Simulated Annealing-Based Large-scale IP Trafic Matrix Estimation

Traffic matrix reflects the volume of traffic that flows between all pairs of sources and destinations in a network. Its element is referred to as an Origin-Destination (OD) pair (or flow). And traffic matrix gives network operators a global aspect of how all the traffic in a large-scale network flows. Thus, with traffic matrix as a key input of traffic engineering and network management, it is very important for network operators to accurately get the traffic matrix in a large-scale network. Unfortunately, as commented in (Papagiannaki et al., 2004), direct measurement of the traffic is not generally practical in the large-scale networks. In 1996, Vardi firstly introduced network tomography method to research the problem that traffic matrix in a network is indirectly measured. Since then, many researchers studied the problem and proposed many solutions(Cao et al., 2000; 2001; Juva, 2007; Soule et al., 2005; 2004; Tan & Wang, 2007; Vardi, 1996; Zhang et al., 2003; 2005). Traffic matrix estimation is so far used by network operators to conduct the network management, network planning, traffic detecting and so on. However, since traffic matrix estimation holds the highly ill-posed properties (Soule et al., 2005; 2004; Tan & Wang, 2007; Vardi, 1996; Zhang et al., 2003; 2005) and especially network traffic is a kind of nonstationary traffic (Cao et al., 2001), this subject is a challenging research problem. The statistical inference techniques are first used to estimate traffic matrix over local area network (LAN). Authors in (Cao et al., 2000; Vardi, 1996) exploited the statistical model to model the OD flows in order to reduce the ill-posed nature of traffic matrix estimation. Zhang et al. (Zhang et al., 2003; 2005) introduced the gravity model into large-scale IP traffic matrix estimation. By the gravity model, they could obtain the prior information about OD flows and then successfully conduct the large-scale IP traffic matrix estimation. Nevertheless, as mentioned in (Juva, 2007; Soule et al., 2005), the statistical inference techniques are sensitive to the prior information, while the gravity model methods still have the larger estimation errors though it partially reduces the sensitivity to the prior information. Especially when the assumptions about OD flows hold, the gravity model methods are found to be more accurate than the statistical inference techniques, while their estimation accuracy decreases more quickly than that of the statistical inference techniques when the assumptions are not exactly true. Hence, this needs to develop a new method to estimate large-scale IP traffic matrix. This chapter provides the reader with a method that large scale IP traffic matrix is estimated accurately by, as envisioned by the authors. It begins by explaining the need for using simulated annealing to estimate the trafficmatrix, how trafficmatrix estimation problem is defined, 5