Input data for communication network design/optimization problems involving multi-hour or uncertain traffic can consist of a large set of traffic matrices. These matrices are explicitly considered in problem formulations for link dimensioning. However, many of these matrices are usually dominated by others so only a relatively small subset of matrices would be sufficient to obtain proper link capacity reservations, supporting all original traffic matrices. Thus, elimination of the dominated matrices leads to substantially smaller optimization problems, making them treatable by contemporary solvers. In the paper we discuss the issues behind detecting domination of one traffic matrix over another. We consider two basic cases of domination: (i) total domination when the same traffic routing must be used for both matrices, and (ii) ordinary domination when traffic dependent routing can be used. The paper is based on our original results and generalizes the domination results known for fully connected networks.