Energy Profiling based Load-Balancing Approach in IoT-Edge for Split Computing

Abstract

In 6G technology, both computation and communication aspects goes hand-to-hand for internet of things (IoT) applications. However, due to the limited computational capability of IoT devices, it is challenging to run complex tasks (like, deep neural networks (DNN) models) in IoT devices. To mitigate this problem, the DNN tasks are partitioned to the nearby nodes (e.g., edge devices) using split computing. However, there is a possibility of edge device running out of battery or get busy with other prioritized applications which abruptly ends DNN task execution. In this situation, DNN task is scheduled to nearest available edge device. To perform the scheduling without restarting the DNN task in new edge device, we propose single task load balancing with prioritization (STLBP) model. Further, we also consider a scenario, where IoT device assigns multiple DNN tasks to different edge devices while the edge devices are assigned with prioritized applications and may run out of battery. To deal with such multiple DNN task among different edge devices, we propose a scheduling mechanism called multiple task load balancing with prioritization (M-TLBP) model. We conduct extensive experiments to compare energy profiling performance of S-TLBP and M-TLBP mechanisms, respectively with respect to the state of art.