基於社交行為對興趣導向訊息之耐延遲網路傳輸策略 - 政大學術集成

全文

(1)國⽴立政治⼤大學資訊科學系 Department of Computer Science National Chengchi University 碩⼠士論⽂文 Master’s Thesis. 學. ‧ 國. 立. 政治大. 基於社交⾏行為對興趣導向訊息之. Nat. sit. y. ‧. 耐延遲網路傳輸策略. n. er. io. An Interest-based Message Dissemination Approach with Social Behavior Consideration in DelayvTolerant Networks al i n Ch engchi U 研究⽣生：陳柏錡指導教授：蔡⼦子傑. 中華民國⼀一零五年七⽉月 July 2016.

(2) 2 . 基於社交⾏行為對興趣導向訊息之耐延遲網路傳輸策略 An Interest-based Message Dissemination Approach with Social Behavior Consideration in Delay Tolerant Networks 研究⽣生：陳柏錡 Student: Po-Chi Chen. 治 Tzu-Chieh Tsai 指導教授：蔡⼦子傑政Advisor: 大. 立. io. sit. y. Nat A Thesis. er. ‧ 國. 碩⼠士論⽂文. ‧. 資訊科學系. 學. 國⽴立政治⼤大學. n. a lDepartment of Computer v Science i Submitted to n C NationalhChengchi i U e n g c hUniversity in partial fulfillment of the Requirements for the degree of Master In Computer Science. 中華民國⼀一零五年七⽉月 July 2016 . 2 .

(3) 3 . 基於社交⾏行為對興趣導向訊息之耐延遲網路傳輸策略摘要耐延遲網路相較於⼀一般的3G、︑､4G或是Wi-Fi網路是相對受限的，只能利⽤用節點之間短暫的相遇時利⽤用Bluetooth以及Wi-Fi direct交流訊息，他不僅沒有穩定的連線能⼒力，頻寬也相對較⼩小。︒｡但耐延遲網路卻因是斷斷續續且可以傳送給現實中. 政治大相遇的⼈人，使他⼈人幫忙攜帶訊息進⽽而傳送⾄至⽬目的⽽而成優點。︒｡因現今社會的⼈人們⼤大立. ‧ 國. 學. 多有固定的⾏行經路線像平⽇日的上下班，亦或是學⽣生⼀一定有著固定的課表和交通路. 線。︒｡另外，有共同興趣的⼈人也會朝著特定的地標前進，像喜歡運動的⼈人會往運動. ‧. 中⼼心集中、︑､喜歡閱讀的⼈人會往圖書館借閱書籍、︑､喜歡⽂文藝的⼈人會賞閱展覽。︒｡這無. y. sit er. io. 屬性有興趣的⼈人們。︒｡. Nat. 形中讓我們可以利⽤用耐延遲網路的⽅方式傳送有特定屬性的廣告訊息給對這特定. al. n. v i n Ch 我們在此論⽂文中提出兩種在耐延遲網路上新的資料傳遞⽅方式，我們集合⼈人們 engchi U. 有各⾃自的喜好並以⼈人們有相同路徑及有著相同興趣的⼈人們會聚集的特性在⼈人群. 中傳送訊息，我們預先⽤用⼿手機蒐集歷史資料並統計⼈人群的移動模型再⽤用以驗證未來節點相遇的情況。︒｡最後，我們將本論⽂文⽅方法與其它資料傳送⽅方法⽐比較評估效能，模擬結果顯⽰示我們提出的傳送⽅方法有較優的傳送成功率與相對較低的資源耗費。︒｡. 關鍵字：耐延遲網路、︑､校園環境、︑､個⼈人資訊、︑､個⼈人興趣、︑､社群關係. 3 .

(4) 4 . An Interest-based Message Dissemination Approach with Social Behavior Consideration in Delay Tolerant Networks . Abstract. Compared with 3G, 4G and Wi-Fi, Delay-Tolerant Networking (DTN) can only have intermittent chance to transmit messages with Bluetooth or Wi-Fi direct. Without. 政治大. a clear end-to-end path and relatively lower bandwidth, routing a message in DTN to. 立. the destination is difficult. But in some particular case, it could be an advantage.. ‧ 國. 學. People around the world have their personal habit and it will be projected on their social life. Also, most people have their own routine to work or to school. Therefore. ‧. we use the social behavior as a foundation feature of our routing algorithm.. y. Nat. io. sit. We propose two new kinds of routing algorithms with our own trace file. On one. er. hand, birds of a feather flock together, so people who have similar interests tend to go. n. a. v. l Cwe combining the personal to the same places. In case of this, n i interests and the trace. hengchi U. file to different buildings where each node locates, we propose the building-based routing algorithm. On the other hand, we think people who have similar interests hang out together more often, so we use the social relationship as a feature and propose social-based routing algorithm. In the end, we compare our algorithms with Epidemic, MaxProp and PRoPHET routing algorithms. The result shows that our algorithms exceed the others in performance. Keywords. Delay-Tolerant Network, DTN, campus environment, personal information, personal interest, social relationship. . 4 .

(5) 5 . 致謝詞轉眼間在政⼤大的時間已經邁向了第七年，從⼤大學⼀一路讀到了研究所，直到現在終於進⼊入尾聲，這段時間在政⼤大資科系中逐漸的學習、︑､成⾧長，從⼀一個完全看不懂程式碼的⾼高中⽣生到能寫出⼀一篇完整的碩⼠士論⽂文也完成了許多的 Android 程式，這段過程真的很感謝許多⼈人的幫助，讓我在政⼤大的這段時間不⽌止學到了必要的知識，也學到了許多在課本上學習不到的⽣生活態度，更交到了很多好朋友。︒｡. 政治大這份論⽂文從發想到確定⽅方向，⼀一步⼀一步的⾛走來都感謝有指導教授蔡⼦子傑⽼老師細⼼心的教導，立. ‧ 國. 學. 讓我可以從⼀一篇篇的論⽂文中找到⾃自⼰己想要研究的⽬目標與⽅方向。︒｡⽼老師在每⼀一次的討論中都能提出我應該注意的地⽅方，以及值得學習的地⽅方，讓我少⾛走了許多彎路。︒｡在⼤大⼀一的程式設計課中，我. ‧. 從蔡⼦子傑⽼老師⾝身上學到了基礎的程式能⼒力，在碩⼠士班的學習過程中也有⽼老師的⼀一路相隨，真的. sit. y. Nat. 很感謝蔡⼦子傑⽼老師這段時間的提攜與指導。︒｡. al. er. io. 在讀研究所的過程中，也很感謝實驗室裡的⼤大家，多虧了有學⾧長們尤其感謝賀翔學⾧長，學. v. n. ⾧長們的熱⼼心指導與經驗的傳承，不管在課業上還是研究上都給予了我相當⼤大的⽀支持，讓我可以. Ch. engchi. i n U. 很快的適應研究所的⽣生活，也要感謝同屆的同學與學弟們，在課業上有問題的時候可以共同討論。︒｡同時也感謝系⽻羽的朋友們以及⼤大學部的學弟妹們讓我在課餘時間能夠參與各種的活動，讓我⼀一同感受到你們的歡笑與淚⽔水。︒｡最後我要感謝我的家⼈人，在這⼀一路的求學過程中是你們給予了我無⽐比的⽀支持以及⿎鼓勵才讓我⾛走到了⼀一天，感謝你們無私的付出，讓我可以沒有後顧之憂的完成碩⼠士學位。︒｡感謝在我⽣生命出現的所有⼈人，沒有你們就沒有今天的我。︒｡希望研究所畢業並不是⼀一個終點，⽽而是⼈人⽣生另⼀一個新⾥里程碑、︑､新起點，在⼈人⽣生的道路上我們可以再次並肩，⼀一起⾯面對所有未知的挑戰。︒｡. . 5 .

(6) 6 . TABLE OF CONTENT CHAPTER 1 Introduction ............................................................................................................................ 9 1.1 Background & Motivation ................................................................................................................................... 9 1.2 Delay-Tolerant Network (DTN) ........................................................................................................................ 9 1.3 Cosine similarity .................................................................................................................................................. 11 CHAPTER 2 Related work ....................................................................................................................... 13 2.1 Social Trace Data ................................................................................................................................................ 13 2.1.1 Reality mining: MIT [11] ......................................................................................................................... 13 2.1.2 Cambridge [12] ............................................................................................................................................ 14 2.1.3 Infocom05, 06 [13] ..................................................................................................................................... 15 2.2 Social-based in Delay-Tolerant Network .................................................................................................... 16 2.2.1 Social-Aware Data Diffusion in Delay Tolerant MANETs [18] ................................................. 16 2.2.2 Social Network Analysis for Routing in Disconnected Delay-Tolerant MANETS [19] .... 17 . 立. 政治大. ‧ 國. 學. ‧. CHAPTER 3 NCCU Trace Data .............................................................................................................. 18 3.1 Form (Selecting Participants) .......................................................................................................................... 18 3.1.1 College ............................................................................................................................................................ 19 3.1.2 Interest ............................................................................................................................................................ 19 3.2 Trace data ............................................................................................................................................................... 19 . y. Nat. sit. n. al. er. io. 4 Routing Approach .................................................................................................................................... 21 4.1 Environment definition ...................................................................................................................................... 21 4.2 Routing strategy ................................................................................................................................................... 21 4.2.1 Direct Contact ............................................................................................................................................... 22 4.2.2 Indirect Contact ............................................................................................................................................ 23 4.2.2.1 Building Based Indirect Routing ........................................................................................................ 23 4.2.2.2 Social Based Indirect Routing ............................................................................................................. 27 . Ch. engchi. i n U. v. CHAPTER 5 Simulation settings ............................................................................................................. 30 5.1 Simulation environment .................................................................................................................................... 30 5.2 Simulation setting ................................................................................................................................................ 31 5.3 Simulation results ................................................................................................................................................ 32 5.3.1 Delivery ratio ................................................................................................................................................ 33 5.3.2 Overhead ........................................................................................................................................................ 34 5.3.3 Feature choosen insight ............................................................................................................................. 35 CHAPTER 6 Conclusion and future work ............................................................................................ 39 Reference ....................................................................................................................................................... 40 . . 6 .

(7) 7 . LIST OF FIGURE Figure 1: Store, Carry, and Forward in DTN........................................................................................ 10 Figure 2: One view of the network created by MIT Reality Mining dataset ........................................ 14 Figure 3: iMote for the experiment ....................................................................................................... 14 Figure 4: Node infected by friend or stranger....................................................................................... 16 Figure 5: Form list ................................................................................................................................ 19 Figure 6: Message forwarding flow ...................................................................................................... 22 Figure 7: The counting of different college of students go to library ................................................... 24 Figure 8: The counting of the interest of library................................................................................... 25 Figure 9: the counting of the new interest of node B............................................................................ 28 Figure 10: One Simulator ..................................................................................................................... 30 Figure 11: NCCU surrounding area ...................................................................................................... 32 Figure 12: Delivery Ratio ..................................................................................................................... 33 Figure 13: Overhead Copy.................................................................................................................... 34 Figure 14: Overhead Ratio.................................................................................................................... 35 Figure 15: Delivery ratio between 2 and 5 interest column in building-based routing ........................ 36 Figure 16: Overhead copy between 2 and 5 interest column in building-based routing ...................... 37 Figure 17: Overhead ratio between 2 and 5 interest column in building-based routing ....................... 37 Figure 18: Delivery ratio between 2 and 5 interest column in social-based routing ............................ 38 . 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. LIST OF TABLE Table 1: Simulation Settings ................................................................................................................. 31 . . 7 .

(8) 8 . LIST OF ALGORITHM Algorithm 1: Calculating probability of which building do students in college go ............................. 25 Algorithm 2: Calculating algorithm of the interests of building .......................................................... 26 Algorithm 3: Calculating the new interest of node (IL) ........................................................................ 27 Algorithm 4: Routing strategy .............................................................................................................. 29 . 立. 政治大. ‧. ‧ 國. 學. LIST OF FORMULA. Formula 1 .............................................................................................................................................. 11 . y. Nat. Formula 2 .............................................................................................................................................. 23 . sit. Formula 3 .............................................................................................................................................. 26 . er. io. Formula 4 .............................................................................................................................................. 28 . al. v i n Ch Formula 6 .............................................................................................................................................. 34 engchi U Formula 7 .............................................................................................................................................. 35 n. Formula 5 .............................................................................................................................................. 33 . . 8 .

(9) 9 . CHAPTER 1 Introduction. 1.1 Background & Motivation In these years, smart phone has become more and more important in people’s lives. Because smart phone is a very powerful device, it can be useful in many situations like sending and receiving e-mails, communicating with friends, acting as a digital calendar to remind us special days, checking. 政治大. daily weather, and virtual wallet. We can see everyone carry a smart phone wherever they go. Smart. 立. phone has become part of our lives.. The issue we concern in is plenty of advertising spam we might receive every time we open our. ‧ 國. 學. email box. Maybe there are some messages we are interested in. However, we don’t have time to check the spam one by one to pick what we want to read, so we ignore them in most cases. If. ‧. messages can be transferred to people who are interested in them, it surely can reduce the overhead. y. Nat. and make better performance. The problem we want to figure out is that when a message is created,. sit. it may have multi-destination. And how do we know where are the destinations, where are they and. er. io. how can we delivery the message to them is what we concern. Transferring the messages through the. al. Internet is not always the best way. First, the Internet is limited to its ability to access the Internet.. n. v i n Second, besides subscribing to specificC channels and receiving h e n g c h i Uarbitrary spam, we can’t get the messages we want. We want to push the messages to wherever the users are despite the ability to access the Internet. To overcome this problem, we think DTN (Delay-Tolerant Networking) is a good choice.. 1.2 Delay-Tolerant Network (DTN) Delay-Tolerant Networking [1] is a dynamic wireless network. Every node may move freely and be organized depending on their social relationship. DTN can provide interoperable communication in challenging environments, which is defined as the network is not always in connect or the network has no end-to-end path. It is an approach of computer network architecture that can use the strategy of store, carry, and forward to transmit messages in the disconnected network environment. In Figure. . 9 .

(10) 10 . 1, when the node is in the environment without network connection, it may convey messages to nearby nodes by using the short distance transmit technique such as Bluetooth or Wi-Fi direct. When the relay node receives the message, it can carry the message until meeting the next proper node to help transmit and forward the message. Via this approach, messages can travel around the environment and be transmitted to the destination. The connection between two nodes in the DTN environments can only keep for just few seconds, so it needs to find appropriate node to help transmit messages in limited time.. 立. 政治大. ‧ 國. 學. . Figure 1: Store, Carry, and Forward in DTN. ‧. Many MANET and some DTN routing algorithms [2] [3] provide forwarding by building and. Nat. sit. y. updating routing tables whenever mobility occurs. And there are two main reasons why we propose a. er. io. routing algorithm based on DTN. First, we think that people have a routine trace everyday, just like most people have to go to work, and students have to go to school. We always get up around the. n. al. Ch. i n U. v. same time in the morning, do the same chores, and most important of all, we commute to our. engchi. destination in almost the same route and in regular time. It reveals that we might meet the same stranger every day, but we don’t even notice. This stranger would be a terrific node in our routing algorithm. Because we can keep meeting this person everyday, we can update the messages with people we meet. Thus, we can know which person is closer to the destination that messages should be transferred to. Second, the Internet only provides us an end-to-end way to transfer the messages. Under this condition, if we want to transmit a message, we have to know where the destination is first. But in some cases, especially in advertising messages, we do not know where all the destinations are when the message is created. If we use the Internet as communication model, there are only three scenarios: (1) Enterprise, which creates the advertising messages, can only send these messages to people who have registered before. (2) People can only transfer messages to their friends. (3) The enterprise can spread the messages randomly, which will cost a lot. But in DTN, we can transfer the message through the node with store, carry and forward strategy. In the previous work . 10 .

(11) 11 . [4], which is published in IEEE magazine, we can see that the research of using the interest as a feature in their routing protocol has an excellent performance. So we want to take it a step further and continue this research. In this thesis, we suppose people who have similar interests tend to go to the same places. For example, people who like sports or exercise will go to the gym or sports field. Moreover, when they are shopping or doing something else, they are more likely to do things related to sports. People who are interested in art will go to see art exhibitions, and people who like reading will go to the library or bookstores. In this thesis, we use this as a feature on our routing algorithm. Finally, we will compare our routing algorithm with classic routing algorithm like Epidemic[5],. 政治大. MaxProp[6], and PRoPHET[7]. And the result will be show at the chapter 5.. 立. ‧ 國. 學. 1.3 Cosine similarity. In our routing algorithm, we have to decide whether people interest the advertisement message. ‧. or not. There are different kinds of similarities in social DTN routing, like Jaccard similarity [8, 10], Euclidean distance [9, 10], Dice Coefficient [10] and Cosine similarity [10, 11]. For simplicity. Nat. sit. y. without losing of generality, we first choose cosine similarity as our indication for social interest. er. io. relation. It is simple and quite easy to perform. It also helps us determine who wants to know the message as the destination of the message. The formula is showing below.. n. al. 𝐶𝑜𝑠𝑖𝑛𝑒 𝑆𝑖𝑚𝑖𝑙𝑎𝑟𝑖𝑡𝑦 =. Ch. engchi. 𝐴, 𝐴! (𝐷). 𝐴 ∙ 𝐴! (𝐷). =. i n U. v. ! !!! 𝐴! ×𝐴! (𝐷)! , 𝑖 ! ! ! ! !!!(𝐴! ) × !!!(𝐴! (𝐷)! ). < 𝑛 (1). In our trace file, we collect five different interests. But we only use two of the interests as the input. Because the rest three interest columns do not show the difference unfortunately. So, we only take two different interests as input column to calculate cosine similarity. The detail will be showing in chapter 5. The interest column will be assigned when a message is created. And all of the nodes in our simulator have their own interests. We can use both the interest column of the message and the interest column of the node to determine whether the node is interested or not. If the node interests the message, we define the node as one of the destination of the message. Detail is in chapter 4. We propose a new DTN routing algorithm, based on the assumption that people who have daily routine and who have similar interests flock together. Then use the cosine similarity to see who is . 11 .

(12) 12 . interested in the message. Finally, we compare our algorithm with the classic routing algorithm, and the result shows that we have a better performance.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. . Ch. engchi. 12 . i n U. v.

(13) 13 . CHAPTER 2 Related work. In the paper [12], the author thinks that social-based routing and location-based routing are slightly different. Due to the difficulty of collecting the real trace data, we can use the social data as a substitution. If we can collect both social-based and location-based data, we can compare them with each other. In such a variety of DTN routing researches, we focus on two types of research (1) Collecting. 政治大 less resource in DTN environment. Three trace data we will discuss below do not consider the 立 condition of what the real society is like. There are different kinds of people with different interests human real movement data. (2) How to use social data to send data to the destination quickly with. ‧ 國. 學. in the real society, and this is what makes a diverse society. The problem we focus on is multi-destination problem like [13]. People may go to different places or do different things. ‧. depending on their jobs and interests. So if we can use a more realistic trace file that we can regard it as a tiny real society. Different kinds of nodes are moving freely in the emulator, which is closer to. sit. n. al. er. io 2.1 Social Trace Data. y. Nat. the real society.. 2.1.1 Reality mining: MIT [14]. Ch. engchi. i n U. v. This experiment was carried out by MIT. The researcher gives 100 NOKIA’s smart phone to 100 students, and the experiment duration is 9 months. Students who participated in the experiment were asked to use smart phones to communicate with other students by Bluetooth, and their trace, contact time, and communicate time were recorded. Via this experiment, we can analyze and predict social activities’ relation with the subjects to know its next movement and social relations. The disadvantage of the experiment was that 75 students were from MIT Media Laboratory, and the other 25 students were from MIT Sloan business school. We think that the composition of participants can’t be a miniature of the real society.. . 13 .

(14) 14 . 立. 政治大. ‧ 國. 學. . sit. y. Nat. 2.1.2 Cambridge [15]. ‧. Figure 2: One view of the network created by MIT Reality Mining dataset. er. io. This experiment was carried out by Cambridge computer lab. The researcher used the equipment. al. n. v i n C h graduate and doctoral freshman and sophomore, and it also included e n g c h i U students.. named iMote to collect the real trace data. In Cambridge05, the experiment separated students into. Figure 3: iMote for the experiment . 14 .

(15) 15 . During the 11 days, 54 students used iMote equipment with Bluetooth technique, which helps to measure and record the main active area, the other students they contact, and the length of communicate time of each student. In the future, this data can be used in social relationship experiment.. 2.1.3 Infocom05, 06 [16]. 政治大. This experiment was held in students and professors who attended Infocom conference in 2005 and 2006. In the beginning, every participant was given equipment named iMote. Because there were. 立. lots of different topics in the conference, every participant would go to listen to the topic they were. ‧ 國. 學. interested in. Thus, we can know every participant’s interests and who they communicate with. In 4 days, 98 people participated in this experiment. Through the experiment, we can know each. ‧. participant’s professional specialty and whether they communicated with other people who have the same research domain., we can use participants’ communicate time to conjecture their social. sit. y. Nat. relationship.. After reviewing previous research, we think we have to select the participants in order to make. io. n. al. er. the trace data more similar to the real world. One of the most important things is to pick who can. i n U. enroll in our experiment. All the details are described in chapter 3.. Ch. engchi. v. In [17], the characteristics of these datasets such as inter-contact and contact distribution have been explored in several studies [18, 19, 20], to which we refer the reader for further background information. In the trace file above, the participants are comprised of one or two particular group and it is not a general distribution pattern. Which will lead to the trace file is not general. And the trace file will be limited. What we want is a trace file that is a miniature of real society. So we will keep an eye on this while we are picking the participants to involve our experiment. And the detail will be described in chapter 3.. . 15 .

(16) 16 . 2.2 Social-based in Delay-Tolerant Network 2.2.1 Social-Aware Data Diffusion in Delay Tolerant MANETs [21] This research proposes a routing algorithm based on different interests of each node. If two nodes have similar interests, which means the similarity of interests has exceeded the threshold, then we define them as friends; otherwise, the two nodes are defined as strangers. Therefore, when two nodes meet, they will exchange interest list and data list. When their interests are similar, they are friends to each other, and they will exchange data that the carrier likes. On the other hand, if they are. 政治大. strangers to each other, they will diffuse data that they are not interested in the message, just like the state shown below.. 立. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. . Figure 4: Node infected by friend or stranger Even if the two nodes are strangers, it doesn’t mean that their friends or people they might meet are not interested in the message. So in our routing algorithm, we give a chance to case like that. We log every node that one node has met before, and we give it a try if the log file reveals they have a high chance to meet each other again and the data also interests them. For example, node A and node B go to work at the same time in the morning, and they take the same transportation. However, they are strangers, so they may not talk to each other. It is a good time to exchange the advertising messages they got while they are on the bus. During the time they are on the bus, they can know the interests of each other, and the interests of people who they usually meet. It helps us know whether other people are good relay nodes or not.. . 16 .

(17) 17 . 2.2.2 Social Network Analysis for Routing in Disconnected Delay-Tolerant MANETS [22] It is difficult to diffuse messages in spare MANETs. All the nodes can move freely. So to find out the most efficient path is the key to this research. Previous research has conducted many theories to discover the best way to route the messages. To overcome this issue, there is Centrality, which can reveal whether the node is connected to the neighbor nodes. In other words, the node is able to know whether there is a path to the destination. They propose a new routing protocol named SimBet. SimBet is based on the betweenness Centrality and Similarity of nodes, and it chooses the. 政治大 gather at some active nodes. Nodes, 立which are in relatively static state, can only transfer the. intermediate node to help carry the messages. But the disadvantage is that all the messages tend to messages and are not able to get the messages they want. In our routing algorithm, we do not want a. ‧ 國. 學. few people to carry most of the messages. We think it might lead to information starving for some nodes. We consider not only the connection between nodes but also the landmark where the node has. ‧. gone before. A node can be the relay node if this node will meet some other nodes which are. y. Nat. interested in the message. Everyone can be the relay node depending on their movement and social. n. er. io. al. . sit. relationship, and we think it is a better way to route the messages.. Ch. engchi. 17 . i n U. v.

(18) 18 . CHAPTER 3 NCCU Trace Data All the participants in trace data we mentioned above are limited to some specific group, either students of particular college or participants in particular conference. For example, the participants in MIT trace data were composed of Media Laboratory and business school students. In the Cambridge trace data, only computer laboratory students were enrolled. Furthermore, the Infocom trace data was collected during the conference, so most of the participants were related to the conference. We think these three trace data above can’t reveal how the real social network works. In real social network,. 治政 interests. We assume that where people usually go depends on大 their jobs and interests, and this issue 立 every participator’s background. We limit the college ratio is what we concentrate on. So we check people are not supposed to do the same thing or the same work all the time, and they have different. ‧ 國. 學. of participator depend on the college ratio of our school. Also, we try our best to control they come from different department while they are in same college. In this case, we can guarantee that. ‧. participators are come from different and do different routine on their own. We try our best to emulate how the nodes move in reality. Then, We refer [23] when we want to create our own trace. Nat. sit. y. file. We not only record the trace file, but also record the self-declared interests of each user. In the. io. er. end, because we don’t have enough participants that can represent the real society to implement the experiment, we limit the environment to our campus. Participators are students in National ChengChi. al. n. University.. Ch. engchi. i n U. v. 3.1 Form (Selecting Participants) When we were building our own trace file, we selected participants first. In order to find the suitable participants, we asked everyone who enrolled in to fill out the form, which asked some basic profile information and the most important thing, participants’ college and interests. Figure 5 shows a part of the real data that we obtained.. . 18 .

(19) 19 . Figure 5: Form list. 3.1.1 College. 政治大 strangers to each other, either. Two participants were assigned to a group, and they must know each 立 All the participants should not come from one specific group, and they should not be all. other or we would not accept them. We also cared about what college they are from. The college. ‧ 國. 學. quantity ratio depended on the real college quantity ratio of total students in NCCU. If the quantity of one college was about to surpass the quantity we wanted, we would not accept the coming group,. ‧. either. In this case, participants in our trace data were distributed to different colleges. Some people. sit er. io. 3.1.2 Interest. y. Nat. knew each other, and others didn’t just like the real society.. al. n. v i n In our algorithm, we had to determine of each message was. So we asked C hwhere the destination U i e h n c were divided into five types, which were the participants what kind of interest they are. Theg interests sports, reading, social, art, and service. The score was limited from 0 to 1, and each scale was at least 0.25. On the other side, when a message was created, it would be assigned to these five interest types accordingly. But when we calculating the average of these interest columns, we find out that the social, arts and service column don’t show the difference. So we can only use sports and reading column as an input to calculate the cosine similarity as we mentioned earlier at Formula (1) to define whether the nodes are interested in the messages or not.. 3.2 Trace data The most important part of trace data is the movement of each node. In order to get the real trace. . 19 .

(20) 20 . data, using smart phones was the best way to collect the information we needed. In traditional work, they created a device to record the user location, but the users may forget to bring the device with them. In reality, carrying another device only to send or receive messages is difficult to implement. In these days, people carry their own smart phone wherever they go, even at the places they are not allowed to use it. In conclusion, we think the smart phone is the best device to implement the DTN routing algorithm. For this reason, we designed an Android app and installed it in each participants’ smart phone. We ran a background service to record GPS position every 10 minutes just like the trace file in paper [24]. If we scan too often, the battery of the smart phone will dry out fast. In our. 治政 continue record the user position, we stipulated a scan every 10大 minutes. 立 we ignored the trace data that were not in the campus and After all the trace data were collected, previous experiment, if we scanned every 5 minutes, the battery couldn’t hold on for a day. To. ‧ 國. 學. normalized the trace data. Because we want the trace data to move smoothly just like real walking but not to disappear in one place and appear in another place suddenly. In order to achieve this goal,. ‧. we had to normalize the trace data and keep the trace track continuous. Finally, when these works were done, our trace data were about to be used. There were 115 available data in our trace data in. Nat. sit. n. al. er. io. can be downloaded at [25].. y. total, and the experiment lasted for two weeks, from 17th Dec to 31st Dec in 2014. All the trace data. . Ch. engchi. 20 . i n U. v.

(21) 21 . 4 Routing Approach. Most traditional routing scenarios only have one single destination, but our goal is to deliver messages to various destinations and reduce the overhead. However, using DTN to transmit a personal message to another person would not be as suitable as the advertising message. Thus, if a company wants to spread an advertising message to as many customers as possible, DTN is a suitable option.. 4.1 Environment definition. 立. 政治大. ‧ 國. 學. First, because the trace data we obtained were from the campus, we limited the simulation environment to the campus, too. When an advertising message is created, we don’t quite know which. ‧. destination we should send it to. All we can do is to find out the people who might be interested, and. y. Nat. push the message to the right person. But in reality, we won’t know whether a node is interested in. sit. the message or not until it bumps into other nodes and exchange metadata. The metadata includes the. er. io. information about the contact node and other’s interests, what nodes the contact node has met, and. al. also what messages the contact node has got so far. When the receiver receives the metadata, it will. n. v i n calculate the cosine similarity between C the interest type of theU h e n g c h i message and the interests of other. nodes. If the result surpasses the threshold, then the message will be delivered. The entire process is termed “contact.”. 4.2 Routing strategy The message forwarding flow is shown in Figure 6. Whenever a message is created, it will be assigned to one node randomly. For instance, we assume the message is assigned to node A. Node A may meet node B along the way to its destination. The time node A and node B spend on exchanging data with each other is called the total “contact time”. When the connection is interrupted or out of the connection area, the data transferring stops. During the contact time, node A will try all kinds of messages in the transferring waiting queue one by one. In next paragraph, there are two kinds of. . 21 .

(22) 22 . contact in our algorithm: direct contact and indirect contact.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 6: Message forwarding flow. 4.2.1 Direct Contact Direct contact is the most common way to transfer the messages. When two nodes meet each other, they will exchange metadata about their interests. And the node will calculate and compare the interest type of the message with the interests of another node it meets. For instance, node A encounters node B, and they exchange the metadata. Node A gets the interests of node B, and it will calculate the cosine similarity (𝐶𝑜𝑠) between the interest of node B and the interest type of message . 22 .

(23) 23 . (𝐼! ). Node A will compare every message that node A has and node B doesn’t have with the interest of node B (𝐼!! ). Formula (2) is shown below. 𝐶𝑜𝑠 𝐼!! , 𝐼! > 𝑇ℎ𝑟𝑒𝑠_𝐷𝐶 (2) If the result of cosine similarity is greater than the 𝑇ℎ𝑟𝑒𝑠_𝐷𝐶, the message will be put in queue and ready to be transferred. The transferring time will last as long as the contact time of two nodes or. 政治大. until all the messages in queue are delivered.. 立. 4.2.2 Indirect Contact. ‧ 國. 學. In addition to direct contact, we expect the contact node can be a relay node, which can carry the. ‧. messages to other nodes that are interested in the message. If we transfer these messages randomly, it will lead to high overhead and doesn’t make our performance better. So we propose two kinds of. er. io. sit. y. Nat. indirect contact routing algorithms, which are base on building and social relationship.. al. 4.2.2.1 Building Based Indirect Routing. n. v i n We use the historical data of whichC building went to forecast the future. First, h e ntheg students hi U c students of different colleges go to different buildings. For example, students of Commerce College have a higher chance to meet one another than any other student of other colleges because they usually go to the classrooms in the Commerce Building. Besides, students of Accounting Department have a much higher chance to meet students of the same department than students of Statistics Department in Commerce College. Second, we assume that students who have similar interests gather in the same building. Just like, students who like to exercise will go to the school gym and students who like to studying like to go to library.. . 23 .

(24) 24 . 立. 政治大. ‧ 國. 學. . ‧. Figure 7: The counting of different college of students go to library. sit. y. Nat. In Figure 7, the students of Language College go to the library more often compared with the students of Science College and Law College. The reason might be that some ancient documents. io. n. al. er. only have print editions, so most of them have to go to the library to get these books. However, the. i n U. v. students of Science College can search the information on the Internet.. Ch. engchi. In order to prove our assumption, we calculate all the probability that students from different colleges go to each building on the campus. We can calculate the interests, which can be defined just like the form we ask the participants to fill out. Suppose that the interest of the building can be counted and defined according to what kinds of people have come before. Figure 8 displays the counting process. We first initialize all the interest of each building to 0. Second, whenever someone goes to the building, we add the interest of that person to the interest of the building. After the entire trace file is checked, we normalize the interest of the building. Finally, we can define the interest of the building.. . 24 .

(25) 25 . 立. 政治大. . Figure 8: The counting of the interest of library. ‧ 國. 學. . When we finish collecting the entire trace file, we calculate how many times students of specific college go to specific building. We think there is logical evidence that students of the same college or. ‧. the same department have a higher chance to go to the same building.. In our trace file there are 8 different colleges in total 115 nodes (𝒮). And there are we define 17. y. Nat. sit. buildings (ℬ) to calculate the probability of college to each building (𝑃𝑟𝑜𝑏!"##$%$. er. io. verify our proposal. The calculating algorithm will be described below.. al. n. v i n Algorithm 1: Calculating probability ofC which building do students in college go U h i e h n gc 𝐼𝑛𝑝𝑢𝑡: 𝑒𝑣𝑒𝑟𝑦 𝑛𝑜𝑑𝑒 𝑖 𝑁! ∈ 𝒮, 𝑡𝑟𝑎𝑐𝑒 𝑓𝑖𝑙𝑒 𝑇𝐹 , 𝑒𝑣𝑒𝑟𝑦 𝑏𝑢𝑖𝑙𝑑𝑖𝑛𝑔 𝑏 ! ∈ ℬ 𝑂𝑢𝑡𝑝𝑢𝑡: 𝑃𝑟𝑜𝑏!"##$%$. !! ,!!. 1: 𝑆𝑒𝑡 𝑒𝑣𝑒𝑟𝑦 𝑃𝑟𝑜𝑏!"##$%$. !! ,!! 𝑡𝑜 0. 2: 𝐹𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑁! ∈ 𝒮 3:. 𝐹𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑡𝑖𝑚𝑒 𝑗 𝑇! 𝑖𝑛 𝑇𝐹. 4:. 𝑖𝑓 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑁! , 𝑇! 𝑖𝑛 𝑏!. 5: 6:. 𝑃𝑟𝑜𝑏!"##$%$. !! ,!!. 𝐶𝑜𝑙𝑙𝑒𝑔𝑒 𝑁! , 𝑏! += 1. 𝑒𝑛𝑑 𝑓𝑜𝑟. 7: 𝑒𝑛𝑑 𝑓𝑜𝑟 8: 𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒 𝑃𝑟𝑜𝑏!"##$%$. . !! ,!! 𝑡𝑜 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 0 𝑤𝑖𝑡ℎ 1, 𝑓𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑏!. 25 . !! ,!! ). and try to.

(26) 26 . Algorithm 2: Calculating algorithm of the interests of building 𝐼𝑛𝑝𝑢𝑡: 𝑒𝑣𝑒𝑟𝑦 𝑛𝑜𝑑𝑒 𝑖 𝑁! ∈ 𝒮, 𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝑜𝑓 𝑁! 𝐼!! , 𝑡𝑟𝑎𝑐𝑒 𝑓𝑖𝑙𝑒 𝑇𝐹 , 𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝑜𝑓 𝑏𝑢𝑖𝑙𝑑𝑖𝑛𝑔 𝑏! (𝐼!! ) 𝑂𝑢𝑡𝑝𝑢𝑡: 𝐼!! 1: 𝑆𝑒𝑡 𝑒𝑣𝑒𝑟𝑦 𝐼!! 𝑡𝑜 0, 𝑏! ∈ ℬ 2: 𝐹𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑁! ∈ 𝒮 3:. 𝐹𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑡𝑖𝑚𝑒 𝑗 𝑇! 𝑖𝑛 𝑇𝐹. 4:. 𝑖𝑓 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑁! , 𝑇! 𝑖𝑛 𝑏! 𝐼!! = 𝐼!! + 𝐼. 5:. 𝑒𝑛𝑑 𝑓𝑜𝑟. 6:. 立. 7: 𝑒𝑛𝑑 𝑓𝑜𝑟. 政治大. ‧ 國. 學. 8: 𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒 𝐼!! 𝑡𝑜 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 0 𝑤𝑖𝑡ℎ 1, 𝑓𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑏!. After we do both Algorithm1 and Algorithm2, we can use them to implement our. ‧. Building-Based Indirect Routing. For instance, when node A encounters node B, node A carries one. y. Nat. message M, but node B doesn’t want message M. Then A will check the interest of every building. sit. and the probability of node B going to each building. If the calculating result surpasses the threshold. n. al. er. io. (𝑇ℎ𝑟𝑒𝑠_𝑚𝑒𝑒𝑡), node A will still transfer the message M to node B. The formula is shown in Formula (3).. Ch. engchi. 𝐶𝑜𝑠 𝐼! , 𝐼!! ∗ 𝑃𝑟𝑜𝑏!"##$%$. i n U. !! , !!. v. > 𝑇ℎ𝑟𝑒𝑠!""#. !! ∈ℬ. 𝐼! 𝐼!! 𝑃𝑟𝑜𝑏!"##$%$ !!, !! 𝑇ℎ𝑟𝑒𝑠!""# . Interest score of the message M Interest score of the building k Probability of the College of the Node b moving to building k Threshold of the meeting probability (3). . 26 .

(27) 27 . 4.2.2.2 Social Based Indirect Routing This is another indirect routing algorithm we propose. We think every one has a routine schedule in a period of time. In most cases, the period is a week. Because most people have to go to work or school on weekdays, it leads to the result that we will do mostly the same task like what we did seven days ago. In our campus scenario, students have to follow their own schedule and go to class accordingly. So if we can record where they were last week, we may forecast where they will go in the near future. To achieve our goal, we define a new interest (IL) that records what kind of people the node met seven days ago. The new interest of node A last week is defined as 𝐼𝐿!! ,!! . The. 政治大. calculating algorithm of the new interest is described below.. 立. ‧ 國. 𝑂𝑢𝑡𝑝𝑢𝑡: 𝐼𝐿. 學. Algorithm 3: Calculating the new interest of node (𝐼𝐿) 𝐼𝑛𝑝𝑢𝑡: 𝑒𝑣𝑒𝑟𝑦 𝑛𝑜𝑑𝑒 𝑖 𝑁! ∈ 𝒮, 𝐼𝑛𝑡𝑒𝑟𝑒𝑠𝑡 𝑜𝑓 𝑁! 𝐼!! , 𝑡𝑟𝑎𝑐𝑒 𝑓𝑖𝑙𝑒 𝑇𝐹. 𝑆𝑒𝑡 𝑒𝑣𝑒𝑟𝑦 𝐼𝐿!! ,!! 𝑡𝑜 0, 𝑁! ∈ 𝑆! , 𝐷! 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑀𝑜𝑛𝑑𝑎𝑦 𝑡𝑜 𝑆𝑢𝑛𝑑𝑎𝑦, 1 ≤ 𝑑 ≤ 14. 2:. 𝐹𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑁! ∈ 𝒮. y. Nat. 𝑖𝑓 𝑁! 𝑚𝑒𝑒𝑡 𝑁! 𝑜𝑛 𝐷! , 𝑁! ∈ 𝒮. 5:. 𝐼𝐿!! ,!! = 𝐼𝐿!! ,!! + 𝐼!!. 6:. 𝐼𝐿!! ,!! = 𝐼𝐿!! ,!! + 𝐼!!. al. n. 7:. io. 4:. sit. 𝐹𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑡𝑖𝑚𝑒 𝑗 𝑇! 𝑖𝑛 𝑇𝐹. er. 3:. ‧. 1:. 𝑒𝑛𝑑 𝑓𝑜𝑟. 8:. 𝑒𝑛𝑑 𝑓𝑜𝑟. 9:. 𝐹𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 𝑁! ∈ 𝒮. Ch. engchi. i n U. v. 10: 𝑁𝑜𝑟𝑚𝑎𝑙𝑖𝑧𝑒 𝐼𝐿!! ,!! 𝑡𝑜 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 0 𝑤𝑖𝑡ℎ 1, 𝑓𝑜𝑟 𝑒𝑣𝑒𝑟𝑦 1 ≤ 𝑑 ≤ 14 For example, node B encountered node C, D and E last Monday. So the new interest of node B (𝐼𝐿!! ,!! ) will be the mean of the interests of node C, D and E, just like what we do when counting the interest of the building. At the Figure 9, we can see the procedure of counting the new interest of B. First, we set the new interest of node B to zero. Second, whenever the other node meets the node B, we add the interest of the other node to the new interest of node B. After the entire meeting are done. We normalize the new interest to between 0 with 1. And it is our output. . 27 .

(28) 28 . 立. 政治大. . Figure 9: the counting of the new interest of node B. ‧ 國. 學. When node A which has a message M meet node B next Monday, node B is not interested in the. ‧. message M. Node A will check the 𝐼𝐿!! ,!! value. If the cosine similarity between the message M. y. Nat. and the new interest 𝐼𝐿!! ,!! surpass the threshold (Thres_Cos). Node A will still transfer the. sit. message M to node B. Then node B will be a carrier of message M and help spread the message M.. er. io. The formula is displayed below.. n. al. i n 𝐶𝑜𝑠C𝐼!h, 𝐼𝐿! ,! > 𝑇ℎ𝑟𝑒𝑠 e n g c h i U!"# !. 𝐼! 𝐼!! 𝑇ℎ𝑟𝑒𝑠!"#. v. !. Interest score of the message M Interest score of the building k Threshold of the Cosine similarity (4). When we simulate our trace file, we only have two weeks of trace data. So, if we want to simulate the first week, we have to use the 𝐼𝐿!! ,!! of the second week. In other words, we have to use the 𝐼𝐿!! ,!!!! in our simulator if we want to simulate the first week. On the other hand, when we want to simulate the trace file of the second week, we use 𝐼𝐿!! ,!!!! .. . 28 .

(29) 29 . There are several kinds of calculating, which shows the detail of the number we get from. It proves that we didn’t put numbers to these attributes arbitrary. The counting formula reveals how we define direct contact(𝐷𝐶), indirect contact(𝐼𝑁𝐶), cosine similarity, probability of going specific building and the social relationship of particular person. After introducing all the counting progress in our routing strategy, we conclude the routing strategy with the routing algorithm what will a node do in our environment below. First, we define if there is a node in our environment encounter another node we set 𝑖𝑠𝐸𝑛𝑐𝑜𝑢𝑛𝑡𝑒𝑟 to true. And it carries many messages (𝑀𝑒𝑠𝑠𝑎𝑔𝑒𝑠). Whenever we can get a message. 治政大together and it is still able to contact 𝑖𝑠𝑆𝑡𝑖𝑙𝑙𝐶𝑜𝑛𝑡𝑎𝑐𝑡(𝑁 ) to true whenever these two nodes are stick 立 to the other node(𝑁 ). And finally define the mode of indirect contact by setting 𝑖𝑠𝐵𝑢𝑖𝑙𝑑𝑖𝑛𝑔𝐵𝑎𝑠𝑒𝑑. (𝑀𝑒𝑠𝑠𝑎𝑔𝑒𝑠. 𝑔𝑒𝑡𝑀𝑒𝑠𝑠𝑎𝑔𝑒) from 𝑀𝑒𝑠𝑠𝑎𝑔𝑒𝑠, we define 𝑀𝑒𝑠𝑠𝑎𝑔𝑒𝑠. ℎ𝑎𝑠𝑁𝑒𝑥𝑡 equal to true. And set !. !. ‧ 國. 學. or 𝑖𝑠𝑆𝑜𝑐𝑖𝑎𝑙𝐵𝑎𝑠𝑒𝑑 to true.. 𝑤ℎ𝑖𝑙𝑒(𝑀𝑒𝑠𝑠𝑎𝑔𝑒𝑠. ℎ𝑎𝑠𝑁𝑒𝑥𝑡 & 𝑖𝑠𝑆𝑡𝑖𝑙𝑙𝐶𝑜𝑛𝑡𝑎𝑐𝑡(𝑁! )) 𝐷𝐶 = 𝐶𝑜𝑠 𝐼!! , 𝐼!. 5.. 𝑖𝑓 𝐷𝐶 > 𝑇ℎ𝑟𝑒𝑠_𝐷𝐶. 7. 8.. n. 6.. al. 𝑇𝑟𝑎𝑛𝑠𝑓𝑒𝑟 𝑚𝑒𝑠𝑠𝑎𝑔𝑒 𝑚. !! ∈ℬ 𝐶𝑜𝑠. 𝐼! , 𝐼!! ∗ 𝑃𝑟𝑜𝑏!"##$%$. 10.. 𝑖𝑓 𝐼𝑁𝐶 > 𝑇ℎ𝑟𝑒𝑠!""# 𝑇𝑟𝑎𝑛𝑠𝑓𝑒𝑟 𝑚𝑒𝑠𝑠𝑎𝑔𝑒 𝑚 𝑒𝑙𝑠𝑒 𝑖𝑓 𝑖𝑠𝑆𝑜𝑐𝑖𝑎𝑙𝐵𝑎𝑠𝑒𝑑. 13.. 𝐼𝑁𝐶 = 𝐶𝑜𝑠 𝐼! , 𝐼𝐿!! ,!!. 14.. 𝑖𝑓 𝐼𝑁𝐶 > 𝑇ℎ𝑟𝑒𝑠!"#. 15. 16.. . engchi. v. 𝑖𝑓 𝑖𝑠𝐵𝑢𝑖𝑙𝑑𝑖𝑛𝑔𝐵𝑎𝑠𝑒𝑑 𝐼𝑁𝐶 = . 12.. Ch. i n U. 𝑒𝑙𝑠𝑒. 9.. 11.. sit. 4.. er. Message m = Messages.getMessage. io. 3.. y. Nat. 2.. ‧. Algorithm 4: Routing strategy 1. 𝑖𝑓 𝑖𝑠𝐸𝑛𝑐𝑜𝑢𝑛𝑡𝑒𝑟 𝑎𝑛𝑜𝑡ℎ𝑒𝑟 𝑛𝑜𝑑𝑒 𝑁!. 𝑇𝑟𝑎𝑛𝑠𝑓𝑒𝑟 𝑚𝑒𝑠𝑠𝑎𝑔𝑒 𝑚 𝐸𝑛𝑑 𝑤ℎ𝑖𝑙𝑒. 29 . !! ,!!. > 𝑇ℎ𝑟𝑒𝑠!""#.

(30) 30 . CHAPTER 5 Simulation settings. 5.1 Simulation environment In our simulation, we use ONE (Opportunistic Network Environment simulator) [26] (shown as Figure 10), which is an open source on github and the map of NCCU (Nation Cheng-Chi University) surrounding area to validate our approach. All nodes in the simulation represent the student of this. 政治大. college, and they walk around according to their class schedule or for some purpose.. 立. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 10: One Simulator. . 30 .

(31) 31 . 5.2 Simulation setting The simulation setting is shown as Table 1. The map area is 3764m x 3420m, which is the main active area of NCCU (Figure 11), and the simulation time is from 12 a.m. to 12 a.m. of the next day. This is about 172800 seconds, it is equivalent to two days. The reason why we choose this time slot is that some students are at the school during the day and others live in the dorms at night. The node data transmission rate is 250KBps, and the transmission rage is 10m. The message size of data is 500KB~1MB, the node buffer size is 100MB, and the message’s TTL is 1080mins, which is equivalent to 18 hours.. 政治大. 6750*5100m. 學. 172800 sec. Data Rate. 250KBps. Radio Range. 10m. Message Size. 500KB~1MB. Buffer Size. 100MB. n. Total Message Created Time to Live. 1080mins. engchi. Table 1: Simulation Settings. . v. i n 62 U. Ch. y. sit. io. al. er. Nat. Simulation Time. ‧. ‧ 國. 立 Area. 31 .

(32) 32 . 立. 政治大. n. al. er. io. sit. y. Nat. 5.3 Simulation results. ‧. ‧ 國. 學 Figure 11: NCCU surrounding area. . i n U. v. In tradition, there are several ways to route the message in DTN environment such as Epidemic,. Ch. engchi. MaxProp[2], and PRoPHET[3]. The routing of Epidemic is by the way of transferring messages to every node where the carriers meet. The overhead of Epidemic is extremely high, but it has a better performance. We want to reduce the overhead as much as we can and prevent the performance from dropping too much. The routing of MaxProp concerns with people they have met before and the sequence of messages to be sent. MaxProp does not keep an eye on which direction the message should be sent to. In our routing algorithm, it will calculate the probability of which relay node gets a higher chance to arrive at the destination. PRoPHET focuses on two nodes meeting each other and which one gets a path that can transfer messages to the destination in a relatively higher probability. However, PRoPHET only works well at unicast case, and it saves all the probability of meeting all the other nodes, which would cost a lot of memory in a giant trace file. In our routing algorithm, we only saves data of the node that has met other nodes before, and it surely helps us reduce the usage of memory. . 32 .

(33) 33 . We simulated our trace file with both Building-based and Social-based routing algorithms, and we used the setting that we have described above. We evaluated the performance by checking the deliver rate and the overhead. If we can get higher deliver rate with lower overhead, it means that we get better performance. The deliver rate is what we cared about most. The results are as follows.. 5.3.1 Delivery ratio Formula (5) is the way we counted the deliver rate. Because we had various destinations for one. 政治大. message (𝑚! ), we had to know how many destinations the message would go to (𝐷𝑒𝑠𝑡𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑁𝑢𝑚) and whether these destinations received the message (𝐷𝑒𝑠𝑡𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑅𝑒𝑙𝑎𝑦𝑒𝑑). The total message. 立. number (𝑇𝑜𝑡𝑎𝑙𝑀𝑒𝑠𝑠𝑎𝑔𝑒𝑁𝑢𝑚) created in one day was 62 messages. After we got the 2 parameters. , ∀𝑚! ∈ 𝑀 . y. 𝑇𝑜𝑡𝑎𝑙𝑀𝑒𝑠𝑠𝑎𝑔𝑒𝑁𝑢𝑚. io. sit. Nat. 𝐷𝑒𝑙𝑖𝑣𝑒𝑟 𝑅𝑎𝑡𝑒 = . ! 𝐷𝑒𝑠𝑡𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑅𝑒𝑙𝑎𝑦𝑒𝑑 !! 𝐷𝑒𝑠𝑡𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑁𝑢𝑚. ‧. ‧ 國. deliver rate.. 學. above, we could calculate the mean deliver rate of messages (𝑀) per day. Finally, we could get the. (5). n. al. er. Figure 12 displays the result that our routing algorithms have a better performance than. i n U. v. MaxProp and PRoPHET, but they still have something to improve to reach the Epidemic.. Ch. engchi. Delivery Ratio 0.8 0.7 . Epidemic . Percentage. 0.6 0.5 . MaxProp . 0.4 . PRoPHET . 0.3 . Building based . 0.2 . Social based . 0.1 0 Day . . Figure 12: Delivery Ratio. . 33 .

(34) 34 . In traditional simulation environment, the nodes can only be influenced by some particular factors like the nodes are in particular group, so there is a key routing feature that can be used. Some of the simulation results could reach the performance of Epidemic. But in reality, there are too many factors to affect people’s behavior. For example, people may catch a cold, so they have to go to the hospital or clinic instead of working space. Furthermore, students may skip the class and go out to get some fun. Besides, we may run into somebody we know and then go to a coffee shop, which is not in our plan. So many factors can have an impact on our lives. It is hard to forecast the future. 政治大. precisely, and this is why we cannot beat the Epidemic routing algorithm.. 立. ‧ 國. 學. 5.3.2 Overhead. In order to have a better understanding, we calculated overhead with overhead copies and. ‧. overhead ratio. The calculating formula is shown below. And we separate into two parts of overhead.. y. sit. io. n. al. − 𝐷𝑒𝑠𝑡𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑅𝑒𝑙𝑎𝑦𝑒𝑑 , ∀𝑚! ∈ 𝑀 𝑇𝑜𝑡𝑎𝑙𝑀𝑒𝑠𝑠𝑎𝑔𝑒𝑁𝑢𝑚 (6). er. Nat. 𝑂𝑣𝑒𝑟ℎ𝑒𝑎𝑑 𝐶𝑜𝑝𝑖𝑒𝑠 = . ! !! 𝑅𝑒𝑙𝑎𝑦𝑒𝑑. i n U. Ch. v. i e n g cChopy Overhead . 30 . Overhead Copy . 25 . Epidemic . 20 . MaxProp . 15 . PRoPHET . 10 . Building based . 5 . Social based . 0 Day . Figure 13: Overhead Copy. . 34 .

(35) 35 ! 𝑅𝑒𝑙𝑎𝑦𝑒𝑑 − 𝐷𝑒𝑠𝑡𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑅𝑒𝑙𝑎𝑦𝑒𝑑 !! 𝐷𝑒𝑠𝑡𝑖𝑛𝑎𝑡𝑖𝑜𝑛𝑅𝑒𝑙𝑎𝑦𝑒𝑑. 𝑂𝑣𝑒𝑟ℎ𝑒𝑎𝑑 𝑅𝑎𝑡𝑖𝑜 = . 𝑇𝑜𝑡𝑎𝑙𝑀𝑒𝑠𝑠𝑎𝑔𝑒𝑁𝑢𝑚. , ∀𝑚! ∈ 𝑀 . Overhead Ratio. 立. Epidemic . 政治大. MaxProp PRoPHET . ‧ 國. Building based Social based . ‧ y. Figure 14: Overhead Ratio. io. sit. Nat. Day. er. 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 . 學. Overhead Ratio. (7). al. v i n C ha little bit lower overheads overhead ratio than Epidemic, and we have than MaxProp and PRoPHET engchi U relatively. Combined with Figure 12, we have better delivery rate. It proves that the Building-Based n. In Figure 13 and Figure 14, we can see that we have lower overhead in both overhead copies and. and Social-Based routing algorithms have a better performance than the three traditional routing algorithms.. 5.3.3 Feature choosen insight The showing results above, we only take sport and reading interest column as the input to calculate the destination. The reason why is that the rest social, art and social interest column do not show the obvious difference in our routing algorithm. We think that is because we have gym building, which relate to sport interest column. And the school library relate to reading interest. But in other hand, we don’t have Art College in our school. The rest social and service interest column cannot be relate to. . 35 . .

(36) 36 . specific building. It leads that these interest columns cannot reveal the difference in our building-based routing algorithm. The comparison we can see the result below.. 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 . 立. 政治大. 2 interest 5 interest . 學. ‧ 國. Percentage. Delivery Ratio. Day. ‧. . Figure 15: Delivery ratio between 2 and 5 interest column in building-based routing. y. Nat. sit. And in the same reason, if we cannot select the right feature, it leads to the extra overhead. When we. al. er. io. try to route the message forward, it is important to select the right feature. To confirm the delivery or. v. n. not depends on the input. If we input the uniformly interest column, the algorithm can’t tell the. Ch. difference. So there are some extra overhead comes behind.. . engchi. 36 . i n U.

(37) 37 . Overhead Copy. Overhead Copy 35 30 25 20 15 10 5 0 . 2 interest 5 interest . 政治大 Day. 立. . Figure 16: Overhead copy between 2 and 5 interest column in building-based routing. 0 . ‧ 國. y. sit er. Overhead Ratio. 0.5 . al. n. 1 . io. 1.5 . ‧. 2 . 學. 2.5 . Nat. 3 . Overhead Ratio. Ch. engchi. i n U. v. 2 interest 5 interest . Day. Figure 17: Overhead ratio between 2 and 5 interest column in building-based routing On the other side, we can see the there are some difference between building-based and social-based routing. There is no such concept like building in the social-based routing. So it works fine in our school environment. Due to this reason, the social-based doesn’t have to be limited in 2 interest column as input. The people who like art can go freely and meet some other people who have similar interests. And so do social and service interest columns. The result we can see below.. . 37 .

(38) 38 . Percentage. Delivery Ratio 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 . 2 interest 5 interest . 政治大 Day . 立. . Figure 18: Delivery ratio between 2 and 5 interest column in social-based routing. ‧ 國. 學. No matter choose 2 or 5 columns as an input, there is no obvious difference between them in. ‧. social-based. In social-based routing, we only use the people who we met before as the input. So no matter whether there are buildings or not it can work perfectly. If we don’t want to use the. Nat. sit. n. al. er. io. substitution.. y. point-of-interest concept as routing algorithm, social-based routing algorithm is a good way as a. . Ch. engchi. 38 . i n U. v.

(39) 39 . CHAPTER 6 Conclusion and future work. In this thesis, we propose Building-Based and Social-Based routing algorithms. We use the trace file and the interests of people, which were collected in our experiment. Nodes in our experiment can not only be a destination node, but also be a relay node. It is helpful to spread messages. When two nodes meet, they will exchange metadata. We check whether the nodes are destinations and whether they are good relay nodes to make sure that the message transferring is efficient. Finally, we evaluate. 政治大 performance. In the future, we will consider using the real social relationship between each node. We 立 think if two nodes know each other, they will have a higher probability to meet again. If this our routing algorithms with other algorithms, and the result shows that our algorithms have a better. ‧ 國. 學. hypothesis is true, it will be a good feature to check whether the node is a good relay node or not. And maybe combined with auto collect personal interest from user profile, text, or something they. ‧. view always these kinds of routing strategy can be used in point-of-interest in the future, which can help us get the messages we want correctly.. n. er. io. sit. y. Nat. al. . Ch. engchi. 39 . i n U. v.

(40) 40 . Reference. [1] K. Fall, “A delay-tolerant network architecture for challenged internets.” in Proc. SIGCOMM, 2003. [2] E. P. C. Jones, L. Li, and P. A. S. Ward, “Practical routing in delay-tolerant networks,” in Proc. WDTN, 2005. [3] A.Lindgren, A.Doria, and O.Schelen, “Probabilistic routing in intermittently connected networks,”. 政治大 [4] Tzu-Chieh Tsai and Ho-Hsiang Chan, “NCCU Trace: social-network-aware mobility trace,” 立 Communications Magazine, IEEE, vol. 53, pp. 144–149, 2015. in Proc. SAPIR, 2004. ‧ 國. 學. [5] A. Vahdat and D. Becker, “Epidemic Routing for Partially Connected Ad Hoc Networks,” Tech. Rep. CS-2000- 06, Duke Univ., July 2000.. ‧. [6] BURGESS, J., GALLAGHER, B., JENSEN, D., AND LEVINE, B. N. MaxProp: Routing for Vehicle-Based Disruption-Tolerant Networks. In Proceedings of IEEE Infocom (April 2006).. y. Nat. sit. [7] LINDGREN, A., DORIA, A., AND SCHELEN, O. Probabilistic routing in intermittently. er. io. connected networks. In The First International Workshop on Service Assurance with Partial and. al. Intermittent Resources (SAPIR) (2004).. n. v i n [8] Zhang, Xiaomei, and Guohong Cao.C "Transient community h e n g c h i Udetection and its application to data forwarding in delay tolerant networks." 2013 21st IEEE International Conference on Network Protocols (ICNP). IEEE, 2013. [9] Bigwood, Greg, et al. "Exploiting self-reported social networks for routing in ubiquitous computing environments." 2008 IEEE International Conference on Wireless and Mobile Computing, Networking and Communications. IEEE, 2008. [10] Sanguankotchakorn, Teerapat, Shradha Shrestha, and Nobuhiko Sugino. "Effective ad hoc social networking on OLSR MANET using similarity of interest approach." International Conference on Internet and Distributed Computing Systems. Springer Berlin Heidelberg, 2012. [11] A. Mei, G. Morabito, P. Santi and J. Stefa, “Social-aware stateless forwarding in pocket switched networks,” in Proc. 30th IEEE Conference on Computer Communications(INFOCOM) mini-conference, 2011.. . 40 .

(41) 41 . [12] K. Zhu, W. Li, and X. Fu. Rethinking routing information in mobile social networks: Location-based or social-based? Elsevier Computer Communications (to appear), 2014. [13] W. Gao, Q. Li, B. Zhao, G. Cao Multicasting in delay tolerant networks: a social network perspective MobiHoc ’09: Proceedings of the 10th ACM International Symposium on Mobile ad hoc Networking and Computing, ACM, New York, NY, USA (2009), pp. 299–308 [14] N. Eagle and A. Pentland. Reality mining: sensing complex social systems. Personal and Ubiquitous Computing, Vol 10(4):255–268, May 2006. [15] P. Hui, People are the network: experimental design and evaluation of social-based forwarding. 治政大 [16] V. Srinivasan, M. Motani, and W. T. Ooi, “Analysis and implications of student contact patterns 立in Proc.ACM MobiCom, Los Angeles,CA, Sep.2006,pp.86–97. derived from campus schedules,” algorithms, Ph.D. dissertation, UCAM-CL-TR-713. University of Cambridge, Comp.Lab., 2008. ‧ 國. 學. [17] P. Hui, J. Crowcroft, and E. Yoneki. Bubble rap: social-based forwarding in delay tolerant networks. Proc. MobiHoc, pages 241–250, 2008.. ‧. [18] A. Chaintreau, P. Hui, J. Crowcroft, C. Diot, R. Gass, and J. Scott, “Impact of human mobility on the design of opportunistic forwarding algorithms,” in Proc. INFOCOM, April 2006.. Nat. sit. y. [19] T. Karagiannis, J.-Y. Le Boudec, and M. Vojnovic, “Power law and exponential decay of inter. er. io. contact times between mobile ´ devices,” in ACM MobiCom ’07, 2007. [20] J. Leguay, A. Lindgren, J. Scott, T. Friedman, and J. Crowcroft, “Opportunistic content. n. al. Ch. i n U. v. distribution in an urban setting,” in ACM CHANTS, 2006, pp. 205–212.. engchi. [21] Y. Zhang, W. Gao, G. Cao, T. L. Porta, B. Krishnamachari, and A. Iyengar, “Social-Aware Data Diffusion in Delay Tolerant MANET,” Handbook of Optimization in Complex Networks: Communication and Social Networks, 2010 [22] E. Daly and M. Haahr, “Social network analysis for routing in disconnected delay-tolerant manets,” in Proceedings of ACM MobiHoc, 2007. [23] Socievole, Annalisa, Floriano De Rango, and Antonio Caputo. "Wireless contacts, Facebook friendships and interests: Analysis of a multi-layer social network in an academic environment." 2014 IFIP Wireless Days (WD). IEEE, 2014. [24] Cabrero, Sergio, et al. "Understanding Opportunistic Networking for Emergency Services: Analysis of One Year of GPS Traces." Proceedings of the 10th ACM MobiCom Workshop on Challenged Networks. ACM, 2015. [25] https://github.com/NCCU-MCLAB/NCCU-Trace-Data . 41 .

(42) 42 . [26] A. Keränen, J. Ott, and T. Kärkkäinen. The ONE Simulator for DTN Protocol Evaluation. In Proceedings of the 2nd International Conference on Simulation Tools and Techniques, March 2009.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. . Ch. engchi. 42 . i n U. v.

(43)