You are here: » Lectures » Model-based Performance Analysis

Model-based Performance Analysis

lecturer: Dorina C. Petriu

The verification of non-functional requirements of software models is possible by extending first the UML models with additional information specific to the property to be evaluated, and then by transforming such an annotated UML model into a formal model that can be analyzed with known analysis techniques and tools. Examples of formal models used for performance analysis are queueing networks, Petri nets, stochastic process algebras, etc. The “UML Profile for Modeling and Analysis of Real-Time and Embedded systems (MARTE)” can be used for adding quantitative performance annotations (such as performance requirements, resource demands made by different software execution steps, number of visits to resources, etc.) to a given UML model, particularly to the architecture, behaviour and deployment views.

The talk will start by discussing the kind of performance annotations that need to be added to UML software models, and the principles for transforming annotated software models into performance models. Such a model transformation must bridge a large semantic gap between the source and the target model. Next, the talk will present the state of the art on model-based performance analysis of annotated UML software models, giving examples from literature and from speaker’s own work. Future research challenges will be discussed, such as: a) using separation of concerns in the performance annotations for different system layers in the context of MDA, which will allow for reusing performance annotations for different underlying platforms; b) applying model-based performance analysis to Aspect-Oriented Models, and c) to Software Product Lines.