Use of Measurements in Worst-Case Execution Time Estimation for Real-Time Systems

Real-time systems are computational systems that are subject to temporal requirements. The increase in processing and efficiency needs has created demand for the use of processors with ever higher levels of complexity. In order to ensure compliance with deadlines, the Worst-Case Execution Time (WCET) of each task should be estimated. The main contribution of this work is an experimental evaluation of the applicability of Measurement-Based Probabilistic Timing Analysis (MBPTA), based on the Extreme Value Theory, in the specific case of an operating system based on the ARINC 653 specification, running on a processor with complex acceleration elements (e.g. cache memory and superscalar pipeline). Although not perfect, the application of EVT proved to be an adequate alternative, with satisfactory results, to the usual practice of using the highest observed value with a safety margin.