Bottom Up Algorithm to Identify Link-Level Transition Probability

Abstract

Network tomography aims to obtain network characteristics by end-to-end measurements. Most works carried out in the past focused on the methods and methodologies to identify some of the characteristics, such as loss rate, delay distribution, etc. which are typical static statistical variables showing long-term network behaviors. In contrast to the previous works, we in this paper turn our attention to dynamic characteristics, e.g. transition probability of each link, which unveil the temporal correlation of traffic flows. Those dynamic characteristics could be more important than those static ones since the temporal information can be used in prediction. Apart from that, those characteristics are essential to many other issues, including the models used in network tomography. To identify transition probabilities by end-to-end measurements and in a real-time manner is a challenging task although the problem can be formulated by a hidden Markov model (HMM). Instead of using Baum-Welch algorithm to identify the transition probabilities because it needs a long execution time, we propose a new method that consider the correlations observed by receivers to obtain the transition probabilities in a simple and real-time manner. The proposed method is equal to a closed form solution that makes it a candidate for real-time network control.