Task Offloading Based on GRU Model in IoT

Abstract

With the rapid growth of the Internet of Things (IoTs), edge computing draws greater attention. Task offloading becomes the main part of edge computing which can affect performance. To reduce the tasks time delay and improve the utilization of the edge server, the task offloading problem can be modeled as a decision-making problem for minimizing the time latency and develop a GRU-based model to predict the computational task offloading. We choose a dataset from Google Cluster and offload the top 1000 tasks for comparison. Compering with existing offloading techniques such as total offloading (TOT), random offloading technique (ROT), and deep learning-based offloading technique (DOT), the GRU-based model can save 15.09% time than TOT, 13.46% time than ROT and 4.25% time than DOT while offloading 1000 tasks on an edge computing system in IoT. Experimental result showed that, compared with other techniques, our proposed GRU-based model is able to reduce the delay of tasks effectively, while increasing the number of tasks and enhancing the offloading performance on edge computing system.