Abstract Within the federated framework of multiprocessor scheduling, a dedicated set of processors is assigned to each real-time task for its sole purpose. When tasks are described in an overly cautious manner (which is common in safety-critical environments), the majority of task invocations will probably exhibit computational demands significantly lower than the worst-case scenario, indicating that they could have been effectively scheduled on far fewer... Within the federated framework of multiprocessor scheduling, a dedicated set of processors is assigned to each real-time task for its sole purpose. When tasks are described in an overly cautious manner (which is common in safety-critical environments), the majority of task invocations will probably exhibit computational demands significantly lower than the worst-case scenario, indicating that they could have been effectively scheduled on far fewer processors than those allocated based on the worst-case runtime prediction. If we could reliably determine during execution when all processors are necessary, the surplus processors could be put into a low-energy “sleep” state when not in use, or repurposed for handling non-real-time tasks in the background. In this paper, we introduce a model for depicting parallelizable real-time tasks in a way that allows us to achieve this. Our model does not necessitate detailed insight into the internal workings of the code represented by the task; instead, it defines each task using a handful of parameters acquired through multiple executions of the code under varying conditions, while recording the run-times. Read More Read Less