Comparative Study of Routing Protocols for Opportunistic Networks

Transcription

1 Comparative Study of Routing Protocols for Opportunistic Networks Majeed Alajeely School of Information Technology Deakin University Melbourne, Australia Asma a Ahmad School of Information Technology Deakin University Melbourne, Australia anahmad@deakin.edu.au Robin Doss School of Information Technology Deakin University Melbourne, Australia robin.doss@deakin.edu.au Abstract Opportunistic networks or OppNets refer to a number of wireless nodes opportunistically communicating with each other in a form of Store-Carry-Forward. This occurs when they come into contact with each other without proper network infrastructure. In OppNets there is no end-to-end connection between the source node and the destination node. OppNets grow from a single node (seed) to become large networks by inviting new nodes (helpers) to join the network. Due to these characteristics, OppNets are subject to real routing challenges. In this paper, we have presented an overview of the main available three families of OppNet routing protocols. Further, we have evaluated one protocol from each family (Epidemic, Direct Delivery and PRoPHET) in terms of complexity and scalability. Simulation results show that for small and medium complexity, the three protocols perform better than large complexity. As for scalability, simulation results show that Epidemic and PRoPHET perform better than Direct Delivery in terms of delivery rates and delays, but at a very high cost while Direct Delivery achieved lower delivery rates with a low cost. Keywords: Opportunistic Networks, Routing Protocols, Complexity, Scalability. I. INTRODUCTION Opportunistic networks or OppNets aim to establish reliable networks where there is no end-to-end connection between the source and destination nodes. These nodes usually have high mobility, low density, limited power and short radio range. With the wide expansion of wireless mobile devices (nodes), the applications of OppNets are growing fast in different fields and include disaster recovery, military deployment, and wildlife tracking. OppNets aim to exploit human relationships and interactions with various mobility patterns to build strong routing protocols. OppNets began with applications like wildlife tracking [1], search and rescue [2], and underwater sensor network applications [3]. One of the routing protocols challenges in OppNets is the high mobility of nodes so the work in [4] focused on and observed the inter-contact time and contact duration between two nodes transmitting opportunistically in order to determine the capacity of the opportunistic network such as the amount of data that can be transferred between two nodes in their contact times. OppNets can exploits the social relationships for building routing protocols as in the study [5] where social networking application indicated nodes contact, the properties of nodes, links, paths, and contact graphs which helps for building forwarding paths. Storage and Energy is another challenges in OppNets where nodes always has limitations in both of them. A study on the storage-delay and energy-delay tradeoffs in frequently disconnected paths between the source and the destination nodes has been done in [6] where nodes store the message and carry it till they meet the destination, or forward it to another intermediate node so it can forward it to the distention node. Routing in OppNets is a challenging problem, due to features such as frequent partitions, long delays and no endto-end path from source to destination. Many protocols have been developed to accommodate these features. Replication based protocols have higher delivery rates and lower delays over forwarding based protocols because of the multiple copies in the network. However, the cost incurred on network resources from replication based protocols is higher than the cost incurred from forwarding based protocols. Hybrid family inherited some features from both replication and forwarding families to come up with new protocols that have better performance in term of delivery rate, delays and cost. Researcher are yet to reach agreement on the best routing approach for OppNets. In this paper we undertake a comparative study between three representative protocols, one from each family. Epidemic [7] is one of the first replication family protocols based on flooding, where a message is flooded through the network without consideration of available resources. Direct Delivery [8] is one of the basic forwarding family protocols where a message is delivered to the destination directly without use of intermediate nodes or heavily exploiting network resources. PRoPHET [9] is one of the hybrid family of protocols that uses both replication and forwarding techniques. The main contributions of this paper are to present an overview of current routing protocols for OppNets and their classifications and to evaluate three of the OppNet routing protocols (Epidemic, Direct Delivery and PRoPHET) in terms of complexity and scalability. Two scenarios were designed to evaluate OppNet routing performance when increasing network load and increasing the number of nodes. Based on /13/$ IEEE 209

2 our results, the network load, network area, and the number of nodes impacts the performance levels of routing protocols in OppNets. The remainder of this paper is organized as follows: In section II, we present an overview of routing in OppNets. In section III, we evaluate three routing protocols in OppNets and in section IV, we present our conclusion and future work. II. OVERVIEW ON ROUTING PROTOCOLS IN OPPNETS All routing protocols in OppNets are based on the idea of Store-Carry-Forward as there is no end-to-end connection between sources and destinations. OppNet routing protocols can be classified in different ways [10]. They can also be classified into three families: replication, forwarding, and hybrid. A. Replication Family Replication based protocols replicate the message on the network so at any given time there is more than one copy of the message on the network. This kind of protocol achieves good delivery rates but can waste network resources. Some of the protocols of the replication family use the flooding technique [7] [11] flood the network with a number of message copies. Epidemic routing [7] is a flooding based protocol. When two nodes are in range, the node with the smaller identifier begins sending its summery vector to the node with the larger identifier, including a table of messages on the node s buffer. After the summery vector exchange, each node can determine if the other node has new unseen messages. The nodes can request the globally unique ID of unseen messages, and as long as they have enough buffer space. Nodes receive and carry massages even if there is no path to the destination at that time. Epidemic routing uses FIFO scheme for managing nodes, buffers. When the buffer is full, and the node is no longer able to store new messages, the node discards the first message that has remained the longest in the buffer. Epidemic routing is used as a benchmark to compare other protocols as Epidemic routing achieves 100% message delivery but with a maximum amount of resources used and high congestion. Spray and Wait [11] has two phases; in the first phase, the source node spray a predefined number of copies to the network, and then in the wait phase the nodes do direct delivery to the destination. Other replication protocols use coding techniques [12] [13] to increase the reliability in packet erasure networks. Network Coding (NC) and Erasure Coding (EC) can be used as efficient techniques to encode the original packets into a data stream of encoded packets. Original packets are reconstructed when the destination node obtains a certain number of encoded packets. NC allows the intermediate node to encode the received packets and uses a small size of encoded blocks to achieve high delivery rates, while EC allows the source node to only encode the packets and the messages together to achieve a big size of encoded blocks with high transmission and a low overhead ratio [12]. B. Forwarding Family In the forwarding family, the message is never replicated on the network, so only one copy of the message is available on the network at any given time. As a result, forwarding based protocols waste less network resources but the delivery rates are less than replication based protocols. The most basic and simplest of forwarding family routing protocols are Direct Delivery (DD) [8], Direct Transmission [14] and First Contact (FC) [15]. DD and Direct Transmission use only one hop instead of a number of hops, meaning the source node directly forwards its message to the destination node. In FC [15], the message is forwarded to the next hop randomly. The message is forwarded to any next hop node if this node has not previously carried this message and is directly deleted from the buffer of the sender node. Forwarding based protocols might also use prediction [14] [16] to get information about the best intermediate nodes that can forward the message. Seek-and-Focus [14] has two phases. First phase is the seek phase where the sender performs random forwarding to neighbour nodes with parameter (P). The second phase is the focus phase where the forwarding is based on the utility of the neighbour node which is based on the recent encounter time. A node is selected to be the next hop when its utility value is more than the threshold. PreS [16] uses an adapted binary spraying scheme for designing a multi-copy routing protocol. PreS is based on the assumption that nodes usually move around main venues and they can exchange messages only if they are on the same main venue. However, the authors did not consider node connections between different venues. Forwarding based protocols can use time varying shortest path based techniques [15] [17]. In [15] four routing algorithms are proposed based on such techniques, Minimum Expected Delay (MED), Earliest Delivery (ED), Earliest Delivery with Local Queue (EDLQ), and Earliest Delivery with All Queues (EDAQ). In MED, the cost of the next hop is calculated as the sum of the average waiting time, propagation delay, transmission delay, and the proactive routing approach is used for message routing. In ED, Dijkstra s shortest path algorithm is used for path calculation at the source node without consideration to the intermediate node buffer size, even though messages might be dropped with a limited buffer. In EDLQ, the next hop delay is calculated using local queue occupancy and the route can be recalculated in each hop. In EDAQ, instantaneous queue sizes can be calculated using queuing oracle. The source node can calculate the best route for the message, and the capacity for all nodes on that path are reserved at the time of sending to ensure adequate time for message movement and accurate prediction of queuing in the network. In DHR [17], the contact information between nodes is combined to achieve an aggregation level. Nodes above this level can maintain information about the time invariant hierarchical network, and the nodes below this level maintain /13/$ IEEE 210

3 information about time variance based on the shortest path construction. Some forwarding based protocols use buffer management and congestion control based techniques [18] [19], as nodes have limited storage and may carry messages for a long time. The message can be forwarded to other nodes on the network, with the node receiving the message either accepting or rejecting the message if it doesn t have enough buffer space. Nodes may place certain restrictions on the total number of messages and size of message on their buffer, and may discard some messages from the buffer after some time. In TBR [18], each node has a priority list or schedule for the messages that will be forwarded/dropped on its buffer based on the message s TTL, hop count, message replication count, and message size. This mechanism efficiently utilizes node buffers with each node maintaining a list of delivered messages. The destination node inserts the ID of delivered message into the delivered list, and when nodes meet each other, acknowledgment messages are exchanged including the list of delivered messages. According to the list, delivered messages will be deleted from their buffers. Messages with an earlier deadline have priority to be forward or delivered. This type of protocol, or TTL based protocols aim to enhance message delivery rates. In EBMP [19] a utility value of each message can be calculated by using message properties such as the number of copies of each message on the network, message age, and remaining TTL of each message. Some forwarding based protocols use social relationship information [20] [21] [22] [23] for selecting the best next hop. In CiPRO [20], when two or more nodes are on the same transmission range, the sender sends a control massage Hm to all the first hop neighbors containing a hashed value. The first hop neighbours then compare their hashed values with the received hashed value to calculate the encounter probability with the destination. Following this they broadcast their Hm to the other first hop neighbours so they can also calculate their encounter probability with the destination and then return it to the first hop node. This node then selects the higher probability value from its neighbours and return it to the sender who will use this information to send the message content. PeopleRank [21] is similar to Google s proposed mechanism of PageRank. Higher weight is given to nodes if they are socially connected to other important nodes on the network. When two nodes meet each other they exchange their current PeopleRank values and the number of social graph neighbours they have then updates their PeopleRank values. Therefore, the more nodes that meet each other, the more their rank value increases. In BUBBLE Rap [22], the context information is the social communities that nodes belong to. Based on the patterns of contacts between nodes the communities are automatically defined and labeled. When a source wants to send a message to the destination, it begins looking for nodes belonging to the same community of the destination node. If such nodes are not found, it will try to forward the message to sociable nodes which have more chances of meeting with the community of the destination node. The idea of FRESH [23] is based on the idea that node that I met 5 minutes ago is probably closer to me than the node I met 5 hours ago. In FRESH, nodes maintain a record of all nodes recently encountered. Nodes forward their messages to neighbouring or intermediate nodes if the nodes encounters the destination more recently than the node itself and so on until the message reaches its destination. C. Hybrid Family In hybrid family, the authors attempt to exploit both forwarding and replication based mechanisms to come up with new and powerful protocols. Some hybrid protocols use the utility replication based idea to select the best neighbour. [24] [9] [25] [26]. The main idea of the delegation forwarding algorithm [24] is the assumption that each node has an associated quality metric. If a node encounters another node with a quality metric higher than other nodes they have seen, then it will transmit the message to the encountered node. However, this will add overhead on the network as the node may have to carry this message for long time. PRoPHET [9] is similar to Epidemic routing [7]. Nodes exchange summary vectors when they meet each other and this contains delivery predictability values which are built in each node. If the node has previously visited a specific location several times before, then it will most likely revisit that location again. This delivery predictability ages with time and has a transitive property. After exchanging summary vectors, the source node transfers messages to other nodes if the delivery predictability of other nodes is higher. PRoPHET achieved good delivery rates with less network overhead. RAPID [25] uses a utility function to assign a utility value to each packet. Packet utility value is based on the average delay metric. RAPID first replicates packets with the highest increase in utility. There are four steps in RAPID. The first is the initialization step which involves the exchange of metadata to help estimate packet utilities. In the second step, the direct delivery step, packets are intended for immediate transmission to neighbour nodes. In the replication step, packets are replicated according to a marginal utility. The RAPID protocol ends in the fourth and final termination step when contacts are broken or all packets are replicated. In DTC [26], the sender node selects the next hop node based on the node utility value that is computed by using a number of variables such as most frequently noticed, future plans, power and rediscovery interval. Some hybrid protocols utilize and improve the Spray and Wait protocol [11] for selecting the best neighbour, such as HiBOp [27], and EBR [28]. In HiBOp [27] the source node sprays a number of message copies with a utility forwarding approach using context information. HiBOp calculates the delivery potential of nodes and looks for nodes that have an increasing match with the known context attributes of the destination. A high match means a high similarity between nodes and destination contexts. Consequently, the message is handed to the destination s community. EBR [28] aims to /13/$ IEEE 211

4 control and limit the flooding of messages. EBR is similar to the Spray and Wait protocol, only that it uses previous contact history in the spray phase to spray messages in the network. Messages are forwarded to nodes with a high encounter rate. EBR implements security measures against a black hole denial of service attack. Improved Epidemic protocols such as MaxProp [29] also fall within the hybrid family. The core idea of MaxProp is to prioritize both the schedule of packets transmitted and dropped as MaxProp assumes limited storage and bandwidth. Each node has a vector of the total number of the nodes in the network. So when two nodes meet, they exchange their vectors and then use these vectors to estimate the shortest path to the destination node. These priorities are based on the likelihood of path to peers and is also based on a number of complementary mechanisms including historical data, hop count, acknowledgments, a head start for new packets, and any lists of previous intermediary nodes. MaxProp uses broadcasted acknowledgements to updates the encountered nodes about the delivered messages so they can delete them from their buffers. Some hybrid protocols use coding techniques to select the next best hop from neighbour nodes such as Network Coding [30]. The main idea of Network Coding is to reduce network overhead. This is a process of allowing and encouraging intermediate nodes to mix data in the network so they can send out packets with linear combinations of information received earlier. A tuples of encoding vectors and information vectors received will be stored and checked in the node s buffer, and the node will then find the packets originally intended for it. Source vectors are grouped to generations in order to limit the size of matrix stored on the nodes buffer. In addition there will be one matrix per generation and combined packets should be from the same generation. Hash functions are used by nodes in order to determine which generation to insert a given packet. The RED algorithm [31] is based on erasure coding and encounter prediction techniques. RED has two parts; the data transmission part which is responsible for the decision of when and where to forward data based on delivery probabilities. The second part is message management which is responsible for the optimal erasure coding parameters based on its delivery probability to accomplish the required data delivery rates while reducing network overhead. III. EVALUATING ROUTING PROTOCOLS IN OPPNETS Using two scenarios, we evaluated three OppNets routing protocols - Epidemic [7], Direct Delivery [8] and PRoPHET [9] in terms of complexity and scalability. Where complexity defines how efficient the protocol functions as the overhead on the network increases, and scalability is the ability of the protocol to scale and deal with the network as it enlarges and the number of nodes increases. We use three metrics for evaluation purposes: 1) Delivery Rate: The ratio of the number of successfully delivered messages to the total number of messages generated. 2) Delivery Delay: The average duration between the times a message is generated at the source and the time the message or its copy is received at the destination. 3) Overhead ratio: The number of copies per generated message in the network. A. Simulation settings To study the above metrics we used two scenarios implemented in the ONE simulator [32]. The following settings apply to both scenarios; however, any changes in the settings will be outlined for each individual scenario in the next section. Each simulation is defined to last for 6 hours, with 0.5 seconds of update intervals. Bluetooth is chosen for connectivity with transmit range of 10 meters, and transmit speed of 2 Mbps. There are 150 active nodes composed of cars and pedestrians. Pedestrians and cars have up to 10 MB of RAM for storage. Pedestrians move at random speeds between 0.5 and 1.5 m/s, cars drive on roads only and move at speeds between km/h, with wait times of secs. Map based movement is used for pedestrians and cars, with a network area of m. Nodes move randomly on roads and walkways with movement warm-up of 100 seconds. There are 3 groups of trams, with 2 trams in each group. Since trams have bigger buffers in their communication devices, their buffers have up to 60 MB of RAM. Map route movement is used for trams to follow a constructed tram line. Trams drive at speeds of 7 10 m/s with wait time 10 30secs at each configured stop. In addition to the Bluetooth interface, a group of trams uses the high speed interface with transmit range of 1000 meters and transmit speed of 10Mbps. Messages are generated every 15 to 25 minutes per node, with message sizes between 250k and 950k, and message time to live of 3 hours. Each point on the graph represents the average of 30 simulation runs. B. Simulation results of Scenario 1 Scenario 1 aims to test the complexity of the 3 protocols. For this scenario, overhead is added to the network by increasing message sizes, and reducing the intervals of message generation to 15 and 25 seconds. Simulation is defined to last for 2 hours. Simulation is run 3 times for each protocol as follows: small sized messages (10KB to 20KB), medium sized messages (50KB to 700KB), and large sized messages (900KB to 3MB). Based on the results of scenario 1, the 3 protocols will be rated depending on their ability to deal with complexity, a protocol is rated with either high, moderate, or low complexity. The graphs in figures 1 and 2 show the delivery rates and delivery delays of the 3 message sizes. 1) Analyzing the results of scenario 1: In fig. 1 1, with small message sizes, the delivery rates are high for Epidemic and PRoPHET. With medium sized messages, the delivery rates decrease by more than 50% for each protocol. With large sized messages, delivery rates drop really low for the 3 protocols. In fig. 2, the delivery delay is the lowest with small sized messages, and quite high for the 3 protocols with both medium and large sized messages. From these results, we can observe that Epidemic and PRoPHET have higher delivery /13/$ IEEE 212

5 Fig. 1. Comparison of Delivery Rates (Scenario 1) Fig. 3. Comparison of Delivery Rates (Scenario 2) Fig. 2. Comparison of Delivery Delay (Scenario 1) Fig. 4. Comparison of Delivery Delay (Scenario 2) rates than Direct Delivery because there is more than one copy of the message on the network. From the results of scenario 1, we conclude that the delivery rates decrease as message sizes increases and the intervals of message generation decreases. With large sized messages, the delivery rates drops to almost 0 for all the 3 protocols. This is because the buffer space for nodes is consumed continuously, and the need to drop messages to free buffer space is required. The aim of scenario 1 was to test the ability of the 3 protocols to deal with increasing network overhead. Based on the results, Epidemic and PRoPHET performed better than Direct Delivery for small and medium sized messages. Epidemic and PRoPHET were rated with a medium level in terms of dealing with network complexity, and a low rating is given to Direct Delivery. C. Simulation results of Scenario 2 Scenario 2 aims to test the scalability of the 3 protocols by increasing the number of nodes and doubling the network area. We simulated 3 network sizes as follows: small (306 nodes composed of 200 pedestrians, 100 cars, and 6 trams), medium (909 nodes composed of 600 pedestrians, 300 cars, and 9 trams), and large (1812 nodes composed of 1200 pedestrians, 600 cars, and 12 trams). Based on the results of this scenario, the 3 protocols will be rated with high, moderate, or low scalability. The graphs in fig. 3-5 show the delivery rates, delivery delays, and overhead cost of the 3 different number of nodes and network size increase. 1) Analyzing the results of scenario 2: In fig. 3, as the number of nodes increases in the network, the delivery rates are high for Epidemic and PRoPHET, and lowest for Direct Delivery. In fig. 4, as the number of nodes increases in the network, the delivery delays decreases for Epidemic and PRoPHET, and remains quite high for Direct Delivery. In fig. 5, as the number of nodes increases in the network, the cost increases for Epidemic and PRoPHET, and remains very low for Direct Delivery. The cost for Direct Delivery remains low because the message is sent directly from the source to the destination and so it does not consume network resources. Also, the reason behind the high delivery delays for Direct Delivery is the single copy for each message. Based on these results, as the network increases in area and number of nodes, we conclude that chances for messages to reach their destinations gets higher. Epidemic and PRoPHET scale with high delivery rates and decreasing delivery delays, but with an increasing overhead cost, so we rated them with medium Fig. 5. Comparison of Overhead Ratio (Scenario 2) /13/$ IEEE 213

6 scalability. Although the cost for Direct Delivery is low, the delivery rates are low with a high delivery delay so we rated it with low scalability. IV. CONCLUSION AND FUTURE WORK The aim of this paper was to research the available protocols of OppNets and their classifications, and to evaluate OppNet routing protocols in terms of complexity and scalability. Two scenarios were designed to evaluate OppNets routing performance in the increasing network load and increasing number of nodes. Based on our results, the network load, network area, and the number of nodes impacts the performance levels of routing protocols in OppNets. Obtained from scenario 1, as the load on the network increases and the intervals of message generation decreases, the performance of the three protocols decreases. Based on the results, Epidemic and PRoPHET perform better than Direct Delivery for small and medium sized messages. Epidemic and PRoPHET achieved a medium rating in terms of dealing with network complexity, and a low rating is given to Direct Delivery. From scenario 2, as the network enlarges in area and number of nodes, Epidemic and PRoPHET performed better than Direct Delivery in delivering messages to the destination, but at a very high cost. Epidemic and PRoPHET were rated with medium scalability. Although the cost for Direct Delivery is low, the delivery rates are low with a high delivery delay and so we have given it a low rating for scalability. Also, from the results of both scenarios, we can see that Epidemic and PRoPHET have higher delivery rates than Direct Delivery because there is more than one copy of the message on the network, while Direct Delivery uses only one copy of the message. However, the incurred network cost of Direct Delivery is very low compared to Epidemic and PRoPHET. A lot of work remains to be done in the future. Complexity and scalability are largely unsolved research issues in OppNet routing. REFERENCES [1] P. Juang, H. Oki, Y. Wang, M. Martonosi, L. Peh, and D. Rubenstein. Energy-efficient computing for wildlife tracking: design tradeoffs and early experiences with Zebranet. In: Proceedings of 10th ASPLOS-X, ACM Press, [2] J. Huang, S. Amjad, and S. Mishra. CenWits: A Sensor-Based Loosely Coupled Search and Rescue System Using Witnesses. In: ACM SenSys, pages , November [3] C. Detweiller, I. Vasilescu, and D. Rus. An Underwater Sensor Network with Dual Communications, Sensing, and Mobility. In: OCEANS Europe, pages 1 6, [4] A. Chaintreau, P. Hui, J. Crowcroft, C. Diot, R. Gass, and J. Scott. Pocket switched networks: real-world mobility and its consequences for opportunistic forwarding. Technical Report UCAM-CL-TR-617, February [5] A. Mtibaa, A. Chaintreau, J. LeBrun, E. Oliver, A. Pietilainen, and C. Diot. Are You Moved by Your Social Network Application?. In: Proceedings of the first workshop on Online social networks, ACM, pages 67 72, August [6] T. Small and Z. Haas. Resource and Performance Tradeoffs in Delay- Tolerant Wireless Networks. In: Proceedings of ACM SIGCOMM Workshop on Delay Tolerant Networking (WDTN), [7] A. Vahdat and D. Becker. Epidemic routing for partially-connected ad hoc networks. Duke University Technical Report, [8] M. Grossglauser and D. Tse. Mobility increases the capacity of ad hoc wireless networks. IEEE/ACM Transactions on Networking (TON, Volume 10 Issue 4,Pages , August [9] A. Lindgren, A. Doria, and O. Schelen. Probabilistic Routing in Intermittently Connected Networks. ACM SIGMOBILE Mobile Computing and Communications Review, Volume 7,Issue 3,Pages 19-20, July [10] Y. Cao and Z. Sun. Routing in Delay/Disruption Tolerant Networks: A Taxonomy, Survey and Challenges. Communications Surveys & Tutorials, IEEE, 15(issue 2): , [11] T. Spyropoulos, K. Psounis, and C. Raghavendra. Multiple-copy Routing in Intermittently Connected Mobile Networks. USC,Tech. Rep. CENG , [12] A. Fujimura, S. Oh, and M. Gerla. Network coding vs. erasure coding:reliable multicast in ad hoc networks. in IEEE MILCOM08, San Diego, California, USA, [13] Y. Wang, S. Jain, M. Martonosi, and K. Fall. Erasure-coding based routing for opportunistic networks. in ACM WDTN 05, Philadelphia,Pennsylvania, USA, [14] T. Spyropoulos, K. Psounis, and C. Raghavendra. Single-copy routing in intermittently connected mobile networks. In: IEEE SECON, pages , [15] S. Jain, K. Fall, and R. Patra. Routing in a Delay Tolerant Network. ACM SIGCOMM04, Oregon, Portland, [16] J. Niu, J. Guo, Q. Cai, N. Sadeh, and S. Guo. Predict and spread: An efficient routing algorithm for opportunistic networking. Wireless Communications and Networking Conference (WCNC),IEEE, [17] C. Liu and J. Wu. Scalable Routing in Cyclic Mobile Networks. IEEE Trans. Par. Distrib. Syst., vol. 20, no. 9, page , [18] A. Prodhan, R. Das, H. Kabir, and G. Shoja. TTL based routing in opportunistic networks. Journal of Network and Computer Applications, volume 34:pages , [19] K. Shin, and S. Kim. Enhanced buffer management policy that utilises message properties for delay-tolerant networks. IET Communications, pages , October [20] H. Nguyen and S. Giordano. Context information prediction for socialbased routing in opportunistic networks. ELSEVIER, Ad Hoc Networks, [21] A. Mtibaa, M. May, C. Diot, and M. Ammar. Peoplerank: social opportunistic forwarding. In: INFOCOM 10: Proceedings of the 29th conference on Information communications, [22] P. Hui, J. Crowcroft, and E. Yoneki. BUBBLE Rap: Social-Based Forwarding in Delay Tolerant Networks. In: Proceedings of ACM MobiHoc 08, [23] M. Grossglauser H. Dubois-Ferriere and M. Vetterli. Age matters: efficient route discovery in mobile ad hoc networks using encounter ages. In: Proceedings 4th ACM international symposium on Mobile ad hoc networking & computing, ACM, pages , [24] V. Erramilli, M. Crovella, A. Chaintreau, and Christophe. Delegation Forwarding. ACM SIGMOBILE Mobile Computing and Communications Review,Volume 7 Issue 3,Pages 19-20, July [25] A. Balasubramanian, B. Levine, and A. Venkataramani. DTN Routing as a Resource Allocation Problem. ACM SIGCOMM Computer Communication Review, Volume 37 Issue 4,Pages , October [26] X. Chen and A. Murphy. Enabling Disconnected Transitive Communication in Mobile Ad Hoc Networks. In Proc of Workshop on Principles of Mobile Computing, colocated with PODC01, pp 21-27, [27] C. Boldrini, M. Conti, J. Jacopini, and A. Passarella. HiBOp: a History Based Routing Protocol for Opportunistic Networks. World of Wireless, Mobile and Multimedia Networks, WoWMoM IEEE International Symposium, [28] S. Nelson, M. Bakht, and R. Kravets. Encounter Based Routing in DTNs. IEEE, [29] J. Burgess, B. Gallagher, D. Jensen, and B. Levine. Maxprop: routing for vehicle-based disruption-tolerant networking. In: Proceedings of INFOCOM 06, pages 1 11, [30] J. Widmer and J. Boudec. Network Coding for Efficient Communication in Extreme Networks. WDTN 05 Proceedings of the 2005 ACM SIGCOMM workshop on Delay-tolerant networking,pages , [31] Y. Wang and H. Wu. Replication-based efficient data delivery scheme (red) for delay/fault-tolerant mobile sensor network (DFT-MSN). Pervasive Computing and Communications Workshops, PerCom Workshops Fourth Annual IEEE International Conference, [32] A. Keranen, J. Ott, and T. Karkkainen. The ONE Simulator for DTN Protocol Evaluation. SIMUTools 09: 2nd International Conference on Simulation Tools and Techniques. Rome, March /13/$ IEEE 214