Traffic vs topology in network clustering: Does it matter?

Network clustering is traditionally accomplished by relying just on the topology information, while a traffic-aware clustering approach has been recently proposed. The latter approach employs traffic matrices to take into account the intensity of the relationship between nodes. The extra effort needed to gather the traffic matrices is warranted if the composition of the clusters obtained in the traffic-based approach is significantly different from that obtained under the topology-based approach. In this paper we compare the outcomes of the two approaches, using the Rand Index as a similarity metric. For a variety of established clustering algorithms, and two large datasets, we show that the two approaches provide significantly different results, since the Rand Index lies far below one.