Data Offloading in Heterogeneous Dynamic Fog Computing Network: A Contextual Bandit Approach

Abstract

The urban environment is a particularly important application scenario for wireless sensor networks. Nevertheless, these environments are often dense and dynamic, and the sensor nodes are resource-constrained and heterogeneous. Hence, reliable data collection and scalable coordination are a challenge. In this paper, we model the data offloading problem as a bandit problem based on the context of the fog networking paradigm—an important extension of the multi-armed bandit. Through this way, we use the heterogeneity of sensor nodes as contextual information so that sensors can collaborate to offload data to the fog nodes and complete the data collection. Moreover, we have improved the algorithm for the dynamic movement characteristics of the nodes in the urban environment, so that the collaborative system has stable performance in the complex and changing urban environment. The analysis and trajectory simulations based on human movement data in urban environments demonstrate that the proposed scheme can significantly reduce the offloading delay and improve the offloading success rate.