Intelligent Multi-Path Selection Based on Parameters Prediction

Abstract

This paper provides a method for multi-path selection based on parameters prediction. In wireless networks, links with different bandwidths induce different end-to-end delay and the packet loss rate characteristics. It means that we should be able to gain some knowledge of the type of links given the end-to-end delay and the packet loss rate. In this work, we use a neural network machine learning method to infer the types of the links. After predicting the types of the links, we can choose the path based on the prediction of the incremental throughput, for example by choosing the path with the largest potential incremental throughput.