An Approach for Model-Driven Design and Generation of Performance Test Cases with UML and MARTE

Abstract

High-quality software needs to meet both functional and non-functional requirements. In some cases, software must accomplish specific performance requirements, but most of the time, only high-level performance requirements are available: it is up to the developer to decide what performance should be expected from each part of the system. In this work, we show several algorithms that infer the required throughput and time limits for each action in an UML activity diagram from a global constraint and some optional local annotations. After studying their theoretical and empirical performance, we propose an approach for generating performance test cases from the activity diagram after it has been implemented as code. Our approach decouples the performance analysis model from the implementation details of the code to be tested. © Springer-Verlag Berlin Heidelberg 2013.