Network Congestion Arbitration and Source Problem Prediction Using Neural Networks

Abstract

Rapidly growing needs for networking in the Internet and in local Intranets make network management increasingly important in today's computer world. We propose using learning techniques to predict network congestion problems before they start impacting the performance of services. In this paper, we focus on using a simple feed-forward neural network to predict severe congestion in a network. We also use neural networks to predict the source or sources responsible for the congestion, and we design and apply a simple control method for limiting the rate of the offending sources so that congestion can be avoided. Unlike the usual TCP/IP flow control, the proposed method is applied only to selected nodes and converges faster to the final rate. The described techniques set the stage for a new wave of network managers that are capable of preventing networking problems instead of repairing them.