Hybrid Probabilistic Timing Analysis

Abstract

A major challenge in time-critical systems is ensuring that a job will be completed before the required deadline. For this purpose, the behavior of the entire software needs to be analyzed. Worst execution time represents the highest possible execution time of a software unit and this metric is used in resource planning of time-critical systems. Recent studies in the related field have mostly included statistical approaches. They complement their measurement-based timing analysis with probabilistic confidence levels obtained with the help of stochastic methods. In the most used methods, the upper limit of the whole program is tried to be determined either by using end-to-end measurements with the help of extreme value theory or by using convolution techniques to measure small program units. Both approaches are problematic. In this study, a hybrid probabilistic timing analysis method is proposed. With this method, the subunits that are the building blocks of the program can be modeled separately and the highest values can be captured with the help of extreme value theory, and then the dependencies between the units are modeled by using copulas and a better boundary value distribution can be found as a result.