Social Network Analysis for Information Flow in Disconnected Delay-Tolerant MANETs by Elizabeth M.Daly and Mads Haahr

1
(No Transcript)
2
Infrastructure of MANETs

• MANETS are without a fixed infrastructure
• Network Graphs in MANETS are rarely or ever
  connected
• MANET routing protocols such as AODV,DSR,DSDV and
  LAR make the assumption that network graph is
  fully connected.

3
Workarounds?

• Exploit the mobility of the network to transfer
  messages.
• Mobility assisted routing that assist the
  store-and-forward model.
• BUT NODES DO NOT HAVE AN EQUAL PROBABILITY OF
  ENCOUNTERING EACH OTHER.

4
Problems

• Not all nodes are equally likely to encounter
  each other.
• Nodes need to assess the probability of
  encountering the destination node.
• Social Networking to the rescue..

5
Social Networking

• Social Networking comes from the observation that
  individuals are linked by a small chain of
  acquaintances.
• Milgrams 1967 experiment and 6 degrees of
  separation.
• Hsu and Helmys research promotes the idea that
  Social Network analysis techniques can be applied
  to disconnected MANETs.

6
Social Networking in MANETs

• In Millgrams experiment, the carriers of
  messages were not randomly selected. The
  participant sent the letter to the person who
  might be perceived as the best carrier.
• Intelligent selection of good carriers based on
  the local information needs to be explored.
• This paper proposes the use of social network
  analysis techniques in order to exploit the
  underlying social structure in order to provide
  information flow from source to destination in a
  DDTM.

7
Related Work

• Data Mule Project
• Message ferrying project
• Time dependent network graphs
• Epidemic Routing
• PRoPHET Routing
• Usage of KALMAN filters in selectively forwarding
  messages.

8
SimBetTS Routing Metric

• SimBetTS Routing metric as proposed in this paper
  is comprised of both a nodes centrality and its
  social similarity.
• SimBetTS Routing improves upon encounter-based
  strategies where direct or indirect encounters
  between source and destination may not be
  available.

9
Social Networks For Information Flow

• Social network analysis is the study of
  relationships between entities and on the
  patterns and implications of these relationships.
  
• Graphs may be used to represent the relational
  structure of social networks in a natural manner.
  
• Each of the nodes may be represented by a vertex
  of a graph.
• Relationships between nodes may be represented as
  edges of the graph.

10
Centrality in Graph Theory

• It is the quantification of the relative
  importance of a vertex within the graph.
• Central node has a stronger capability of
  connecting other network members.
• Three widely used centrality measures are
  Freemans degree, closeness, and betweenness
  measures

11
Centrality Measures

• Freemans degree
• - It is the number of direct ties that involve a
  given node.
• - Segregates the popular from the unpopular
  nodes.
• - Degree centrality (for a node pi) is
  calculated as follows

12
Centrality Measures.

• Closeness
• - Measures the reciprocal of the mean geodesic
  distance, which is the shortest path between the
  node and all other reachable nodes.
• - Closeness centrality can be measured by

13
Centrality Measures.

• Betweenness
• - This centrality measures the extent to which a
  node lies on the geodesic paths linking other
  nodes.
• - It can be regarded as how much a node can
  facilitate communication to other nodes in the
  network.
• - It can be calculated as follows

14
Ego Networks

• Freemans centrality metrics become difficult to
  evaluate in a network with large node
  populations.
• This is the reason for the proposal of ego
  networks..
• Ego networks can be defined as a network
  consisting of a single actor (ego) together with
  the actors they are connected to (alters) and all
  the links among those alters.
• Ego network analysis can be performed locally by
  individual nodes without complete knowledge of
  the entire network.

15
Ego Networks vs. Sociocentric Networks

• Degree centrality can easily be measured for an
  ego network where it is a simple count of the
  number of contacts.
• Closeness centrality is uninformative in an ego
  network, since by definition an ego network only
  considers nodes directly related to the ego node.
• Betweenness centrality in ego networks has shown
  to be quite a good measure when compared to that
  of the sociocentric measure.

16
Betweenness Centrality in Ego-Networks
In the figure below, the rankings of the
betweenness are identical for Sociocentric and
Egocentric.
17
Drawbacks of routing based on Centrality

• Central nodes suffer serious traffic congestion.
• Centrality does not take into account the time
  varying nature of the links and the availibility
  of the link in the network

18
Strong Ties for Information Flow

• Tie strength is a quantifiable property that
  characterizes the link between two nodes.
• Brown and Reingen found that strong ties were
  more likely to be activated for information flow
  when compared to weak ties.
• Tie strength factors in the time varying nature
  of the links.

19
Tie-Strength Indicators

• Frequency
• Intimacy/Closeness
• Long Period of time (Longetivity)
• Reciprocity
• Recency
• Multiple Social Contexts
• Mutual-Confiding Trust

20
Usage of Tie-Strength Indicators

• Tie strength indicators can be used for
  information flow to determine which contact has
  the strongest social relationship to a
  destination.
• Messages can be forwarded through links
  possessing the strongest relationship, link
  representing a strong relationship will be more
  likely to be activated for information flow.

21
Utility of Weak ties

• Granovetter emphasized that weak ties lead to
  information dissemination between groups.
• Weak ties can act as bridges
• Need is there to find routing based on a
  combination of strong ties and identified bridges.

22
Tie Predictors

• Tie strength evaluates already existing
  connections whereas predictors use information
  from the past to predict likely future
  connections.
• A network is said to show clustering if the
  probability of two nodes being connected by a
  link is higher when the nodes in question have a
  common neighbor.

23
Common Neighbor Metrics

Here score is the future collaboration between
nodes x and y. N(x) and N(y) are neighbors of x
and y respectively. P(x,y) weighs the rarer
neighbors more heavily than common neighbors.
24
Usage of Tie-Predictors

• Tie predictors can be used in order to predict
  the evolution of the graph and evaluate the
  probability of future links occurring.
• Tie predictors may be used not only to reinforce
  already existing contacts but to anticipate
  contacts that may evolve over time.

25
Routing based on Social Metrics

• In this paper it is proposed that the combination
  of centrality, tie strengths, and tie predictors
  are highly useful in routing based on local
  information when the underlying network exhibits
  a social structure.
• The combined metric is known as SimBetTS utility.

26
Sim-Bet-TS utility

• Sim(Similarity)Bet(Betweenness)TS(Tie-Strength)
• When two nodes meet, they exchange a list of
  encountered nodes.
• this list is used to locally calculate the
  betweenness utility, the similarity utility, and
  the tie strength utility.
• Each node then examines the messages it is
  carrying and computes the SimBetTS utility of
  each message destination.
• Messages are then exchanged where the message is
  forwarded to the node holding the highest
  SimBetTS utility for the message destination node.

27
Betweenness in SimBetTS
The betweenness centrality is calculated by
computing the number of nodes that are indirectly
connected through the ego node.

Betweenness in SimBetTS relies on the Adjacency
Matrix representation ofthe ego-nodes and its
neighbors.
28
Similarity in SimBetTS

• The adjacency matrix allows for the calculation
  of similarity for nodes that have been met
  directly, but during the exchange of the nodes
  contact list, information can be obtained with
  regard to nodes that have yet to be encountered.

In the above similarity calculation Nn and Ne are
the set of contactsheld by node n and e,
respectively.
29
(No Transcript)
30
Node-Utility in SimBetTS

• The aim is to select the node that provides the
  maximum utility for carrying the message. This is
  achieved using a pairwise comparison matrix on
  the normalized relative weights of the attributes
• The attributes here are the Similarity,
  Betweenness and Tie-Strength

31
Node-Utility in SimBetTS

Here U is TSUtil,SimUtil,BetUtil
32
(No Transcript)
33
Performance of SimBetTS

• SimBetTS Routing achieves the highest delivery
  performance by combining the three metrics. The
  load distribution shows that routing based on the
  combined metrics reduces congestion on highly
  central nodes.

34
Performance of SimBetTS
35
SimBetTS Routing Protocol Behavior

• SimilarityTie-Strength The sending node has a
  nonzero tie strength value for the destination
  node and has a nonzero similarity to the
  destination node

36
SimBetTS Routing Protocol Behavior

• Similarity The sending node has never
  encountered the destination node resulting in a
  zero tie strength.

37
SimBetTS Routing Protocol Behavior

• None The sending node has never encountered the
  destination node and has no common neighbors,
  hence a zero similarity value.

38
Conclusions from the Routing Protocol Behavior

• The message is forwarded to nodes with a high
  betweenness and social similarity, until a node
  with a high tie strength for the destination node
  is found.
• In all cases, the tie strength utility for the
  final hop is the highest contributing utility
  value.
• In all cases, the betweenness utility value is
  much reduced in its influence of the forwarding
  decision as the message is routed closer to the
  destination

39
SimBetTS vs. Epidemic/PRoPHET

• Single-copy and multicopy SimBetTS Routing show
  higher message delivery when compared to that of
  PRoPHET.
• Multicopy SimBetTS Routing achieves delivery
  performance similar to that of Epidemic Routing
  with short path lengths and low end-to-end delay.
  
• The use of replication comes at the cost of an
  increased number of forwards and an increase in
  control data. However, when compared to that of
  Epidemic Routing, these metrics are still
  relatively low in terms of overhead

40
Questions???