A Measurement-Based Model for Parallel Real-Time Tasks

Under the federated paradigm of multiprocessor scheduling, a set of processors is reserved for the exclusive use of each real-time task. If tasks are characterized very conservatively (as is typical in safety-critical systems), it is likely that most invocations of the task will have computational demand far below the worst-case characterization, and could have been scheduled correctly upon far fewer processors than were assigned to it assuming the worst-case characterization of its run-time behavior. Provided we could safely determine during run-time when all the processors are going to be needed, for the rest of the time the unneeded processors could be idled in low-energy "sleep" mode, or used for executing non-real time work in the background. In this paper we propose a model for representing parallelizable real-time tasks in a manner that permits us to do so. Our model does not require us to have fine-grained knowledge of the internal structure of the code represented by the task; rather, it characterizes each task by a few parameters that are obtained by repeatedly executing the code under different conditions and measuring the run-times.