Precise Continuous Non-Intrusive Measurement-Based Execution Time Estimation

Abstract

Precise estimation of the Worst-Case Execution Time (WCET) of embedded software is a necessary precondition in safety critical systems. Static methods for WCET analysis rely on precise models of the target processor’s micro-architecture. Measurement-based methods, in contrast, rely on exhaustive measurements performed on the real hardware. The rise of the multicore processors often renders staticWCET analysis infeasible, either due to the computational complexity or due the lack of necessary documentation. Current approaches for (hybrid) measurement-based WCET estimation process the trace data offline and thus need to store large amounts of data. In this contribution, we present a novel approach that performs continuous online aggregation of timing measurements. This enables long observation periods and increases the possibility to catch rare circumstances. Moreover, we incorporate the execution contexts of basic blocks. We can therefore account for typical cache behaviour, without being overly pessimistic.