A Framework for the Simulation of Complex Real-Time Systems

147 p.Thesis (Ph.D.)--University of Illinois at Urbana-Champaign, 1997.Simulation is often used in the analysis of complex computing systems. The design and validation of complex real-time systems demands capabilities not found in existing simulation environments. We categorize these capabilities into three major functional areas: search control, the execution engine, and output analysis. Search control is an important topic often treated lightly in simulation environments for real-time systems. The search controller chooses values for variable input parameters and launches simulations to determine the performance of the system under different sets of parameter values. The search control framework presented in this thesis allows both conventional design search and also validation search; the latter is required for the deterministic validation of real-time systems. Previous simulation environments did not make it easy to incorporate the wide variety of different scheduling paradigms and algorithms found in large, modern real-time systems. The execution engine framework described in this thesis provides a simple, flexible architecture for incorporating complex schedulers that address one or more dimensions of scheduling. Furthermore, the execution engine framework allows a fully dynamic simulation where tasks migrate, schedulers are changed on-the-fly, etc. The framework also allows hierarchical specification and execution of the system being modelled, which facilitates both information hiding and hierarchical scheduling. The framework thus allows an integrated approach to the analysis of complex real-time systems. No simulation is useful, however, without the ability to easily extract useful high-level information from the voluminous set of low-level events which comprises the basic output of a discrete-event simulator. Output analysis is usually given only cursory treatment in existing simulation environments for real-time systems. The output analysis framework allows the specification of patterns of events which are detected in linear time as the simulation executes. Statistics computed when patterns are matched are used for run-length control and to help direct the search. Finally, a reference implementation of the framework has been created and used in a sample case study. The reference implementation validates the architectural claims made for the framework, while its use in the case study validates the reference implementation itself.U of I OnlyRestricted to the U of I community idenfinitely during batch ingest of legacy ETD