Modeling is an important activity in the software development pro-cess whose output are design artefacts that describe the resultingsoftware from a high-level perspective. Recent research investi-gates the role of models at runtime and the results indicate thatanalysts perform better at observing the behaviour of a runningsystem if they can utilize models during the analysis. However,setting up a system which allows the analysis of its behaviour atruntime using models involves many challenges regarding the mod-eling environment, the introspection infrastructure, the traceabilitymanagement and the analysis integration. This paper summarizesdesign alternatives for implementing systems with manual analysissupport by investigating recurring concepts like patterns, model-ing habits, languages, middlewares and development techniquesfound in approaches that utilize models at runtime. We organizethe gained knowledge as patterns in a pattern language which cap-tures various issues and their solution alternatives, including theirbenefits and liabilities. The pattern language consists of modelingpatterns for setting up the models and the environment for theanalyst, introspection patterns for extracting data from the runningsystem, traceability patterns for relating the extracted data withthe models, and analysis patterns for processing the extracted datausing the models. We demonstrate the application of the patternlanguage based on the implementation of a robot system.

Top