A supercomputing framework for the evaluation of real-time analysis and optimization techniques

Abstract

Generic framework to massively generate synthetic distributed system models.Automatic application of real-time analysis and optimization techniques.Extensible framework independent of any particular metamodel or real-time tool.Supercomputer usage enables large studies for a better validation of techniques.Different generation methods covering a variety of system characteristics. The evaluation of new approaches in the analysis and optimization of real-time systems usually relies on synthetic test systems. Therefore, the development of tools to create these test systems in an efficient way is highly desirable. It is usual for these evaluations to be constrained by the processing power of current personal computers. For example, in order to assess whether a specific technique generally performs better than others or whether the improvement observed is constrained to a limited set of circumstances, a vast set of examples must be tested, making the execution infeasible in a common PC. In this paper, we present a framework that defines the building blocks of a tool to enable the validation of real-time techniques, through the efficient execution of massive evaluations of event-driven synthetic distributed systems. Its main characteristic is that it can leverage the computing power of a supercomputer to perform large studies that otherwise could not be performed with a PC. The framework also defines different generation methods so that the generated systems can cover a wide range of characteristics that can be present in different application domains. As a case study, we also implement this framework based on a previously developed prototype.