Energy optimization of streaming applications in IoT on NoC based heterogeneous MPSoCs using re-timing and DVFS

The Multiprocessor System-on-Chip (MPSoC) computing architectures are widely adopted in modern embedded systems for real-time applications due to their high performance, reliability, and Quality-of-Service (QoS). Green computing or energy-efficient task scheduling is a critical technological challenging facet in an energy constrained embedded systems because higher energy consumption limits the lifetime of the computing platform and causes an increased carbon footprint. In this paper, we investigate energy-aware task scheduling on Dynamic Voltage and Frequency Scaling (DVFS) enabled Network-on-Chip (NoC) based Heterogeneous MPSoCs (HMPSoCs). We transform the intra-data dependencies into inter-data dependencies of the tasks with precedence constraints represented by Directed Acyclic Graph (DAG). We further implement Energy-efficient Task Scheduling Heuristic (ETSH) algorithm embedded with a list scheduler to perform energy-aware task scheduling while considering the energy performance profiles of the processors and task deadlines. The observed results on 5 real-world and 5 synthetic Task Graphs (TGs) adopted from Embedded Systems Synthesis (E3S) benchmarks suit demonstrate that ETSH outperforms state-of-the-art technique. Concisely, it achieves 20% and 38% average energy-efficiency with and without using coarse-grained software pipelining respectively