Adaptation and graceful degradation of control system performance by task reallocation and period adjustment

In this paper, we consider resource adaptation in multiprocessor real-time control systems running periodic control tasks, where the main resource is processor utilization, and the measure of utility is control performance. Processor utilization of a periodic task can be changed by changing the task's period. To estimate the reward rates for running a task at different periods we will use a well-understood concept from optimal control theory, the performance index. We present and evaluate two algorithms for reallocating resources, which are of low complexity in order that they can be run quickly for fast recovery from failure. We also consider transient scheduling effects of changing periods and moving tasks when reallocating resources. By simulation, we show that the algorithms provide results that are close to optimal with randomly generated task sets. We demonstrate the calculation of reward rates for different frequencies of an example application task, and also demonstrate the calculation of the reward rate for a shadow task for the example application task.