Subtask-Aware Energy Allocation Algorithm for Parallel Applications Scheduling on Heterogeneous Computing Systems

Heterogeneous computing systems have been extensively applied in recent years due to their ability to greatly enhance the computation capability of parallel applications. While this kind of system can greatly increase the computation capability of parallel applications, it is usually constrained by energy cost. Therefore, to achieve a tradeoff between computing ability and energy cost, the optimization of the schedule length under the constraint of energy cost has been paid more and more attention. In this paper, a subtasks-aware task scheduling strategy is proposed, and we defined the concepts of subtasks execution time ratio (SER) and subtask impact factor (SIF) to allocate energy of tasks reasonably. Specifically, the computational load of all subtasks of each task in the application is considered as an influence on the performance of the algorithm. Concurrently, to minimize the makespan of applications and make the allocation of energy more equitable, this paper presents a new method of pre-distribution of weighted energy. Our experimental results show that this method is effective in reducing the scheduling length under energy cost limits in parallel applications.