OpenMLPerf - linguagem específica de domínio gráfica para testes de desempenho em sistemas web

Abstract

Test a system performance is one of the most important tasks in software development process and evolution. Most of the performance tests currently applied consist on testing an implemented system, which make difficult to predict problems, such as load capacities and bottlenecks. Thus, it is recommended to find mechanisms for modeling and specifing the system information before the implementation. A possible solution is the use of models which represent unchangeable aspects from the system domain, such as infrastructure features or users behavior. Use models to abstract, define and model aspects from a domain, nowadays become the most commonly used approaches to solve a problem in a specific context which meets with one of Model-Driven Engineering (MDE) principles, which are to use models in all software cycle-life. One way to specify and model a domain is using Domain-Specific Languages (DSL). DSL are small languages, with limited expressiveness to a determined domain. Different from general purpose languages, such as python and JAVA, DSL are not used to implement systems for any domain, being restricted just to their initial purpose. In this work, we propose the reimplementation of a graphical DSL, named as Canopus, to model performance tests for web systems. In Canopus implementation has used comercial licenses technologies, this restricted its distribution and evolution. Thus, is proposed a new version of this DSL, named OpenMLPerf, will be released in open source license allowing the use, improvement and evolution. To select of DSL support development tool, we performed a Systematic Mapping Study, where we found fifty-three (53) DSL supporting tools. We choose Ecore, from Eclipse Modeling Framework (EMF), and Sirius framework to implement the language metamodel and modeling language. An empirical evaluation was performed to verify the effort required by the user to model a performance test scenario using the OpenMLPerf approach in comparison a UML profile for performance test. The results present evidences a very close effort by the participants who model using both approachs, however, OpenMLPerf approach was affected by performance issues related to Eclipse platform.