In this paper, we characterize the delay profile of an Ethernet cross-traffic network statically loaded with one of the ITU-T network models and a larger Ethernet inline traffic loaded with uniformlysized packets, showing how the average time interval between consecutive minimum-delayed packets increases with increased network load. We compare three existing skew-estimation algorithms and show that the best performance is achieved by solving a linear programming problem on "de-noised" delay samples. This skew-estimation method forms the basis of a new sample-mode algorithm for packet delay variation filtering. We use numerical simulations in OPNET to illustrate the performance of the sample-mode filter in the networks. We compare the performance of the proposed PDV filter with those of the existing sample minimum, mean, and maximum filters and observe that the sample-mode filtering algorithm is able to match or outperform other types of filters, at different levels of network load.