Towards a parallel template catalogue for software performance predictions

Software Performance Engineers evaluate quality attributes (like response time) of software rich systems based on architectural models during early design time. Thereby, they use model-based approaches to analyze the software's behaviour and resource consumption. One prominent approach is the Palladio Component Model (PCM) which has been researched for over a decade. However, when it comes to multicore support or massive parallel execution the approaches still suffer from major drawbacks. Drawbacks include inaccurate prediction models, insufficient modelling languages, and missing tool support, to name just some of them. In this paper, we focus on overcoming the last two drawbacks, namely improving the modelling language and the tool support. We present a template catalogue for parallel performance patterns in software performance predictions. We use the example of a parallel-loop in the Palladio Component Model to exemplify our idea.