Delay-Optimal Task Offloading for Dynamic Fog Networks

Abstract

Fog computing is a promising paradigm to perform low-latency computation for supporting the internet of things (IoT) applications. It enables provisioning resources and services to be closer for end users. Limited by the computing and storage resources, end users offload the computation-intensive tasks to the nearby fog nodes. However, due to mobility feature of the fog nodes, it's challenging to realize efficient task offloading. We rigorously formulate the task offloading problem for dynamic fog networks as an online stochastic optimization problem, and design offloading policies when the network is in stationary status and non-stationary status. When the fog network is in stationary status, we propose task offloading for the stationary status (TOS) algorithm to minimize the long-term average offloading delay. When the fog network is in non-stationary status, we propose two algorithms as task offloading for the non-stationary status using a sliding window (TON-SW) and task offloading for non-stationary status using a discount factor (TON-D) to minimize the average offloading delay. Besides, learning regret bounds of our algorithms are given. Numerical simulations show that our algorithms achieve a significant performance improvement compared to the upper-confidence bound (UCB) algorithm.