Real-time systems are subjected to strong timing requirements and consequently, timing analyses are necessary to check their temporal correctness.
Worst-Case-Execution-Time (WCET) analyses are of crucial importance, since they enable to check whether all results are delivered in-time. Timing Anomalies (TAs) characterize particular execution phenomena that jeopardize the predictability of the underlying systems, and may threaten the soundness of the WCET analyses. In this talk, we will focus on counter-intuitive TAs arising from scheduling effects. A timing anomaly occurs when a locally faster execution (e.g., due to a cache hit) finally leads to a larger execution time. Beyond the intuitive understanding, various formal definitions were proposed so as to capture this situation. The definitions mainly differ in two ways: the nature of thephenomenon against its consequences on WCET analyses, and the granularity adopted to define locally faster executions. Moreover, the existing definitions are often based on theoretical hardware models that do not refer to the components of the concrete architecture. We base this talk on a concrete assessment framework, from which we go over the various definitions. We argue that TAs are a concrete-hardware property and we show that none of the existing definitions is able to precisely detect TAs on a simple Out-of-Order (OoO) pipeline.