Statistical characterization of execution time through simulation

In todaypsilas modern embedded real-time systems forming a deterministic and reliable schedule for task execution remains one of the most pressing demands on the designers of these systems. Almost any scheme utilized in the composition of these schedules will surely rely on the availability of the worst-case execution times (WCET) of the main processes in the system. WCET estimation has been effectively relegated, with a few exceptions, to the domain of static offline analysis. The main reason for such a consignment is the lingering impracticality associated with attempting to provide sufficient coverage for all of the possible system elements and permutations of factors one would need to simulate and observe in order to employ a measurement based approach. This paper describes the development of a methodology relying on the routines and tools available from the statistically based design of experiments (DOE) in the modeling and simulation of the performance of these processes. The analysis done demonstrates the strength of DOE in distinguishing between those parameters which have a significant effect on the performance of the system from other less influential factors. It also lays the foundation for the long overdue incorporation of the use of empirical data in the design and testing of systems.