This thesis presents a methodology for the tool-assisted generation of performance models from software designs. First, the author has defined the Core Scenario Model (CSM) as an intermediate model to capture performance information from a software behaviour specification. Second, he has examined in detail the problem of transforming CSMs derived from UML annotated with MARTE performance stereotypes into performance models.
The thesis presents a set of algorithms for transforming CSMs into other CSMs in order to: enforce the correctness of the associations defined in the metamodel; clean up
CSMs with minor syntactic flaws; and normalize CSMs from heterogeneous software designs in order facilitate the generation of performance models. The thesis also presents algorithms for weaving CSM aspect sub-models and for generating Layered Queueing Network (LQN) and Queueing Network (QN) models from CSMs.
Three substantial case studies of service systems defined in UML, automatically generated as CSMs with tools developed by others, and then automatically generated LQNs using a CSM2LQN tool based on Eclipse and which implements the algorithms developed by the author.
The advantages of
the methodology presented here are that it captures emergent system behaviour and its associated resource use in a manner that accounts for blocking interactions and does not lose the performance impact of the layered resource architecture.