Task Offloading and Approximate Computing in Solar Powered IoT Networks

Abstract

This letter considers approximate computing and task offloading in a solar powered Internet of Things (IoT) network. Specifically, it addresses the novel problem of minimizing the energy consumption of IoT devices by either offloading their tasks or executing these tasks in approximate mode. To this end, this letter outlines a novel mixed integer linear program (MILP) that computes the minimum total energy required to execute tasks. It optimizes four key factors: (i) task offloading decision of devices, (ii) use of approximate computing by devices, (iii) channel allocation, and (iv) virtual machine (VM) assignment. Further, it outlines a novel solution that determines these factors using channel gain and energy arrival estimates obtained from digital twins (DTs). The results show that our DT-based solution uses at most 1.62x more energy than MILP.