Comparing centrality indices for network usage optimization of data placement policies in fog devices

Abstract

This paper presents an optimization policy for data placement in fog architectures to reduce the network usage. The proposal is based on modeling the devices (sensors, gateways, fog devices and the cloud provider) through complex weighted networks using the hop count between the fog devices and the data sources (sensors). Centrality indices are evaluated to select the fog devices that store the data generated by the sensors of the same type. The experiments were conducted to evaluate the improvements obtained with 3 centrality indices (eigenvector, betweenness centrality, and current flow) and four different types of network topologies (Barabasi-Albert, Grid, Random Euclidean, and Lobster). The results were validated by comparing between the network usage obtained with our placement proposal and the case of storing all the data in the cloud provider. Important benefits were measured in the case of the eigenvector centrality, reducing the number of transmitted packets to only the 31.84% of packets that are transmitted when the data is placed in the cloud provider. The other two centrality indices showed worst results.