Worst-Case Scheduling of Software Tasks - A Constraint Optimization Model to Support Performance Testing

Real-Time Embedded Systems (RTES) in safety-critical domains, such as maritime and energy, must satisfy strict performance requirements to be deemed safe. Therefore, such systems have to be thoroughly tested to ensure their correct behavior even under the worst operating conditions. In this paper, we address the need of deriving worst case scenarios with respect to three com- mon performance requirements, namely task deadlines, response time, and CPU usage. Specifically, we investigate whether this worst-case analysis can be ef- fectively re-expressed as a Constrained Optimization Problem (COP) over the space of possible inputs to the system. Solving this problem means finding the sets of inputs that maximize the chance to violate performance requirements at runtime. Such inputs can in turn be used to test if the target RTES meets the expected performance even in the worst case. We develop an OPL model for IBM ILOG CP OPTIMIZER that implements a task priority-based preemptive scheduling, and apply it to a case study from the maritime and energy domain. Our validation shows that (1) the input to our model can be provided with rea- sonable effort in an industrial setting, and (2) the COP effectively identifies test cases that maximize deadline misses, response time, and CPU usage.