Revisiting Timing Anomalies in Predictable In-Order Pipelines

The quality of timing guarantees ensured through worst-case-execution time analysis and schedulability tests - required to be both sound and precise - is directly influenced by the predictability properties of the execution platform. A platform is considered predictable when safe and precise bounds can be computed through analysis tools. Counter-intuitive and Amplification Timing Anomalies (TAs) are detrimental to predictability and thus may make it much harder/impossible to compute such bounds. In order to address this issue, research has followed two orthogonal approaches, (i) designing predictable execution platforms and (ii) characterizing counter-intuitive TAs through formal definitions. However, predictable designs rarely apply any formal definitions of timing anomalies. This paper aims at investigating precisely this relationship. We first show how a previously proposed definition of counter-intuitive TAs can be applied to the predictable in-order processor SIC. We then extend this approach in order to provide the first formal definition of both counter-intuitive and amplification effects. The proposed definitions are then evaluated on a regular in-order processor as well as the predictable SIC core using a systematic approach that allows to assess their applicability and relevance. Finally, we prove, for the first time, the absence of some, but not all, TA effects in SIC.