A Comparison of Opportunistic Connection Datasets

Abstract

Opportunistic networking differs from more conventional architectures by the lack of existing network infrastructure, which can cause intermittent connectivity or increased communication delay between nodes. From a message routing perspective, solving these problems require a different set of techniques than those used in more traditional network schemes. Forwarding algorithms in these scenarios aim to improve performance metrics such as message delivery ratio and message delay time, while trying to keep the number of message copies small. A common approach used for testing the performance of opportunistic protocols relies on existing opportunistic contact traces. These datasets are widely available on the Internet, and provide a convenient way of simulating realistic usage scenarios. As such, studying the contact patterns between nodes can lead to useful observations to take into account on future experiments. This paper presents the results of a study on four different datasets. First, we describe the main characteristics of each trace. Then, we propose a graphical representation of the contact behavior for each pair of nodes. The next step was to perform an analysis in terms of the distribution of connectivity among nodes, having found that the contacts follow a roughly lognormal distribution and noting that a small group of nodes is usually much more popular than the rest. We have also made a temporal analysis over the duration of each collection experiment. It was noticeable that individual nodes have very similar contact patterns over time, as well as revealing some cyclic variation over time (namely over weekends). Using dataset derived time-varying graph models, a significant performance decrease was achieved with simple remotion of few critical nodes.