An AOP-based Performance Evaluation Framework for UML Models

Abstract

Performance is a key aspect of non-functional software requirements. While performance cross-cuts much of the software functionality, it is frequently difficult to express in traditional software development representations. In this paper we propose a framework for evaluating software performance using aspect-oriented programming (AOP) and examine its strengths and limitations. The framework provides a mechanism for supporting software performance evaluation prior to final software release. AOP is a promising software engineering technique for expressing cross-cutting characteristics of software systems. We consider software performance as a cross-cutting concern since it is not confined only a few modules, but often spread over multiple functional and non-functional elements. A key strength of our framework is the use of a code instrumentation mechanism of AOP - AspectJ code for performance analysis is separated from Java code for functional requirements. Java code is executable regardless of Aspect J code and can be woven together with AspectJ code when performance is evaluated. Our performance evaluation modules, written in AspectJ are semi-automatically or automatically generated from the UML [1] models with annotated performance profiles. The AspectJ code generator facilitates performance evaluation by allowing performance requirements that have been specified in UML models to be analyzed. The UML diagrams can then be improved by reflecting the feedback from the results of the performance analysis.