Efficiently Measure the Topologies of Large-Scale Networks Under the Guidance of Neural Network Gradients

Abstract

In this letter, we propose a novel network topology measurement method called the neural network gradient-guided method (NGM), which can use traceroutes to efficiently collect topological information (IPs and links) by selecting appropriate destination IPs. In this method, the mapping between probing destination IPs and probing revenues is modelled into a multiclassification problem using a neural network model, and networks are iteratively probed under the guidance of the neural network gradients. Experimental results in real networks demonstrated that NGM can collect more IPs and links by comparing with the methods that randomly selects IPs and subnets for probing.