Measurement of OS services and its application to performance modeling and analysis of integrated embedded software

Performance analyses for embedded software construction with existing components require knowledge of performance characteristics of both application software and operating system (OS) services, especially those services that are critical for real-time applications. Since end users normally do not control the structure and implementation of OS services, but have to use them to meet the system-level performance constraints, it is essential and critical to characterize the performance of OS services with measurements. As such measurements are taken for performance analysis, not for comparison, the measurement methods should be different from those traditionally, used for comparison. In this paper we present an end-to-end method for measuring the performance of timing and scheduling services in selected real-time OSs for the performance modeling and analysis. The proposed method takes the factors of both OS implementations and application configurations into account to obtain the measured performance close to what applications will experience at runtime. The results have shown that the performance characteristics of OS services can be measured without instrumenting the kernel source code, and hence, can be reused for the analysis of a family of applications.