Research On Performance Prediction Method And Autonation From Uml Software Architectures

With the increase of large-scale and highly complex software, software performance is paid more and more attention. Forecasting the performance of software system can not only predict the bottlenecks of system performance, but also help to get the best design scheme, reducing cost greatly.The paper searches a model-based performance prediction method of UML software architectures. Firstly, we describe software architecture by UML model, using UML use case diagram, activity diagram and component diagram to describe select the UML use case diagram, activity diagram and component diagram to describe the system's external inputs, the execution path and resources respectively. Secondly, to add performance requirements in UML model, a performance document called SPT is used and stereotypes and tagged values are introduced to translate UML model into UML SPT model. Third, we research on the transformation algorithm that transforms the system's UML SPT model into a queueing network model, whose UML activity diagram model contains fork or joint nodes. Then, solve the queueing network model to get performance parameters and bottlenecks, which can determine whether the software system design meets performance requirements, and point out the entry point for system's future improvement.Finally, we provide detail design of this automated tool for predicting system performance of UML software architecture and use the tool to predict the performance of reconfigurable routing and switching platform management system, which shows the feasibility and validity of our performance prediction methods and tool from UML software architectures.