RIM: Relative-importance based data forwarding in people-centric networks

https://doi.org/10.1016/j.jnca.2015.12.007Get rights and content

Abstract

The fast penetration of mobile phones has arisen the requirement to share content (e.g., news, photo, music, video clips, etc.) among devices. To improve the efficiency of content sharing, recent works select nodes with high centrality in the system to cache and forward contents, resulting in a bias towards the most popular nodes. However, these nodes are not the appropriate candidates for target nodes, since the globally powerful nodes may have little influence on some local communities where the targets belong. Interestingly, we observe that nodes with low global centrality but high relative importance to the targets bear most weight on content allocation. Motivated by this observation, we exploit the relative importance of a node with respect to a group of nodes to guide the allocation process. We quantify the relative importance of nodes using graph spectrum theory, we then propose RIM (Relative IMportance), a novel data forwarding scheme to improve the allocation efficiency. By applying RIM on three real people-centric scenarios, the evaluation results show that RIM achieves significantly better mean delivery delay and cost than the state-of-the-art solutions, while achieving delivery ratios sufficiently close to those by Epidemic under different message TTL requirements.

Introduction

The first decade of this century witnessed a continual advancement in microelectronic technology, which enables vendors to integrate an increasing number of sensors (e.g., gyroscope, accelerometer and high resolution cameras) in mobile phones. Combining with the wireless communication functions, such as the Bluetooth and WiFi, the mobile phones open the door for a new sensing paradigm, participant or people-centric sensing (Lane et al., 2008, Ganti et al., 2011, Yuan and Liu, 2015), where users with smart phones form a mobile social network to collaboratively sense the things around them, and forward the sensory data for the purpose of sharing local contents of interest or improving global awareness of issues, etc.

Several impressive people-centric sensing applications have emerged in recent years. SignalGuru (Koukoumidis et al., 2011) is a system that uses windshield-mounted phones to sense current traffic signals, opportunistically communicate and predicate traffic signal schedule patterns. CoCam (Toledano et al., 2013) is a framework for smart phones that enables uncoordinated real-time multimedia (image/video) sharing between different mobile users. Other applications of people-centric sensing include SensOrchestra (Cheng et al., 2012), a collaborative sensing for symbolic location recognition, and CenseMe (Miluzzo et al., 2008), an interesting application for sharing personal sensing presence with their social graph, etc.

The success of the exciting applications requires an efficient content sharing mechanism. Unfortunately, the mobility of this paradigm makes it challenging to distribute contents among users, due to the lack of stable paths between node pairs. Early works with a store-carry-and-forward manner (Fall and Farrell, 2008, Conti and Kumar, 2010) would be of use in the above scenarios. This scheme exploits the contact of nodes to perform a peer-to-peer content exchange with others, and strives to achieve content delivery even when the connectivity of the network is intermittent. That is, when two nodes encounter each other, a data forwarding decision is generally made by adopting various heuristics, such as inferring the likelihood of forwarding the content (e.g., Lindgren et al., 2003, Balasubramanian et al., 2010, Gunawardena et al., 2011), employing the contact locations (e.g., Leguay et al. (2006)), or focusing on the contact frequencies (Erramilli et al., 2007). These solutions select relays and cache contents based on the prediction of physical contact metrics1 of nodes and neglect the impact of social contact metrics on content allocation. In fact, the sensing devices may lose connection when people move around (Augustin et al., 2007, Gao and Cao, 2010). Hence, the social contact metrics achieved by the complex network analysis (Marsden, 2002) capture the inherent characteristics of human mobility, and should be used to guide the allocation process.

Recently, there are a few attempts to explicitly make use of the social metrics to formulate the opportunistic forwarding decision. Among them, SimBet (Daly and Haahr, 2007), Bubble (Hui et al., 2011) and PeopleRank (Mtibaa et al., 2010) are three representative works. Although the detailed forwarding schemes may be different, all of them are motivated by the following two important observations from society: (1) people with closer relationship tend to reside in communities and (2) people within a community may have different popularity. As such, the increasingly “popular” or “central” nodes are more probably chosen as carriers to relay contents between disconnected communities (Mtibaa et al., 2010, Pietilainen and Diot, 2012), until a node belonging to the same community with the destination is reached (Daly and Haahr, 2007, Hui et al., 2011). Intuitively, information about community structure and node popularity enables them to outperform well-known opportunistic forwarding algorithms that are not explicitly “social-based”.

Nevertheless, all of the three protocols prefer to use global measures of node centrality (e.g., exploiting ego networks in Daly and Haahr, 2007, betweenness centrality in Hui et al., 2011 and PageRank Brin and Page, 1998 algorithm in Mtibaa et al., 2010), in that each node is ranked with respect to all other nodes in the network. We argue that these solutions paying more attention to the whole network topology and neglecting local dynamic behaviors miss many interesting features. This is mainly because: (1) In real world it is found that people have different average number of contacts in different social cliques. The same person in one clique may be sociable, having many contacts with others, while in another clique he/she may be more taciturn. Such dynamic behaviors have also been seen in mobile social networks, where people are more frequently to contact their friends, while they meet accidentally with strangers (Yuan et al., 2013). If one tried to characterize such a network by statistics of the total number of contacts a person has, one would be missing the features of the network, such as the dynamics of epidemic diffusion, resulting in a biased understanding of the content delivery process; (2) To validate whether the popular nodes or the relative important nodes bear most weight on content delivery, we observe their roles in content dissemination by gradually removing the same amount of the two types of nodes, respectively. Figure 1 shows some interesting results at NCSU scenario (NCSU is a dataset we used in this paper, please refer to Section 4.1). The first one is that the relative important nodes have a big impact on the mean delivery delay. Removing them increases the transmission delay almost by 25%. This finding contradicts the previous results (Daly and Haahr, 2007, Hui et al., 2011, Mtibaa et al., 2010). The second one is that the removal of the relative important nodes leads to a sudden, sharp increase in transmission delay conditioned on the source and destination within a community (as shown in Fig. 1(b), when the fraction of removing relative important nodes varies from 25% to 30%, the transmission delay increases suddenly), while removing the popular nodes only results in a relatively stable increase in transmission delay. As a result, the relative importance metric provides fine-grained relations among nodes. It therefore is helpful to make informed forwarding decisions (e.g., a node is just a desired relay, if it exhibits a highly relative importance to the destination׳s community partners).

To do so, we first employ the theory of graph spectrum to measure the relative importance metric. We then develop a data forwarding scheme by jointly considering the relative importance and the community structure, to improve the efficiency of content sharing. We summarize our contributions as follows:

  • We evaluate the performance of data forwarding based on the relative importance metric. To the best of our knowledge, this is the first attempt to integrate this new social metric into data forwarding.

  • We propose an online method to compute node׳s relative importance, which makes our work more applicable. We also detect the overlapped community structure by effectively distinguishing the bridging nodes from other nodes, and exploit the community structure to label the community partners of destination.

  • We observe that the community structure heavily impacts the message dissemination, and there exists obvious “phase transition” in message diffusion process. We simultaneously analyze nodes׳ roles in message diffusion, and observe that the “bridging nodes” play a bigger role in message dissemination than other nodes׳ categories.

  • We formulate the strength of relationship between nodes as a Decayed Sum Problem (Cohen and Strauss, 2006), and use a Decayed Aggregation Graph (DAG) to model the dynamic of network topology.

  • We implement RIM and compare it to several state-of-the-art works through three real opportunistic networking scenarios. Our extensive evaluation results show that, overall, the RIM outperforms other solutions, especially in terms of mean delivery delay and cost. For example, it achieves up to a 70% improvement in mean delivery delay over Prophet (Lindgren et al., 2003) and 40% over Bubble (Hui et al., 2011), and has a reduction of cost by up to 2 and 3 factors compared to Bubble and Prophet respectively in the same scenario.

The remainder of this paper is organized as follows. Section 2 overviews the problem and network model. Section 3 describes the forwarding scheme. In Section 4, we make performance evaluation. After briefly reviewing the related work in Section 5, we draw our conclusions in Section 6.

Section snippets

Centrality and relative importance metrics

Node centrality measures the global importance of a node relative to all other nodes in the network (i.e., how popular a person is within a social network). In this paper, we propose and investigate a related but somewhat different social metric, called node relative importance, which is defined as follows.

Definition 1

(Relative Importance) Node relative importance refers to the extent to which a node belongs to a community.

Freeman (Freeman, 1979) proposed three most widely used methods to estimate node

Relaying algorithm

In this section, we first present our data forwarding scheme RIM in Section 3.1, and then discuss how to quantify the relative importance metric and detect the overlapped community structure in 3.2 Evaluating relative importance, 3.3 Detecting the overlapped community structure, respectively.

Data-sets and experimental results

We first analyze the overlapped community structures underlying the data-sets we used, and then compare the performance of RIM with two state-of-the-art works: Bubble and Prophet together with the Epidemic algorithm as benchmark. Bubble is a well-known social-based forwarding algorithm and Prophet is currently an IETF draft (DTN Research Group, 2012). Results of Epidemic algorithm provide us the upper and lower bounds of important performance evaluation metrics: mean delivery delay, cost and

Related work

Opportunistic forwarding scheme provides an efficient solution to achieve content sharing in people-centric networks. A lot of data forwarding algorithms have been proposed in the past few years. We classify them into the following two categories based on the contexts they used.

Conclusion

In this paper, we deal with the content sharing issue in people-centric sensing networks. By analyzing the roles of different node categories in content dissemination, we find that nodes with low global centrality but high relative importance to the target nodes bear most weight. Motivated by this observation, we propose RIM, a relative-importance based data forwarding algorithm, to improve the efficiency of content allocation. We first formulate the strength of relationship between nodes as a

Acknowledgement

We acknowledge the support of the National Natural Science Foundation of China under Grant no. U1404602, the Science and Technology Foundation of Henan Educational Committee under Grant no. 14A520031, the Dr. Startup Project of Henan Normal University under Grant no. qd14136 and the Young Scholar Program of Henan Normal University. We also wish to thank the reviewers for their valuable comments.

References (57)

  • P. Yuan et al.

    Hotspot-entropy based data forwarding in opportunistic social networks

    Pervasive Mob Comput

    (2015)
  • U.G. Acer et al.

    Timely data delivery in a realistic bus network

    IEEE Trans Veh Technol

    (2012)
  • C. Augustin et al.

    Impact of human mobility on the design of opportunistic forwarding algorithms

    IEEE Trans Mob Comput

    (2007)
  • A. Balasubramanian et al.

    Replication routing in DTNsa resource allocation approach

    IEEE/ACM Trans. Netw

    (2010)
  • Betsy G, Shashi S. Time-aggregated graphs for modeling spatio-temporal networks. In: Proceedings of ACM CoMoGIS; 2006....
  • Burgess J, Gallagher B, Jensen D, Levine BN. MaxProp: routing for vehicle-based disruption-tolerant networks. In:...
  • Cheng HT, Sun FT, Buthpitiya S, Griss M, SensOrchestra: collaborative sensing for symbolic location recognition. In:...
  • F.R.K. Chung

    Spectral Graph Theory

    (1997)
  • A. Clauset et al.

    Finding community structure in very large networks

    Phys Rev E

    (2004)
  • M. Conti et al.

    Opportunities in opportunistic computing

    Computer

    (2010)
  • Daly E, Haahr M. Social network analysis for routing in disconnected delay-tolerant MANETs. In: Proceedings of ACM...
  • Ding C, He, X. K-means clustering via principal component analysis. In: Proceedings of ACM ICML; 2004. p....
  • DTN Research Group. Extensions of probabilistic routing protocol for intermittently connected networks. Downloaded from...
  • Erramilli V, Chaintreau A, Crovella M, Diot C. Diversity of forwarding paths in pocket switched networks. In:...
  • Erramilli V, Crovella M, Chaintreau A, Delegation forwarding. In: Proceedings of ACM MobiHoc; 2008. p....
  • K. Fall et al.

    DTN: an architectural retrospective

    IEEE J Sel Areas Commun

    (2008)
  • R.K. Ganti et al.

    Mobile crowdsensing: current state and future challenges

    IEEE Commun Mag

    (2011)
  • Gao W, Cao G, Fine-grained mobility characterization: steady and transient state behaviors. In: Proceedings of ACM...
  • Cited by (9)

    • A trustworthiness-enhanced reliable forwarding scheme in mobile Internet of Things

      2019, Journal of Network and Computer Applications
      Citation Excerpt :

      More than one thousand people daily were attracted for 2 weeks as the event was very popular. The site is particularly small and each participant spent less than 3 h. Several studies, e.g., (Kaveevivitchai and Esaki, 2010; Rallapalli et al., 2010; Yuan et al., 2016), have applied it into different research fields to explore its characteristics such as intra/inter contact distribution. In this experiment on the State Fair Dataset, we choose the source-destination pairs randomly.

    • MATEM: A unified framework based on trust and MCDM for assuring security, reliability and QoS in DTN routing

      2018, Journal of Network and Computer Applications
      Citation Excerpt :

      These assumptions, however, restrain the behavior of DTNs where connectivity is intermittent and communication is opportunistic in nature. Again some social-aware routing protocols (Boldrini et al., 2010; Gao et al., 2012; Ciobanu et al., 2014b; Jedari and Xia, 2013; Xiao et al., 2014; Singh et al., 2016; Yuan et al., 2016) have been made available in the literature for message forwarding in opportunistic DTNs. These protocols have been developed from a social networking perspective and try to optimize the social characteristics of mobile users for message forwarding.

    • Routing strategy based on centrality in opportunistic networks

      2019, Proceedings of the 2019 6th International Conference on Computing for Sustainable Global Development, INDIACom 2019
    • A Social-Based DTN Routing in Cooperative Vehicular Sensor Networks

      2018, International Journal of Cooperative Information Systems
    View all citing articles on Scopus
    View full text