Multi-valued performance metrics for real-time embedded systems

To rapidly explore the design space of a real-time embedded system, it is essential to be able to efficiently analyze the timing behaviors of different system architectures. This includes not only determining if a design can satisfy all the timing constraints but also comparing the timing performance of different designs for tradeoff purposes. Understanding the exact timing behavior of a large system can be computationally prohibitive. Previous work in this area has mostly focused on producing a yes/no answer to the schedulability of a system architecture under the worst-case scenario. This not only often leads to overly pessimistic designs, but also provides no insight as how to rank different architectural designs with respect to their timing performance. In this paper, we present several metrics that may be used to measure the timing performance of a design. The metrics were analyzed using workloads from both real-world task systems and randomly generated task systems. A superior metric has been identified through analysis of large sets of experiments. We also show, through an example, how this metric can be used effectively during a design exploration process.