A social overlay-based forwarding scheme for mobile social networks Sun-Kyum Kim 1 • JunYeop Lee 1 • Sung-Bong Yang 1 Published online: 18 December 2015 Ó Springer Science+Business Media New York 2015 Abstract In mobile social networks, forwarding is a challenging problem since there may be no end-to-end paths. Existing schemes using overlays are not applicable to pure mobile social networks since they use environments different from the mobile social networks. To resolve this problem, we propose a novel forwarding scheme that considers the common neighbor similarity and contact probability. In the proposed scheme, the more nodes have in common, the more likely they are to be friends, and higher contact probabilities between nodes increase the chances they will arrive at the destination. The proposed scheme constructs an overlay using two layers with the common neighbor similarity and the contact probability and forwards messages to the destinations through the overlay. Experimental results show that the proposed scheme outperforms most known forwarding schemes in terms of balancing the network traffic and transmission delay. Keywords Mobile social network Overlay Forwarding Common neighbor similarity Contact probability 1 Introduction As the advances of wireless communication technologies and the popularity of mobile devices, this emerging para- digm is applicable to various types of mobile sensing such as mobile cloud computing [1–3], mobile crowdsourcing [4], and wireless communications such as wireless sensor networks [5], multi-hop wireless networks [6], cognitive radio networks [7], green communications [8, 9], vehicular networks [10], heterogeneous networks [11], software-de- fined networks [12], mobile ad hoc networks [13, 14], delay tolerant networks [15, 16], social networks [17], and mobile social networks [18–20]. Among them, mobile social networks (MSNs; also known as opportunistic net- works [21, 22] or pocket switched networks [23]) have attracted a lot of attention for wireless communication paradigms. MSNs are advanced networks of mobile ad hoc networks (MANETs) and delay-tolerant networks (DTNs) with social properties of human behaviors [24]. Due to the short transmission range and node mobility, such networks suffer from intermittent connectivity and frequent discon- nections in sparse environments. For this reason, MSN applications are used for the Sami Network Connectivity Project [25], Zebranet [26], the Shared Wireless Info-Sta- tion (SWIM) [27], the N4C project [28] including web caching, email/not-so-instant messengers, hiker’s applica- tions, meteorological data transfer applications, animal tracking applications, and TIRE applications [29], provid- ing the connection between technology and the users. Since there may be no stable end-to-end paths in MSNs, the forwarding or routing schemes [30] for ad-hoc net- works are so inaccessible that forwarding or routing has become a challenging problem in MSNs. The forwarding mechanism operates hop by hop in a store, carry, and forward manner, where the relay nodes store and carry & Sung-Bong Yang [email protected]Sun-Kyum Kim [email protected]JunYeop Lee [email protected]1 Department of Computer Science, Yonsei University, Seoul, Korea 123 Wireless Netw (2016) 22:2439–2451 DOI 10.1007/s11276-015-1162-2
13
Embed
A social overlay-based forwarding scheme for mobile social ...algo.yonsei.ac.kr/international_JNL/social_overlay_forwarding.pdf · delay tolerant networks [15, 16], social networks
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
A social overlay-based forwarding scheme for mobile socialnetworks
Sun-Kyum Kim1• JunYeop Lee1 • Sung-Bong Yang1
Published online: 18 December 2015
� Springer Science+Business Media New York 2015
Abstract In mobile social networks, forwarding is a
challenging problem since there may be no end-to-end
paths. Existing schemes using overlays are not applicable
to pure mobile social networks since they use environments
different from the mobile social networks. To resolve this
problem, we propose a novel forwarding scheme that
considers the common neighbor similarity and contact
probability. In the proposed scheme, the more nodes have
in common, the more likely they are to be friends, and
higher contact probabilities between nodes increase the
chances they will arrive at the destination. The proposed
scheme constructs an overlay using two layers with the
common neighbor similarity and the contact probability
and forwards messages to the destinations through the
overlay. Experimental results show that the proposed
scheme outperforms most known forwarding schemes in
terms of balancing the network traffic and transmission
delay.
Keywords Mobile social network � Overlay �Forwarding � Common neighbor similarity � Contactprobability
1 Introduction
As the advances of wireless communication technologies
and the popularity of mobile devices, this emerging para-
digm is applicable to various types of mobile sensing such
as mobile cloud computing [1–3], mobile crowdsourcing
[4], and wireless communications such as wireless sensor
3), … U(i, m)}. Whenever Ni meets Nj, Ni evaluates Ei, Si,
Ii, Ui, exchanges and updates its information vectors with
Nj for clustering during the warm-up period. At this time,
Ni also computes P(j, i) at the same time to reduce extra
communication between them. After the warm-up period
and message generation, each node Ni constructs an over-
lay for forwarding. Since information vectors are changed
dynamically, each node Ni exchange the information vec-
tors with the destination and its constructed overlay
information in the forwarding process.
4.5 Construction of overlay
The more nodes have in common, the more likely they are
to be friends, and when nodes have higher contact proba-
bilities, they are more likely to meet the destination. These
are the characteristics of the common neighbor similarity
and contact probability. SOF constructs an overlay con-
sisting of two asynchronous social layers with these metrics.
Note that the two social layers are different from each
other. Both types of social layers are complementary since
these layers consist of a lot of social information for con-
tact such as the number of common neighbor nodes, the
total contact count, the contact duration and the inter-
contact time. Mei et al. [44] found that nodes with similar
social information tend to have contacts more often in
MSNs. Therefore, it is also possible to construct an overlay
by considering the social information or relationships
regarding the common neighbor similarity and contact
probability among nodes to improve the performance of
MSN forwarding.
Before constructing the overlay, each node, Ni, builds
two layers, the common neighbor similarity-based layer
and the contact probability-based layer, using the social
information it obtained. When Ni meets another node, Ni
exchanges and updates the social similarity and the contact
probability by Eqs. (1), (2), and (3). Then, Ni builds clus-
ters in the common neighbor similarity-based layer and
creates a social graph in the contact probability-based
layer. Ni sorts the obtained similarities to the destination in
ascending order and extracts the last score, which is the
highest similarity score to the destination in the sorted
similarities. Ni divides the sorted nodes into clusters of
equal size according to the range of the similarity scores
and the number of clusters. The highest similarity score
that a node has in each cluster becomes the cluster’s sim-
ilarity (i.e., delegated similarity).
Afterwards, Ni constructs an overlay by combining the
similarity-based layer and the contact probability-based
layer. The construction of the overlay is implemented only
in the last cluster in the similarity-based layer that the
destination belongs to. Ni filters the nodes with higher
contact probabilities by virtually linking the nodes
belonging to a cluster to one another. Since the network has
no servers for global information, Ni updates its own
overlay information including the common neighbor sim-
ilarities and contact probabilities even though the warm-up
period has ended.
Figure 1 shows the construction of an overlay of Ni. We
assume that Ni and Di are the source and the destination,
respectively. The similarity score of Ni to Di is six. Ni
builds the social similarity-based layer. Since the maxi-
mum similarity to Di is 20 and the number of clusters is set
to four, the four clusters are evenly built at intervals of five
similarity score units. Ni belongs to the second cluster. The
delegated similarity of each cluster is 5, 10, 15, and 20.
Then, only the nodes in the last cluster, which has the
destination, link to one another according to the contact
probabilities.
Algorithm 1 describes how to construct an overlay. dindicates the number of clusters as a parameter. maxValueiand maxLeveli denote the highest similarity score of Ni for
clustering and the maximum cluster level, respectively. Fi
is the cluster levels of nodes encountered by Ni. Ri is the
score range for clustering of Ni, which is calculated by
maxValuei, maxValuei, and the number of encountered
nodes. m0 denotes the total number of encountered nodes of
Ni. After sorting Si in ascending order according to the
similarity to the destination Di, Ni extracts the last (i.e., the
maximum) similarity score and keeps it in maxValuei. Ni
then divides the encountered nodes into clusters according
to Ri and stores cluster levels (i.e., the social similarity-
based layer information) in Fi in Lines 7–14. Finally, Ni
links the nodes to one another.
Wireless Netw (2016) 22:2439–2451 2443
123
The selection of the number of clusters, d, could affect
the performance significantly. If d is large, each cluster has
higher network traffic in terms of communications since
each cluster has a large number of nodes. On the other
hand, if d is small, each cluster has a lower transmission
delay because each cluster has a small number of nodes.
Thus, we perform extensive experiments to find a proper d(explained in Sect. 5.3.1).
4.6 Message forwarding
In this section, we describe how each node forwards a
message in SOF. Each node, Ni, uses either the cluster-
based forwarding or the probabilistic forwarding. In the
cluster-based forwarding, Ni forwards the message to Nj
belonging to a cluster with a higher delegated similarity.
Note that when nodes are clustered at a higher similarity
score, they are more likely to be similar to the destination;
that is, they are more likely to meet the destination fre-
quently. On the other hand, when Ni encounters only Nj
belonging to a cluster with a lower delegated similarity, Ni
does not forward the message to Nj.
When a message arrives at a node belonging to the
cluster with the highest delegated similarity (where the
destination is), the received node, Ni, forwards the message
to an Nj with a higher contact probability until the message
is delivered to the destination. Such forwarding is called
probabilistic forwarding. Since Ni uses the clustering-based
forwarding inter-clusters and probabilistic forwarding in
the last cluster, the messages can be delivered reliably and
the network traffic can be reduced while maintaining an
acceptable delay.
Fig. 1 Construction of an
overlay
Algorithm 1. Pseudo code for constructing the overlay cluster01: ConstructingOverlay(δ){02: Sort Si in ascending order according to the similarity to the destination Di;03: maxValuei← the last (highest) value in the sorted Si;04: maxLeveli ← δ;05: for n ← 1 to maxLeveli06: Ri[n] ← (maxValuei / maxLeveli) * (n+1);07: for n ← 1 to 08: if Ri[0] > S(i, n) { // the lowest delegated similarity09: Fi[n] ← 1;10: for k ← 1 to maxLeveli – 1 11: if Ri[k] = S(i, n) && Ri[k+1] > S(i, n) { // the mid-value delegated similarity12: Fi[n] ← k+1;13: if S[maxLeveli -1] < S(i, n) && maxValue > S(i, n) // the highest delegated similarity14: Fi[n] ← maxLeveli;15: Link node n to one another belonging to Fi[n] = maxLevel;16: }
2444 Wireless Netw (2016) 22:2439–2451
123
Figure 2 illustrates how message forwarding is done
based on Fig. 1 in SOF. Each cluster has a delegated
similarity, DS, such as 5, 10, 15, and 20. The source node,
Ni, and the destination node, Di, each belongs to the clus-
ters with a DS of 10 and a DS of 20. The large circles and
the small circles indicate clusters and nodes, respectively.
Using cluster-based forwarding, Ni forwards the message to
nodes belonging to the clusters with a DS of 15 or 20 and
does not forward to nodes belonging to the cluster with a
DS of 5 since Ni’s cluster has a higher DS than the cluster
with a DS of 5. When a node belonging to the cluster with a
DS of 20 receives the message, the node forwards to the
other nodes with a higher contact probability that belong to
the same cluster until the message is delivered to the des-
tination using probabilistic forwarding. At this time, each
node forwards the message to the nodes with higher contact
probabilities that belong to the cluster with the highest DS.
The forwarding algorithm in SOF is outlined inAlgorithm
1. Assume that Ni has a message destined for Di. Since all
values in the information vectors are changed dynamically,
Ni updates its information vectors whenever Ni meets Nj
during the forwarding process. Ni exchanges Ni’s informa-
tion vectors including Di and Fi with Nj. Ni then updates the
similarity scores, Si, and the contact probability, P(i, j), with
I(i, j) andU(i, j).Ni also updates each P(i, k) for k = i = j in
Pi using Eq. (3) with Pj. Next, Ni (re-)constructs an overlay
using Si andPi. Finally,Ni compares its cluster levelFi. Since
Ni also exchanges Fi, the forwarding process is implemented
according to the source node’s cluster level. IfNj belongs to a
clusterwith a higher delegated similarity thanNi,Ni forwards
the message to Nj. Otherwise, Ni does not forward the mes-
sage to Nj. However, if they belong to the same cluster with
the highestDS,Ni compares its contact probability toDi. IfNj
has a higher contact probability, Ni sends the message to Nj.
Fig. 2 Message forwarding in
SOF
Algorithm 2. Pseudocode for SOF forwarding 01: // After discovery of an encountered node Nj02: SOF_Forwarding() {03: Exchange_Information (Ni, Di, Fi); // Exchange information vectors of Ni, adding Di and Fi04: Update_SimilarityScore(Si); // Update the similarity score of Ni05: Update_ContactProbability(Pi); // Update the contact probability of Ni
06: ConstructingOverlay(δ); // Update an overlay07: if Di = Nj08: Forward the message to Nj;09: else10: if Fi [Ni] < Fi [Nj] // Cluster-based forwarding11: Forward the message to Nj;12: else13: if Fi [Ni] = Fi [Nj] && Fi [Ni] = maxLeveli // Probabilistic forwarding14: if P(Ni, Di) < P(Nj, Di) 15: Forward the message to Nj;16:}
Wireless Netw (2016) 22:2439–2451 2445
123
5 Experimental results
5.1 Simulation environments
Our simulation model follows the model in [41]. We exploit
the network simulator NS-2 v2.35 [53] for the simulations.
The network area and the community size are set to 450 m x
450 m and 150 m x 150 m, respectively. The number of
communities is four among nine grids. Each community has
10 or more nodes. The number of nodes is set to 40, 50, 60,
and 70. The communication range of a node varies from 10
to 50 m. The node movement follows the HCMM, which is
a widely used mobility pattern in MSN simulations. In the
HCMM, we are not allowed to manipulate the input data to
our advantage. The velocity of a node ranges from 1 to 9 m/
s. Since the HCMM does not have social information, we
need the warm-up period to obtain the information to sim-
ulate the non-zero knowledge forwarding schemes; thus, the
total simulation time is 8000 s, including the 1000 s warm-
up period for constructing an overlay. The number of
clusters varies from 1 to 10 to construct the overlay for our
extensive experiments. In our simulator, each node issues
one message after the warm-up period. After transmitting a
message to other nodes, the originating node does not delete
the message. We run each scheme 20 times to determine the
average of the results. Table 2 lists the parameters used in
our simulation.
We evaluate the proposed scheme using the following
three performance metrics:
• Delivery ratio: the ratio of the number of delivered
messages to the total number of messages issued.
• Network traffic: the total number of messages sent and
received.
• Delay: the time needed for a message to travel from the
source to the destination.
• Hop count: the average number of hops required for a
message to travel from the source to the destination.
5.2 Forwarding schemes for simulations
For 8000 s, we simulate and compare the proposed scheme,
SOF, with non-clustering schemes such as Epidemic, Wait,
SimBet, ProPhet, and the common neighbor similarity-
based scheme. During this period, except for Wait, the
schemes achieve a delivery ratio of one.
• Epidemic [35]: The source node spreads messages to
any node it encounters.
• Wait: The source node waits to send a message until it
meets the destination node.
• SimBet [42]: The source node sends a message to a
node with a higher SimBet utility value through
considering the similarity between the source node
and the destination and the betweenness centrality.
• ProPhet [47]: The source node sends a message to a
node with a higher contact probability, p(a,b).
• Common: The source node sends to a node with a
higher social similarity by using Eq. (1).
5.3 Simulation results
5.3.1 Effect of the number of clusters
We examine the performance of SOF in terms of the number of
clusters, d, in the common neighbor similarity-based layer when
the communication range is 10 m.This simulation indicates how
many clusters are appropriate for the performance.We need this
result since there is one overlay and the performance depends on
d. Figure 3(a)–(c) shows the network traffic, the transmission
delay and the hop count of SOF with the number of clusters
increasing from1 to10.As shown inFig. 3(a),whend is smaller,
SOF has low network traffic whereas SOF has higher network
traffic when d increases. Such phenomena arise because as dincreases, SOF hasmany clusters that each node sendsmessages
to.On theother hand, inFig. 3(b) asd increases, the transmission
delay of SOF decreases since SOF has a lot of nodes joining to
forward and the messages to the destination node quickly. In
Fig. 3(c), since thenodes act as bridgenodes, asd increases,SOFhas a larger number of hop counts.
According to Fig. 3(a)–(c), it turns out that as d increases
and each node communicates with many other nodes, the
network traffic of SOF increases, the transmission delay
decreases, and the number of hop counts increases. However,
since SOF appropriately uses both the social cluster-based
forwarding and probabilistic forwarding, its traffic amount is
still smaller than other schemes.Clearly, a suitabled is four forthis environment, since we get relatively lower traffic (about
180,000) as well as acceptable delay and hop counts (delay of
less than 600 s and hop count of about two). However, an
appropriate number of clusters should be chosen according to
a given environment.
Table 2 Parameters in our simulation
Parameter Value (default)
Network area 450 9 450 m2
Community size 15,050 m2
Number of grids 9
Number of communities 4
Number of nodes 40, 50, 60, 70, (40)
Communication range 10, 20, 30, 40, 50, (10) m
Velocity of nodes 1–9 m/s
Warm-up period 1000 s
Number of clusters 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, (5)
Simulation time 8000 s
2446 Wireless Netw (2016) 22:2439–2451
123
5.3.2 Effect of the simulation time
We evaluate the performance of all schemes based on the
simulation time. Figure 4(a)–(c) show the simulation
results for various simulation times. The number of nodes
and the communication range are set to 40 and 10 m,
respectively. We set the number of social clusters, d, tofour as a result of the previous simulation in Sect. 5.3.1. In
Fig. 4(a), most schemes achieve the maximum delivery
ratio faster than Wait due to multiple copies of messages
being used. Since Wait does not distribute a message and
instead waits for the message to reach its destination, it
Fig. 3 Effect of the number of clusters, d. a Network traffic, b transmission delay, c hop count
Fig. 4 Effect of the simulation time. a Delivery ratio, b network traffic
Wireless Netw (2016) 22:2439–2451 2447
123
needs a greater amount of time to reach the 1.0 delivery
ratio. Owing to sufficient copies of messages by cluster-
based forwarding and probabilistic forwarding, SOF
achieves 1.0 slightly slower than other schemes. Each
scheme undergoes a high amount of network traffic as the
simulation time increases [Fig. 4(b)]. Wait is better than
SOF because a single copy of the message is maintained.
However, other schemes are worse than SOF, since SOF
only distributes the messages to the nodes belonging to
clusters with higher delegated similarities and only com-
municates with the nodes belonging to the cluster with the
highest delegated similarity score; their contact probabili-
ties consider frequency, longevity and regularity. Thus, as
the simulation time passes the network traffic difference
between SOF, the non-clustering schemes increases while
maintaining comparable delivery ratios.
5.3.3 Effect of the number of nodes
We evaluate the performance of the schemes as the number
of nodes increases. The communication range is set to
10 m. The number of clusters, d, is four. Figure 5(a) shows
the network traffic when the number of nodes increases. As
expected, except for Wait and SOF, the schemes’ traffic
greatly increases. SOF in particular has lower network
traffic than the other schemes. In SOF, nodes only com-
municate with nodes belonging to the clusters with higher
delegated similarity even if the number of nodes increases.
Since nodes forward the messages between clusters (except
for the last cluster), the clusters act like large-sized nodes.
Thus, SOF has lower network traffic. Figure 5(b) shows the
transmission delay. The delay for most schemes decreases
as the number of nodes increases. However, Wait has
similar patterns regardless of the number of nodes. ProPhet
and SOF decrease slightly in terms of the transmission
delay since these schemes need to find the nodes with
higher contact probabilities. However, since the actual
gap between SOF and the other scheme is somewhat large,
it is acceptable. Even though the number of nodes in
the network increases, SOF maintains the balance between
the network traffic and transmission delay well.
Figure 5(c) shows the hop counts of the schemes.Wait does
Fig. 5 Effect of the number of nodes. a Network traffic, b transmission delay, c hop count
2448 Wireless Netw (2016) 22:2439–2451
123
nothing until each destination is encountered; hence, it has
a hop count of one. Basically, as the number of nodes
increases, the hop counts of all of the schemes also
increase. However, SOF has a smaller number of hop
counts than the other schemes because it handpicks nodes
belonging to the clusters with higher delegated similarities
to send the messages to.
5.3.4 Effect of the communication range
Lastly, we examine the effect of the nodes’ communication
range on each scheme. The number of nodes and the
number of clusters, d, are set to 40 and 4, respectively. In
Fig. 6(a), as the communication range increases, the net-
work traffic of SOF and the other schemes increases. SOF
traffic is higher thanWait but lower than the other schemes.
Furthermore, the actual gap between SOF and the other
schemes is large since SOF nodes do not distribute the
messages to nodes belonging to lower delegated similari-
ties. Figure 6(b) shows the transmission delay. Most
schemes experience a decreased delay as the number of
nodes increases. The transmission delays of SOF and Wait
decrease significantly. Figure 6(c) shows the hop counts.
As the communication range increases, the hop counts of
all of the schemes basically increase. However, the hop
counts of SOF and ProPhet decrease. This is because in
both schemes only the nodes with higher contact proba-
bility to the destination participate in the delivery of
messages within their communication range. Interestingly,
in ProPhet, when the communication range increases, the
network traffic slowly increases, however, the transmission
range slowly decreases and the hop count decreases
because nodes forward the messages to other nodes with
higher contact probabilities within their communication
range. Since SOF uses both cluster-based forwarding and
probabilistic forwarding, the network traffic of SOF slowly
increases, the transmission delay decreases sharply, and the
hop count also decreases. SOF benefits from the advan-
tages of both kinds of forwarding methods. Consequently,
SOF maintains the network traffic and transmission delay
levels due to the existence of sufficient copies of the
messages in the constructed overlay.
Fig. 6 Effect of the communication range. a Network traffic, b transmission delay, c hop count
Wireless Netw (2016) 22:2439–2451 2449
123
6 Conclusion
We proposed an effective forwarding scheme for MSNs,
SOF, which effectively uses an overlay. SOF constructs an
overlay by using social information and takes advantage of
cluster-based forwarding and probabilistic forwarding in
the overlay that is built. In cluster-based forwarding, a
source node belonging to a cluster with a high delegated
similarity forwards a message. A node that receives a
message in the cluster with the highest delegated similarity
only sends the message to another node with a higher
contact probability through probabilistic forwarding. Since
SOF distributes the messages to the nodes in the clusters, it
reduces the network traffic in comparison to the non-
clustering schemes; the transmission delay is acceptable in
comparison to Wait. In future work, we plan to study more
enhanced, dynamic, forwarding schemes with varying
resources in MSNs.
Acknowledgments This research was supported by the Basic Sci-
ence Research Program through the National Research Foundation of
Korea (NRF) funded by the Ministry of Education, Science and
Technology (2013R1A1A2011114).
References
1. Rahimi, M. R., Venkatasubramanian, N., & Vasilakos, A. V.
(2013). Music: Mobility-aware optimal service allocation in
mobile cloud computing. In Proceedings of IEEE Sixth Interna-
tional Conference on Cloud Computing (CLOUD) (pp. 75–82).
2. Rahimi, M. R., Ren, J., Liu, C. H., Vasilakos, A. V., &
Venkatasubramanian, N. (2014). Mobile cloud computing: A
survey, state of art and future directions. Mobile Networks and
Applications, 19(2), 133–143.
3. Feng, Z., Zhu, Y., Zhang, Q., Ni, L. M., & Vasilakos, A. V.
(2014). TRAC: Truthful auction for location-aware collaborative
sensing in mobile crowdsourcing. In Proceedings of INFOCOM
(pp. 1231–1239).
4. Vasilakos, A. V., Li, Z., Simon, G., & You, W. (2015). Infor-
mation centric network: Research challenges and opportunities.
Journal of Network and Computer Applications, 52, 1–10.
5. Song, Y., Liu, L., Ma, H., & Vasilakos, A. V. (2014). A biology-
based algorithm to minimal exposure problem of wireless sensor
networks. IEEE Transactions on Network and Service Manage-
ment, 11(3), 417–430.
6. Meng, T., Wu, F., Yang, Z., Chen, G., & Vasilakos, A. V. (2016).
Spatial reusability-aware routing in multi-hop wireless networks.
IEEE Transactions on Computers, 65(1), 244–255. doi:10.1109/
TC.2015.2417543.
7. Youssef, M., Ibrahim, M., Abdelatif, M., Chen, L., & Vasilakos,
A. V. (2014). Routing metrics of cognitive radio networks: A