Traffic Matrix Estimation Methods Based On Additional Link Load Measurements

To design, control, manage the network successfully, to provide necessary quality of services to the users, network operators need to know more details on the network. Traffic matrices are critical inputs for many network traffic engineering tasks and it could provide a strong support for network planning, congestion controlling, traffic anomaly detection, faults detection, network management and so on.As the network becomes larger, more heterogeneous, more distributed, measure traffic matrix directly is very difficult. It would cost a lot of money. Also, due to the deficiencies of the technology, measure traffic matrix direcly puts a great burden on the equipments, the manipulation of the data would be exhaustive for the network, the transmission of the massive data of the traffic flows and storage of them is a difficult problem. As to measure traffic matrix directly is so difficult and it is so important for the network operators to have traffic matrix, many researchers and industries are studing how to measure traffic matrix indirectly. As the link load measurements and the routing information are much more easier to get than the the traffic matrix in the network and the traffic mtrix, routing matrix, link load measurements have relationships on each other, many studies on how to measure traffic matrix indirectly are based on these measurements.This thesis makes an explicit introduction on how to get traffic matrix, especially on the gravity model of traffic matrix and how to infer the traffic matrix with the gravity model. This thesis proposes two methods to estimate traffic matrix according to the information get by people. One method considers when we cannot differentiate whether the edge link is peering link or access link, how to estimate the traffic matrix. In this situation, we propose an Iterative Information-Theoretic Approach (IITA) to estimate traffic matrix. The IITA could get similar performances with the Minimum Mitual Information (MMI) which needs to differentiate the characteristics of the edge links, and the MMI is one of the best methods in estimating traffic matrix. Another method to traffic matrix is considering when we could measure some traffic flows directly. We propose an Information-Theoretic Approach with Partial Direct Measurements (ITAPDM) to estimate traffic matrix. The ITAPDM could get much better performance than the MMI do. Also the ITAPDM gives the network carriers some guidelines when they could measure some traffic flows directly.