On the analysis of the timing behaviour of time randomised caches

Abstract

Time Randomised caches (TRc), which can be implemented at hardware level or with software means on conventional deterministic cache designs, have been proposed for real-time systems as key enablers for Probabilistic Timing Analysis (PTA) and in particular its measurement-based variant: Measurement-Based Probabilistic Timing Analysis (MBPTA). A key parameter of MBPTA is the number of runs required to ensure representativity of the execution time measurements taken at analysis time with respect to execution times that can occur during system deployment, so that MBPTA can trustworthily be applied. In this thesis, we propose several methods to determine whether the number of observations taken at analysis, as part of the normal MBPTA application process, capture the cache events significantly impacting execution time and Worst-Case Execution Time (WCET). If this is not the case, our techniques provide the user with the number of extra runs required so that cache events are captured ensuring trustworthiness on MBPTA provided WCET estimates. Our techniques have been evaluated using a set of synthetic benchmarks and a real avionics application.