Quantitative Evaluation of UML State Machines Using Stochastic Petri Nets

Abstract

Todays technical systems tend to be very complex and usually are embedded in an asynchronous, concurrent, and distributed world. Besides the functional correctness of a system the observation of aspects like dependability, reliability, and timeliness becomes mandatory. Cost-awareness plays also an important role. It is desirable to detect incorrect and inefficient behavior in the very early stages of the system development process in order to revise design decisions easier and cheaper. Thus, methods and tools are needed that support designing and modeling as well as performance evaluation. This work focuses on the use of the Unified Modeling Language (UML) for modeling aspects of real-time systems. UML gained increasing acceptance over the last years and is well established as a standard for modeling discrete event and software systems. Within UML, this work concentrates on State Machines. In combination with the Profile for Schedulability, Performance, and Time UML State Machines are well suited for modeling object life cycles including quantitative system aspects. The direct quantitative evaluation of the resulting UML State Machine models is extensive. Therefore, an indirect approach transforming UML State Machines into Stochastic Petri Nets is proposed. Thus, a re-use of well established methods is enabled. Existing Petri Net tools like TimeNET can be used for analysis and simulation. In this context, a formal semantics for UML State Machines by means of Stochastic Petri Nets is given. Based on the resulting Stochastic Petri Nets the quantitative evaluation of the UML State Machines is performed. As a result of this work TimeNET has been extended by a new State Machine net class. This allows modeling and evaluation within the same tool, since Stochastic Petri Nets are already supported. The application of the presented approach to the upcoming European Train Control System with its high demands for safety, timeliness, and reliability shows the applicability of the approach.