Sensitivity of Traffic Matrix Estimation Techniques to Their Underlying Assumptions

Abstract

In this paper we study the traffic matrix estimation problem. Based on the nature of additional information that is used to make the problem solvable, we identify two major groups among the estimation methods proposed in literature: The gravity model based methods and the second moment methods. All methods make some assumptions in order to deploy the extra information in the estimation process. We study the sensitivity of the estimation accuracy to these underlying assumptions. The gravity model methods are found to be more accurate when the assumptions hold, but on the other hand their accuracy declines faster than that of the second moment methods when the assumption is not exactly true. In addition, we propose a novel estimation technique which incorporates both sources of extra information. This method is shown in many cases to outperform the current estimation methods relying only on one or the other.