> By Lamia Benamara and ClĂ©mence Magnien

We study here the behavior of nodes in dynamic contact networks.

We use the Rollernet dataset, which has been collected during a rollerblading tour in Paris [1]. This dataset consists of contact logs between Intel iMote nodes distributed on 62 participants. Each iMote performed regular scans (every 15 seconds) and registered the Mac addresses of the responding devices around, thus logging the moments at which the corresponding persons are close to each other. If it receives answers from a given node to all the packets it sent during a given period of time, we consider that the two nodes are in contact during this period. The total duration of the tour is about three hours, composed of two sessions of 80 minutes, interspersed with a break of 20 minutes.

In this video, we observe all contacts in a sliding observation window of 800 s, represented by the vertical lines in the inset (the green vertical line represents the beginning and the red one represents the end of this window). The plots in the inset correspond to the number of all contacts (in orange color) and the number of distinct contacts (in blue color) seen in the corresponding observation window.

In the main plot, each point corresponds to one node of the network (the label represents its identifier). The coordinates of the point are the mean (x-axis) and the standard deviation (y-axis) of the durations of this node’s contacts that were observed in this observation window.

First, we observe that there is a correlation between the number of contacts seen in a given window and the mean and the standard deviation of the nodes’ contacts’ durations. We can observe an heterogeneous behavior, with some large values, at the beginning of the measurement (where we have few contacts). When the observation window is approximately between 1200 and 5000 seconds (see the inset), we observe that the values of the mean and the standard deviation tend to stabilize. This is related to the number of contacts observed during the observation window: we clearly see that there are a large number of contacts during this time. After this, we observe that the number of observed contacts decreases and the values of the standard deviation and/or the mean of some nodes are very high. We also observe that different nodes have very different behaviors. For example, the node 53 has a large mean and standard deviation, which means that the majority of its contacts are long, but it has also some very short contacts.

When we perform the same study by considering different observation window lengths, the observations obtained depend on this length: when we take an observation window of 1500 seconds, the observations obtained are less heterogeneous and more stable than the ones obtained for an observation window of 800 seconds. For an observation window of 2500 seconds, they are even more stable, but we do not observe a very large difference with the ones obtained for 1500 seconds. This allows us to have different and complementary points of view on this dataset.

[1] Pierre-Ugo Tournoux, Jeremie Leguay, Farid Benbadis, Vania Conan, Marcelo Dias de Amorim, John Whitbeck: The Accordion Phenomenon: Analysis, Characterization, and Impact on DTN Routing. INFOCOM 2009: 1116-1124.