Probabilistic performance risk analysis at system-level

We present a novel hybrid approach for performance analysis of a system design. Unlike other approaches in this area, in this paper we do not focus on the determination of pessimistic best-case and worst-case quantities of system properties. Our proposed analysis methodology determines qualitative numbers between best-case and worst-case of system properties and quantifies them with probabilities. For this issue, we combine local coarse-grained profiling and formal system-level analysis models in a hybrid approach for an early quantitative determination of qualitative system properties. Our approach considers the control-flow of communicating processes and the impact of blocking communication instances on the temporal behavior of the entire system during formal analysis. This can be used for determining the global system performance. The application of our new methodology leads to an inclusion of probabilities concerning system properties and allows an early performance risk estimation of a design with regard to predefined system requirements and constraints.