Runtime Performance Evaluation of a Non-Preemptive Cooperative Multithreading Framework Through Tracing

Abstract

In the aerospace and automotive domains, there is a growing trend towards delegating more tasks to embedded software, employing sophisticated algorithms and machine learning based solutions. As a result of this trend, the complexity of embedded software is escalating rapidly. Classical performance analysis methods, such as static worst-case execution time analysis, struggle to cope with this complexity without providing prohibitively over-approximated upper bounds. In this paper, we introduce a tracing-based performance analysis approach tailored to data flow space applications. We illustrate how traces are leveraged to extract arrival curves, minimum distance functions, and execution times. We showcase the utility of tracing in design decisions using an aerospace use case, e.g., optimising the number of cores to reduce end-to-end latency. Furthermore, we extracted and presented debugging information graphically. While our tracing-based performance analysis may introduce overhead on the extracted timing properties, such as worst-case execution time, this overhead is bounded by 6.5%. Finally, we demonstrated the efficacy of our proposed tracing based analysis approach through its application in a space application scenario.