attracted by high communication density and repelled by high node weights. The goal is to reduce task switching times by avoiding too lightweight tasks and to relieve the bus by keeping inter-task communication low. To obtain tasks correctly, we define formal composition operations for merging nodes in a function network. For each operation, we prove that it preserves specification semantics in terms of causality of block and node executions, respectively.