Probabilistic simulation-based analysis of complex real-time systems

Abstract

Many industrial real-time systems have evolved over a long period of time and were initially so simple that it was possible to predict consequences of adding new functionality by common sense. However as the system evolves the possibility to predict the consequences of changes becomes more and more difficult unless models and analysis method can be used. Moreover, traditional real-time models, e.g., fixed priority analysis, may be too simple for accurately capturing a complex system's characteristics. For instance, assuming worst-case execution time may not be realistic. Hence, analyses based on these models may give an overly pessimistic result. In this paper we describe our approach to introducing analyzability into complex real-time control systems. The proposed method is based on analytical models and discrete-event based simulation of the system behavior based on these models. The models describe execution times as statistical distributions which are measured and calculated in the existing system. Simulation will not only enable models with statistical execution times, but also correctness criterion other than meeting deadlines, e.g., nonempty communication queues. The simulation result is analyzed by specifying properties in a probabilistic property language. The result of such an analysis is either of probabilistic nature or boolean depending on how the property is specified. Having accurate system models enable analysis of the impact on the temporal behavior of e.g., customizing or maintaining the software.