Predicting the Behavior of a Highly Configurable Component Based Real-Time System

Abstract

Software components and the technology supporting component based software engineering contribute greatly to the rapid development and configuration of systems for a variety of application domains. Such domains go beyond desktop office applications and information systems supporting e-commerce, but include systems having real-time performance requirements and critical functionality. Discussed in this paper are the results from an experiment that demonstrates the ability to predict deadline satisfaction of threads in a real-time system where the functionality performed is based on the configuration of the assembled software components. Presented is the method used to abstract the large, legacy code base of the system software and the application software components in the system; the model of those abstractions based on available architecture documentation and empirically-based, runtime observations; and the analysis of the predictions which yielded objective confidence in the observations and model created which formed the underlying basis for the predictions.