Schedulability-Driven Partitioning of Heterogeneous Real-Time Systems

During the development of a real-time system, the main goal is to nd an implementation that satis es the system's speci ed worst-case timing constraints. Often, the most cost-e ective solution is a heterogeneous implementation, where some parts of the functionality are implemented in software, and the rest in hardware, using application-speci c circuits. Hardware/software codesign allows the designer to describe the complete system homogeneously, and thereafter divide it into separate hardware and software parts. This thesis is a contribution to hardware/software partitioning of real-time systems. It proposes an automatic partitioning of a set of real-time tasks in order to meet their deadlines. A key issue when verifying timing constraints is the analysis of the task scheduler. Therefore, an extension of xed-priority scheduling theory is proposed, which is suitable for heterogeneous implementations. It includes an optimal task priority assignment algorithm. The analysis uses information about the execution time of the tasks in di erent implementations, and a method for estimating these data is also proposed. The analysis results are used to guide the partitioning process, which is based on a branch-andbound algorithm. This work has been supported by The Swedish National Board for Industrial and Technical Development (NUTEK). Department of Computer and Information Science Linkoping University S-581 83 Linkoping Sweden