- Open Access
- Total Downloads : 17
- Authors : R. S. Pravena, B. Leena
- Paper ID : IJERTCONV3IS16119
- Volume & Issue : TITCON – 2015 (Volume 3 – Issue 16)
- Published (First Online): 30-07-2018
- ISSN (Online) : 2278-0181
- Publisher Name : IJERT
- License: This work is licensed under a Creative Commons Attribution 4.0 International License
Distributed Cache Network Central Location Technique for Efficient Data Access
1R. S. Pravena, 2B. Leena
1PG Scholar,2Assistant Professor
Sri Guru Institute of Technology,Coimbatore
Abstract: Disruption tolerant networks are characterized by low node density, node mobility, and lack of global network information. Most of research efforts in DTNs focus on data forwarding, but only limited work has been done by an efficient data access to mobile consumers . A novel scheme to support cooperative caching in DTN and a novel sleep scheduling method to reduce the delay of alarm broadcasting from any sensor node in WSNs. we design two determined traffic paths for the transmission of alarm message. When a critical event occurs, an alarm is quickly transmitted along one of the traffic paths to a center node, and then as soon as, it broadcast by the center node along another path without collision, which enables the sharing and coordination of cached data among multiple nodes and reduces data log on delay.In network, the basic idea is to purposely cache data at a set of network central locations which can be easily accessed by other nodes. An efficient scheme that ensures appropriate NCL selection based on a probabilistic selection metric and coordinates multiple caching nodes to optimize the tradeoff between data accessibility and caching overhead.
Keywords- Disruption tolerant networks, Caching overhead, NCL.
Disruption tolerant networks consist of mobile devices that contact each other. Due to the low node density and unpredictable node mobility, only intermittent network connectivity exists in DTN, and the subsequent difficulty of maintaining end to end communication links makes it necessary to use carry and forward methods for data transmission. Examples of such networks include groups of individuals moving in disaster recovery areas, military battlefields, or urban sensing applications. In such networks, node mobility is exploited to let mobile nodes carry data as relays and forward data opportunistically when contacting others. The key problem is how to determine the appropriate relay selection strategy. Although forwarding schemes have been proposed in DTN there is limited research on providing efficient data access to mobile users, despite the importance of data accessibility in many mobile applications. For example, it is desirable that smart phone users can find interesting digital content from their nearby peers. In vehicular ad hoc networks, the availability of live traffic information will be beneficial for vehicles to avoid traffic delays. In these applications, data are only requested by mobile users whenever needed, and requesters do not know data locations in advance. The destination of data is unknown when data are generated. This communication paradigm differs from publish
subscribe systems in which data are forwarded by broker nodes to users according to their data subscriptions. Appropriate network design is needed to ensure that data can be promptly accessed by requesters in such cases.
A common technique used to improve data access performance is caching. To cache data at appropriate network locations based on query history, so that queries in the future can be responded with less delay. Although cooperative caching has been studied for both web based applications and wireless ad hoc networks to allow sharing and coordination among multiple caching nodes, it is difficult to be realized in DTN due to the lack of persistent network connectivity. First, the opportunistic network connectivity complicates the estimation of data transmission delay, and furthermore makes it difficult to determine appropriate caching locations for reducing data access delay. This difficulty is also raised bythe incomplete information at individual nodes about query history. Second, due to the uncertainty of data transmission, multiple data copies need to be cached at different locations to ensure data accessibility. The difficulty in coordinating multiple caching nodes makes it hard to optimize the tradeoff between data accessibility and caching overhead as shown in figure 1.,
A novel scheme to address the challenges and to efficiently support cooperative caching in DTNs basic idea is to intentionally cache data at a set of Network Central Locations, each of which corresponds to a group of mobile nodes being easily accessed by other nodes in the network
Figure 1 Wireless adhoc networks and Disruption tolerant networks
Each NCL is represented by a central node, which has high popularity in the network and is prioritized for caching data. Due to the limited caching buffer of central nodes, multiple nodes near a central node may be involved for caching, and it ensure that popular data are always cached nearer to the central nodes via dynamic cache replacement based on query history. Our detailed contributions are listed as follows:
Develop an efficient approach to NCL selection in DTN based on a probabilistic selection metric. The selected NCLs achieve high chances for prompt response to user queries with low overhead in network storage and transmission.
A data access scheme to probabilistically coordinate multiple caching nodes for responding to user queries optimize the tradeoff between data accessibility and caching overhead, to minimize the average number of cached data copies in the network.
A utility based cache replacement scheme to dynamically adjust cache locations based on query history, and the scheme achieves good tradeoff between the data accessibility and access delay.
Data forwarding in DTN originates from Epidemic routing, which floods the entire network and some later studies focus on proposing efficient relay selection metrics to approach the performance of Epidemic routing with lower forwarding cost, based on prediction of node contacts in the future. Some schemes do such prediction based on their mobility patterns, which are characterized. In some other schemes, node contact pattern is exploited as abstraction of node mobility pattern for better prediction accuracy based on the experimental and theoretical analysis of the node contact characteristics. The social network properties of node contact patterns, such as the centrality and community structures, have also been also exploited for relay selection in recent social based data forwarding schemes.
Data access in DTN, on the other hand, can be provided in various ways data can be disseminated to appropriate users based on their interest profiles. Publish subscribe systems were used for data dissemination, where social community structures are usually exploited to determine broker nodes. In other schemes without brokers, data items are grouped into predefined channels, and are disseminated based on users subscriptions to these channels. Caching is another way to provide data access.
Cooperative caching in wireless ad hoc networks
Distributed determination of caching policies for minimizing data access delay has been studied in DTNs assuming simplified network conditions. In it is assumed that all the nodes contact each other with the same rate. In users are artificially partitioned into several classes such that users in the same class are identical. In data are intentionally cached at appropriate network locations with generic data and query models, but these caching locations are determined based on global network knowledge support cooperative caching in a fully distributed manner in DTNs, with heterogeneous node contact patterns and behaviors.
WIRELESS ROUTING PROTOCOL
DYNAMIC SOURCE ROUTING
The Dynamic Source Routing protocol is a simple and efficient routing protocol designed specifically for use in multi-hop wireless ad hoc networks of mobile nodes.DSR allows the network to be completely self- organizing and self-configuring, without the need for any existing network infrastructure or administration. Dynamic Source Routing is a reactive routing protocol that uses source routing to send packets. It uses source routing means that the source must know the complete hop sequence to the destination.
Each node maintains a route cache, all routes it knows are stored. The route discovery process is initiated only if the desired route cannot be found in the route cache. To limit the number of route requests propagated, a node processes the route request message only if it has not already received the message and its address is not present in the route record of the message. DSR uses source routing, the source determines the complete sequence of hops that each packet should traverse. This requires that the sequence of hops is included in each packet's header. A negative consequence of this is the routing overhead every packet has to carry. One big advantage is that intermediate nodes can learn routes from the source routes in the packets they receive. Since finding a route is generally a costly operation in terms of time, bandwidth and energy, this is a strong argument for using source routing. The protocol is composed of the two main mechanisms of Route Discovery and Route Maintenance, works together to allow nodes to discover and maintain routes to arbitrary destinations in the ad hoc network as shown in figure 2. The protocol allows multiple routes to any destination and allows each sender to select and control the routes used in routing its packets, for use in load balancing or for increased robustness.
in which each node caches pass by data based on data D F
popularity, so that queries in the future can be responded with less delay. Caching locations are selected incidentally
among all the network nodes. Some research efforts have A C
been made for caching in DTNs, but they only improve
data accessibility from infrastructure network such as WIFI B E
access points or Internet. Peer toper data sharing and access among mobile users are generally neglected.
DSR Header packet
Figure 2 Dynamic source routing
PROPERTIES OF DSR
DSR uses the key of source routing. Intermediate nodes do not need to maintain up to date routing information in order to route the packets they forward. There is also no need for periodic routing advertisement messages, which will lead to reduce network bandwidth overhead, particularly during periods when little or no significant host movement is taking place. Battery power is also conserved on the mobile hosts, both by not sending the advertisements an by not needing to receive them, a host could go down to sleep instead. This protocol has the advantage of learning routes by scanning for information in packets that are received.
The uncertainty of data transmission, multiple data copies need to be cached at Different locations to ensure data accessibility. The difficulty in coordinating multiple caching nodes makes it hard to optimize the tradeoff between data accessibility and caching overhead. A utility based cache replacement scheme to dynamically adjust cache locations based on query history, and our scheme achieves good tradeoff between the data accessibility and access delay. A data access scheme to probabilistically coordinate multiple caching nodes for responding to user queries furthermore optimize the tradeoff between data accessibility and caching overhead, to minimize the average number of cached data copies in the network. An efficient approach to NCL selection in DTNs based on a probabilistic selection metric. The selected NCLs achieve high chances for prompt response to user queries with low overhead in network storage and transmission. The utility values in Social Cast are linked to movement patterns and co-location with other hosts: as the basic assumption is that hosts which have same interest spend time co-located, the Social Cast routing aims at exploiting as carrier for messages hosts which have been co-located often with the interested subscribers.
The design of caching strategy in wireless ad hoc networks benefits from the assumption of existing end to end paths among mobile nodes, and the path from a requester to the data source remains unchanged during data access in most cases. Such assumption enables any intermediate node on the path to cache the pass-by data. A data item or a query is described by a set of keywords over a keyword space so that caching nodes can determine the appropriate data that a user is interested in. In these scenarios, data requesters are randomly distributed in the network. We focus on efficiently utilizing the available node buffer to optimize the overall caching performance, which is measured by the successful ratio and delay for mobile users to access different data items. The central nodes representing NCLs are selected the network administrator is responsible for notifying each node in the network about the information of NCLs via cellular or satellite links. Since each node is
only notified about the identifiers of central nodes, this notification is cost effective without producing noticeable communication overhead, even in cases where the central nodes frequently change. Requester multicasts a query to central nodes of NCLs to pull data, and a central node forwards the query to the caching nodes. Multiple data copies are returned to the requester, and we optimize the tradeoff between data accessibility and transmission overhead by controlling the number of returned data copies. Critical event monitoring in wireless sensor networks MANET where only a small number of packets need to be transmitted most of the time. When a critical event occurs, an alarm message should be broadcast to the entire network as soon as possible. To prolong the network lifetime, some sleep scheduling methods are always employed in MANETs, resulting in significant broadcasting delay, especially in large scale MANETs.
Disruption tolerant networks
Network central locations
NCL load balancing
DISRUPTION TOLERANT NETWORKS
The design of caching strategy in wireless ad hoc networks benefits from the assumption of existing end to end paths among mobile nodes, and the path from a requester to the data source remains unchanged during data access in most cases. Such assumption enables any intermediate node on the path to cache the pass-by data. C forwards all the three queries to data sources A and B, and also forwards data d1 and d2 to the requesters. In case of limited cache space, C caches the more popular data d1 based on query history, and similarly data d2 are cached at node K. In general, any node could cache the pass-by data incidentally.
Basic solution to improve caching performance in DTNs is to restrain the scope of nodes being involved for caching. Instead of being incidentally cached anywhere, data are intentionally cached only at specific nodes. These nodes are carefully selected to ensure data accessibility, and constraining the scope of caching locations reduces the complexity of maintaining query history and making caching decision.
THE BIG PICTURE
A data item or a query is described by a set of keywords over a keyword space so tht caching nodes can determine the appropriate data that a user is interested in. In these scenarios, data requesters are randomly distributed in the network. An efficiently utilizing the available node buffer to optimize the overall caching performance, which is measured by the successful ratio and delay for mobile users to access different data items.
NETWORK CENTRAL LOCATIONS
NCL SELECTION METRIC
The data transmission delay between two nodes A and B, indicated by the random variable Y, is measured by the weight of the shortest opportunistic path between the two nodes. In practice, mobile nodes maintain the information about shortest opportunistic paths between each other in a distance-vector manner when they come into contact. The maintenance of such network information is expensive in DTNs due to the lack of persistent end-to- end network connectivity. Selecting NCL with the assumption of complete network information from the global perspective. Afterwards Distributed NCL selection methods that efficiently approximate global selection results and can operate on individual nodes in an autonomous manner.
TRACE BASED VALIDATION
Traces record contacts among users carrying mobile devices in conference sites and university campuses. The mobile devices, including Mica2 sensors or smart phones, are distributed to users being participated into the experiment. Devices equipped with Bluetooth interface periodically detect their peers nearby, and a contact is recorded when two devices move close to each other. Devices equipped with Wi-Fi interface search for nearby Wi-Fi APs and associate themselves to APs with the best signal strength. A contact is recorded when two devices are associated with the same AP. The detected contacts are recorded in the local storage of mobile devices.
A network warm-up period is reserved for nodes to collect information and calculate their pair-wise contact rates as described in Section 3.2, and central nodes are selected after the warm-up period ends. Data are unable to be intentionally cached at the NCLs during the warm-up period. Instead, data are incidentally cached by nodes in the network. The central nodes representing NCLs are selected, the network administrator is responsible for notifying each node in the network about the information of NCLs via cellular or satellite links. Since each node is only notified about the identifiers of central nodes, this notification is cost effective without producing noticeable communication overhead, even in cases where the central nodes frequently change. Note that the central nodes are selected due to their popularity in the network, rather than their computation or storage capabilities. Therefore, in general, we assume that the central nodes have similar capabilities in computation, data transmission, and storage with other nodes in DTNs.
A network warm-up period is reserved for nodes to exchange and maintain necessary information about opportunistic paths to others. However, a longer warm-up period is needed for distributed NCL selection because multi-hop opportunistic data transmission is required for distributed maintenance of such information. Afterwards,
each node in the network autonomously calculates the value of its NCL selection metric according to and broadcasts this value to the network. After a predefined broadcasting period, a node having received these values then selects the nodes with the K highest metric values as the central nodes representing NCLs.The first and more straightforward method is to extend the broadcasting period, so that each node is aware of the metric values of all the other nodes in the network. However, this method may be impractical in some mobile applications with strict requirements of timeliness. Another alternative is to opportunistically correct suboptimal NCL selections when nodes contact each other.
A data source generates data it pushes data to central nodes of NCLs, which are prioritized to cache data. One copy of data is cached at each NCL. If the caching buffer of a central node is full, another node near the central node will cache the data decisions are automatically made based on buffer conditions of nodes involved in the pushing process.
Requester multicasts a query to central nodes of NCLs to pull data, and a central node forwards the query to the caching nodes. Multiple data copies are returned to the requester, and we optimize the tradeoff between data accessibility and transmission overhead by controlling the number of returned data copies. Utility based cache replacement is conducted whenever two caching nodes contact and ensures that popular data are cached nearer to central nodes. We generally cache more copies of popular data to optimize the cumulative data access delay. We also probabilistically cache less popular data to ensure the overall data accessibility.
5.3.1 CACHING LOCATION
The opportunistic path weight to the central node as relay selection metric for such data forwarding, and a relay forwards data to another node with a higher metric than itself.
Figure3 Determining caching location at NCLs.
This Compare and Forward strategy has been widely used for efficient data forwarding. According to Definition 1 on opportunistic path, this strategy probabilistically ensures that each forwarding reduces the remaining delay for data to be delivered to the central node as shown in figure 3.
Node may request data, and hence, data requesters are randomly distributed in the network. Requester multicasts a query with a finite time constraint to all the central nodes to pull data, and existing multicast schemes in DTNs. The query broadcast finishes when query expires. Each caching node at NCLs maintains up to date information about query history for cache replacement.
A probabilistic scheme to address these challenges and optimize the tradeoff between data accessibility and transmission overhead. Our basic idea is that, having received the query, a caching node probabilistically decides whether to return the cached data to the requester as shown in figure 4,
Figure4 Pulling data from the NCLs.
Different strategies are used for this decision, according to the availability of network contact information.
The locations where it is cached are dynamically adjusted via cache replacement. This replacement is based on data popularity, and generally places popular data nearer to the central nodes of NCLs as shown in figure 5,
Figure 5 Normal cache replacement and possible data removal
Traditional cache replacement strategies such as LRU, which removes the least-recently-used data from
cache when new data are available, are ineffective due to its over simplistic consideration of data popularity. Greedy Dual Size calculates data utility by considering data popularity and size simultaneously, but cannot ensure optimal selection ofcached data.
NCL LOAD BALANCING
The central nodes cache the most popular data in the network and respond to the frequent queries for these data. The central nodes are also responsible for broadcasting all the queries they receive to other caching nodes nearby as shown in figure 6,
Figure 6 NCL load balancing
However, such functionality may quickly consume the local resources of central nodes that include their battery life and local memory. In addition, we would like our caching schemes to be resilient to failures of central nodes.
SELECTING THE NEW CENTRAL NODE
A central node fails or its local resources are depleted, another node is selected as a new central node. Intuitively, the new central node should be the one with the highest NCL selection metric value among the current non- central nodes in the network.
ADJUSTMENT OF CACHING LOCATIONS
Movement is achieved via cache replacement when caching nodes opportunistically contact each other. Each caching node at the original NCL recalculates the utilities of its cached data items with respect to the newly selected central node. In general, these data utilities will be reduced due to the changes of central nodes, and this reduction moves the cached data to the appropriate caching locations that are nearer to the newly selected central node.
PERFORMANCE OF NCL LOAD BALANCING No Cache : Caching is not used for data access and each query is only responded by data source.
Random Cache: In which every requester caches the received data to facilitate data access in the future.
Cache Data: Cooperative caching in wireless ad hoc networks, and lets each relay in DTNs cache the pass-by data based on their popularity.
Bundle Cache: Packs network data as bundles and makes caching decision on pass-by data by considering the node
contact pattern in DTNs, so as to minimize the average data access delay.
The center node computes the sleep scheduling according to the proposed scheduling scheme and broadcasts the scheduling to all the other nodes. The implementation of obtaining topology and broadcasting scheduling is introduced in Section Experiments of supplementary file, which can be found on the Computer Society Digital Library.
For the critical event monitoring in a WSN, sensor nodes are usually equipped with passive event detection capabilities that allow a node to detect an event even when its wireless communication module is in sleep mode. Upon the detection of an event by the sensor, the radio module of the sensor node is immediately woken up and is ready to send an alarm message.
Time of sensor nodes in the proposed scheme is assumed to be locally synchronous, which can be implemented and maintained with periodical beacon broadcasting from the center node the broadcasting delay is independent of the length of the duty cycle, but it increases linearly with the number of the hops. The broadcasting delay is independent of the density of nodes. The energy consumption is very low as nodes wake up for only one slot in the duty cycle during the monitoring. Scheduling method includes two phases:
Any node which detects a critical event sends an alarm packet to the center node along a predetermined path according to level-by-level offset schedule.
The center node broadcasts the alarm packet to the entire network also according to level-by-level offset schedule.
The traffic paths from nodes to the center node as uplink and define the traffic path from the center node to other nodes as downlink, respectively. Each node needs to wake up properly for both of the two traffics.
Therefore, the proposed scheduling scheme should contain two parts:
establish the two traffic paths in the WSN;
calculate the wake-up parameters (e.g., time slot and channel) for all nodes to handle all possible traffics.
SIMULATION AND PERFORMANCE
A. Simulation parameters
Number of nodes
Size of the network
Two ray model
Duration of simulation
B.SIMULATION WINDOWS C.NODE DISTRIBUTION
Figure 7 Distribution of nodes
Figure7 shows the formation of nodes and the distribution of nodes by using the network simulator. Here totally 100 nodes are considered and the distribution is done in orderly manner. All the nodes are unique and in active condition and thus perform their own function, having cache node in each of them.
UPDATING OF NODES
To minimize the broadcast delay, we establish a breadth first search (BFS) tree for the uplink traffic and a colored connected dominant set for the downlink traffic, respectively.
Figure 8 Updating nodes
Figure 8 shows distribution of nodes ,all the nodes move to the assigned place based on the values and all the nodes are aligned and it is put into active state by updating the nodes.
SELECTION OF SOURCE AND DESTINATION
Figure 9 Selection of source and destination
Figure 9, shows the selectionof source and destination by using node number present inall the nodes. In this state, since sleep scheduling is used only the node which are required will be in active state. Other node will be in idle state. The idle state nodes will move to the active state when it is required.
Figure 10 Data transmission
Figure 10 shows after choosing the source and the destination, the data path is selected and the path is verified by checking the NCL and then the data are transmitted. once the data is transmitted, the information will be updated automatically in the cache memory and in the NCL.
DATA TRANSMISSION RATE
Figure 11 Data transmission rate
Figure 11 shows the transmission of data rate from source to destination in bytes, during the packet transmission process data rate will shown in way through which it is passing. After the transmission process, cache will be updated automatically, which is useful for other data transmission purpose.
PACKET DELIVERY RATIO
Figure 12 Packet Delivery Ratios versus Events
Figure 12 shows packet delivery ratio. Graph is plotted for number of data packets delivered. In this the performance of Least Recently Used, First In First Out and Sleep Scheduling Network Central Location methods are compared and values are obtained.
NCL LOAD BALANCING
Figure13 NCL load balancing versus data life time
Figure 13 shows NCL load balancing. . Graph is plotted for NCL load balancing. In this the performance of
Least Recently Used, First In First Out and Sleep Scheduling Network Central Location methods are compared and values are obtained.
Figure14 Energy consumption versus life time
Figure 14 shows energy consumption. . Graph is plotted for energy consumption. In this the performance of Least Recently Used, First In First Out and Sleep Scheduling Network Central Location methods are compared and values are obtained.
Figure15 cache replacement versus number of data packets
Figure 15 shows cache replacement. Graph is plotted for number of data packets and number cache replacement. In this the performance of Least Recently Used, First In First Out and Sleep Scheduling Network Central Location methods are compared and values are obtained.
Figure 16 Delay versus packets
Figure 16 shows delay of packets. . Graph is plotted for delay in packet delivery. In this the performance of Least Recently Used, First In First Out and Sleep Scheduling Network Central Location methods are compared and values are obtained.
The performance of Dynamic source routing protocol on the basis of parameters are Packet Delivery Ratio, NCL Load Balancing, Energy consumption, Cache Replacement and Delay.
By comparing these parameters with the Sleep Scheduling Network Central Location and existing scheme First In First Out and Least Recently Used techniques. It has been reached to a conclusion that Sleep Scheduling Network Central Location is better routing performance than the existing scheme.
Table 1 Performance Analysis
Load balancin g (bytes)
Energy Consumption (joules)
Cache Replacemen t
Dela y (ms)
CONCLUSION AND FUTURE ENHANCEMENT
A novel scheme to support cooperative caching in DTN basic idea is to intentionally cache data at a set of NCL, which can be easily accessed by other nodes. Appropriate NCL selection based on a probabilistic metric and coordinates caching nodes to optimize the tradeoff between data accessibility and caching overhead. Extensive simulations show that our scheme greatly improves the ratio of queries satisfied and reduces data access delay, when being compared with existing schemes. Sleeping scheme could essentially decrease the delay of alarm
broadcasting from any node in WSN. Moreover, the alarm broadcasting delay is independent of the density of nodes in WSN. Theoretical analysis and conducted simulations showed that the broadcasting delay and the energy consumption of the proposed scheme is much lower than that of existing methods.In future enhancement Sleep scheduling algorithm is used it only uses the nodes which are required for the process, and the nodes which are not in use are kept in sleep state. So it reduces the traffic and increases packet delivery ratio. In this paper sleep scheduling algorithm is used to transfer the information in a single network. In the future enhancement, sleep scheduling algorithm is used to transfer the information and cover long distance area. So it will be useful to cover larger area in shorter time.
REFERENCESA Cooperative Caching for Efficient Data Access in Disruption Tolerant NetworksWei Gao, Member, IEEE, Guohong Cao, Fellow, IEEE, Arun Iyengar, Fellow, IEEE, and Mudhakar srivatsa, Member, ieee ieee transactions on mobile computing, vol. 13, no. 3, march 2014.
A. Balasubramanian, B. Levine, and A. Venkataramani, DTNRouting as a Resource Allocation Problem, Proc. ACMSIGCOMM Conf. Applications, Technologies, Architectures, andProtocols for Computer Comm.
A. Chaintreau, P. Hui, J. Crowcroft, C. Diot, R. Gass, and J. Scott,Impact of Human Mobility on Opportunistic ForwardingAlgorithms, IEEE Trans. Mobile Computing, vol. 6, June 2007.
B. Tang, H. Gupta, and S.R. Das, Benefit-Based Data Caching inAd Hoc Networks, IEEE Trans. Mobile Computing, vol. 7 Mar. 2008.
E. Yoneki, P. Hui, S. Chan, and J. Crowcroft, A Socio-AwareOverlay for Publish/Subscribe Communication in Delay TolerantNetworks, Proc. 10th ACM Symp. Modeling, Analysis, and Simulation of Wireless and Mobile Systems .
F. Li and J. Wu, MOPS: Providing Content-Based Service inDisruption-Tolerant Networks, Proc. Intl Conf. Distributed ComputingSystems, .
H. Zhu, L. Fu, G. Xue, Y. Zhu, M. Li, and L.M. Ni,Recognizing Exponential Inter-Contact Time in VANETs,,2010.
I. Psaras, L. Wood, and R. Tafazolli, Delay-/Disruption- TolerantNetworking: State of the Art and Future Challenges, technicalreport, Univ. of Surrey, 2010.
J. Burgess, B. Gallagher, D. Jensen, and B. Levine, MaxProp:Routing for Vehicle-Based Disruption-Tolerant Networks, Proc.IEEE INFOCOM, 2006.
J. Reich and A. Chaintreau, The Age of Impatience: OptimalReplication Schemes for Opportunistic Networks, Proc. ACMFifth Intl Conf. Emerging Networking Experiments and Technologies(CoNEXT), 2009.
J. Zhao, P. Zhang, G. Cao, and C. Das, Cooperative Caching inWireless P2P Networks: Design, Implementation, and Evaluation,IEEE Trans. Parallel & Distributed Systems, vol. 21, no. 2, Feb. 2010.
L. Yin and G. Cao, Supporting Cooperative Caching in Ad HocNetworks, IEEE Trans. Mobile Computing, vol. 5, no. 1Jan. 2006.
M.J. Pitkanen and J. Ott, Redundancy and Distributed Caching inMobile DTNs, Proc. ACM/IEEE Second Workshop Mobility in theEvolving Internet Architecture 2007.
P. Cao and S. Irani, Cost-Aware WWW Proxy CachingAlgorithms, Proc. USENIX Symp. Internet Technologies andSystems, 1997.
P. Costa, C. Mascolo, M. Musolesi, and G. Picco, Socially AwareRouting for Publish-Subscribe in Delay-Tolerant Mobile Ad HocNetworks, IEEE J. Selected Areas in Comm.June 2008.
P. Hui, J. Crowcroft, and E. Yoneki, Bubble Rap: Social- BasedForwarding in Delay Tolerant Networks, Proc. ACM Mobile Hoc.
S. Ioannidis, L. Massoulie, and A. Chaintreau, DistributedCaching over Heterogeneous Mobile Networks, Proc. ACMSIGMETRICS Intl Conf. Measurement and Modeling of ComputerSystems, 2010.
S. Martello and P. Toth, Knapsack Problems: Algorithms andComputer Implementations. John Wiley & Sons, 1990.
S.M. Ross, Introduction to Probability Models. Academic, 2006.
W. Gao and G. Cao, User-Centric Data Dissemination inDisruption Tolerant Networks, 2011.
W. Gao, G. Cao, A. Iyengar, and M. Srivatsa, SupportingCooperative Caching in Disruption Tolerant Networks, Proc.Intl Conf. Distributed Computing Systems (ICDCS), 2011.