Predictive OS Modeling for Host-Compiled Simulation of Periodic Real-Time Task Sets

Abstract

With the increasing complexity of embedded software, host-compiled simulators have been introduced to address the need for a fast simulation environment. However, designers pay the price for higher performance with a loss in timing accuracy. In this letter, we introduce a novel predictive OS model to provide fast software simulation with accurate scheduling of periodic real-time tasks. The OS model predicts the next preemption point by monitoring system state, and automatically and optimally adjusting the granularity of back-annotated delays. We evaluated our simulator on a range of periodic task sets. Our observations show that we can achieve the same 99% accuracy as a simulation at 1 μs granularity with an average 230× speedup.