Workload model specifications and adaptive scheduling of semi-hard real-time controls

The authors investigate a class of real-time control systems in which a variety of types of time constraint are imposed on computational activities. In addition, only a subset of critical activities must always meet the time constraints within the allowable bounds; still others do not have time constraints. Also, the top-level system objectives, which determine the criticalness and time constraints of the execution of activities, can be dynamically altered by the system operators or other subsystems. Therefore, specification and scheduling of workload must effectively incorporate the dynamic high-level system objectives and time constraints and reconcile these objectives with contention for limited system resources. The authors define a workload model for specification of activities which consume the system resources and classify activities based on a taxonomy of constraints on the arrival times of an activity's instance. An advanced sensor system is used to illustrate how to use this workload model to specify time constraints of computational activities in system resources. The types of constraints that occur in such systems do not correspond to the classical periodic activities in the real-time system literature. Simulation studies show the shortcomings of some conventional scheduling algorithms. The authors have derived and verified guidelines for tuning scheduling parameters to achieve desired performance objectives.<>