Routing for context-addressable messages using Predefined Concept Connection Map in delay tolerant network

Abstract

This article presents methods of nodes discovery and message routing for a context-based addressing in Delay and Disruptive Tolerant Networks. The context-based addressing is the novel way of descriptive addressing that is based on ontology model. The context-based addressing enables description of each node by a set of concepts. The main idea of this article is to present the method of the addressees discovery that uses specialized graph composed of concepts. It is called Predefined Concept Connection Map. The Predefined Concept Connection Map describes physical communication abilities between context-based addressed nodes. In this article, ns-2 simulations are presented. They are based on specially designed simulation model which supports context-based address assigned for each Delay and Disruptive Tolerant Network node.

Keywords

Delay and disruptive tolerant networks wireless ad hoc networks message routing wireless communications predefined context model concept maps

Introduction

Nowadays, Delay and Disruptive Tolerant Networks (DTN)¹ are dynamically explored. The DTN are characterized by a changeable structure of the network. The most important property of DTN is that the path between the message source and the message destination usually does not exist due to the permanent network partitions. The communication in the DTN network is done by ferrying messages between separate sub-networks based on mobility of nodes. Each node tires to exchange messages when the opportunity of connectivity with other networks exists. This style of working is described as a store-carry-and-forward paradigm. The nodes which are storing and moving messages are called message ferries or mules.

Those mechanisms, which are relatively simple and well known to the old-fashioned human environment like post or courier services, cannot be easily implemented in an environment of highly mobile wireless networks. Although the technology progress is very fast, the communication and nodes storing capability are still an important problem to be solved. Therefore, the proper routing algorithm which saves restricted node resources is fundamental for the DTN network creation.

Due to acceptable disruptions in the structure of the network, the DTN may be used in many possible applications. It can be applied in rescue and emergency situations, where traditional communication infrastructure is unserviceable, for example, in earthquake which breaks all wired connections, in small-populated and hardly accessible terrains, in railway and metro tunnels, or in the modern warfare theatre. The small size of the node makes it possible to be carried by rescuers or other incident participants.

Although the nature of applications scenarios is usually very dynamic, the behaviour of participants is predictable and often repetitive.² For this reason, almost all emergency situations are handled in a similar way. This means that the nodes-actors behaviour can be formalized in a form of procedures and described in details. In the next step, this description can be formed as a predefined formal behaviour model.

The spontaneous, changeable, and disruptive structure of the DTN favours the applicability of context-based addressing. Recently, many works, especially in sensor networks area, have been focused on descriptive-based methods of nodes addressing.^3,4 The idea of this addressing style is to describe dynamically addressees by expressing their properties in a descriptive way. It can characterize not only node attributes but also a whole environment. The address description can be easily specified at the source using formalized keywords like ‘Firefighter in the Tunnel’, ‘Victim at the Medical concentration point’, ‘Rescuer at the Incident place’, and so on. An example of this address is presented in Figure 1.

Figure 1.

Simple example of context-based address.

The basic assumption of context-based address is that each node in the network has access to a common context model. The context model for address generation is an ontology. Each of elementary parts of the address belongs to this ontology. Each address is constructed by combining of class, relations, and individuals. The rules of context-based address construction and resolving were presented in our previous works by Domaszewicz et al.³ and Koziuk et al.⁴

In our work, the routing is done by handling classes. The classes are equivalent of all elementary concepts, and these are structured in the predefined model. The model is based on one selected property. For the DTN-routing purpose, this property is characterized as the ability of the connection between different nodes semantically described by elementary concepts.

The model is created a priori based on observation of roles connectivity between nodes. The structure of the model is not variable during entire life of the network and fully reflects the communication relationships between concepts.

The major contribution of this article is the algorithm for addressees discovery and context-based addressed messages routing in DTN using specialized Predefined Concept Connection Map (PCCM). The proposed algorithm will be referred to as Concept-Based Routing for DTN (CBR-DTN) in the remaining part of the article.

Related work

Although presented algorithm is novel, some of its aspects are related to the existing solutions. Although the main DTN-routing strategies were designed for identifier-based addresses, some algorithms are important background to this solution. The most simple solution, often used as a reference, is called Epidemic Routing (ER).⁵ In ER, message routing is done by forwarding messages to every node, for which the message is new. Although ER is very reliable, the solution is ineffective because of the huge number of stored copies of one message. On the other hand, some solutions try to restrict the number of exchanged message copies. One of the reference solutions’ group, which is based on one copy and simple random forwarding policy, is called Random Walk (RW).⁶ Sometimes the number of copies can also be restricted to some copies, which are multiplied in the source of the message and forwarded separately.⁷

The second aspect related to this solution is the nodes mobility modelling and prediction and social properties of the nodes used in DTN routing. The models of nodes mobility are important for forwarding messages in DTN network. In real scenarios, ferry nodes (robots or humans) are mobile in an ordered and non-random way. Therefore, the nodes routes awareness can be helpful in effective routing strategies creation. Many works are related to the creating routes of mobile nodes,⁸ some algorithms are worked using geographical positioning,^9,10 while others are based on nodes mobility observations as changeable graph.^11,12 Solutions based on social properties of the nodes go further. Social properties are describing the social abilities of the nodes: the number of neighbours, the periodical communication opportunities, and so on. The social relations are used in many routing-addressed works like Sarafijanovic-Djukic et al.,¹³ Leguay et al.,¹⁴ Burns et al.,¹⁵ and Yoneki et al.,¹⁶ and some of them are concentrating on the context and the mobility prediction.^17–21 In this area, active research is made on spatial–temporal models that describe the distribution of nodes in space and time, such as moving individual nodes in the network.²² Some works are related to the social relations of nodes.^23,24 Major group of models focuses on the most accurate fidelity devotion mobility of people, especially in the context of space. An example of this is the work based on the collection of predefined realistic locations as in Hsu et al.,²⁵ or meetings at certain times resulting from the schedule of human activity,^26,27 sometimes extending this in the context of human social behavior.²⁸ The advanced modelling approaches which merge the complex properties of human mobility and sociality^29,30 were made based on this assumption. All these solutions are related to the proposed CBR-DTN because of nodes behaviour modelling. In opposite to the proposed CBR-DTN, in these works, the mobility modelling and behaviour prediction are based mainly on formalized mathematical models. None of the known solutions used semantic models for message routing; in specific, they did not use the idea of using elementary concepts model as context description. Therefore, CBR-DTN routing idea should be considered as novel in DTN networks.

The context-based addressed messages routing

PCCM

The usage of DTN is usually restricted to specific applications. The scenarios in specific applications are very often similar. Moreover, the behaviour of most of the actors (like rescuers, victims, auxiliary stuffs, and spectators) is very similar as well. Based on previous scenarios analyses, this property allows developing a specialized domain-specific context model, for example, for emergency scenarios in tunnels and urban disaster accidents.

One of the main assumptions for the context model formulation is that the model is composed of elementary units called concepts. The concept is a representation of the context snippet. The nodes are described using at least one concept, and there is no limitation to the number of nodes described by the same concept.

Each elementary concept can be structured in the model, based on the real nodes characteristics. In DTN networks, the crucial characteristic of a node is its connection ability. The connection ability is defined as a neighbouring node that can be connected to this node during its movement. Assuming specific behaviour of the nodes, it is possible to define a map-model. The map-model includes all elementary concepts that represent the node and are linked with concepts that describe nodes in the neighbourhood.

For example, during fire incident in the railway tunnel, the ‘Firefighter’ and the ‘Rescuer’ are working at the ‘Incident place’. They are both fighting with fire and rescuing the injured ‘Victim’. During the rescue actions, the ‘Victim’ is transported to the ‘Medical Point’ as soon as possible, where the ‘Victim’ is waiting for specialized help. The help is provided by ‘Doctor’. All elementary concepts and corresponding ‘in touch’ communication ability can be formed in the model called PCCM. This simple scenario and corresponding PCCM are presented in Figure 2.

Figure 2.

The simple example of the scenario based on emergency procedures in Central Europe and the corresponding Predefined Concept Connection Map model.

In case many concepts are describing one node (e.g. ‘The Firefighter at the Incident Place’, that is, an address composed of two concepts ‘Firefighter’, ‘Incident Place’), each concept is described in PCCM separately and is linked with other concepts based on the same rules as in the case of separate nodes.

The presented model is very useful in routing decisions of context-based addressed messages due to its simplicity. The main idea of message routing is to select the best message ferry based on abstract PCCM. The algorithm is triggered when new message is received by node or when there is a new node detected in the neighbourhood. In the first step of the algorithm, the destination context-based address is decomposed into elementary concepts and described as the destination set. The location of the destination set of elements is identified in PCCM, and the shortest paths between destination set and the current node context description are specified.

The main task of the message routing algorithm of each node is to select right concepts at the shortest path. These concepts should be in-tough in the PCCM to the node’s own description. In the next step, selected concepts are used in routing table searches of physically available neighbouring ferry nodes described by these concepts. If a result of the search is positive, the message is forwarded to the neighbouring node. If the result is negative, the message is stored in the current node.

The simple example of the message routing is presented in Figure 2. The ‘Rescuer at the Incident Place’ sends message to the ‘Doctor in the Incident Place’. The algorithm will work as follows: at first moment, the message will be forwarded to node described as ‘Victim’. Due to lack of connectivity with other nodes, the message will be stored at this node until the victim is moved to the ‘Medical Point’. The node corresponding to the ‘Medical Point’ is closer to the destination address in PCCM, so the message will be forwarded to this node. If the connectivity with ‘Doctor’ occurs, the message will be forwarded to him. The algorithm of the context-based ferries selection and message routing is presented in Algorithm 1.

Algorithm 1.

CBR-DTN Routing Algorithm.

CBR_DTN_ALGORITHM(addr)

2. AFL ←∅, BFL ←∅;

3. my_metric ← MIN_DISTANCE_BETWEEN(addr, myAddr);

4. N ← GET_NEIGHBOUR_CONCEPTS(myAddr);

5. for each n ∈ N ∧ WAS_NOT_READ(n) do

6. rel_metric ← MIN_DISTANCE_BETWEEN(addr, n);

7. if rel_metric < my_metric then

8. BFL ← BFL ∪ GET_BEST_FORWARDER(n);

9. else if rel_metric = my_metric then

10. AFL ← AFL ∪ GET_BEST_FORWARDER(n);

11.

12.if AFL = ∅∧ BFL = ∅ then

13. ADD_TO_BUFFER(bundle);

14.else if BFL ≠∅ then

15. for each forwarder ∈ BFL do

16. SEND_MESSAGE(forwarder);

17.else

18. ADD_TO_BUFFER(bundle);

19. for each forwarder ∈ AFL do

20. SEND_MESSAGE(forwarder);

In detail, the algorithm looks as follows:

In the first step of the algorithm,

Two lists of forwarding nodes are initiated:

BFL: best forwarding nodes list for available neighbouring nodes which are ‘better then itself’ at the shortest path at the PCCM;

AFL: auxiliary candidate for forwarding nodes list contains the nodes which are neighbours and have the same metric counted in number of concepts in PCCM as the current node does.

There is counted metric between concepts of current node and the destination address;

There is a selected set of concepts which are neighbouring to the current node.

In the second step of algorithm, BFL and AFL sets are filled based on local metrics between concepts. The my_metric represents the minimal distance between addresses of the current node and destination, and the rel_metric is recounted distance value between destination and neighbour. After finishing BFL and AFL filling process, the storing and forwarding decision is made. If there are no candidates to forward, the message is stored. If there are better candidates, the message is sent. If there are candidates in BFL, the message is doubled, one is sent, and the other is stored.

Routing maintenance

One of the tasks of routing maintenance is to provide information necessary for routing decisions such as neighbours’ availability and the context changes. The information is obtained using periodical broadcast messages sent by each node. Every message contains the following: its own context description as set of elementary concepts, list of available neighbours, as well as their concepts. Any changes in routing tables or in context descriptions are the reason for triggering a routing process for stored messages. Each stored message is reviewed and the decision on forwarding and/or storing is made.

Simulation results

The simulations were made using the ns-2³¹ simulation tool with CMU-Monarch wireless extension. For the DTN purpose, a specialized dedicated scenario was proposed. The main idea of the dedicated scenario was to achieve disruption and partly nodes isolation in the defined area. Therefore, the simulation plane was divided into equal square pieces called sub-planes. Each sub-plane had a randomly selected unique context description and contained only nodes which were described by these concepts. Due to the network disruption requirement, the proposed size of each sub-plane was bigger than the nodes radio range. For all nodes within each sub-plane, random-waypoint mobility model (RWM) was assumed. The location of destination and source nodes was chosen randomly. All simulation parameters are presented in Table 1.

Table 1.

Simulation parameters.

Parameter	Value
Simulation time	7200 s
Number of sub-planes {X, Y}	{3 × 3, 4 × 4, 5 × 5, 6 × 6, 7 × 7, 8 × 8}
Size of each sub-plane	250 m × 250 m, 375 m × 375 m, 500 m × 500 m
Radio range	250 m
Mobility model	RWM restricted to within each sub-plane
Vertical speed	0 up to 10 m/s
Number of nodes in the sub-plane	1
Number of destination nodes	5
Number of sources	1
Number of simulations per one measure point	50
Confidence level	95%

RWM: random-waypoint mobility model.

The simulations were done on the basis of two different variants of proposed message routing algorithm. The first one, called CBR-DTN, was based on the best delivery ratio. The second, called CBR-DTN-S, was optimized on minimization of the total storage memory usage ratio. The object of CBR-DTN was to store and forward messages to neighbouring nodes which are equal or closer to destinations (greater number of nodes which are storing the message, and therefore higher fast delivery probability), while the CBR-DTN-S preferred only closer nodes (less number of engaged nodes, but smaller delivery probability).

Both algorithm versions were tested on the basis of two different types of PCCM model. First of them is called ‘complex’. It maps the simulated concept grid with both connections, on edges and corners of each rectangular sub-plane. The second model, called ‘simple’, provides only connections on the edges. The simple example of automatically generated simulation plane – the grid of nodes described by a single elementary concept (for simplicity letters) and corresponding ‘complex’ and ‘simple’ PCCM models is presented in Figure 3.

Figure 3.

The example of simulation plane and the ‘complex’ and the ‘simple’ PCCM models.

CBR-DTN and CBR-DTN-S were compared with two other solutions. In order to compare them, the most popular and very simple ER⁵ and a single-copy non-sophisticated routing called RW⁶ were chosen.

The main goal of simulations was to verify the characteristic of the addressees discovery and of the context-based messages forwarding algorithm. Therefore, different metrics that verify routing algorithm characteristics were selected. Metrics are based on (a) nodes count as an equivalent of computing cost and effectiveness, (b) time count as a metric of communications effectiveness, and (c) packet count as energy consumption cost and communication effectiveness. The minimal values of metric are better.

The leading metric is based on number of engaged nodes during the addressees discovery phase related to addressees discovered, that is, ‘Flooded/Discovered’; the metric presents a cost of discovering one node related to number of nodes which are engaged in discovering. The best performance is when the number of nodes flooded is the smallest and when the number of nodes discovered is the highest; therefore, the best solution should perform the smallest results. The simulation results of our algorithms are presented in Figure 4.

Figure 4.

The ratio of nodes flooded to addressees discovered for different sizes of sub-networks.

Results presented in Figure 4 are related to ER and RW. The observations show that for all dimensions of sub-planes and for all network sizes, the performance of proposed variants of CBR-DTN algorithm is better than in case of comparable solutions. The values of the metric are growing proportionally to the number of the nodes in the network, but the spread is related to the routing strategy. The best performance of CBR-DTN results from the number of shortest paths in PCCM. The ER algorithm engages all nodes available in the network, and therefore it is ineffective. The ineffectiveness of the RW solution is caused by the unidirectional random strategy.

The second metric is the number of nodes in which the message is stored, divided by the total number of nodes in the network called ‘Number of ferries/all nodes’ ratio (see Figure 5). The metric presents the overall cost of message routing counted as the number of engaged nodes. It has been observed that the performance of the CBR-DTN algorithm depends on the PCCM model. The best performance (the lowest value) was achieved using the more detailed ‘complex’ PCCM for both versions of CBR-DTN and CBR-DTN-S. The relative values of the ratio decrease because of disproportion of the slowly increasing number of ferry nodes related to the faster increasing total number of nodes in the network per each simulation point.

Figure 5.

The message ferry nodes to all nodes in the network for different network dimensions.

Both the minimal number of ferry nodes and the most proper selection decision have influence on the total time of message delivery (see Figure 6). The total time depends not only on the nodes’ mean vertical speed and communication radius but also on the shortest path of used ferries between message source and destinations. Therefore, the mean time of delivering messages by CBR-DTN and CBR-DTN-S was very similar to ER, which in this context is the best possible solution. The results of delivery time significantly depend on communication radius. The less the radius, the more significant will be the node speed, and the delivery time will increase.

Figure 6.

Mean total time of message delivery to addressees.

The last metric is presented in Figure 7. The metric is based on the total number of data and maintenance packets sent in the network, that is, ‘Data and Maintenance packets sent’. The metric presents the communications cost, as well as, indirectly, energy effectiveness related to the number of data messages sent per each second. The effectiveness of CBR-DTN and CBR-DTN-S grows along with frequency of the number of data messages sent.

Figure 7.

Data and maintenance packets sent.

Conclusion

This article presented the specialized PCCM idea and addressees discovery and routing for descriptive addressing like context-based addressing in DTN. The proposed algorithm with all versions was extensively analysed using the ns-2 simulation tool. Simulation results show that the proposed algorithm has a better performance than in case of comparable solutions (ER and RW), and it is very attractive in DTN environment.

In this article, we focused on the presentation of the PCCM idea for messages routing in DTN network. We skipped considerations for complex context-based address resolution by reasoner utilities in mobile devices. This was the subject of our previous work.^3,4,32 Our PCCM model is defined a priori. The more accurate model will result in the better performance. The more accurate model has its consequences in a significant and non-linear increase in its size. It is obvious that this results in a significant increase in computational load. However, we believe that due to the rapid growth of mobile device performance, solutions that now look complex and slow will work very well in the near future.

Footnotes

Academic Editor: Leandro Villas

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship and/or publication of this article.

References

Fall

. Messaging in difficult environments. Technical report IRB-TR-04-019, December 2004, Intel Corp.

Gonzlez

Hidalgo

Barabsi

A-L

. Understanding individual human mobility patterns. Nature 2008; 453: 779–782.

Domaszewicz

Koziuk

Schoeneich

. Context-addressable messaging with ontology-driven addresses. In: ODBASE 2008, Monterrey, Mexico, 11–13 November 2008, vol. 5332, pp.1470–1480. Heidelberg: Springer.

Koziuk

Domaszewicz

Schoeneich

. Mobile context-addressable messaging with DL-Lite domain model. In: EuroSSC 2008, Zurich, 29–31 October 2008, vol. 5279, pp.168–181. Heidelberg: Springer.

Vahdat

Becker

. Epidemic routing for partially connected ad hoc networks. Technical report CS-2000-06, 2000, CS Department, Duke University, Durham, NC.

Dhillon

Van Mieghem

. Comparison of random walk strategies for ad hoc networks. In: Med-Hoc-Net 2007, Corfu, 13–15 July 2007, pp.196–203.

Spyropoulos

Psounis

Raghavendra

. Spray and wait: an efficient routing scheme for intermittently connected mobile networks. In: WDTN ‘05. Proceeding of the 2005 ACM SIGCOMM workshop on delay-tolerant networking, Philadelphia, PA, 22–26 August 2005. New York: Association for Computing Machinery.

Zhao

Ammar

Zegura

. Controlling the mobility of multiple data transport ferries in a delay-tolerant network. In: INFOCOM 2005. Proceedings of the IEEE 24th annual joint conference of the IEEE computer and communications societies, Miami, FL, 13–17 March 2005, pp.1407–1418. New York: IEEE.

Lai

Liu

. DTGR: disruption-tolerant geographic routing for wireless ad hoc networks. Simulation 2006; 82: 399–411.

10.

Lebrun

Chuah

C-N

Ghosal

. Knowledge-based opportunistic forwarding in vehicular wireless ad hoc networks. In: 2005 IEEE 61st vehicular technology conference VTC ’05, Stockholm, 30 May–1 June 2005,vol. 4, pp.2289–2293. New York: IEEE.

11.

Jain

Fall

Patra

. Routing in a delay tolerant network. In: Proceedings of the 2004 conference on applications, technologies, architectures, and protocols for computer communications, Portland, OR, 30 August–3 September 2004, vol. 34. pp.145–158. New York: Association for Computing Machinery.

12.

Zhao

. Socially-aware publish/subscribe system for human networks. In: 2010 IEEE wireless communications and networking conference, Sydney, NSW, Australia, 18–21 April 2010. New York: IEEE.

13.

Sarafijanovic-Djukic

Piorkowski

Grossglauser

. Island hopping: efficient mobility-assisted forwarding in partitioned networks. In: 2006. SECON ‘06. 2006 3rd annual IEEE communications society on sensor and ad hoc communications and networks, Reston, VA, 28–28 September 2006, vol. 1. pp.226–235. New York: IEEE.

14.

Leguay

Friedman

Conan

. Evaluating mobility pattern space routing for DTN’s. In: WDTN ‘05. Proceeding of the 2005 ACM SIGCOMM workshop on delay-tolerant networking, Philadelphia, PA, 22–26 August 2005. pp.276–283. New York: Association for Computing Machinery.

15.

Burns

Brock

Levine

. MV routing and capacity building in disruption tolerant networks. In: Proceedings IEEE 24th annual joint conference of the IEEE computer and communications societies, Miami, FL, 13–17 March 2005, pp.398–408. New York: IEEE.

16.

Yoneki

Hui

Chan

S-Y

. A socio-aware overlay for publish/subscribe communication in delay tolerant networks. In: Proceedings of the 10th ACM symposium on modelling, analysis, and simulation of wireless and mobile systems, Chania, 22–26 October 2007, pp.225–234. New York: Association for Computing Machinery.

17.

Musolesi

Hailes

Mascolo

. Adaptive routing for intermittently connected mobile ad hoc networks. In: Proceedings of the sixth IEEE international symposium on world of wireless mobile and multimedia networks, Giardini Naxos, 13–16 June 2005, pp.183–189. New York: IEEE.

18.

Musolesi

Mascolo

. Evaluating context information predictability for autonomic communication. In: Proceedings of the 2006 international symposium on world of wireless, mobile and multimedia networks, 2006, pp.495–499, https://www.cl.cam.ac.uk/∼cm542/papers/acc2006.pdf

19.

Boldrini

Conti

Passarella

. Exploiting users’ social relations to forward data in opportunistic networks: the HiBOp solution. Perv Mobile Comput 2008; 4: 633–657.

20.

Cuevas

Jaho

Guerrero

. OnMove: a protocol for content distribution in wireless delay tolerant networks based on social information. In: ACM CoNEXT 2008 student workshop, Madrid, 2008, http://cgi.di.uoa.gr/∼istavrak/publications/OnMove_camera_ready.pdf

21.

Costa

Mascolo

Musolesi

. Socially-aware routing for publish-subscribe in delay-tolerant mobile ad hoc networks. IEEE J Sel Area Comm 2008; 26: 748–760.

22.

Song

Koren

Wang

. Modelling the scaling properties of human mobility. Nat Phys 2010; 6: 818–823.

23.

Chaintreau

Hui

Crowcroft

. Impact of human mobility on opportunistic forwarding algorithms. IEEE T Mobile Comput 2007; 6: 606–620.

24.

Gao

Zhao

. Multicasting in delay tolerant networks: a social network perspective. In: Proceedings of the tenth ACM international symposium on mobile ad hoc networking and computing MobiHoc ’09, New Orleans, LA, 18–21 May 2009, vol. 6, pp.299–308. New York: Association for Computing Machinery.

25.

Hsu

W-J

Spyropoulos

Psounis

. Modelling spatial and temporal dependencies of user mobility in wireless mobile networks. IEEE/ACM T Netw 2009; 17: 1564–1577.

26.

Ekman

Keranen

Karvo

. Working day movement model. In: Proceedings of the 1st ACM SIGMOBILE workshop on mobility models, mobility models’08, Hong Kong, China, 26–30 May 2008, pp.33–40. New York: Association for Computing Machinery.

27.

Zheng

Hong

Liu

. Agenda driven mobility modelling. Int J Ad Hoc Ubiq Co 2010; 5: 22–36.

28.

Boldrini

Passarella

. HCMM: modelling spatial and temporal properties of human mobility driven by users social relationships. Comput Commun 2010; 33: 1056–1074.

29.

Karamshuk

Boldrini

Conti

. SPoT: Representing the social, spatial, and temporal dimensions of human mobility with a unifying framework. Perv Mobile Comput 2014; 11: 19–40.

30.

Karamshuk

Boldrini

Conti

. Human mobility models for opportunistic networks. IEEE Commun Mag 2011; 49: 1–18.

31.

NS2 Project, 2014, http://www.isi.edu/nsnam/ns/

32.

Schoeneich

Domaszewicz

Koziuk

. Concept-based routing in ad-hoc networks. In: Proceedings of the 10th international conference on distributed computing and networking (ICDCN ‘09) (ed. Garg

Wattenhofer

Kothapalli

), 2009, pp.43–48, Berlin; Heidelberg: Springer-Verlag, http://dx.doi.org/10.1007/978-3-540-92295-7_8