CHAPTER 1 Introduction
1.4 Organization
Figure 2: Centrality relationship 1.2.4 Similarity
Similarity is a measurement of the degree of separation. The number of common neighbors between people in social networks can calculate it. In a social relationship, there is a higher probability of two people connected when they have a common neighbor. In a network, the probability of two nodes being connected by a link is higher when they have a common neighbor. When the neighbors of nodes are unlikely to be in contact with each other, diffusion can be expected to take longer than when the similarity is high. Many researches are proposed by community detection algorithms [25]–[27] are available for identifying social similarity from the contact graph of DTNs.
1.2.5 Friendship and Selfishness
Friendship is concept that describes close personal relationships. In social behaviors, it has been shown that people often contact others who have the same interests and actions.
Therefore, the friendship in DTNs can be roughly determined by using either contact history between two nodes or common interests between two nodes. Selfishness can measure the selfish behaviors of nodes controlled by rational entities. A selfish node may drop other’s messages and excessively replicate its own messages to increase its own delivery rate. In DTNs, selfishness can describe the selfish behaviors of DTN nodes controlled by rational entities [28][29].
1.3 Our Goal
The interaction between people through social behaviors and message transmissions rely on geographic position information. We design a routing protocol integrating the concept of geographic routing and social-based routing in DTNs.
1.4 Organization
The rest of this thesis is organized as follows. Chapter 2 reviews related work about
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
social-based and geographic routing protocol. Our algorithm in detail is described in Chapter 3.
Chapter 4 discusses performance evaluation results. Chapter 5 in final summarizes our works and discusses future work.
‧
Message-ferry-based routing protocols, Opportunity-based routing protocols, Prediction-based routing protocols and Social-based routing protocols. We will briefly describe the feature used by these protocols to make routing decision and conclude their strengths and weaknesses.2.1 Message-ferry-based routing Protocol
In message-ferry-based [20][21] methods, systems usually employ extra mobile nodes as ferries for message delivery. The trajectory of these ferries is controlled to improve delivery performance with store-and-carry. However, controlling these nodes leads to extra cost and overhead.
2.2 Opportunity-based Protocol
In this category[3][4], it uses redundancy messages to make the delivery success. In order to make sure that there will at least one copy of message reach the destination, this kind of protocol is using a great number of redundancy copies. The advantage of flooding-based routing protocol is easy to implement and needs no information of the networks. On the other side, it needs a bigger buffer to cache more copies to get a good delivery ratio.
2.2.1 Epidemic Routing Protocol
In this routing protocol[3], there is a unique identifier in data message. When nodes encounter one-hop neighbor, they would exchange all the identifier number in its buffer. If there is any identity it doesn’t have, it will ask for transmitting these messages from the contact node. With this rule, nodes do not need any information about the networks and all of them will get all messages if resources are unlimited. When resources are infinite, Epidemic routing will get the minimal delay latency and maximal delivered message counts. Epidemic is
‧
also as a benchmark to evaluate the performance in DTNs.
2.2.2 Direct Contact Routing Protocol
In this protocol, nodes will not transmit any message until its radio range covers the destination. Since it transmits nothing but the last one to goal, it needs the minimal resource of a node. On the other side, it might take the longest delay latency because of only one transmission node for a message. Between mobile nodes and fixed gateways, direct contact has been purposed that will increase throughput and decrease resource usages [5].
2.3 Prediction-based Routing Protocol
In this routing category, it is using the information about networks to make the best relay selection. Network topology is the major property they are used. They collect or calculate information like location information and historical contact recorders to choose the best relay node from encounter nodes. The advantage is that it reduces the consumption of resource. But the information is difficult to obtain complete topology information.
2.3.1 PROPHET Routing Protocol
The gradient routing is using a weighted value to recognize if the contact node is a good relay to the message. Each node has a metric table about all of possible destinations. Messages are only transmitted when the contact node has higher probability to the given destination of the message. This kind of protocol needs more information about networks than a location-based one and has to maintain the metric table of each node for probability computing.
Lindgren et al. propose a probability routing protocol called PROPHET [7] using history of encounters and transitivity information.
2.4 Social-based routing protocol
Recent studies [9] have shown that the social relations of the carrying users will directly affect the contact opportunities between the devices (i.e., nodes). Those people’s social relations are less volatile than mobility, and therefore can be used for better forwarding
‧
attempted to apply social network analysis techniques to uncovering these social relations and make forwarding decisions based on them.
2.4.1 SimBet Routing Protocol
In SimBet [10] routing uses social centrality and similarity to assess the utility of a node for forwarding. Messages are forwarded to the newly encountered node with higher utility value than the current one. It presents a multidisciplinary solution based on the consideration of the so- called small world dynamics which have been proposed for economy and social studies and have recently revealed to be a successful approach to be exploited for characterizing information propagation in wireless networks. To this purpose, some bridge nodes are identified based on their centrality characteristics, i.e., on their capability to broker information exchange among otherwise disconnected nodes. Due to the complexity of the centrality metrics in populated networks the concept of ego networks is exploited where nodes are not required to exchange information about the entire net- work topology, but only locally available information is considered.
Figure 3: The SimBet Routing 2.4.2 Bubble rap Routing Protocol
In Bubble Rap [11], the community detection techniques are explicitly used to identify the communities of tightly connected nodes. Messages are forwarded to the newly encountered node
C1 C2
u
a b c v
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
with higher centrality value until a node that belongs to the destination’s community is reached.
Then any node with higher local centrality within the community is selected for forwarding.
When a node s has a message with destination of d, it first bubbles the message up based on the global centrality, until the message reaches a node that is in the same local community Cd as the destination d. This procedure is shown as blue arrows in Figure 4. After the message reaches d’s community at node u, Bubble Rap Forwarding switches to the second phase that uses members of Cd as relays. This forwarding strategy continues to bubble up the message through the local community based on local centrality until the destination is reached.
Figure 4: The Bubble Rap Routing 2.4.3 SMART Routing Protocol
SMART (Secure multilayer credit-based incentive) scheme [25] [26] uses credits to provide incentives to selfish nodes. This scheme allows the credits to be transferred/distributed by the current intermediate nodes without the involvement of the sender. Such scheme adopts a novel layer concatenation technique to withstand cheating actions of selfish nodes. However, the
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
applications. Even two optimization techniques are proposed to improve the overall efficiency, their overheads are still considerable.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
CHAPTER 3 SAGE Routing
It is simple to explain the concept of social community into DTNs to explore interactions among wireless devices, due to wireless devices are usually carried by people. It is said that devices within the common friends or the same community have higher chances to encounter each other.
Therefore the community structure of friendship relation could help a routing protocol to choose better forwarding relays to improve the chance of delivery. Besides, we also consider the real geographic information to enhance our routing protocol.
In this section, we will introduce our design in detail. At first, we give an overview of the protocol and describe details in the subsections below. Consider both social behavior and node mobility, we purpose a social ally with geographic enhancement (SAGE) routing protocol.
‧
special mobile nodes called social allies by social properties. Then social ally receives message, it will search next level social allies help forward messages. Due to Six Degrees of Separation (Figure 5), a chain of “a friend of a friend” statements can be made to connect any two people in a maximum of six steps. We will search social allies level in a maximum of six levels.
In fact, forwarding message relies on geographic information. When two nodes encounter, the messages will be forwarded from one to another if the node moves toward to social ally or destination.
3.1 System Model
Figure 6: Two layers in SAGE routing
We design a Social Ally with Geographic Enhancement (SAGE) routing protocol. It divides network topology into two sub layers: social layer and geographic layer (in Figure 6).
In social layer, we design a Social Ally Select Algorithm to select a set of social allies. And then in geographic layer, we design a Geographic Messenger Forwarding Algorithm for
‧
forwarding decision.
We now describe how to transmit message to destination. Without loss of generality, consider node S, which intends to send message a message to node D. The steps below:
A. In Social Layer,
A.1 S looks up its node relation matrix and local activity matrix.
A.2 According to the result, S selects n nodes to become social allies.
A.3 S calculates social level that connects any two nodes.
A.4 S creates n+1 messages in message buffer for transmitting to social allies or D.
A.5 If social level less than 6, Go to geographic layer.
B. In Geographic Layer,
B.1 When S meets Node A, S compares the geographic probability that from S to D and the geographic probability that from A to D.
B.2 If the probability that from A to D is bigger than that from S to D, it means that A is a good receiver for relay the message.
B.3 S transmits message to node A.
B.4 When message transmits to social ally, then go to social layer.
The following section discusses the Social Ally Select Algorithm and Geographic Messenger Forwarding Algorithm in more detail.
3.2 Leverage Algorithm
We model a DTN as a set of mobile nodes in a pedestrian environment. A node is a pedestrian who carries a mobile device equipped with GPS and 802.11 communication modules.
People have known their friendship relationship, so we have known node relation matrix and local activity matrix in our research. We also assume every node belongs to at least one community.
‧
the GPS information of both its moving route and messages’ destinations to packet in beacon message for contact broadcast. . The destination of each message is within a fixed community.
3.2.1 Social Layer
Figure 7: Social Ally Select Algorithm
We propose a Social Ally Select Algorithm in social layer. We have known node relation matrix and local activity matrix. We record relationships among nodes in node relation matrix, such as friends, family members, colleague and fans, etc. We record node’s community activity in local activity matrix (Table 1). It is represented a node could move many
‧
observation is called homophile phenomenon. If Nodes with close relationship such as family members, colleagues, friends and fans, the SFND can be defined as
if(RelationOfNode(Node, Destination) == KINDRED) Friendship is 1. If nodes’ relation is colleague relation, we define the Social Friendship is 0.7.
There is the second highest score due to colleagues have the same region. If nodes’ relation is friend relation, we define the Social Friendship is 0.5. It is represented friends are usually long term social characteristics but the weak relation to geography. If nodes’ relation is fans relation, we define the Social Friendship is 0.3. It is represented fans are usually the weak social characteristics and the weak relation to geography. In addition to the above, we define the Social Friendship is 0.
Then, we define the Community Activity (CAND) to calculate probability that node meets destination at the same community. The CAND can be defined as:
if( NodeN.Community[i] == NodeD.Community[j] ) CAND =NodeN.Probability[i]*NodeD.Probability[j]
Finally, Social ally Ranking (SRND) represents the closer friendship between N and D at the same community. SRND can be defined as
‧
candidates to be social ally.CurrentNode.(Threshold N)new =
CurrentNode.(Threshold N)old * "#$% %&'#(%) * "#$% +#,#-.%)
"#$% %&'#(%)
Otherwise, the threshold is decreased if there are no enough nodes to be social allies and node chooses destination node to be social ally. It is considered some nodes are not popular in the community or social relation. They could not find enough friends to help them forward
CurrentNode.(Threshold ND)new =
CurrentNode.(Threshold ND)old * Y Time expired
end
In Social Ally Select Algorithm, we search a set of special mobile nodes called social allies by SRND. If SRND is bigger than the threshold, N could select to be social ally. We select n social allies and we add the social ally information to the message buffer. The algorithm below:
For each Node_i do begin
‧
if(Node_i.SRND > CurrentNode.Threshold ND &&
CurrentNode.count < Social_ally_count )
We create messages in message buffer for transmitting to social allies or destination. If node needs n social allies, we create n+1 message information in message buffer. We always add a message about the destination node is a social ally. Due to six degrees of separation, the social level in a maximum is 6. If the social level is bigger than 6, we stop to find social ally.
The message information is about serial number, source node, social level, message, social ally and destination node. For example, there are three nodes and one destination to be social allies in Table 2.
Serial Number Source Node
Social
Serial Number Source Node
Social
Serial Number Source Node
Social
Serial Number Source Node
Social
‧
the result, S selects n nodes to become social allies. S calculates social level that connects any two nodes. S creates n+1 messages in message buffer for transmitting to social allies or D. If social level less than 6, we execute Geographic Messenger Forwarding Algorithm in geographic Messenger Forwarding Algorithm.3.2.2 Geographic Layer
Figure 8: Geographic Messenger Forwarding Algorithm
In Geographic layer, we design a Geographic Messenger Forwarding Algorithm to select neighbors’ for forwarding messages. Without loss of generality, consider node A, which meets node B and decides to forward message, we calculate geographic probability (Pgeo) that can be defined as
Pgeo(B) = αCommunityActivity(B) * βDirection(B) * γVelocity(B)
Here, Direction (B) is the angle between Destination and B’s destination. As the Figure 9 shows below,
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 9: Message Relay Policy
8 ： The direction from node A to its message m(a)’s destination.
9 ： The direction from node B to its destination.
θ ： The angle between
8 and
9
A compares the geographic probability that from A to D and the geographic probability that from B to D. If the probability that from B to D is bigger than that from A to D, it means that B is a good receiver for relay the message. A transmits message to node B, until encounter social ally or D.
Consider the social level, the algorithm is below:
In Geographic layer, when S meets Node A, S compares the geographic probability
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
message. S transmits message to node A. When message transmits to social ally, then go to social layer to execute social ally selection algorithm.
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
CHAPTER 4
Simulation and Results
In this section, we would like to evaluate our SAGE routing, and contrast its performance against Epidemic routing, PROPHET routing, and two social-based routing, SimBet and Bubble Rap. Additionally, we also want to see how the routing parameters, i.e., social ally and geographic parameters, impact the performance of our SAGE routing. We will first introduce the setup in our simulations.
The major object of DTNs routing protocol is to maximize delivery ratio on a rapidly changing topology. By considering energy, nodes’ lifetime is also an important issue to compare. We evaluate our design in four metrics:
l Messages delivery ratio
l Overhead ratio = :;<=>;? @=AB;CD *E;<FG;H;? @=AB;CD E;<FG;H;? @=AB;CD
l End-to-end delay
l Routing efficiency = I;DD=J; ?;<FG;H> H=CFK LG;HM;=? H=CFK ∗ OP?*CK*;P? ?;<=>
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
4.1 Simulation Setup
We use a map-based model of a small part of Helsinki for The ONE simulator [24]
(version 1.4.1). All nodes are pedestrian and can only move on the roads of the map. When a node reaches its destination, it will randomly choose a new location of the map as its next destination and move to there by shortest path algorithm. Figure 10 is a snap shot of a simulation.
Figure 10: A snap shot of the ONE simulator
‧
4.1.1 General Settings of Simulator
Table 3: Parameters of simulation setting
The simulation time is 43200 seconds (12 hours). There is a warm up time for PROPHET set as 1000 seconds and all data will not count in evaluation during warm up period. The moving speed of nodes is 0.5 m/s to 1.5 m/s. The data transmission rate is 1 Mbps and the transmission range is 100m. The message size will randomly generate between 512 KB to 1 MB .Its generation interval is 5 seconds to 10 seconds and randomly chooses a node to own the message. The TTL is 5 hours. Each message randomly chooses a community of the
Map Size 4500m x 3400m
Simulation Time 43200 sec (12 hr) Number of Isolated Nodes 40, 80, 120
Number of Groups
4, 2, 1
Transmission Rate 1 Mbps
TX Radio Range 100 m
Message Size 512KB ~ 1MB (random) Interval of Msg. creation 5s ~ 10s (Total = 6190)
Buffer Size (Byte) 250 MB
Node Speed 0.5 m/s ~ 1.5 m/s
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
map as its destination before generated. The buffer size is 250 MB in each node. Table 3 is the parameters we described above.
4.1.2 Parameters about SAGE Routing Protocol
We evaluate our design in Figure 11. Each node moves cross 3 communities. There are 6 communities, In this section, we would like to know the social ally parameter and geographic probability parameter that impact the performance of our SAGE routing protocol.
Figure 11: Community setting
4.1.2.1 Social Ally
We assume our geographic parameter: α=1, β=0, γ=0. We consider Social ally
‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
parameter from 0 to 6. In Figure 12, they show the more social allies that the higher delivery ratio. It is represented more social allies that mean the more message copies for the different social allies. Also, we found that n=3+1 had the trend of ascent in delivery ratio.
Figure 12: Delivery Ratio
We observed the social ally parameter that impacts the delay latency. In Figure 13 shows that there is a marked drop in delay latency when the social allies are more than 3 (n=3+1). Also, we found that the more the number of nodes have the lower delay latency, due to consider more nodes have more social allies have been chosen quickly.
Delivery Ratio
0.5
0.3
‧
Figure 13: End-to-end delay latency
In Figure 14, they show the overhead Ratio of SAGE routing protocol;. The more social allies that the higher overhead ratio. It is represented more social allies that mean the more message copies for the different social allies. Also, we found that n=4+1 had the trend of ascent in overhead ratio.
Average Delay
‧
3 (n=3+1), there is better routing efficiency. Since n=3+1 had the trend of ascent in delivery ratio and there is a marked drop in delay latency, finally we selected Social ally parameter is equal to 3.‧ 國
立 政 治 大 學
‧
N a tio na
l C h engchi U ni ve rs it y
Figure 15: Routing Efficiency
4.1.2.2 Geographic parameter
We Selected Social ally parameter is equal to 3 by section 4.1.2.1 result. In this section,
We Selected Social ally parameter is equal to 3 by section 4.1.2.1 result. In this section,