FISTE

Abstract

The goal of traffic engineering is to achieve a target Quality of Service (QoS) while maximizing network utilization. While determining the QoS for end-to-end paths in a network under self-similar traffic models is difficult, end-to-end network performance analysis is still essential in providing QoS to networks such as Virtual Private Networks (VPN) and Peer-to-Peer (P2P) networks. The Fast Importance Sampling based Traffic Engineering (FISTE) approach proposed in this article is a prediction-based approach that maps the ingress traffic levels of a network to the QoS of end-to-end path(s) in the network. Because FISTE is a hybrid of simulation analysis and closed-form analysis, it can treat a complex network as a black box. When we combined Simulated Annealing (SA) with FISTE, the resulting approach can provide a traffic engineering solution so that multiple end-to-end QoS requirements are satisfied while the network resource utilization is maximized. FISTE originated from the concept of Importance Sampling (IS), and our approach differs from the previous Importance Sampling based approaches since this is the first time that IS is applied to multi-queue systems under Fractional Gaussian Noise (FGN) input and traffic engineering.