Content uploaded by Ciprian Dobre
Author content
All content in this area was uploaded by Ciprian Dobre on Aug 22, 2014
Content may be subject to copyright.
Opportunistic Networks: A Taxonomy of Data
Dissemination Techniques
Radu Ioan Ciobanu1, Ciprian Dobre1*
* Corresponding author.
1 Computer Science Department, Faculty of Automatic Control and Computers,
University Politehnica of Bucharest, Romania
radu.ciobanu@cti.pub.ro, ciprian.dobre@cs.pub.ro
Abstract. When mobile devices are unable to establish direct communication, or
when communication should be offloaded to cope with large throughputs, mobile
collaboration can be used to facilitate communication through opportunistic net-
works. These types of networks, formed when mobile devices communicate only
using short-range transmission protocols, usually when users are close, can help
applications still exchange data. Routes are built dynamically, since each mobile
device is acting according to the store-carry-and-forward paradigm. Thus, contacts
are seen as opportunities to move data towards the destination. In such networks
data dissemination is usually based on a publish/subscribe model. Opportunistic
data dissemination also raises questions concerning user privacy and incentives. In
this we present a motivation of using opportunistic networks in various real life
use cases, and then analyze existing relevant work in the area of data dissemina-
tion. We present the categories of a proposed taxonomy that captures the capabili-
ties of data dissemination techniques used in opportunistic networks. Moreover,
we survey relevant techniques and analyze them using the proposed taxonomy.
Keywords: Opportunistic network, taxonomy, dissemination, mobile devices.
1 Introduction
In the past years, mobile devices (such as smartphones, tablets, or netbooks) have
become almost ubiquitous, which has lead to the advent of several new types of
mobile networks. Such networks are composed almost entirely of mobile devices,
and differ considerably from the classic wired networks both in terms of structure,
but also in regard to the protocols and algorithms used for routing, forwarding and
dissemination. Since there isn't a stable topology, nodes in mobile networks are
not aware of a global structure and have no knowledge of their relationship with
other nodes (like proximity, connection quality, etc.). Each node is only aware of
information about the nodes that it is in contact with at a certain moment of time,
2
and may act as data provider, receiver and transmitter during the time it spends in
the network. Thus, a node can produce data, carry it for other nodes and transmit
it, or receive it for its own use.
One type of such mobile networks that has been deeply researched in recent
years is represented by opportunistic networks (ONs). They are dynamically built
when mobile devices collaborate to form communication paths while users are in
close proximity. Opportunistic networks are based on a store-carry-and-forward
paradigm [1], which means that a node that wants to relay a message begins by
storing it, then carries it around the network until the carrier encounters the desti-
nation or a node that is more likely to bring the data close to the destination, and
then finally forwards it.
One of the main challenges of opportunistic networks is deciding which nodes
should the data be relayed to in order for it to reach its destination, and do it as
quickly and efficiently as possible. Various types of solutions have been proposed,
ranging from disseminating the information to every encountered node in an epi-
demic fashion [2], to selecting the nodes with the highest social coefficient or cen-
trality [3]. Prediction methods have also been employed [4], based on the
knowledge that the mobile nodes from an opportunistic network are devices be-
longing to humans, which generally have the same movement and interaction pat-
terns that they follow every day. The analysis of contact time (duration of an en-
counter between two nodes) and inter-contact time (duration between consecutive
contacts of the same two nodes) has also been used in deciding a suitable relay
node. Aside from selecting the node that the data will be forwarded to, research
has also focused on congestion control, privacy, security, or incentive methods for
convincing users to altruistically participate in the network.
An important topic in opportunistic networks is represented by data dissemina-
tion. In such networks, topologies are unstable. Various authors proposed different
data-centric approaches for data dissemination, where data is pro-actively and co-
operatively disseminated from sources towards possibly interested receivers, as
sources and receivers might not be aware of each other and never get in touch di-
rectly. Such data dissemination techniques are usually based on a pub-
lish/subscribe model. In this article we analyze existing work in the area of data
dissemination in opportunistic networks. We analyze different collaboration-based
communication solutions, emphasizing their capabilities to opportunistically dis-
seminate data. We present the advantages and disadvantages of the analyzed solu-
tions. Furthermore, we propose the categories of a taxonomy that captures the ca-
pabilities of data dissemination techniques used in opportunistic networks. Using
the categories of the proposed taxonomy, we also present a critical analysis of four
opportunistic data dissemination solutions. To our knowledge, a classification of
data dissemination techniques has never been previously proposed.
The rest of the paper is structured as follows. Section 2 highlights the motiva-
tion of our work, along with the significance and applicability of opportunistic
networks in real life. Section 3 presents relevant contributions in the research area
of opportunistic networks. Section 4 proposes the categories of a taxonomy for
analyzing and comparing data dissemination techniques in opportunistic networks.
In Section 5 we survey and critically analyze, using the proposed taxonomy, four
3
relevant dissemination techniques. In Section 6 we conclude and present future re-
search directions of our work.
2 Motivation
There are many real life scenarios where opportunistic networks may be em-
ployed. One such scenario is represented by disaster management [5]. When a dis-
aster (such as an earthquake, a tsunami, an explosion, etc.) occurs, regular com-
munication might be down, because of potential damage to the wireless access
points. In this situation, mobile devices can form a new opportunistic infrastruc-
ture that can be used for communication. The devices can be employed either for
the use of the rescue teams, or even to look for survivors among the debris. If the
survivors have their devices on and running the opportunistic software, then they
might able to communicate with other devices close by and signal the rescue
teams with the owner's position. Another potential use is the communication be-
tween the rescue teams, by placing temporary access points at key places in the
area of the disaster, and using them to relay data.
Opportunistic networks have an important use in situations where the contacts
between mobile devices happen often and for longer periods of time. Such a situa-
tion occurs in crowded places, like a music concert or an amusement park. In these
cases, ONs may be used to broadcast the location of mobile device owners to in-
terested receivers. For example, a child at an amusement park might carry a
smartphone that opportunistically broadcasts the child's encrypted location, by
leveraging the neighboring devices. The child's parents have a smartphone of their
own which receives the broadcast from encountered nodes, thus being aware of
the child's location at any point in time. Similarly, events regarding the bands that
will be playing or any other announcements can be disseminated to participants of
a music concert or any other type of social event with high participation. Such de-
sign principles also lead to the creation of fully-distributed social networking ser-
vices [6], which were lately proposed as an alternative to solving problems related
to data ownership and privacy problems associated with today’s centralized ap-
proaches.
Another potential practical use of opportunistic networks is in regard to floating
content in areas such as open city squares [7], where mobile nodes enter a geo-
graphical zone (called an anchor zone), spend some time in it and then leave.
While in the anchor zone (which gives the physical boundaries of the network),
the mobile devices produce content and opportunistically replicate it to other
nodes. These nodes may copy the data either if they need it for themselves, or if
they transport it for the benefit of other nodes. A node that exits the anchor zone
deletes the data, so the availability of the floating content is probabilistic. When
dealing with floating content, ON-specific algorithms are employed and adapted
for making the replication decisions. The best nodes should be chosen for replicat-
ing data, while at the same time congestion should be avoided.
4
We also believe that a platform for supporting generic context-aware mobile
applications such as CAPIM [8] can fully benefit from opportunistic network inte-
gration. CAPIM (Context-Aware Platform using Integrated Mobile services) is a
solution designed to support the construction of next-generation applications. It in-
tegrates various services that collect context data (location, user’s profile and
characteristics, environment information, etc.). The smart services are dynamically
loaded by mobile clients, which take advantage of the sensing capabilities provid-
ed by modern smartphones, possibly augmented with external sensors. The data is
collected and aggregated into context instance. Such a framework can be em-
ployed in an academic environment as a means to facilitate the interaction be-
tween students, professors and other faculty members. The communication could
be performed opportunistically, instead of used a fixed infrastructure such as WiFi
or 3G, in order to limit the bandwidth used and to save battery life (since Blue-
tooth consumes less power than both WiFi and 3G). CAPIM would be used to dis-
seminate announcements to students, signal their location, the classes they take,
the interactions they have, etc.
Smart cities are cities that monitor and integrate the conditions of all their criti-
cal infrastructures (such as roads, bridges, tunnels, rails, subways, airports, sea-
ports, communications, water, power, etc.) in order to better optimize their re-
sources and plan their preventive maintenance activities [9]. They connect the
physical, IT, social and business infrastructures, for the purpose of leveraging the
collective intelligence of the cities. Opportunistic networks are the logical means
of achieving at least a part of a smart city infrastructure, since they can be em-
ployed to perform communication between various parts of a smart city. For ex-
ample, the traffic lights system can be opportunistically connected to a service that
offers information about traffic jams, crowded roads, accidents, etc., so it can
adapt to the conditions of the environment.
More recently, ONs have been used for other purposes, such as targeted adver-
tising. One such example is MobiAd [10], an application that presents the user
with local advertisements in a privacy-preserving manner. The ads are selected by
the phone from a pool of advertisements which are broadcast on the local mobile
base station or received from local WiFi hotspots. Information about ad views and
clicks is encrypted and sent to the ad channel in an opportunistic way, via other
mobile devices or static WiFi hotspots. This helps ensure privacy, since the other
nodes won't discover which ads were viewed, and the ad provider can't know
which user saw what ad.
Thus, we have shown that there are a multitude of potential uses for opportunis-
tic networks in various situations. Therefore, it is important to have an overview
of the capabilities of ONs, and this is the main reason for the taxonomy of data
dissemination methods in opportunistic networks that we propose in Section 4. A
differentiation should be made between the various available techniques, since
specific algorithms should be employed in specific situations.
5
3 Related Work
A taxonomy for analyzing forwarding techniques in opportunistic networks sepa-
rates them between algorithms without infrastructure and algorithms where the ad-
hoc networks exploit some form of infrastructure to forward messages [1]. Algo-
rithms without infrastructure can be further divided into algorithms based on dis-
semination (like Epidemic, Meeting and Visits and Networking Coding), and algo-
rithms based on context (like CAR and MobySpace). Algorithms that exploit a
form of infrastructure can also be divided into fixed infrastructure and mobile in-
frastructure algorithms. In case of fixed infrastructure algorithms (like Infostations
and SWIM), special nodes are located at specific geographical points, whereas
special nodes proposed in mobile infrastructure algorithms (like Ferries and
DataMULEs) move around in the network randomly or follow predefined paths.
An alternative taxonomy [11] separates the forwarding methods according to their
knowledge about context. Accordingly, there are three types of dissemination ap-
proaches: context-oblivious, partially context-aware and fully context-aware.
Some of the most popular forwarding algorithms nowadays are BUBBLE Rap [3],
Propicman [12], HIBOp [13] or dLife [14].
A thorough analysis of opportunistic networking is presented in [15]. The au-
thors present details regarding the architecture of the Haggle project and give var-
ious solutions to forwarding and data dissemination techniques. Haggle is a Euro-
pean Commission-funded project that designs and develops solutions for
opportunistic networks communication, by analyzing all aspects of the main net-
working functions, such as routing and forwarding, security, data dissemination
and mobility traces and models [16]. Also discussed in [15] is security in terms of
opportunistic networking, along with applications such as mobile social network-
ing, sharing of user-generated content, pervasive sensing or pervasive healthcare.
Several papers exclusively treat the problem of data dissemination in opportun-
istic networks. The Epidemic approach is presented in [2]. In [17], a dissemination
technique based on publish/subscribe communication and communities is de-
scribed, while [18] and [19] propose a wireless ad hoc podcasting system based on
opportunistic networks. A multicast distribution method is presented in [20], while
ContentPlace, a system that exploits dynamically learned information about users'
social relationships to decide where to place data objects in order to optimize con-
tent availability, is presented in [21]. These methods are further analyzed in Sec-
tion 5. To compare them we apply the categories of the proposed taxonomy.
4 A Taxonomy for Dissemination Techniques
In this Section we introduce the categories of the proposed taxonomy for data dis-
semination techniques in opportunistic networks (as seen in Figure 1). This taxon-
omy was created by analyzing the existing methods in the literature and highlight-
ing the common characteristics between various methods.
6
Fig. 1. A taxonomy for data dissemination techniques.
According to the proposed taxonomy, data dissemination algorithms can be
categorized by the organization of the network on which they apply. In general, in
an opportunistic network no assumption is made on the existence of a direct path
between two nodes that wish to communicate. Nonetheless, some dissemination
algorithms may exploit certain nodes called "hubs" and build an overlay network
between them. The hubs are the nodes with the highest centrality in each commu-
nity, where a node's centrality is the degree of its participation in the network.
There are several types of node centrality relevant to data dissemination in oppor-
tunistic networks (such as degree centrality, betweenness centrality or closeness
centrality), that will be detailed later. The algorithms that build an overlay network
based on hubs fall under the category of data dissemination algorithms with infra-
structure. However, relying on an infrastructure might be costly to maintain (due
to the large number of messages that have to be exchanged to keep it) and also
highly unstable, especially in case of networks that contain nodes with a high de-
gree of mobility. Considering this aspect, many data dissemination algorithms as-
sume that the opportunistic network is a network without infrastructure. The net-
work organization is relevant for a data dissemination algorithm because it
directly influences the data transfer policies.
7
The actual nodes that participate in an opportunistic network play an important
part in the way a data dissemination algorithm works. Consequently, the proposed
taxonomy also categorizes dissemination techniques according to node character-
istics. A first characteristic of a node in an opportunistic network is the node state.
Depending on its implementation, a dissemination technique can either follow a
stateful, a stateless or a hybrid approach. An approach that maintains the state of a
node requires control traffic (e.g. unsubscription messages in a publish/subscribe-
based algorithm) that can prove to be expensive. Moreover, it suffers if frequent
topology changes occur. On the other hand, a stateless approach does not require
control traffic, but has unsatisfactory results if event flooding is used. The hybrid
approach takes advantage of both the stateful and the stateless approaches.
Another important characteristic of a node in an opportunistic network is node
interaction. As stated before, nodes in an opportunistic network generally have a
high degree of mobility, so the interaction between them must be as fast and as ef-
ficient as possible. The reason for this is that contact duration (the time interval
when two network devices are in range of each other) may be extremely low. Ac-
cording to the proposed taxonomy, there are three basic aspects of node interac-
tion, the first one being node discovery. Depending on the type of mobile device
being used, the discovery of nodes that are in the wireless communication range
can be done in several ways, but it is usually accomplished by sending periodical
broadcast messages to inform neighboring nodes about the device's presence.
When two nodes come into wireless communication range and make contact, they
each have to inform the other node of the data they store. Therefore the second as-
pect of node interaction is content identification, meaning the way in which nodes
represent the data internally and how they "declare" it (usually using some form of
meta-data descriptions). Nodes may also advertise the channels they have data
from or they may present a hash of the data they store. The final subcategory of
node interaction is data exchange, which is the way two nodes transfer data to and
from each other. This refers not only to the actual data transferring method, but al-
so to the way data is organized or split into units. The three node interaction steps
presented here may also be done asynchronously for several neighboring nodes,
and the way they are implemented affects the performance of a data dissemination
algorithm.
As stated in [22], an interesting use case for opportunistic networks is the shar-
ing of content available on mobile users' devices. In such a network, users them-
selves may generate content (e.g. photos, clips) on their mobile devices, which
might be of interest to other users. However, content producers and consumers
might not be connected to each other, so an opportunistic data dissemination
method is necessary. Because there can be many types of content, each having dif-
ferent characteristics, the proposed taxonomy also classifies data dissemination al-
gorithms according to the content characteristics.
An important aspect of the actual content is its organization. Most often, con-
tent is organized into channels, an approach used for publish/subscribe-based data
dissemination. The publish/subscribe pattern is used mainly because communica-
tion is based on messages and can be anonymous, whilst participants are decou-
pled from time, space and flow. Time decoupling takes place because publishers
8
and subscribers do not need to run at the same time, while space decoupling hap-
pens because a direct connection between nodes does not have to exist. Further-
more, no synchronized operations are needed for publishing and subscribing, so
nodes are also decoupled from the communication flow. The approach allows the
users to subscribe to a channel and automatically receive updates for the content
they are interested in. Such an organization is taken further by structuring chan-
nels into episodes and enclosures.
Aside from the way content is organized at a node, the proposed taxonomy cat-
egorizes data dissemination techniques by content analysis. Content analysis rep-
resents the way in which the algorithm analyzes a certain content object and de-
cides if it will fetch it or not. There are two reasons a node might download a
content object from another encountered node: it is subscribed to a channel that
the object belongs to, or the node has a higher probability of being or arriving in
the proximity of another node that is subscribed to that channel than the node that
originally had the information. Not all dissemination algorithms analyze the data
from other nodes: some simply fetch as much data as they can, until their cache is
full, like Epidemic routing [2], while others only verify if they do not already con-
tain the data or if they have not contained it recently. More advanced content
analysis can be accomplished by assigning priorities (or utilities) to each content
object from a neighboring node. In this way, considering the amount of free cache
memory, a node can decide what and how many content objects it can fetch from
another node. A node can also calculate the priority for its own content objects,
and advertise only the priorities. Thus, a neighboring node can choose the data
that maximizes the local priority of its cache. One method of computing priorities
is based on heuristics that compare two content objects. Heuristics can compare
content objects by their age, by their hop count or by the number of subscribers to
the channel the object belongs to. A more complex approach to computing the
value of priorities is to use a mathematical formula that assigns weights to various
parameters. This method is used especially in socially-aware dissemination algo-
rithms, where users are split in communities, and each community is assigned an
individual weight (more about socially-aware algorithms will be presented in the
next paragraph).
The final category of the proposed taxonomy is the social awareness. Recently,
the social aspect of opportunistic networking has been studied, because the actual
nodes in an opportunistic network are represented by humans. They are the carri-
ers of the mobile devices, so the human factor is an important dimension that must
be considered by data dissemination algorithms. When designing such an algo-
rithm, it is important to know that user movements are conditioned by social rela-
tionships. The first subcategory of social awareness is represented by socially-
unaware algorithms, which do not assume the existence of a social structure that
governs the movement or interaction of the nodes in an opportunistic network. Da-
ta dissemination techniques of this type may be as simple as spreading the content
to all encountered nodes, but they can also take advantage of non-social context
information such as geographical location.
Most of the recent data dissemination techniques that are aware of the social
aspect of an opportunistic network are community-based. Such dissemination al-
9
gorithms assume that users can be grouped into communities, based on strong so-
cial relationships between users. Even though there are several proposed represen-
tations of social behavior, the caveman model [22] is by far the one mostly used,
along with its variations. Users can belong to more communities (called "home"
communities), but can also have social relationships outside of their home com-
munities (in "acquainted" communities). Communities are usually bound to a geo-
graphical space (static social communities), but they may also be represented by a
group of people who happen to be in the same place at the same time (e.g. at a
conference - temporal communities). According to this model, users spend their
time in the locations of their home communities, but also visit areas where ac-
quainted communities are located. As previously stated, a utility function may be
used to decide which content objects must be fetched when two nodes are in range
of each other. In a community-based approach, each community would be as-
signed a weight, and the utility of a data object would be computed according to
the community its owner comes from and the community of the (potentially) in-
terested nodes.
One step that has to be executed before designing a community-based dissemi-
nation algorithm is the community detection. There are several methods used for
organizing nodes from an opportunistic network into communities. One way is to
simply classify nodes based on the number of contacts and contact duration of a
node pair according to a threshold value, while another approach would be to de-
fine k-CLIQUE communities as unions of all k-CLIQUEs that can be reached
from each other through a series of adjacent k-CLIQUEs [17].
The phase following the detection of existing communities is the design of a
community structure. All nodes in a community can be identical (from the per-
spective of behavior), but there are also situations where certain nodes are more
important in the dissemination scheme. As previously described, some data dis-
semination algorithms use network overlays constructed using hubs or brokers
(e.g. nodes with the highest centrality in a community). The advantage of such an
approach is that only nodes having a high centrality transfer messages to other
communities. When a node wants to send a content object, it transfers it to the hub
(or to a node with a higher centrality, which has a better chance of reaching the
hub). The hub then transfers the object to the hub of the destination's community,
where it eventually reaches the desired destination. The structure of a community
has a high relevance in classifying data dissemination techniques, because a well-
structured community can speed up the dissemination process significantly.
5 Critical Analysis of Dissemination Algorithms
In this Section we analyze the properties of four techniques for disseminating data
in an opportunistic network, using the categories of the proposed taxonomy. The
presented study evaluates the most relevant recent work in data dissemination al-
gorithms. We also apply the proposed taxonomy to analyze and differentiate be-
tween the presented data dissemination techniques.
10
Authors of [17] present a publish/subscribe data dissemination solution that us-
es a Socio-Aware Overlay created on top of user-centric detected communities.
The second data dissemination solution [18], [19] uses a Wireless Ad Hoc Pod-
casting system based on opportunistic networks. The next method is called the
DTN Pub/Sub Protocol (DPSP) and is an efficient publish/subscribe-based mul-
ticast distribution method for opportunistic networks [20]. The final analyzed sys-
tem for data dissemination is ContentPlace [21]. It is a system that exploits dy-
namically learned information about users' social relationships to decide where to
place data objects in order to optimize content availability.
5.1 Socio-Aware Overlay
The Socio-Aware Overlay algorithm [17] is a data dissemination technique that
creates an overlay for an opportunistic network with publish/subscribe communi-
cation, composed of nodes having high centrality values that have the best visibil-
ity in a community. The data dissemination technique assumes the existence of a
network with infrastructure, built by creating an overlay comprising of representa-
tive nodes from each community. The dissemination of subscriptions is done, to-
gether with the community detection, during the node interaction phase, through
gossiping. The gossiping dissemination sends each message to a random group of
nodes, so from a node state point of view, the Socio-Aware algorithm takes a hy-
brid approach. In order to choose an appropriate hub (or broker) in a network, the
algorithm uses a measurement unit called node centrality.
Node discovery is performed through Bluetooth and WiFi devices, while there
are two modes of node interaction, namely unicast and direct. The former is simi-
lar to Epidemic routing, while the latter provides a more direct communication
mechanism like WiFi access points. From the standpoint of content organization,
the Socio-Aware algorithm is based on a publish/subscribe approach. At the data
exchange phase, subscriptions and unsubscriptions with the destination of com-
munity broker nodes are exchanged, as well as a list of centrality values with a
time stamp. When a broker node changes upon calculation of its closeness central-
ity, the subscription list is transferred from the old one to the new one. Then, an
update is sent to all the brokers. During the gossiping stage, subscriptions are
propagated towards the community's broker. When a publication reaches the bro-
ker, it is propagated to all other brokers, and then the broker checks its own sub-
scription list. If there are members in its community that must receive the publica-
tion, the broker floods the community with the information.
The Socio-Aware algorithm is a socially-aware community-based algorithm,
that has its own community detection method. This method assumes a community
structure that is based on a classification of the nodes in an opportunistic network,
from the standpoint of another node. A first type of node is one from the same
community, having a high number of contacts of long/stable durations. Another
type of node is called a familiar stranger and has a high number of contacts with
the current node, but the contact durations are short. There are also stranger nodes,
11
where the contact duration is short and the number of contacts is low, and finally
friend nodes, with few contacts, but high contact durations.
In order to construct an overlay for publish/subscribe systems, community de-
tection is performed in a decentralized fashion, because opportunistic networks do
not have a fixed structure. Thus, each node must detect its own local community.
The authors propose two algorithms for distributed community detection, named
Simple and k-CLIQUE. In order to detect its own local community, a node inter-
acts with encountering devices and executes the detection algorithm. The detec-
tion algorithm is done in the data exchange phase of the interaction between
nodes. Each node accomplishes the content identification by maintaining infor-
mation about the encountered nodes and contact durations (represented as a map
called the familiar set) and the local community detected so far. When two nodes
meet, a data exchange is performed, with each node acquiring information about
the other's familiar set and local community. Each node then updates its local
community and familiar set values, according to the algorithm used. As more
nodes are encountered over time, the shape of the local community may be modi-
fied.
5.2 Wireless Ad Hoc Podcasting
The Wireless Ad Hoc Podcasting system [18], [19] extends podcasting to ad-hoc
domains. The purpose is the wireless ad-hoc delivery of content among mobile
nodes. Assuming a network without infrastructure, the wireless podcasting service
enables the distribution of content using opportunistic contacts whenever podcast-
ing devices are in wireless communication range. From the standpoint of content
organization, the Ad Hoc Podcasting service employs a publish/subscribe ap-
proach. Thus, it organizes content into channels, which allows the users to sub-
scribe and automatically receive updates for the content they are interested in.
However, the channels themselves are divided into episodes and enclosures. Fur-
thermore, enclosures are also divided into chunks, which are transport-level small
data units of a size that can typically be downloaded in an individual node encoun-
ter. The reason for this division is the need for improving efficiency in the case of
small duration contacts. The chunks can be downloaded opportunistically from
multiple peers, and they are further divided into pieces, which are the atomic
transport units of the network.
For node interaction, when two nodes are within communication range they as-
sociate and start soliciting episodes from the channels they are subscribed to.
Since data is not being pushed, the nodes have complete control over the content
they carry and forward. Node discovery is done by using broadcast beacons sent
periodically by each node. Content identification is performed to identify channels
and episodes at the remote peer that the current node is subscribed to. Two nodes
in range exchange a Bloom filter hash index that contains all channel IDs that each
node offers. Then each node checks the peer's hash index for channels it is sub-
scribed to. The data exchange phase begins if one of the nodes has found a match-
12
ing channel. In this case, it starts querying for episodes. In order to perform con-
tent analysis, the Wireless Ad Hoc Podcasting system proposes three different
types of queries, employed according to the channel policy: a node requests any
random episodes that a remote peer offers, a node requests episodes from the peer
that are newer than a given date starting with the newest episode, or a node re-
quests any episodes that are newer than a given date starting with the oldest epi-
sode.
When two nodes meet, and neither has content from a channel the other is sub-
scribed to, several solicitation strategies are employed [18]. They are used to in-
crease the probability of a node having content to share with other nodes in future
encounters. The solicitation strategies proposed are Most Solicited, Least Solicit-
ed, Uniform, Inverse Proportional and No Caching. The Most Solicited strategy
fetches entries from feeds that are the most popular. The Least Solicited strategy
does the opposite, by favoring less popular feeds. The Uniform strategy treats all
channels equally, by soliciting entries in a random fashion, and has the advantage
of being easy to implement. The Inverse Proportional strategy maintains a history
list and solicits a feed with a probability which is inverse proportional to its popu-
larity. Finally, No Caching is more of a benchmark for other strategies than a
strategy itself, and assumes that a device has no public cache at all and that it
stores or distributes only content from the fields it is subscribed to. Experiments
show that the Uniform strategy has the best overall performance, while Inverse
Proportional is the best one in regards to fairness.
5.3 DPSP
Authors of [20] propose a probabilistic publish/subscribe-based multicast distribu-
tion infrastructure for opportunistic networks based on DTN (Delay Tolerant Net-
working). The protocol uses a push-based asynchronous distribution delivery
model. The idea is that nodes in the opportunistic network replicate bundles to
their neighbors in order to get the bundle delivered by multiple hops of store-
carry-and-forward.
As its name states, DPSP has a content organization based on a channel sub-
scription system, where users subscribe to channels and senders publish content.
Although from the network organization standpoint, DPSP assumes no infrastruc-
ture, the nodes in the network are divided into three categories: sources, sinks and
other nodes. Sources are the nodes that send content (in the form of bundles of da-
ta) to channels, while sinks subscribe to channels and receive information from
them. The rest of the nodes are not interested in specific bundles, but they store,
carry and forward bundles and subscriptions.
The node interaction phase has several steps. When two nodes meet, content
identification is performed through the exchange of subscription lists. An entry in
a subscription list contains the channel's URI, the subscription's creation time, its
lifetime, the number of hops from the original subscriber to the current node, and
an identifier for the subscription. Then, each node builds a queue of bundles to
13
forward to the peer, and uses a set of filters to select the best. The selected bundles
are subsequently sorted according to their priorities, and the data exchange stage
is performed by sending the bundles one by one until the contact finishes or the
queue becomes empty.
In this approach, a set of filters is used in order to select the best bundles in a
queue. Because the DPSP protocol is socially-unaware, the filters used do not
consider the organization of users into communities. There are three filters that
handle the content analysis and that can be used in any combination: Known Sub-
scription Filter, Hop Count Filter and Duplicate Filter. The Known Subscription
Filter removes bundles nobody is interested in, the Hop Count Filter removes
bundles that are too old, while the Duplicate Filter removes bundles that the peer
has already received. Content analysis is also performed when the remaining bun-
dles from a queue are sorted according to their priorities. Four heuristics are used
to assign priorities to bundles: Short Delay, Long Delay, Subscription Hop Count
and Popularity. Short Delay prefers newer bundles, Long Delay prefers older bun-
dles, Subscription Hop Count sorts bundles according to hop count, and the Popu-
larity heuristic sorts bundles by the number of nodes subscribed to the bundle's
channel. The authors noticed that the Short Delay heuristic performs better with
respect to delivery rates than the other heuristics.
5.4 ContentPlace
ContentPlace [21] deals with data dissemination in resource-constrained opportun-
istic networks, by making content available in regions where interested users are
present, without overusing available resources. To optimize content availability,
Content Place exploits learned information about users' social relationships to de-
cide where to place user data. The design of ContentPlace is based on two as-
sumptions: users can be grouped together logically, according to the type of con-
tent they are interested in, and their movement is driven by social relationships.
For performance issues, ContentPlace assumes a network without infrastruc-
ture. When a node encounters another node it decides what information seen on
the other node should be replicated locally. When two nodes are in range, they
have to discover each other. The node discovery is not specified, but since the
nodes are mobile devices it is probably done by WiFi or Bluetooth periodic broad-
casts. For content identification, nodes advertise the set of channels the local user
is subscribed to upon encountering another node. ContentPlace defines a utility
function by means of which each node can associate a utility value to any data ob-
ject. When a node encounters another peer, it selects the set of data objects that
maximizes the local utility of its cache. Due to performance issues, when two
nodes meet, they do not advertise all information about their data objects, but in-
stead they exchange a summary of data objects in their caches. Finally, the data
exchange is accomplished when a user receives a data object it is subscribed to
when it is found in an encountered node's cache.
14
Content organization in ContentPlace is done through channels to which users
can subscribe. Consequently, unsubscription messages are not necessary, so a
stateless approach is used for the nodes. ContentPlace is a socially-aware, com-
munity-based data dissemination algorithm. To have a suitable representation of
users' social behavior, an approach that is similar to the caveman model [22] is
used, that has a community structure which assumes that users are grouped into
home communities, while at the same time having relationships in acquainted
communities. For content analysis nodes compute a utility value for each data ob-
ject. The utility is a weighted sum of one component for each community its user
has relationships with. The utility component of a data object for a community is
the product of the object's access probability from the community members, by its
cost (which is a function of the object's availability in the community), divided by
the object's size. Community detection, like at the Socio-Aware Overlay, uses the
algorithms described in [23].
By using weights based on the social aspect of opportunistic networking, Con-
tentPlace offers the possibility of defining different policies. There are five poli-
cies defined: Most Frequently Visited (MFV), Most Likely Next (MLN), Future
(F), Present (P) and Uniform Social (US). MFV favors communities a user is most
likely to get in touch with, while MLN favors communities a user will visit next. F
is a combination between MLN and MFV, as it considers all the communities the
user is in touch with. In the case of P, users do not favor other communities than
the one they are in, while at US all the communities the users get in touch with
have equal weights.
5.5 Analysis Results
This Section presents a critical analysis of the four described protocols, according
to the proposed taxonomy. The results of this analysis is presented in Figure 2.
Fig. 2. Critical analysis of four dissemination techniques.
According to our analysis of the four solutions, only one assumes that the net-
work over which data dissemination is performed has an infrastructure. The Socio-
Aware Overlay algorithm builds an overlay infrastructure using the nodes with the
15
highest centrality from each community. However, opportunistic networks gener-
ally contain nodes with a high degree of mobility, which make the task of creating
and maintaining an infrastructure very hard to accomplish. The reason for this is
that nodes may change communities very often (or they may not belong to a
community at all), thus complicating the community detection phase. Further-
more, a device that is considered to be the central node (or hub) of a community
may be turned off (due to different circumstances, like battery depletion), leaving
the nodes in the hub's community without an opportunity to send messages to oth-
er communities, until a new hub is elected. Given these reasons, we believe that an
approach that does not assume the existence of an infrastructure should be further
considered.
The characteristics of a node from an opportunistic network play an important
role in the structure of a data dissemination algorithm. Node characteristics refer
to the way a node's state is represented and the way nodes interact when they are
in contact. As stated in Section 4, the approach a data dissemination algorithm can
take in regard to node state can be either stateless, stateful, or hybrid. Of the pro-
tocols we analyzed, ContentPlace chooses a stateless approach, while the Socio-
Aware Overlay uses a hybrid representation of a node's state. The authors of the
other two algorithms do not specify the node state, but we assume a stateful ap-
proach, because of the way the content is represented (for example, DPSP main-
tains subscription lists, for which node state is required). According to [17], a hy-
brid approach is the preferred solution because it takes advantage of both stateful
and stateless approaches. Such an approach would not suffer under frequent topol-
ogy changes, while at the same it would not require a large amount of control traf-
fic.
The interaction between nodes has three steps that have been presented in detail
in Section 4: node discovery, content identification and data exchange. Node dis-
covery is usually done in the same way for all algorithms analyzed, but it may dif-
fer according to the type of devices that are present in the network. In case of the
Socio-Aware Overlay and ContentPlace, the discovery is performed by using the
Bluetooth or WiFi capabilities. The Ad Hoc Podcasting algorithm uses broadcast
beacons, while the authors of DPSP do not mention a particular discovery method.
It is a good approach to use the existing capabilities from the wireless protocols,
but a data dissemination algorithm should try to extend the battery's life as much
as possible. For example, when the battery is low, the broadcast beacons should be
sent at larger time intervals.
Content identification, meaning the way in which nodes represent the data in-
ternally, also has a big impact in the efficiency of a data dissemination technique.
The Socio-Aware Overlay maintains information about the encountered nodes and
the duration of contacts, Ad Hoc Podcasting uses a Bloom filter hash index that
contains all channel IDs, DPSP exchanges subscription lists and ContentPlace ad-
vertises the set of channels a node is subscribed to. The most efficient method is
using Bloom filters, because they are space efficient data structures of fixed size
that avoid unnecessary transmissions of data that the receiver has already received
[24].
16
Data exchange should also be performed in a manner that optimizes the dura-
tion of a transfer. The nodes from the Socio-Aware Overlay exchange subscrip-
tions and lists of centrality values, Ad Hoc Podcasting exchanges episodes or
chunks, DPSP uses bundles and ContentPlace nodes exchange data objects. The
smaller the data unit, the bigger is the chance of a transmission to successfully fin-
ish, even in opportunistic networks where contact durations are very small. There-
fore, one of the best approaches is the one employed by Ad Hoc Podcasting,
where data is split into episodes and chunks.
The type of content organization that best suits opportunistic networks is the
publish/subscribe pattern. The reason for this is that participants are decoupled
from time, space and flow. Interested users simply subscribe to certain channels
and receive data whenever the publishers post it. Publishers and subscribers do not
have to be online at the same time, and it is not necessary that a direct connection
exists between them. Consequently, all the analyzed data dissemination techniques
organize their content according to a publish/subscribe approach. Content can also
be analyzed in order for a node to decide what to download from an encountered
peer. The Ad Hoc Podcasting technique uses five solicitation strategies that aim to
increase the probability of a node having content to share with other nodes. DPSP
has three filters used to select the best bundles in a queue and four heuristics that
sort the remaining bundles. Finally, ContentPlace computes a utility function
based on every community a node is in relationship with. The ContentPlace ap-
proach performs the best, because it takes advantage of the social aspect of oppor-
tunistic networking.
According to [15], human social structures are at the core of opportunistic net-
working. This is because humans carry the mobile devices, and it is the human
mobility that generates communication opportunities when two or more devices
come into contact. Social-based forwarding and dissemination algorithms reduce
by about an order of magnitude the overhead, compared to algorithms such as Ep-
idemic routing. Therefore, the social aspect has a very important role in the effi-
ciency of a data dissemination technique in an opportunistic network. Social
awareness is based on the division of users into communities, which are defined as
groups of interacting individuals organized around common values within a shared
geographical location. Thus, an important step for socially-aware dissemination
algorithms is community detection. Of the techniques we studied, only the Socio-
Aware Overlay proposes its own community detection algorithms, called Simple
and k-CLIQUE. ContentPlace uses similar algorithms, while Ad Hoc Podcasting
and DPSP are socially-unaware. As far as community structure goes, the Socio-
Aware Overlay splits the nodes in a community from the standpoint of another
node, according to the contact duration and number of contacts, while Content-
Place adopts a model similar to the caveman model. We consider that the future of
data dissemination algorithms should be based on a socially-aware approach to
take advantage of the human aspect of opportunistic networking.
After analyzing the four data dissemination techniques, we can conclude that
there is no single best approach, but each algorithm provides certain aspects that
offer advantages over the other implementations. In the next phase we plan to ex-
17
tend this work and propose a dissemination algorithm that uses the advantages of
all analyzed solutions for maximum efficiency.
6 Conclusions and Future Work
In this article, we analyzed existing relevant work in the area of data dissemination
in opportunistic networks. We began by highlighting the use of ONs in real life
and describing some potential scenarios where they can be applied. Then, we pre-
sented the categories of a proposed taxonomy that captures the capabilities of data
dissemination techniques used in opportunistic networks. Moreover, we critically
analyzed four relevant data dissemination techniques using the proposed taxono-
my. The purpose of the taxonomy, aside from classifying dissemination methods,
has been to analyze and compare the strengths and weaknesses of the analyzed da-
ta dissemination algorithms. Using this knowledge, we believe that an efficient da-
ta dissemination technique for opportunistic networks can be devised. We believe
that the future of opportunistic networking lies in the social property of mobile
networks, so a great deal of importance should be given to this aspect.
In the future, we aim to propose and implement an opportunistic mobile wire-
less solution for communication based on the conclusions of our analysis. Such a
solution can be used together with a context-aware platform for developing appli-
cations designed for mobile devices, with a focus towards recommendation and in-
formation of events towards users (such as the dissemination of academic events
to all academic members). Such a solution might help in disseminating data be-
tween users having similar interests, even without the presence of dedicated wired
access points and with lower costs than long-range mobile telecommunication pro-
tocols such as 3G or WiMAX. We believe that such a solution should be socially-
aware, splitting nodes into communities (such as teachers, students, or students
from the same group). An infrastructure may also be considered, built from nodes
that are in contact with many communities (such as teachers or teaching assis-
tants). Moreover, content should be exchanged between nodes based on the device
owner's preferences, using context-aware data.
Acknowledgement
This work was partially supported by project “ERRIC -Empowering Romanian
Research on Intelligent Information Technologies/FP7-REGPOT-2010-1”, ID:
264207.
18
References
1. Pelusi, L., Passarella, A., Conti, M.: Opportunistic networking: data forwarding in dis-
connected mobile ad hoc networks. IEEE Communications Magazine, 44, pp. 134-141
(2006)
2. Vahdat, A., Becker, D.: Epidemic routing for partially-connected ad hoc networks.
Technical report, Duke University (2000)
3. Hui, P., Crowcroft, J., Yoneki, E.: BUBBLE Rap: Social-Based Forwarding in Delay-
Tolerant Networks. In: Mobile Computing, IEEE Transactions on, vol. 10, no. 11, pp.
1576-1589 (2011)
4. Ciobanu, R.I., Dobre. C.: Predicting encounters in opportunistic networks. In: Proceed-
ings of the 1st ACM workshop on High performance mobile opportunistic systems (HP-
MOSys '12), pp. 9-14. ACM, New York, NY, USA (2012)
5. Lilien, L., Gupta, A., Yang, Z.: Opportunistic Networks for Emergency Applications
and Their Standard Implementation Framework, In: Performance, Computing, and
Communications Conference, IPCCC 2007, pp. 588-593. IEEE International (2007)
6. Thilakarathna, K., Petander, H., Mestre, J., Seneviratne, A.: Enabling mobile distributed
social networking on smartphones. In: Proc. of the 15th ACM international conference
on Modeling, analysis and simulation of wireless and mobile systems (MSWiM’12), pp.
357-366, Paphos, Cyprus (2012).
7. Desta, M.S., Hyytiä, E., Ott, J., Kangasharju, J.: Characterizing Content Sharing Proper-
ties for Mobile Users in Open City Squares. In: 10th Annual IE0EE/IFIP Conference on
Wireless On-Demand Network Systems and Services (WONS). Banff, Alberta, Canada
(2013)
8. Dobre, C., Manea, F., Cristea, V.: CAPIM: A context-aware platform using integrated
mobile services. In: International Conference on Intelligent Computer Communication
and Processing (ICCP), pp. 533-540. IEEE (2011)
9. Chourabi, H., Nam, T., Walker, S., Gil-Garcia, J.R., Mellouli, S., Nahon, K., Pardo,
T.A., Scholl, H.J.: Understanding Smart Cities: An Integrative Framework. In: Proceed-
ings of the 2012 45th Hawaii International Conference on System Sciences (HICSS '12),
pp. 2289-2297. IEEE Computer Society, Washington, DC, USA (2012)
10. Haddadi, H., Hui, P., Henderson, T., Brown, I.: Targeted Advertising on the Handset:
Privacy and Security Challenges. Human-Computer Interaction Series. Springer (2011)
11. Conti, M., Crowcroft, J., Giordano, S., Hui, P., Nguyen, H.A., Andrea, P.: Routing is-
sues in opportunistic networks. Middleware for Network Eccentric and Mobile Applica-
tions, pp. 121-147 (2009)
12. Nguyen, H.A., Giordano, S., Puiatti, A.: Probabilistic Routing Protocol for Intermittent-
ly Connected Mobile Ad hoc Network (PROPICMAN). In: World of Wireless, Mobile
and Multimedia Networks, 2007 (WoWMoM 2007), pp. 1-6 (2007)
13. Boldrini, C., Conti, M., Jacopini, J., Passarella, A.: HiBOp: a History Based Routing
Protocol for Opportunistic Networks. In: World of Wireless, Mobile and Multimedia
Networks, 2007 (WoWMoM 2007), pp. 1-12 (2007)
14. Moreira, W., Mendes, P., Sargento, S.: Opportunistic routing based on daily routines. In:
World of Wireless, Mobile and Multimedia Networks (WoWMoM), pp. 1-6 (2012)
15. Conti, M., Giordano, S., May, M., Passarella, A.: From opportunistic networks to oppor-
tunistic computing. IEEE Communications Magazine, 48, pp. 126-139 (2010)
16. Su, J., Scott, J., Hui, P., Crowcroft, J., De Lara, E., Diot, C., Goel, A., Lim, M.H., Up-
ton, E.: Haggle: seamless networking for mobile applications. In: Proceedings of the 9th
international conference on Ubiquitous computing, UbiComp ’07, pp. 391-408. Spring-
er-Verlag, Berlin, Heidelberg (2007)
19
17. Yoneki, E., Hui, P., Chan, S., Crowcroft, J.: A socio-aware overlay for publish/subscribe
communication in delay tolerant networks. In: Proceedings of the 10th ACM Symposi-
um on Modeling, analysis, and simulation of wireless and mobile systems, MSWiM '07,
pp. 225-234. ACM, New York, NY, USA (2007)
18. Lenders, V., Karlsson, G., May, M.: Wireless ad hoc podcasting. In: 4th Annual IEEE
Communications Society Conference on Sensor Mesh and Ad Hoc Communications and
Networks, pp. 273-283 (2007)
19. Lenders, V., May, M., Karlsson, G., Wacha, C. Wireless ad hoc podcasting.
SIGMOBILE Mob. Comput. Commun. Rev., 12, pp. 65-67 (2008)
20. Greifenberg, J., Kutscher, D.: Efficient publish/subscribe-based multicast for opportun-
istic networking with self-organized resource utilization. In: Proceedings of the 22nd In-
ternational Conference on Advanced Information Networking and Applications - Work-
shops, pp. 1708-1714. IEEE Computer Society, Washington, DC, USA (2008)
21. Boldrini, C., Conti, M., Passarella, A.: Design and performance evaluation of Content-
Place, a social-aware data dissemination system for opportunistic networks. Computer
Networks, 54, pp. 589-604 (2010)
22.Wu, J., Watts, D.J.: Small worlds: the dynamics of networks between order and ran-
domness. SIGMOD Rec., 31, pp. 74-75 (2002)
23. Hui, P., Yoneki, E., Chan, S.Y., Crowcroft, J.: Distributed community detection in delay
tolerant networks. In: Proceedings of 2nd ACM/IEEE international workshop on Mo-
bility in the evolving internet architecture, MobiArch '07, pp. 71-78. ACM, New York,
NY, USA (2007)
24. Bjurefors, F., Gunningberg, P., Nordstrom, E., Rohner, C.: Interest dissemination in a
searchable data-centric opportunistic network. In: Proc. European Wireless Conference,
EW 2010, pp. 889-895. IEEE, Piscataway, NJ, USA (2010)
