ate traffic congestion in two-layer networks. In this al-
gorithm, the optimal path in physical layer is selected
by considering the node degree of two layers. Lenando
et al.[19] designed Epsoc that incorporates node de-
gree centrality into limited flooding routing. In EpSoc,
the Time-to-Live (TTL) for a message is adjusted ac-
cording to the degree centrality of a forwarding node,
and the number of copies of a message is controlled by
message blocking mechanism. Igarashi and Miyazaki
[20] presented a DTN routing algorithm, which uses
community and centrality to control the message for-
warding at each forwarding node. Guo et al.[21] used
the “gravity” and betweenness centrality of a node for
collaborative routing, where the “gravity” between a
pair of nodes is computed based on the distance be-
tween them and the residual energy. By using the
“gravity” and betweenness centrality, the throughput
can be increased, and the transmission errors can be
reduced.
Since centrality captures the relative importance of
nodes in a network, forwarding messages to nodes
with high centrality can obviously improve data deliv-
ery ratio and reduce latency to some extent. However,
centrality of nodes does not consider the social rela-
tionships between nodes. Sometimes high centrality
of nodes does not necessarily mean high probability
of contact with the intended destination node, as is the
case when the destination node is at the edge of a net-
work. Thus, other social features should also be con-
sidered when designing more effective opportunistic
routing algorithms.
2.2 Similarity-based Routing
It is observed that social networks display high degree
of transitivity. In real life, everyone has his/her own
social features. Two people would have heightened
probability of acquaintance if they share one common
acquaintance or if they share similar hobbies, careers
or geographical locations. Watts and Strogatz showed
that real-world networks exhibit strong clustering or
network transitivity [22]. Similarity, which is a metric
to characterize the grouping nodes sharing common
interests or connections, was proposed and utilized in
[23].
Daly and Haahr [23] leveraged the betweenness cen-
trality and similarity to design their SimBet for mes-
sage forwarding. They combined Ego Networks with
SNA and considered only the local information of
nodes when extracting social features. A message is
forwarded to a node with high betweenness central-
ity and high similarity to the destination node, which
can increase the successful delivery ratio. Simbet has
been shown with good performance in message deliv-
ery, even close to Epidemic routing when the cost is
controlled. However, its performance may be affected
by potential traffic congestion at central nodes.
LABEL, proposed by Hui and Crowcroft [24], is
one of the earliest routing protocols that apply social
features to opportunistic routing. Assume that each
node has a label to inform other nodes of its member-
ship. When a pair of nodes meet, the node carrying
the message compares their labels. If the encountered
node and the destination node have the same label, the
message will be forwarded to the encountered node;
otherwise, the current node continues to carry the mes-
sage. The disadvantage of LABEL is that the delivery
ratio will be relatively low when a message can only
tolerate a short delay while the distance to the destina-
tion is relatively far away. This is because it may be
difficult for the source node to encounter the members
from the community of the destination node when the
destination node is far away.
Zhang et al.[25] applied Mobile Edge Computing
(MEC) to OMSNs in order to reduce the computa-
tional pressure in routing and forwarding process and
designed a corresponding routing algorithm named
FRRF. This algorithm is based on fuzzy reasoning sys-
tems and information entropy and considers the move-
ment and similarity of mobile devices to determine the
transmission priority of mobile devices and compares
the transmission priority between mobile devices in a
network to select the best relay nodes. Specifically,
in this algorithm, two mobile devices meet to collect
and update their respective state information, and then
unload their respective state information to the nearest
vehicle with more powerful computing capability to
accurately compute the comprehensive similarity be-
tween mobile devices and determine the special trans-
mission relationship between mobile devices. Exper-
iments show that FRRF is efficiency in energy con-
sumption, delay, and transmission efficiency.
Lin et al.[26] proposed a routing algorithm, namely
cosSim, based on the cosine similarity of packets be-
tween nodes. For ease of computing, the packet sets in
a network are represented as vectors. The cosine dis-
90 China Communication · February 2021