Know Thy Neighbor: Towards Optimal Mapping of
Contacts to Social Graphs for DTN Routing
Theus Hossmann, Thrasyvoulos Spyropoulos, and Franck Legendre
Computer Engineering and Networks Laboratory
ETH Zurich, Switzerland
lastname@tik.ee.ethz.ch
Abstract—Delay Tolerant Networks (DTN) are networks of
self-organizing wireless nodes, where end-to-end connectivity
is intermittent. In these networks, forwarding decisions are
generally made using locally collected knowledge about node
behavior (e.g., past contacts between nodes) to predict future
contact opportunities. The use of complex network analysis has
been recently suggested to perform this prediction task and
improve the performance of DTN routing. Contacts seen in the
past are aggregated to a social graph, and a variety of metrics
(e.g., centrality and similarity) or algorithms (e.g., community
detection) have been proposed to assess the utility of a node to
deliver a content or bring it closer to the destination.
In this paper, we argue that it is not so much the choice
or sophistication of social metrics and algorithms that bears
the most weight on performance, but rather the mapping from
the mobility process generating contacts to the aggregated social
graph. We first study two well-known DTN routing algorithms
– SimBet and BubbleRap – that rely on such complex network
analysis, and show that their performance heavily depends on
how the mapping (contact aggregation) is performed. What is
more, for a range of synthetic mobility models and real traces,
we show that improved performances (up to a factor of 4 in
terms of delivery ratio) are consistently achieved for a relatively
narrow range of aggregation levels only, where the aggregated
graph most closely reflects the underlying mobility structure.
To this end, we propose an online algorithm that uses concepts
from unsupervised learning and spectral graph theory to infer
this “correct” graph structure; this algorithm allows each node
to locally identify and adjust to the optimal operating point, and
achieves good performance in all scenarios considered.
I. INTRODUCTION
The Delay Tolerant Networking (DTN) paradigm has been
proposed to support emerging wireless networking applica-
tions, where end-to-end connectivity cannot be assumed for
technical reasons (e.g., propagation phenomena, and node mo-
bility) or economical reasons (e.g., lack of infrastructure, low
power nodes) [1], [2], [3]. To cope with this, opportunistic or
mobility-assisted routing algorithms have been proposed [4],
[5]: messages are forwarded one hop at a time, only when two
nodes are in contact (i.e., move within transmission range);
without full or any knowledge of future contact opportunities,
a forwarding decision normally aims to simply increase the
delivery probability at every step.
To combat the inherent uncertainty of future contact oppor-
tunities, many protocols forward in parallel multiple replicas
of the same content [6] or resort to coding (network coding,
erasure coding). Nevertheless, node mobility (and resulting
contact opportunities) are not entirely random. Instead, weak
or strong patterns are present. To this end, numerous utility-
based routing schemes attempt to differentiate nodes that
are more likely to deliver content or bring it closer to the
destination [7].
Among them, a number of schemes implicitly assess the
strength of (“social”) ties between nodes. For example [8] uses
time of last encounter, and [9] uses contact frequency as a hint
on the similarity of mobility patterns. [10], [11] use instead a
metric much akin to degree centrality to identify nodes that
are highly mobile/social; the former scheme is reminiscent
of search in scale-free networks [12], while the latter uses
centrality to choose which relays to “spray” a limited budget of
message replicas to. However, these simple metrics may only
capture one facet of the underlying mobility process, which
can hinder good contact predictions.
Complex network analysis [13] (CNA) has recently been
proposed as a more generic and powerful tool to formulate
and solve the problem of future contact prediction in DTNs.
Past observed contacts between nodes are aggregated into a
social graph, with graph edges representing (one or more)
past meetings between the vertices. An edge in this graph
conveys the information that two nodes often encounter each
other either because they have a strong social tie (friends),
or because they are frequently co-located without actually
knowing each other (familiar strangers); thus, existence of an
edge intends to have predictive capacity for future contacts.
Two recently proposed routing protocols, SimBet and Bub-
bleRap [14], [15], make explicit use of CNA metrics and algo-
rithms in order to highlight a node’s position in the aggregated
social graph, and assess its utility to act as a relay for messages
destined to other nodes in the graph. Although the detailed
mechanisms of the two protocols differ (see next Section),
they are both based on the same principles: they assume
that nodes naturally reside in mobility-related communities
(e.g., class, work, home). Increasingly “central” or “well-
connected” nodes in the graph are then chosen as carriers to
relay content over different communities, until a node that
shares many neighbors with the destination [14], (i.e., belongs
to the destination’s community [15], [16]) is reached. These
protocols have been reported to often outperform well-known
DTN routing schemes that are not explicitly “social”.
Nevertheless, it is not well understood under what con-
ditions these protocols and their individual components can
achieve the suggested performance, nor is it why. What is
more, it is actually not (just) the choice or sophistication of
social metrics or algorithms that bears the most weight on
performance, but rather the mapping from the mobility process
generating contacts to the aggregated social graph. This
mapping presents a tradeoff, where some information about