Iterative Histogram-Based Performance Analysis of Embedded Systems

Precise execution time profiles (ETPs) are an extremely helpful instrument to assist software and system designers in analyzing the performance and timing behavior of embedded systems. Previously, we have presented an approach that exploits embedded trace units of modern system-on-chip. It allows us to compute execution time histograms during the runtime of the system under test (SuT). These histograms can easily be converted into ETPs. In this contribution, we show an extended version of this method that uses the information gathered in previous runs of the SuT to refine the binning used for the collection of the histograms. We show that these improved histograms deliver much more insight.