How far are we from the definition of a common software performance ontology?

Abstract

The recent approaches to software performance modeling and validation share the idea of annotating software models with information related to performance (e.g. operational profile) and transforming the annotated model into a performance model (e.g. a Stochastic Petri Net). Up to date, no standard has been defined to represent the information related to performance in software artifacts, although clear advantages in tool interoperability and model transformations would stem from it. This paper is aimed at questioning whether a software performance ontology (i.e. a standard set of concepts and relations) is achievable or not. We consider three meta-models defined for software performance, that are the Schedulability, Performance and Time profile of UML, the Core Scenario Model and the Software Performance Engineering meta-model. We devise two approaches to the creation of an ontology: (i) bottom-up, that extracts common knowledge from the meta-models, (ii) top-down, that is driven from a set of requirements.