Software performance engineering in the context of model-driven development

点击翻译按钮获取中文摘要

Model-Driven Development (MDD) is an evolutionary step in the software field that changes the focus of software development from code to models. MDD is based on abstraction and automation: abstraction separates the model of the application under construction from underlying platform models, and automation is used to generate the code from models. The platform models (PM) are reusable and layered, each providing services to the layers above, and requiring services of the layers below. The Object Management Group (OMG) uses the copyrighted term Model-Driven Architecture (MDA) to describe the initiative of producing standards (such as UML, MOF, XMI and CWM) for MDD.The long-term objective of the proposed research is to integrate software performance engineering techniques in the MDD process. Performance (defined here as the degree of providing timeliness) is a pervasive quality of any software system that is determined by all its layers: application, middleware, operating system, hardware (which may be described as layered "platforms" in MDD). A performance model would allow us to reason about the performance properties of the system at the level of assemblies of components/platforms early in the life cycle, when the correction of performance problems is easier and more effective. While previous research of the applicant was concerned with the automatic transformation of a monolithic UML model to a performance model, the proposed research will extend the transformation techniques to a set of inter-related UML models that have to be transformed and composed: a platform-independent model (PIM) of the application and reusable models of the underlying platforms (PMs).There are three short term goals. The first addresses model transformation techniques for generating an overall performance model from a set of input PIM + PMs annotated with performance information. The second goal deals with separating the performance contribution of each platform through performance annotations, which can be composed later. The third goal applies the proposed methods and techniques to two case-studies.