DSTM: Dual-Strategy Framework for Energy and Thermal Management on Heterogeneous Multicore Systems-on-Chip

With the increasing demand of modern smart devices, they are becoming more sophisticated and computationally demanding. Since their processors, GPUs and accelerators have to handle computationally intensive tasks like video processing, image recognition and real-time analytics etc., they generate significant amount of heat. If this heat is not managed properly, it can lead to a range of issues including thermal throttling, reduced performance, system instability, and even permanent hardware damage. In this paper we have developed a Dual Strategy Thermal Management (DSTM) scheduling technique for heterogeneous Systems-on-Chip (SoC) based smart devices which in its first phase classify the applications based on their characteristics and resource demands and allocate them to the appropriate cores. After that it continuously monitors CPU utilization and temperature of the cores in real-time and dynamically migrate the tasks between cores when thresholds are exceeded. For testing the efficiency of proposed work, traditional scheduler is replaced with DSTM scheduler in Apalis iMX8 Quadmax octa-core heterogeneous processor and targeted with a surveillance application consisting of both multimedia and non-multimedia tasks. Experimental results demonstrate a system-wide improvement in power by 23.9% and temperature profile of the processing cores which not only increased the efficiency of the system but also decreased the thermal incidents while maintaining the performance of the application.