Performance Stress Testing of Real-Time Systems Using Genetic Algorithms

Reactive real-time systems must react to external events within time constraints: Triggered tasks must execute within deadlines. Through performance stress testing, the risks of performance failures in real-time systems are reduced. We develop a methodology for the derivation of test cases that aims at maximizing the chances of critical deadline misses within a system. This testing activity is referred to as performance stress testing. Performance stress testing is based on the system task architecture, where a task is a single unit of work carried out by the system. The method developed is based on genetic algorithms and is augmented with a tool, Real Time Test Tool (RTTT). Case studies performed on the tool show that it may actually help testers identify test cases that are likely to exhibit missed deadlines during testing or, even worse, ones that are certain to lead to missed deadlines, despite schedulability analysis assertions.