WCET Analysis of Modern Processors Using Multi-Criteria Optimisation

The Worst-Case Execution Time (WCET) is an important execution metric for real-time systems, and an accurate estimate for this increases the reliability of subsequent schedulability analysis. Performance enhancing features on modern processors, such as pipelines and caches, however, make it difficult to accurately predict the WCET. One technique for finding the WCET is to use test data generated using search algorithms. Existing work on search-based approaches has been successfully used in both industry and academia based on a single criterion function, the WCET, but only for simple processors. This paper investigates how effective this strategy is for more complex processors and to what extent other criteria help guide the search, e.g. the number of cache misses. Not unexpectedly the work shows no single choice of criteria work best across all problems. Based on the findings recommendations are proposed on which criteria are useful in particular situations.