BLM-DTO: Bandit Learning and Matching based Distributed Task Offloading in Fog Networks

Abstract

This paper proposes an algorithm called BLM-DTO that allows each fog node (FN) to implement the task offloading operations in a distributed manner in the fog computing networks (FCNs). Fundamentally, BLM-DTO leverages the principle of matching game theory to achieve the stable matching outcome based on preference relations of two sides of the game. Due to the dynamic nature of fog computing environment, the preference relation of one-side game players is unknown a priori and achieved only by iteratively interacting with the other side of players. Thus, BLM-DTO further incorporates multi-armed bandit (MAB) learning using Thompson sampling (TS) technique to adaptively learn their unknown preferences. Extensive simulation results demonstrate the potential advantages of the proposed TS-type offloading algorithm over the ϵ-greedy and upper-bound confidence (UCB)-type baselines.