Power-Aware Task Assignment for Priority-Driven Distributed Real-Time System
Abstract
Power consumption has long been a limiting factor in portable devices and is of growing concern for non-portable systems. Both increased parallelism and dynamically varying voltage and speed of processors are being used to decrease power consumption. For hard real-time applications with dynamically changing task sets, online algorithms are needed to assign tasks to processors in a way that reduces overall energy usage. This paper compares several heuristic algorithms, for assigning periodic tasks to processors in partitioned hard real-time multiprocessor systems using priority-driven DVS algorithms. Which task assignment heuristic performs best depends on the DVS algorithm being used and communication costs between processors. The Min∆P task assignment algorithm, introduced in this paper, produces lower energy usage than other heuristic task assignment algorithms evaluated, when tasks are scheduled with static EDF. With more aggressive DVS algorithms, task assignment has little effect on total energy consumed, suggesting that an algorithm with high schedulability, like Best Fit is most appropriate for use with these DVS algorithms.