Stochastic Prediction and Feedback Control of Router Queue Size in a Virtual Network Environment.

摘要

Modern congestion and routing management algorithms work well for networks with static topologies and moderate QoS requirements. However, these algorithms may not be suitable for modern military networks with fluid dynamic topologies and traffic demands that include mobile networks with many assets. These secure networks require a high level of Quality of Service (QoS) that must adapt to changing demands with no service interruptions. The idea of a network weatherman was developed by Stuckey to predict router queue size by using an extended Kalman filter. He modeled and exercised his design in OPNET with positive results. The goal of this research is to investigate the use of queue size predictions and network weatherman and to determine the effectiveness of three types of filters to predict future traffic demand in a virtual network environment. These filters are an extended Kalman filter, an unscented Kalman filter, and a basic filter. These queue size predictions will be used to implement a network controller to improve the performance of information technology (IT) networks and formulate some type of context awareness and cognitive process in the management of networks by reducing packet loss.