Routing in Opportunistic Networks Chapter 1: Social-based Routing Protocols in Opportunistic Networks Ying Zhu and Yu Wang University of North Carolina at Charlotte © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 1 Outline Introduction Social Properties Social-based Routing Conclusion © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 2 Routing in Opportunistic Networks Intermittent Connectivity in OppNets “Store and Forward“ No connection available? Store & carry the data Make forwarding decision based on certain routing strategy © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 3 Routing in Opportunistic Networks OppNet Routing Strategies: Based on mobility pattern Unpredictable mobility High overhead Based on social characteristics Long term Less volatile Low overhead This chapter focuses on social-based routing © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 4 Outline Introduction Social Properties Social-based Routing Conclusion © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 5 Social Graph Social Graph: A global mapping of everybody and how they are related Vertices: people Edges: social ties Different social relationships, i.e. friends, co-workers Intuitive source for many social metrics Sometime is hard to directly obtain © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 6 Contact Graph Contact Graph: Recording contacts seen in the past Vertices: Mobile nodes which are carried by people Edges: One or more past meetings Indicate node’s relationships in OppNets People with close relationships tend to meet more often, more regular and with longer duration © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 7 Social Properties: Community Community: A group of interacting users Devices within same community have higher chances encounter each other © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 8 Social Properties: Community Community Detection Methods: Minimum-cut method Hierarchical clustering Girvan-Newman algorithm Modularity maximization The Louvain method Clique based method © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 9 Social Properties: Centrality Centrality: Social importance of its represented node in a social network Degree centrality The number of links upon a given node Betweenness centrality The number of shortest paths passing via given node Closeness centrality An inverse of node’s average shortest distance to all other nodes © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 10 Social Properties: Centrality Degree centrality a->3, b->4, others->1 Betweenness centrality a->18, b->24. others->0 Closeness centrality a->2/3, b->3/4, c/d/e->6/13,f/g->3/7 © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 11 Social Properties: Similarity Similarity: A measurement on degree of separation A simple way to define: Number of common neighbors between nodes in social/contact graph Similarity between a and c is 1 c and e is 3 © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 12 Social Properties: Friendship Friendship: Close personal/contact relationships In OppNets, friends may have: Long-lasting contacts Regular contacts Common interests Similar actions Different ways to define © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 13 Outline Introduction Social Properties Social-based Routing Conclusion © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 14 Label Routing Label Routing [Hui & Crowcroft, 2007] Small label for each node (its social group) Only forward messages to nodes which has same label with destination or directly to destination Requires little information Easy to implement Long delay © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 15 SimBet Routing SimBet Routing [Daly & Haahr, 2007] SimBet utility, a weighted combination of betweenness centrality and similarity Forward message to node with larger SimBet utility with destination © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 16 SimBet Routing SimBet uses local centrality & betweenness to reduce overhead may lead to inaccurate “bridge” identification Node u will not pass message to node a considers local SimBet utility © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 17 Bubble Rap Forwarding Bubble Rap Forwarding [Hui, Crowcroft, Yonek, 2008] global centrality: across whole network local centrality: within local community © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 18 Bubble Rap Forwarding Bubble-up on global centrality Forward message to the node with higher global centrality Until it reaches a node belongs to the same local community as destination Bubble-up on local centrality Use nodes within destination’s community as relays Choose the ones with higher local centrality When destination only belongs to communities whose members are all with low global centrality, BubbleRap may fail. © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 19 Social-Based Multicasting Social Based Multicasting [Gao, et al. 2009] Cumulative contact probability of node i: N is the total number of nodes in network T is the total time period λi,j is average contact rate of Possion process for node pair (i,j) © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 20 Social-Based Multicasting Single-data multicast Destinations are uniformly distributed All nodes need to be contacted within T Select minimal number of relay nodes Using cumulative contact probabilities Considered as unified knapsack problem Multi-data multicast Relay and destination in different communities: Forwarding via gateways (G1, G2) Relay and destination in same community: Same as single-data multicast © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 21 Homophily Based Data Diffusion Homophily Based Data Diffusion [Zhang & Zhao, 2009] When contact time too short or buffer is limited, need consider data propagation orders Friends usually share more common interests than strangers (Friendship is user defined) Diffuses the most similar data of their common interests to friend first Diffusing start from the data most different from their common interests to strangers © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 22 Friendship Based Routing Friendship Based Routing [Bulut & Szymanski, 2010] Social pressures metric(SPM) between i and j: f(t) denotes the remaining time to the first encounter of node i and j after time t T denotes the total time period Describes the average forwarding delay © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 23 Friendship Based Routing Link quality: An inverse of SPM Bigger link quality represents closer friendship Construct friendship community based on link quality Forward message to node in the same friendship community with destination Forward message to node with stronger friendship to destination than current node © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 24 Social-aware and Stateless Routing Social-aware and Stateless Routing (Sane) [Mei et al., 2011] People with similar interests tend to meet more often Interest profile for node u: K-dimensional vector Iu Cosine similarity: If cosine similarity betwween encounted node and destination is larger than a threshold, forward message Stateless & Scalable © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 25 User-Centric Data Disseination User-Centric Data Disseination [Gao & Cao, 2012] Interest profile of node i: Pij : prob. of user i interested in jth keyword A data item is described by the importance of ki Probability of node i interested in data D: © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 26 User-Centric Data Disseination Centrality value of node i for data dk at t≤Tk: Tk: Time constraint of data dk Ni: Set of nodes whose information is maintained by i Cij(Tk-t): Prob. of node i can forward dk to j within Tk-t Ci(k)(t): Expected number of interesters i can encounter during Tk-t © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 27 User-Centric Data Disseination Node i is selected as relay for data dk only if: NRk(t): The number of selected relays for dK at time t NIk(t): The number of interesters will receive dk by Tk , estimated at time t © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 28 Sociability-Based Routing Sociability Based Routing [Fabbri and Verdone, 2011] Sociability indicator: Evaluate node’s forwarding ability The node’s number of encounters with all other nodes in the network over a period T Nodes which frequently encounter many different nodes have high degree of sociability Good forwarder: Nodes with high sociability Forward packet to the most sociable node © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 29 Summary Social-based routing uses one or multiple social properties to make forwarding decision © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 30 Outline Introduction Social Properties Social-based Routing Conclusion © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 31 Conclusion Social-based approaches are promising for OppNets None of these approaches guarantee perfect routing performance Performance of routing protocol in OppNets depends heavily on mobility model, environment, node density, social structure, and many other facts Universal routing solution for all Oppnet application scenarios is extremely hard For particular Oppnet applications, specific routing protocols and mobility/social models are needed © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 32 Future Directions Are there new social characteristics better than existing ones? How to combine multiple social properties efficiently? How to model and extract accurate social characteristics in dynamic OppNets? How to combine social-based approaches with other type of routing stratigies? ... © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 33 Thanks for your attention! © Y. Zhu and Y. Wang @ University of North Carolina at Charlotte, USA 34