Static Worst-Case Analyses and Their Validation Techniques for Safety-Critical Systems

Abstract

The reliable operation of systems with both timing and energy requirements is a fundamental challenge in the area of safety-critical embedded systems. In order to provide guarantees for the execution of tasks within given resource budgets, these systems demand bounds of the worst-case execution time (WCET) and the worst-case energy consumption (WCEC). While static WCET analysis techniques are well established in the software development process of real-time systems nowadays, these program analysis techniques are not directly applicable to the fundamentally different behavior of energy consumption and the determination of the WCEC. Besides the missing approaches for WCEC bounds, the domain of worst-case analyses generally faces the problem that the accuracy and validity of reported analysis bounds are unknown: Since the actual worst-case resource consumption of existing benchmark programs cannot be automatically determined, a comprehensive validation of these program analysis tools is not possible.This summary of my dissertation addresses these problems by first describing a novel program analysis approach for WCEC bounds, which accounts for temporarily power-consuming devices, scheduling with fixed real-time priorities, synchronous task activations, and asynchronous interrupt service routines. Regarding the fundamental problem of validating worst-case tools, this dissertation presents a technique for automatically generating benchmark programs. The generator combines program patterns so that the worst-case resource consumption is available along with the generated benchmark. Knowledge about the actual worst-case resource demand then serves as the baseline for evaluating and validating program analysis tools. The fact the benchmark generator helped to reveal previously undiscovered software bugs in a widespread WCET tool for safety-critical systems underlines the relevance of such a structured testing technique.