摘要：The Measurement-Based Probabilistic Timing Analysis (MBPTA) infers probabilistic Worst-Case Execution Time (pWCET) estimates from measurements of tasks execution times; the Extreme Value Theory (EVT) is the statistical tool that MBPTA applies for inferring worst-cases from observations/measurements of the actual task behavior. MBPTA and EVT capability of estimating safe/pessimistic pWCET rely on the quality of the measurements; in particular, execution time measurements have to be representative of the actual system execution conditions and have to cover multiple possible execution conditions. In this work, we investigate statistical dependences between execution time measurements and tasks with multiple runtime operational modes. In the first case, we outline the effects of dependences on the EVT applicability as well as on the quality of the pWCET estimates. In the second case, we propose the best approaches to account for the different task execution modes and guaranteeing safe pWCET estimates that cover them all. The solutions proposed are validated with test cases.
关键词：Measurement-Based Probabilistic Timing Analysis; probabilistic Worst-Case Execution Time; Extreme Value Theory; Execution Time Measurements Representa
