Energy optimization for deadline-constrained parallel applications on multi-ECU embedded systems

Abstract

Multi-ECU (electronic control unit) systems are widely used in drones and electric vehicles. The trend of sharing resources by multiple tasks is inevitable due to the sensitive costs and space. However, the random arrival nature of tasks makes the scheduling difficult to be designed. This paper studies the task scheduling on multi-ECU embedded platforms that assume to accept more than one kind of tasks with random arrival. First, we propose an integer-linear-programming-based dynamic DAG scheduling algorithm (ILPS), which can efficiently reduce energy overheads while satisfying deadline constraints of all tasks. In addition, an offline-training-online-using strategy is proposed to speed up the generation of scheduling policies, for the use of ILPS in practice. Second, we propose a DVFS-available energy saving algorithm (DAES), which decreases energy consumption by finding an appropriate frequency for each task in an iterative manner, whilst satisfying deadline constraints. Results of extensive experiments illustrate that our algorithm is better than existing methods in terms of energy consumption and scheduling length.