An energy-aware gradient-based scheduling heuristic for heterogeneous multiprocessor embedded systems

In this paper, we propose a heuristic static energy-aware scheduling algorithm for scheduling tasks with precedence constraints on a heterogeneous multiprocessor embedded system consisting of processing elements equipped with dynamic voltage scaling capabilities. While most energy-aware scheduling algorithms in the literature assume that the mapping of the tasks to the processors is known and consider only task ordering and voltage scaling, our algorithm takes into consideration all three factors using the concept of energy gradient. Higher values of energy gradient result in larger reduction in the energy consumption together with smaller increase in the makespan of the schedules. We compare our algorithm to a genetic algorithm in the literature and show that although our algorithm does not consider intra-task voltage scaling, it still provides an average energy savings of about 4% while reducing the optimization time by more than 93%. These energy savings are more significant for larger task graphs.