Research On Dynamic Traffic Matrix Measurement And Application In IP/MPLS Based Backbone

Traffic Engineering can control and optimize network resources utility. In particular, it is generally desirable to ensure that subsets of network resources do not become over utilized and congested while other subsets along alternate feasible paths remain underutilized. Multi-protocol label switch (MPLS) can offer fast packet switching and bridge over physical and IP layers. Recent developments in MPLS open possibilities to address some of the limitations of IP systems concerning traffic engineering. Historically, effective traffic engineering has been difficult to achieve in public IP networks. One particular shortcoming of conventional IP systems is inadequacy of measurement functions. The limitations of intradomain routing control functions are another issue with conventional IP systems.In this dissertation, we study on traffic engineering for IP/MPLS based backbone. We focus on dynamic traffic matrix measurement and effective traffic control research. Our work is on foundation of two funds. We present a model for deriving dynamic traffic matrix and apply traffic matrix to solve bandwidth fragmentation problem and to balance traffic in the network. We render a multi-field packet classification algorithm and based on it we software realize a traffic measurement system that can measure different service and data objectives in real-time. Firstly, we propose a traffic measurement model for deriving dynamic traffic matrix within IP/MPLS framework. When we obtain the dynamic traffic matrix, we can know traffic at each link of the networks. The main idea of the model is applying traffic measure at each ingress node of network whereas concerning little at core nodes. We present mathematic descriptions for our model and prove its correctness by network flow theorem. When flow equilibrium no longer keep due to packet loss as network in heavy-load situation, we infer formulas by utilizing queuing theory to correct the model. Also we show architecture, extensions to label distribution protocol for routing information, management information base and some algorithms for computing traffic matrix. Lots of simulations prove that our model is effective. To guarantee quality of service (QoS) in networks, most service providers make resource reservations and usually overbooking for specific services. However, it causes bandwidth fragmentation that will waste bandwidth resource. We propose an approach to solve the bandwidth fragmentation by applying dynamic traffic matrix. Compare with the traditional constrained shortest path first (CSPF) routing, our approach makes bandwidth waste induced by bandwidth fragmentation less. Meanwhile, to make traffic control more balance, we present a routing algorithm based on notion of load balance degree that describes equilibrium of utilization at each link of the network. The routing algorithm also applies traffic matrix obtained by our model. Simulation results show that the algorithm benefits traffic control for traffic engineering comparing with CSPF. Also we propose a multi-field packet classification algorithm that includes preprocessing and classification phases. After preprocessing, the algorithm produces a proper data structure that benefits for classification. Then we realize a traffic measurement system by software in IP network based on such classification algorithm. In the end, we conclude our research and shed light on future works.