Performance Modelling of Smart Cyber-Physical Systems

Abstract

Context: the dynamic nature of complex Cyber-Physical Systems (CPS) introduces new research challenges since they need to smartly self-adapt to changing situations in their environment. This triggers the usage of methodologies that keep track of changes and raise alarms whether extra-functional requirements (e.g., safety, reliability, performance) are violated. Objective: this paper investigates the usage of software performance engineering techniques as support to provide a model-based performance evaluation of smart CPS. The goal is to understand at which extent performance models, specifically Queueing Networks (QN), are suitable to represent these dynamic scenarios. Method and Results: we evaluate the performance characteristics of a smart parking application where cars need to communicate with hot-spots to find an empty spot to park. Through QN we are able to efficiently derive performance predictions that are compared with long-run simulations, and the relative error of model-based analysis results is no larger than 10% when transient or congestion states are discarded. Conclusion: the usage of performance models is promising in this domain and our goal is to experiment further performance models in other CPS case studies to assess their effectiveness.