Vehicular networks are large-scale communication systems envisioned to exploit wireless technologies to interconnect moving cars. The principle is to let vehicles 'talk' to each other or to some roadside infrastructure through wireless links, so as to build autonomous networks in urban areas or along highways.
Vehicular networks are foreseen to provide drivers with real-time information on potential dangers, traffic conditions, and travel times. By exploiting such data, drivers (or fast-reacting built-in controllers if order-of-milliseconds response times are required) could prevent cars from taking part in chain accidents or being stuck in massive traffic jams. They could also receive in-advance notification of approaching out-of-sight emergency vehicles, while data collection from the network would allow applications such as real-time traffic control, adaptive traffic light management, and liability attribution in case of accidents.
Today - more than a decade after the allocation of dedicated frequency bands in the USA - vehicular networks are developing into real-world systems, under the push of standardization bodies (that led to the definition of dedicated standards such as IEEE 802.11p, IEEE 1609, OSI CALM-M5 and ETSI ITS) and the automobile industry.
As in any other communication network, the properties of vehicular networks are primarily driven by the network topology, i.e., the way nodes (the vehicles in our case) are linked to each other to form the 'connectivity graph' of the network. Knowledge of the connectivity graph properties allows to determine the capabilities of the network, its strengths and weaknesses, and to derive rules for the design of dedicated protocols.
In the case of vehicular networks, such connectivity graph is especially challenging to characterize, since it is very dynamic (as cars travel at high speed, links are estabilshed and tore down quickly) and has variable spatiotemporal properties (e.g., the topology of a vehicular network during rush hours and at night are very diverse, and so are those observed in city centers and along peripheral roads).
This notwithstanding, vehicular network connectivity graphs remain a rather unknown object when considered at very large scales. In urban environments in particular, even basic questions stay unanswered, such as: is the vehicular network well connected or highly partitioned? Which size can clusters of multi-hop connected vehicles attain? Which is the internal structure of such clusters? How sparse or dense are single-hop communication neighborhoods? How do all these network connectivity features vary in time? How do they depend on the geographical location?
Our aim is to characterize the instantaneous topology of a vehicular network and provide answers to the questions above. To that end, we focus on the case study of Cologne, Germany, a typical middle-sized European city. The choice was dictated by the availability of a large-scale realistic dataset of the road traffic, describing the per-second position of each car in a 400 square km area during 24 hours. Part of such dataset is publicly available at http://kolntrace.project.citi-lab.fr. By exploiting tools from complex network science, we have already drew conclusions on the significant limitations of the topology in terms of connectivity, availability, reliability and navigability, at both network and component levels. We also unveiled the underlying structure of the vehicular network in the Cologne scenario, composed of vehicles gathered into small cliques which are then connected to each other in a weak, intermittent fashion. Overall, the vehicular network topology appears unfit to medium- and long-range delay-bounded data transfers, for which cars should most probably resort to traditional cellular communication. The vehicular network seems instead to best fit delay-tolerant transfers within localized areas.
Focus