Integrating a Performance Analysis Kit into Model-Driven Development

Abstract:

Model-driven development is a generative programming technique
in which domain-specific features and engineering decisions are described
using models of desired properties and behavior at various levels
of abstraction. Producing models at the requisite level of specificity is
an expensive undertaking. Generally, organizations offset the costs associated
with generative techniques like model-driven development by
reusing high-level domain-specific models across members of a product
family. Most generative techniques focus on the automation of design
and implementation activities. However, we believe that the increased
level of formalism required by model-driven development can also be
leveraged during other activities like analysis, verification, testing, maintenance
and deployment. Furthermore, as design and implementation
activities become increasingly automated, we feel that verification and
testing activities must leverage the models and the generative engine in
order to be feasibly address the generated systems. This paper presents
a high-level view of a performance analysis kit as an example of a verification
activity that can be integrated into model-driven development.
By automating simulation code generation and instrumentation, data
collection and data reduction, and by using existing models as inputs
into the performance analysis kit, we hope to make it feasible to conduct
performance analyses in an ongoing manner throughout the development
life cycle. Our research will involve the creation of the model-driven performance
analysis kit, its integration into a model-driven development
environment, and an evaluation of the kit’s utility during the design and
implementation of a complex distributed system.