Delay Tolerant Network - Journal

1

Abstract—The interest in challenged networks is increasing

and many researchers seek reliable end-to-end connectivity under harsh environments, specifically long propagation delay, high error rates, low data rate, and intermittent connectivity. The concept of a Delay Tolerant Network was introduced to provide challenged networks based upon reliable transmission and interoperability with an overlay network. In this paper, we present a comprehensive overview of Delay Tolerant Network and introduce a study case about the implementation of such a network. This paper is designed to encourage the exploration of this field by presenting the basic concept and a study. Keywords: Delay Tolerant Networks, Routing, Security, Challenge Network

I. INTRODUCTION ANY evolving wireless networks such as military, space, and underwater networks have

characteristics different from the Internet. For example, the instability of the links, long propagation and queuing delays, extremely asymmetric data rate, and high link error rates. For these reasons, new communication protocols are required to reduce the number of retransmissions while providing reliable transmissions despite the high error rates and long delays [3].

A Delay/Disruptive Tolerant Network (DTN) is designed to provide interoperable communications between a wide range of networks with different performance standards, environments, and reliability in case of failure of hardware (network) and or software (protocol) [2, 3]. A DTN must accommodate long delays between and/or within

Manuscript was received on March 12, 2011, and then improved on May

15, 2011 based on comment of Prof. G. Q. Maguire Jr. This work was done while Authors were master student of Communication

Systems, at KTH Royal Institute Of Technology. Laili Aidi contributed for part II History, part III Driving Force, part IV Protocol and Overlay Architecture, part VII Routing, part IX Technology and part X Implementation. Jung Changsu contributed for part I Introduction, part V Network Architecture, part VI Bundle and Encapsulation, part VIII Security, part XI Study Case: KioskNet System, and part XII Conclusion.

The work was submitted to Prof. G. Q. Maguire Jr. as assignment of IK2555 - Wireless and Mobile Network Architectures class. The authors are solely responsible for the contents of this work.

the network.

II. HISTORY DTNs were originally conceived to support the

Inter-planetary Internet (IPN) [3]. There was growing demand for a new network architecture to support communications in the context of long propagation delay, low data rates, and intermittent connectivity. The Interplanetary Internet initiative tried to find a solution and suggested a new network architecture to support reliable transmission between a station on the Earth and satellites, with an overlay network [5].

The Internet Research Task Force (IRTF) DTN Research Group and the DARPA (Defense Advanced Research Project Agency) Disruption Tolerant Networking program advanced this concept. The IRTF DTN Research Group generalized the concept of an Interplanetary Internet into challenged networks. DARPA was interested in the development of protocols for transmitting bundles to DTN nodes [18].

III. DRIVING FORCE

A. Challenged Network Some of the characteristics of a challenged

network are mentioned below [2, 7]: 1. Path and Link

• High Error Rate and Asymmetric Data Rates. The transmission rates may be low; the latencies may be large, connectivity may be intermittent, high mobility in combination with weak signal strength and aggravating circumstances result in high link-error rate making end-to-end reliable communication difficult. Furthermore, due to the intermittent connectivity, data rates may be low or highly asymmetric; the return channel may be unavailable, and the elapsing time between a

Delay Tolerant Network Laili Aidi Jung Changsu

School of Information and Communication Technology KTH, Stockholm, Sweden {aidi, changsu}@kth.se

M

2

request and response may be hours rather than milliseconds, and communications that do arrive may exhibit a high error rate [19].

• Disconnection. The lack of end-to-end connectivity may be more common than connectivity, due to high host or network mobility (satellite passes, moving vehicle, etc). Motion of other objects or interference, opportunistic (nodes arrive in communication range due to random walk), dynamic membership of individual nodes, low-duty-cycle, and unpredictable operation are common attributes (for example in low-capability devices such as might be used in sensor networks).

• Long and Variable Delay. The queuing time in this network could be extremely large, hard to estimate, and source-initiated transmission might be expensive because it is limited [3]. Thus, there is a need to store data in a buffer or queue for a potentially long period at each router, if there is no direct path to the destination node [19].

2. Network • Interoperability Consideration / Intermittent

Connectivity. Interoperability on a large scale is rarely designed in challenged networks [3]. This is because these networks tend to be simple and local in scope. Partitions may occur because of geographic distance, lack of radio signal strength or other factors [19]. Because they are deployed on limited memory and power devices, thus cross communication all over that links has not become a primary feature. There is the problem of frequent failure, low reliability, and or congestion. Thus, there is no guarantee of discontinuous end-to-end connectivity.

• Security. The approach to only secure the endpoints of the network is not sufficient due to the link capacity limit. Thus, the access to the service should be protected at the earliest point in the topology.

3. End System • Limited Longevity. The end node, which

frequently a highly integrated, low-power consumption, low-cost device, may not last for

long usage, due to environmental dangers or power exhaustion. It is not feasible to utilize conventional end-to-end acknowledgment schemes to verify delivery, because the network may be disconnected for a long period, in fact the round-trip or one-way delivery time may exceed the sending node’s lifetime. Therefore it is better to delegate the delivery of traffic to another to any other party.

• Low Duty Cycle Operation. The transmission schedule should receive a special consideration in the routing decision, as the duty cycles of the node may be low, in order to achieve reasonable longevity of the entire network. Thus, the communication pattern is often scheduled in advance due to the limited power.

• Limited Resources. There is frequently limitation in memory and processing capability in the node, hence, if the network is designed for reliability, then the end-node should empty their retransmission buffers quickly, rather than wait for an end-to-end acknowledgment.

Figure 1. A Challenged Network’s Characteristics

[3]

B. Interoperability TCP's handshake and slow start mechanism are

sources of further obstacles in a challenged network with a long-delay. Although there are other improved protocols (SCTP, HSTCP, etc), that can multiplex units of application data for multiple sessions over a single-layer connection (association), multiple round trips are still required

3

in order to transmit application data for session setup [12]. There are several approaches that could adapt the Internet to challenged networks unusual characteristics [2]: 1. Using Link-repair approach. This approach

enables Internet protocols to operate over a comparatively well-performing physical infrastructure, by maintaining the end-to-end reliability and fate-sharing model of the Internet.

2. Attaching network-specific proxy agent at the edge of the Internet. This approach provides access from the Internet to the challenged network and vice versa. Moreover, in order to achieve interoperability

between large diverse networks, with extreme environments and network partitioning, these approaches above will not be adequate or desirable [2]. Attaching a proxy agent does not provide a solution when using a challenged network for data transit purposes, such as to access remotely deployed conventional networks (Intranet) via challenged networks (as an intermediate network).

Thus, there is a need to define a new standard that provides end-to-end communication through multiple regions in a disconnected network, which can tolerate errors and large variable-delay environment using a generalized suite of protocols. Another approach has been suggested, based on a message-oriented reliable overlay architecture, forming an internetwork of challenged Internets [2]. The design is based on the interoperability properties of the classical Internet design, the robust non-interactive delivery semantics of electronic mail, and a subset of the classes of service of a postal system.

C. Store-and-Forward message switching A store-and-forward based DTN was originally

designed to support an Interplanetary Internet (IPN). It is designed to operate above the existing protocols in various network architectures and to use store-and-forward message switching, where the data is transferred from a node, held until this node has a scheduled transfer, and then forwarded it to another potentially dissimilar network [2]. Intermittent connectivity in the Internet can lead to packet loss and even termination of a session (in the

case of session oriented protocols), but a store-and-forward DTN isolates the delays and can hide this intermittent connectivity at the cost of storage, potentially duplicated messages, and increased delay.

This type of DTN is based on in-network storage, retransmission, name based late binding, and routing that is tolerant of network partitioning [2]. The delivery semantics mechanism, asynchronous messaging and postal mail as Class of Service (CoS) are also widely used in the current network applications such as voicemail and email [14]. Furthermore, the links in such a DTN are expected to be diverse, including Radio Frequency (RF), Ultra-Wide Band (UWB), and Free-Space optical, and/or Acoustic (Sonar or Ultrasonic) technology [3, 7].

IV. PROTOCOL AND OVERLAY ARCHITECTURE A. Bundle Layer

This bundle layer architecture was proposed by Kevin Fall, before it was developed into RFC 4838 by the DTN Research Group [3, 14]. As can be seen in figure 2, in this DTN architecture, there is a bundle layer that ties the Application and Transport Layer and all other low-layers into region specific layers. Each regional layer may be a different type of network, but the bundle layer enables them to communicate regardless of the network types, by transmitting bundles using store-and-forward message switching mechanism across or on top of the various regional layers [1, 3, 16].

Figure 2. The Bundle Layer [3]

4

This approach enables to communication across

multiple regions without any given effect on the upper-layer application. This Bundle Service Layering can provide general-purpose delay-tolerant protocol services: custody transfer, segmentation and reassembly, end-to-end reliability, end-to-end security, and end-to-end routing. Below are the different CoSs, which are provided by DTN Bundle Layer [3]: 1. Custody Transfer is the acknowledgement of a

custodial-acceptance from a node to its previous custodian. This is necessary to implement of retransmission responsibility delegation, so the sender node can transfer its retransmission responsibilities to another node, this freeing resources to be used for another bundle.

2. Return Receipt is the confirmation to the source or reply-to node that the destination node has received the bundle.

3. Custody-Transfer Notification is the notification to the source or reply-to node whenever a node accepts a bundle’s custody transfer.

4. Bundle-Forwarding Notification is the notification to the source or reply-to node whenever a bundle is forwarded to the next node.

5. Priority of Delivery with a value of: Bulk, Normal, or Expedited.

6. Authentication is a procedure to verify the identity of the sender and message’s integrity. The figure below illustrates some of CoSs

described above:

Figure 3. DTN CoS [3]

B. Licklider Transmission Protocol Licklider Transmission Protocol (LTP) is a

retransmission-based reliability protocol that runs over a link with extremely long message round-trip times and/or frequent interruptions in connectivity [12]. This protocol is designed as a reliable "convergence layer" protocol, underlying the DTN Bundle protocol [22]. LTP is point-to-point oriented, while the Bundle protocol moves bundles end to end.

V. NETWORK ARCHITECTURE A. Region

As the objective is to interconnect different networks asynchronously, a DTN can utilize the regional networks, where each network is seen as a region with its specific communication protocol [3, 7]. Region is used to interconnect boundaries between nodes in different network protocols, addressing standard. It is identified by a region ID, which is knowable by the other regions of that DTN [3].

B. Node Each node in a DTN might be a host, router, or

gateway. These entities act as source, destination, or forwarder [3]. 1. Host. A host sends or receives bundles (i.e., it is

the source and or destination of bundle transfer), and requires storage to queue bundles. It needs optional custody transfer capacity for retransmission [3].

2. Router. A router forwards each bundle to another node in the same DTN region, and may optionally support custody transfer. A router requires storage to store incoming packets before forwarding these to outgoing links because [2]: • There is no guarantee of that next hop link is

currently available. • Asymmetric data rate between sender and

receiver • Retransmission due to the high error rate link

3. Gateway. A gateway is an interconnection point that forwards bundles to other DTN regions with different protocol stacks by supporting interoperability. This gateway must have storage

5

for reliable delivery and perform mapping between different transport layers. It also performs authentication and check arriving data before forwarding it [2, 3].

VI. BUNDLES AND ENCAPSULATION

A. Bundles Bundles are messages, which have a bundle

header, control information, and a source-application’s user data. Control information specifies the ways of handling, storing, and disposing of user data. The Bundle layer inserts the bundle header. This header is intended for fragmentation of the message [3].

B. Fragmentation The bundle in the Bundle layer may be divided

into several bundle fragments (smaller routable units so as increase improve the possibility of delivery and increase performance [22]. These fragments will be reassembled at the final destination. The message fragmentation in DTN is based on dynamics mechanisms, which could be [14, 20]: 1. Proactive fragmentation. This approach is used

where the contact volume can be predicted to optimize that contact’s usage. The application data may be separated into smaller blocks and transmitted as independent bundles. Only the final destination is responsible for reassembling these bundles.

2. Reactive fragmentation. This approach is used when there is a disconnection while transmitting fragmented bundles. When a node receives a partially fragmented bundle, this node marks the bundle as a fragment and carries out normal forwarding. The previous-hop node can recognize the delivery of abnormal fragmentation via a convergence layer protocol and it creates a reactive fragment to send the remaining bundles to eliminate waste of partially fragmented bundles [14, 22].

VII. ROUTING A. Name and Address

DTN adopts name tuples, which consist of two variable length strings as addresses for delivering messages to its destination. Name tuples have the form of: {Region Name, Entry Name}. The region name is globally unique and translated in DTN gateways to route messages to the specified region. Using its hierarchical region structures can reduce the size of DTN forwarding tables and support additional flexibility due to the variable length strings [2].

The entity name is an identifier and can be resolved within the specified region, thus it does not need to be globally unique [16]. When messages traverse heterogeneous regions, only its region name is used for identifying its destination region. The and then entity name is only translated within the destination region. This late binding has two advantages in DTN source nodes [5]: 1. The source node can generate and deliver

messages without any knowledge of each different regional identifier systems. As a consequence, various regions can add new naming and addressing systems without changing their regions.

2. There is no delay for mapping the destination to a globally unique address when generating the message.

B. Knowledge Oracles

There are 2 important terms when talking about DTN routing [14]: 1. Contact means a period of time (interval) during

which network connectivity is strictly positive, and the delay and capacity can be considered to be constant. There are several types of contacts:

• Scheduled Contact. A scheduled contact may exist between a base station somewhere on earth and a low earth orbiting relay satellite, as it can be predicted when the link between them will be available and for how long it will be available.

• Opportunistic Contact. Opportunistic contact occurs when two entities are present the same

6

place, where there is neither scheduled nor predicted contact available.

• Predicted Contact. A predicted contact is not scheduled, but predictions of its existence can be made by analyzing previous observations.

2. Contact’s volume means the product of the capacity and the duration of the contact (i.e., the volume of data that can be transferred during the contact).

There are also sets of abstract knowledge

oracles, which encapsulate particular knowledge about the network [33, 34]. These oracles are defined in order to understand the fundamental trade-off between performance and knowledge, which is required by different routing algorithms [33]. If contacts and its contact’s volumes are known ahead of time, intelligent routing and forwarding decisions can be made (optimally for small networks). Below are the set of abstract Contact Knowledge Oracles: 1. Null Contact Oracle, this happen when no

question about contact opportunity can be answered, it represents no knowledge about network topology [26].

2. Contacts Summary Oracle can provide long-term network topology or contacts aggregate statistics, thus enabling calculation of the average waiting time until the next contact. This racle only has partial knowledge; hence it can respond with time-invariant or summary characteristics about contacts (for example, the expected average time between contact occurrences and average contact duration) [26, 33, 34].

3. Complete Contacts Oracle can answer any questions regarding network topology or contacts between two nodes at any time. Thus it can specify the exact time when a contact will occur, the duration, capacity and delay of the contact, etc. A contacts summary oracle can be constructed using the Contacts Oracle, but not vice versa [26, 33, 34].

4. Queuing Oracle can give information about instantaneous buffer occupancies (queuing) at any node at any time and route around congested nodes. It is the most difficult oracle to realize

distributed system, because the queuing oracle is affected by both new arriving messages and the routing algorithm choices [33].

5. Traffic Demand Oracle can answer any questions regarding the present or future traffic demand. It provides the set of messages injected into the system at any time [33].

Moreover, Group Membership Oracles to

encapsulate particular knowledge about the group dynamic, as described below: 1. Local Membership Oracle can only answer

questions regarding group membership of the node itself [26].

2. Delayed Membership Oracle can answer any questions regarding membership of an endpoint at a specific time [26, 34].

3. Complete Membership Oracle can answer any question regarding membership of all nodes at any time [26, 34].

Based on those oracles, we can classify the

routing algorithms in DTN into several classes [33]: 1. Zero Knowledge is a class of algorithms that do

not utilize any oracle, thus they may perform poorly.

2. Complete Knowledge is a class of algorithms that utilizes all the oracles (contacts summary, complete contact, queuing and traffic demand).

3. Partial Knowledge is a class of algorithms that uses one or more of the other oracles (congestion, queuing). The message is routed independently based on the future traffic demand.

C. Strategy The traditional routing objective is to select a

path that minimizes some simple metric (e.g. the number of hops). However, the most desirable objective of routing in DTN is not immediately obvious, although the natural objective is to maximize the probability of message delivery because of its challenges [33].

There is a need to define a new routing protocol for the DTN architecture, because the assumptions (continuous connectivity, low delay, and low packet loss), which are used in traditional routing protocols

7

(distance vector or link state) are no longer valid in DTN network [16]. The DTN routing topology is a time-varying multi-graph where there is no guarantee of the availability of the next hop's link for a certain or long time period [20]. In this architecture, the data rate between sender and receiver could be asymmetric, as one node might be much faster than another and the characteristic of high error rate link might also cause retransmission.

DTN Nodes route the bundle along the path. As illustrated in figure 4 below, the DTN Gateway has the same double-stack as a DTN Router. However, the gateway utilize different lower-layer protocols allowing them to bridge between regions that use different lower-layer protocols and taking inter-region routing responsibility, while the DTN Router supports intra-region routing [3]. DTN routers need storage for queuing because of the fundamental limitations discussed above.

Figure 4. DTN Protocol stacks and routing [3] Routing computation in DTN becomes

challenging because the delivery paths through the graph are lossy, and contact intervals and volumes are unknown precisely ahead of time [14]. These issues still become an active area in the (emerging) research of DTN. The performance of different routing algorithms can be compared using performance metrics, as shown below [17]: 1. Message delivery ratio, defines number of

unique multicast bundles that successfully arrive completely at all the receivers over the total number of bundles which are expected to be received.

2. Data efficiency, defines ratio between the unique bundles received by the receivers and the total data traffic.

3. Overall efficiency, defines ratio between the unique bundles received by the receivers and

total traffic generated (both data and control packets) in the networks;

4. Average message delay, defines the average of the end-to-end bundle delivery latencies for each algorithm. The routing strategy in DTN itself is classified

into 2 categories, which is implemented in different DTN, where each of them has different characteristic and numerous approaches for its routing protocol, as shown below [16]: 1. Deterministic Routing. This strategy is build

based on the assumption that the next nodes and the connection between them are known. Thus, protocols that use this strategy are implemented in deterministic or predictable topologies.

2. Stochastic Routing. This strategy is built based on assumption that the network behavior is unpredictable. The protocols that use this strategy depend on decisions regarding where and when to forward messages and implemented in stochastic and time-evolving topologies. A simple approach could be just forwarding the message any node that is reachable, or based on history data, mobility patterns, etc.

More specific Unicast routing techniques in

DTN and comparisons can be found in [27] and [33].

D. Anycast and Multicast

The goal of multicast routing is to reach all nodes in the group, while Anycast routing is to reach at least one node from a particular group. Thus, both of them need mechanism to guide replication, forwarding, and buffer management decisions [31]. Anycast and Multicast in a DTN are challenging due to unpredictability of network connectivity, long delivery delay, and limited storage capacity characteristic [26, 30]. They also have to deal with dynamic group membership, because the group membership may change during the bundle delivery; introducing ambiguity in Anycast and Multicast semantics. There are several semantic models in DTN for Anycast and Multicast [26, 30, 34]: 1. Temporal Interval Membership (TIM) /

Temporal Membership (TM) model:

8

membership is determined by the time interval (membership interval), thus this model is a time-based definition of group membership.

2. Temporal Point Membership (TPM) / Temporal Delivery (TD) model: the membership is determined by both the membership interval and delivery interval that indicates the time period during the message should be delivered to the intended receivers.

3. Current Membership (CM) / Current-Model Delivery (CMD) model: the message should be delivered to the node that is a current member of the group when the message arrives to it. This is what makes this semantics different from the other two above, where receivers are not required to be group members at the time of message delivery. The figure below illustrates semantic models

described above:

Figure 5. DTN Semantic Models [26, 34]

The important point in order to achieve Anycast

in DTN is to expose knowledge to the routing protocol about the groups in the network so it can directly act based on that knowledge [31]. Different from Unicast DTN, where the bundle destination is determined when it is generated; in Anycast DTN, the destination can be any one of a group of nodes and both the path to a group member and the destination can change dynamically during routing the process, depending on the current device movement situation. There are several Anycast routing strategies that have been proposed [31]: 1. Single-copy technique / Forwarding-based.

This technique is generally much less wasteful of

network resources, because only a single copy of the message can exist in network storage [33]. The message can be held until the destination is found or be forwarded through intermediate node via a utility metric [31]. However, in general, this technique also limits the message delivery rates in many DTN. There are several approaches based on this technique:

• Expected Multi-Destination Delay for Anycast (EMDDA) that utilizes the uncontrolled random movement of the node. The Anycast routing is determined by evaluating different routing metrics (Practical Expected Delay / PED) for selecting forwarding nodes [30]. This metric characterizes the expected delay of taking different paths with corresponding probability of connectivity between the nodes. However, this approach does not consider network traffic during its routing selection and assumes all nodes are stationary, except for a few mobile nodes that act as message carriers, thus providing a very constrained environment for evaluation.

• The Anycast genetic algorithms to make route decisions [32]. This approach assumes that all mobility is deterministic and known ahead of time, which is not always true for DTN [31].

2. Multi-copy techniques / Replication based. This technique increases the message delivery rates, since multiple copies of a message exist in the network. However, it uses more network resources and inherently is not scalable. There are several approaches based on this technique [31]:

• Flooding-based protocols, appropriate in non-resource-constrained environments, thus approach does not place a limit on the number of times a message can be replicated, and focuses on smart buffer management and transmission ordering techniques to handle potentially large numbers of replicates. Example protocols include as Epidemic, ProPHET, MaxProp, and RAPID [31].

• Quota-based protocols, suitable for resource-constrained environments, thus there are a hard limit on the number of times a message is

9

allowed to be replicated. This is done by attaching a quota flag to every message, so the total number of replicas from a message never exceeds its original quota. Example protocols include Spray and Wait, Spray and Focus, and EBR [31].

Multicast DTN has a semantic model where the

bundle needs to be buffered in the node's storage until it is deleted because of buffer overhead or time expiration [26, 34]. This approach reduces delay and improves bundle availability. Thus, other nodes (except the source) can handle the join request and a node can send its buffered bundle to the new members when contact is available. This is necessary to overcome network partitions and handle delay of join requests. The node also maintains the local forwarding state for each of its buffer bundles to determine the route or next hop [34]. This Forwarding State is computed and updated based on the node's Contact and Group membership knowledge. This forwarding state is necessary because the data is forwarded in the units of a bundle and intended receiver depends on the semantic model that is used.

The traditional Multicast methods for Internet and mobile ad hoc networks are not suitable in DTN, because of the frequent network partitions and sparse connectivity among nodes making the of complexity to maintaining a source-rooted multicast tree during a multicast session too high. Additionally, the application data would suffer from large end-to end delivery latencies [17]. Moreover, the traditional approaches may fail to deliver a message when the link is highly unavailable. There are several existing routing approaches for supporting multicast communications in a DTN: 1. Unicast-based routing (UBR) / Unicast-

Multicast (U-Multicast), here the Unicast transfer mechanism is used to realize a multicast service, for example the source sends a copy of the message to every intended receiver, this message encapsulates the original multicast message [17, 26, 34]. The source node buffers the multicast message and sends new Unicast messages when it learns of new intended receivers, then removes this message after it is

transmitted to the next hop. The destination node will de-capsulate the message and forwards the original multicast message to the intended receiver according to the message's delivery constraints. This approach has the worst delivery ratio and routing efficiency because it sends a separate copy of the messages to each receiver, which significantly increases contention for node storage and transmission opportunities, and results in message drops.

2. Broadcast-Based Routing (BBR) / epidemic routing, here the message will be flooded throughout all the nodes in the network in order to reach the intended receivers [26, 34]. This approach achieves the highest delivery ratio that does not require any knowledge about contact or membership [26]. It also has the lowest delay because messages are flooded to all nodes, hence always following the shortest path [34]. However, it has the lowest routing efficiency because of message redundancy.

3. Tree-Based Routing (TBR), here the message is forwarded along a tree that is rooted at the source and reaches all receivers [26, 34]. The message is only duplicated at nodes that have more than one outgoing path.

4. Group-Based Routing (GBR) uses the forwarding group concept, in order to increase the chance of delivery. Thus the message is flooded within the forwarding group [26, 34]. Along with BBR, this approach also achieves the highest delivery ratio, because the message may be forwarded to receivers via multiple paths, which better exploits available contact opportunities (contact summary oracle).

The figure below illustrates routing approaches described above:

Figure 6. DTN Routing Approach: (a) UBR (b)

BBR (c) TBR (d) GBR [26, 34]

10

5. Static tree-based routing (STBR) constructs the static shortest path tree from the source to the estimated intended receivers of a message, starting at the message generation time [26, 34]. Thus, if a message misses a contact with a node then it needs to wait for the next opportunity to connect to this node, which may significantly increase the message delay and disallows nodes from utilizing more accurate local information to the forward message using a better path.

6. Dynamic Tree-Based Multicasting Algorithm enables the node to dynamically determine the next-hops of a message based on its view of network conditions (local queuing information or newly available contact information) [34]. Since this algorithm can adapt to network conditions, it performs slightly better than STBR.

• DTBR (Dynamic Tree-Based Routing), here the upstream node assigns the receiver list for its downstream nodes based on its network condition view [26]. The downstream nodes can only forward bundles to the receivers in the list. However, this technique assumes each node has complete knowledge or summary of the link states in the network. Unfortunately, this is hard to achieve in practical scenarios.

• On-demand Situation-aware multicast (OS-multicast), here a unique multicast tree is constructed for each bundle and the tree is adjusted at each intermediate DTN node according to the current network conditions [17]. The node dynamically adjusts the initially constructed tree when it receives a bundle, based on its view of network condition views. Thus it has a smaller delay, better message delivery ratios, any newly discovered path will be quickly utilized, and achieves higher efficiency when the probability of link unavailability is high and the duration of link downtime is large. However, the downside of this approach is that, the receiver may receive multiple copies of the same bundle and relies on opportunistic connectivity among nodes for delivery.

Figure 7. DTBR and OS-Multicast [17, 24]

7. Context Aware Multicast Routing (CAMR)

[17] is a density based adaptive multicast routing scheme, which uses opportunistic connectivity and additional information, for example, node location and node velocity. This scheme increases average message delivery delay. However, it provides efficiency and high delivery ratio with reasonable data efficiency, especially when the network becomes sparser.

Figure 8. Conceptual performance of DTN muticast routing approaches in different levels of knowledge

[17]

VIII. SECURITY A. Issues

Typically a DTN has very limited resources such as transmission bandwidth, storage, and processing cycles. Therefore, some restrictions should be placed on accessing this network and delivering messages without authorization and authentication. Moreover, even authorized applications should be restricted when they attempt to access services that they are not allowed to use. In DTN, there are two different security aspects that should be considered, these are LTP security and Bundle protocol security [16].

11

1. LTP Security. LTP is a point-to-point protocol thus the upper or lower layer will handle most security concerns. For this reason, LTP only supports data integrity with LTP authentication and LTP cookie mechanism, which uses a random number to make a DoS attack more difficult [16].

2. Bundle Layer Security. The Bundle Layer is vulnerable to attack because it is an overlay network. Thus the bundle layer encounters many threats and should consider lower layer’s security issues [23].

The bundle layer protocols can be attacked by multiple underlying network components because bundles may travel across multiple networks. During this journey, bundles are modified intentionally and new bundles are inserted by underlying networks. To prevent unauthorized insertion of bundles, DTN nodes should have the ability to detect and delete these unexpected bundles [23]. Network accesses and resource consumptions from unauthorized objects can cause serious threats because of the node's limited resources, particularly storage and battery power [23]. The bundle layer can be exploited by increasing resource consumption, for example by a DoS attack. Due to the long latency in DTN, the damage may be worse than traditional networks. To avoid unexpected resource consumption, we should only accept authenticated messages and drop others [16, 23].

B. Bundle Security Authentication is carried out in forwarding nodes

(routers and gateways) and these nodes also verify the authenticity of sender’s information to protect network resources from unauthorized traffic as early as possible. This also differentiates DTN networks from other networks, which mutually authenticate the user identities and message integrity, but the router that forwards the traffic itself is not authenticated [3].

The DTN also has a unique mechanism when using public-key cryptography, where both users and forwarders have their own key-pairs and certificates. A sender uses its private key to sign

bundles and create bundle-specific signature. After verifying the sender’s identity and CoS in the forwarding node, the forwarding node replaces the sender’s signature with its own and forwards the data [3].

Figure 9. Security steps [3].

C. Bundle Security Protocol Specification

In the “Bundle Security Protocol Specification” [29], the functionality of data integrity and confidentiality are provided as the bundle security protocol. There are four security blocks in this specification [29]: 1. Bundle Authentication Block (BAB), supports

a function for assurance of the authenticity and integrity of the bundle travelling along a single hop from forwarder to intermediate receiver. The operation of this block is to protect a bundle on a hop-by-hop basis unlike other security blocks. Currently BAB only defines a shared-key Hash Message Authentication Code (HMAC) and the key does not need to be unique but is only required to be shared between nodes.

2. Payload Integrity Block (PIB), the authenticity and integrity of the payload from the PIB security-source to the PIB security-destination. Any node, which is located in between the security-source and the security-destination, can examine the authentication information.

3. Payload Confidentiality Block (PCB) specifies that the encryption of the payload be done wholly or partially by the PCB security-source for the purpose of protecting the content while being transported to the PCB security-destination.

4. Extension Security Blcok (ESB) provides security not for payload blocks, but rather for non-payload blocks in a bundle so ESB is not applicable to PIB and PCB that are related to payload security blocks. The ESB is located in

12

the same position where it should be protected in the bundle [29].

D. Open Issues in Bundle Security The bundle security protocol is still under

development and some critical issues still remain. We will present some issues for the bundle security protocol below [23]: 1. The level of flexibility. The bundle security

protocol can combine applications of the confidentiality and integrity services flexibly but it prevents insecure combinations of application such as including plain-text signatures. Therefore, the level of flexibility is an open issue in the bundle security protocol. This flexibility may support a Virtual Private Network (VPN), but the complexity can cause high costs for implementation and be insecure [23].

2. Key Management. No key management schemes exist in DTN deployments. In fact, existing schemes need hard coding. Additionally, replacement of trusted keys in all nodes makes it difficult to adopt this scheme. One possible solution is to adopt the resurrecting duckling scheme that is suggested for ad-hoc networks. If a node has a “close encounter” with another node, the node can exchange a key through a trustworthy channel and store it for later use. If DTN nodes have enough space in their storage, these nodes may flood public keys whenever nodes encounter each other [23].

3. Canonicalization of bundles. The definition of a canonical bundle form is an open issue for data integrity. The sender and receiver require the same bytes of bundle of signature for integrity, but the bundles may be changed during traversal between nodes [23].

IX. TECHNOLOGY A. Model and Feasibility

The trend of DTN seems to be more of an analytical research, mobility model, etc [11]. The ongoing work is to extend the DTN architecture to smart mobile phone-based mobile ad hoc networks (MANETs) so that a node can effectively use multiple communication links and networks. This

implementation is intended to reduce complex operations by the user, enabling them to perform data communication operations seamlessly and more effectively in terms of delay, intermittent environments, etc. This is not possible using the TCP/IP based architectures [7].

Currently, the DTN architecture research is carried on by several research groups, such as [3, 15]: The Internet Research Task Force’s Delay-Tolerant Networking Research Group (DTNRG), The InterPlaNetary (IPN) Internet Project, NASA Jet Propulsion Laboratory (International Space Station Research), Google Laboratory, Intel Research Corporation, SPARTA, The MITRE Corporation, Distributed Systems Group - Trinity College Dublin, ISTRAC - ISRO, University of California - Berkeley (UCB), University of California - Santa Barbara (UCSB) , University of Southern California (USC), Helsinki University of Technology, Luleå University of Technology, University of Massachusetts Amherst, etc.

The wireless DTN technology also may be diverse and implemented using several technologies, such as [3]:

1. Radio Frequency (RF) 2. Ultra-Wide Band (UWB) 3. Acoustic (Sonar, Ultrasonic) 4. Free-space Optical Communications (FSOC) is

an extreme example of the directional antenna mobile ad hoc network (MANET) [25]. The networking design issues in FSOC come from the challenges in pointing, acquisition, and tracking. It becomes extremely difficult because of long ranges and mobility on rugged terrain, and there is a resultant resource allocation problem's precision of pointing requirement, need to be done with an optical laser head in one connection service. Thus, there is a need in FSOC for topology control. Since the DTN approaches are designed to overcome the network with intermittent connectivity, thus it can be fundamental to solve the FSOC networking problem.

B. Relevant Standard 1. Homing-pigeon-based DTN (HoP-DTN), an

experimental method in RFC 1149: Standard for the Transmission of IP Datagrams on Avian

13

Carriers [8] and RFC 2549: IP over Avian Carriers with Quality of Service [9]

2. Licklider Transmission Protocol - Security Extensions, IETF RFC 5327, experimental [10]

3. Licklider Transmission Protocol - Specification, IETF RFC 5326, experimental [11]

4. Licklider Transmission Protocol - Motivation, IETF RFC 5325, informational [12]

5. Bundle Protocol Specification, IETF RFC 5050, experimental [13]

6. Delay-Tolerant Network Architecture, IETF RFC 4838, informational [14]

X. IMPLEMENTATION The development of DTN networks has been

more sophisticated and detailed in terms of protocols and approaches, analysis of stateless routing algorithms and has thus far found no commercial use [11]. DTN is currently being studied in space networks and other research environments (such as Earth Applications) in order to exploit stressed, disconnected, disrupted networks, preventing failures, enhance safety and security, and to learn new knowledge [15]: 1. The Outer Space / Deep-Space Networks:

Inter Planetary Networks. The Interplanetary Internet (IPN) is a computer network in space, and an example of wireless network outside of the Internet [3]. The communication outside of the Internet is done by using independent networks, that each might be run on power-limitation, specialized communication that is not be able to exchange information. They also have different link delay and connectivity, data-rate asymmetry, error rates, addressing and reliability mechanism, QoS and trust boundary, etc [3].

The IPN is defined as a network of regions, such as a terrestrial Internet region, a surface of the planet region, or ground-to-orbit region, etc. Each region has its own communication in terms of security, resource maintenance, etc [6]. It is a store-and-forward network that runs over interplanetary distances, might be disconnected, and run over a wireless backbone with error-prone links and delays ranging to minutes or hours, when a connection even exists [4].

DTN is used to increases the robustness of the communication network and timeliness of data returned from operating space assets, so it can reduce risk and cost, increase safety and science return, and improve operational awareness [15]. Additionally, DTN can reduce human labor costs through automation of communications operations.

Figure 10. The Challenged Network Examples [3]

2. Terrestrial Civilian Network. Even though the

Outer-Space implementation is the primary beneficiary of the DTN research, many terrestrial network implementations are used and contributed to DTN research as well:

• Drive-by Vehicular and Ferry based Networks [21] include DakNet, Message ferry (hybrid between MANET and DTN), Village network.

• Mule Networks / Node in a box: Hagle, Zebranet (tracking wild animal in wildlife, manage their habitat effectively by attaching wireless sensor node, collecting location data and opportunistically reporting their history when they are in radio range of base station [2, 19, 22]), Sámi Network Connectivity (Reindeer herd tracking by the Saami tribesmen in Arctic Circle [22]), SWIM, Mobile Ad Hoc Network, AUDTWMN (Water monitoring application Test bed for DTN research [22]), Carrier Pigeons (Implemented by Bergen Linux users group: RFC 1149, RFC 2549).

• Challenge Link / Flakynet, for example in the Remote / Developing region: Tier, Seismic monitoring (Early warning system against earth quakes, volcano and landslides [22]), SenDT (implemented by Trinity College Dublin Ireland to monitor lakes in Ireland [22]), UUCP [21].

14

3. Mobile Sensor Networks (Acoustic underwater networks). This network is designed to exist in a large-scale area and nodes per network [5]. It has the characteristic of an extremely limited end-node power, memory, CPU capability and scheduled communication between nodes. There is also an implementation of “proxy” nodes in the network in order to translate the data in the Internet protocols to its native protocols.

4. Wireless Military Tactical / Battlefield networks. The network may run in environment with several causes for disconnection, such as mobility, environmental factors, intentional jamming, etc [5]. There is a strong requirement for infrastructure protection and the sensor or field agent may be deployed over limited wireless coverage. The challenge also increases because the data traffic may be delayed as it may have to compete for a limited bandwidth with other high priority services, and there is a concern in the battery and wide physical dispersant of agent. Thus, wireless connectivity could not be continuously maintained among all agents [15, 19].

XI. STUDY CASE: KIOSKNET SYSTEM A. Overview

The KioskNet system was developed by the University of Waterloo for providing low cost Internet services to the poorest villages of developing countries using the DTN concept. This system uses vehicles to deliver data from villages to Internet gateways and provides various kinds of services to rural residents, for example, birth, marriage, and death certificates; medical consultation, and agricultural problems. This system should have some essential requirements for the reliable connections and the low cost of maintenance. Especially, the cost is a very critical issue for sustaining this system in those remote regions. Moreover, deploying this system had great challenges because of many obstacles such as limited electrical power, dust, mechanical damages, computer viruses, frequent failure of kiosk computers and network connections [24].

KioskNet challenged to make a robust system with two key concepts. The first one is the adoption of a single-board-computer, low-cost and low-power kiosk controller for wireless communication using a vehicle. A vehicle delivers data to a gateway or receives data from a gateway. This ‘mechanical backhaul’ can allow Internet access without the cost of equipments such as satellite dishes and towers in remote areas. Second, KioskNet uses refurbished PCs, which use boot images from the kiosk controller that can offer a very secure boot images virus-free. In addition, the refurbished PC does not have a hard disk to avoid a hard disk failure and viruses. Aside from two key concepts, KioskNet has a few characteristics like low-cost (70$/kiosk/month), low power (6~8W), a LiveCD and free software [24].

B. Component Below are the components of KioskNet network:

1. Kiosks. Every kiosk has its own kiosk controller. A kiosk

controller utilizes recycled PCs to provide some functions, those are, a network boot function, a network file system, user management, and network connectivity through dial-up, GSM/GPRS, Very Small Aperture Terminal (VSAT), or mechanical backhaul. A kiosk controller is constantly possesses a wireless network interface or other connectivity, which are mentioned above [24].

This system considered two types of users who can access a kiosk controller for their connection. First, most users are expected to use cheap recycled PCs (terminals) to connect the system. In this case, a kiosk controller provides these diskless PCs with a network boot image and applications by means of Network File System (NFS). Second, some other users who are government officials, NGO members or wealthier residents access kiosks with their own mobile devices. Unfortunately, this system’s software does not support connectivity to these users because of some technical issues.

If kiosks are located in the same geographical area, these kiosks consist of a KioskNet region for routing and certification [24]. Figure 11 represents that a single server entity in KioskNet administrates two regions in this system.

15

2. Ferries. This system defines ferries as many types of vehicles like a car, buses, motorcycles or trains by means of supporting various connectivity options. A ferry has a cheap computer powered by a vehicle’s own battery and the computer has a 20-40GB hard disk and a WiFi network interface. Ferries contact with kiosk controllers and gateways using opportunistic ways for the time duration of 20 seconds to 5 minutes. During this communication period, ferries can transfer 10-150MB of bundles using a store-and-forward mechanism [24]. 3. Gateways. A gateway is always connected to the Internet through DSL or broadband Internet access. This connection of a gateway is possible with its WiFi network interface. Moreover, it has storage to receive data from a ferry and to upload the data to the proxy through the Internet [24].

4. Proxy. This system is expected to support communication between a kiosk user and the Internet but legacy servers cannot provide connectivity when the network has long delays and disconnections like this situation. Therefore, a special proxy is needed for supporting communication with a legacy server. A proxy should be divided by two halves so that one half set up a session for disconnection-tolerant connection with applications within a kiosk controller or mobile devices. The other half establishes a connection with legacy servers instead of intermittently connected users. For data forwarding from a half to the other half within proxy, application plug-ins are required, for example, a SMTP plug-in for sending mail content to a legacy mail server [24]. 5. Legacy Server. The legacy servers are typical servers supporting applications such as IMAP, SMTP and HTTP with TCP/IP protocol [24].

Figure 11. KioskNet overview [36]

C. Security Architecture

The ultimate security goal of this system is to offer the best possible security service to the whole components of KioskNet such as the infrastructures, users and terminals. To meet security requirements and reliable operation, this system needs four distinct entities, which are KioskNet Franchisers, KioskNet Franchisees, KioskNet Users and Application service providers [24, 28]. 1. Entities. We present Security Entities of this

system as below: • KioskNet Franchisers. Franchisers are public

or private organizations such as non-governmental organizations (NGOs). The franchisers own and check the integrity of their KioskNet infrastructure components such as gateways, ferries, Kiosk controllers and proxies. The basic function of franchisers is to detect the improper usage of infrastructures by any entities [24, 28].

• KioskNet Franchisees. Franchisees are private organizations or licensed individuals. The responsibilities of franchisees are to operate their kiosk terminals and protect terminals from malicious software [28].

• KioskNet Users. Users can access KioskNet services and applications that franchisees own and support [28].

• Application Service Providers (ASPs). These entities are licensed by franchisers for providing their applications to a KioskNet as an example of banking services to local residents [28].

2. Certificate. All entities have unique credentials containing a 2048-bit RSA key and a Public Key

16

Certificate. The public key of a franchiser is certified by a secure root CA server at the University of Waterloo using its own private key and this signature is stored as the form of an X.509 certificate. In turn, franchisers issue the certificates of franchisees and ASPs. When users register and create their identification at their kiosks, franchisees certify the users in an automatic way. In addition, franchisers also issue unique credentials to KioskNet’s gateways and ferries. A public key database, preserved at the proxy is used to periodically broadcast public key certificates for users, franchisees and ASPs all over a franchiser’s region and these certificates are also copied to all kiosk controllers [24].

3. Infrastructure integrity. Digital signatures are used on all remote commands and software updates from franchiser for ensuring the security of this infrastructure. In this system, kiosk controllers are very vulnerable to attacks so franchisers do not give root access authorities to kiosk controllers to prevent them from modifying the software and accessing private information [24].

4. Protecting recycled PCs. For protecting recycled PCs from viruses and malicious software, these PCs can only boot using read-only disk images stored in kiosk controllers through NFS protocol. This boot images can be modified and updated by franchiser administrative staffs [24].

5. User data protection. User data is only stored in kiosk controllers and this system offers encrypted virtual space for each user’s home directory. This file system is encrypted with the user’s password and mounted when users login at a terminal with their password. Moreover, users can access their encrypted home directories using the Linux DM-Crypt disk encryption module. With this mechanism, attackers who have a root authority cannot modify other users’ data [24].

6. Communication privacy and integrity. Before delivering user data to the kiosk controller, the encryption and signing of user data is performed at kiosk terminal for supporting privacy and authenticity. In traditional systems’ case, they

use public key encryption for ensuing end-to-end secure communication like SSL but this approach is very difficult to be applied to this delay-tolerant environment due to handshake for generating a shared key. Therefore, the KioskNet generates random 256 bit keys using AES-CBC (Advanced Encryption Standard Cipher Algorithm in Cipher Block Chaining Mode). The recipients encrypt this key using its public key and decrypt the data after decrypting the AES key by using their own private keys [24].

XII. CONCLUSION The main goal of DTN architecture is to provide

interoperability between different kinds of networks in wide-ranging regions even though this network has many limitations such as long delay, intermittent connectivity, limited power, and high error rates. This architecture originated from the Interplanetary Internet architecture then it is spread to the challenged networks more generally.

In this paper, we have summarized an overview of Delay Tolerant Networking, especially about routing and security. In addition, we give an example of the design and implementation of the KioskNet for better understanding of DTN. Moreover, we think that the implementation of DTN is a good solution to supply people in remote areas with the Internet service for their communication to the world even though DTN still has many open issues.

REFERENCES [1] Jimmy Ray. “Delay Tolerant Networking: How to use Twitter and Hulu

on long Space Missions to Mars”. [Online]. Available: http://www.networkworld.com/community/node/43557 [Feb. 16, 2011]

[2] Kevin Fall. “A Delay-Tolerant Network Architecture for Challenged Internets”. [Online]. Available : http://www.dtnrg.org/docs/papers/IRB-TR-03-003.pdf [Feb. 16, 2011]

[3] Forrest Warthman. “Delay Tolerant Networks (DTNs): A Tutorial Version 1.1”. [Online]. Available: http://www.dtnrg.org/docs/tutorials/warthman-1.1.pdf [Feb. 16, 2011]

[4] Scott Burleigh, Vinton Cerf, Robert Durst, Kevin Fall, Adrian Hooke, Keith Scott, Howard Weiss. (2003, Oct). “The Interplanetary Internet: A Communications Infrastructure for Mars Exploration” Journal title. [Online]. IAC-02-Q.3.3.03. 2. Available: http://www.ipnsig.org/reports/IAF-Oct-2002.pdf [Feb. 16, 2011]

[5] Burleigh, S., Hooke, A., Torgerson, L., Fall, K., Cerf, V., Durst, B., Scott, K., and Weiss, H. Delay-tolerant networking: an approach to interplanetary internet. IEEE Communications Magazine 41, 6 (June 2003), 128–136.

17

[6] Andrian Hooke. Class Lecture, Topic: “Interplanetary Internet”. The Ground System Architectures Workshop (GSAW), University of Southern California, March. 4, 2003. [Online] Available: http://sunset.usc.edu/gsaw/gsaw2003/s3/hooke.pdf [Feb. 16, 2011]

[7] Farid Farahmand. “Delay Tolerant Networks: Challenges and Applications”. [Online]. Available: http://www.engr.uconn.edu/~tehrani/teaching/CE_Seminar/Farahmand.pdf [Feb. 18, 2011]

[8] Standard for the Transmission of IP Datagrams on Avian Carriers, RFC 1149, April 1990

[9] IP over Avian Carriers with Quality of Service, RFC 2549, April 1999 [10] Licklider Transmission Protocol - Security Extensions, RFC 5327,

September 2008 [11] Licklider Transmission Protocol - Specification, RFC 5326, September

2008 [12] Licklider Transmission Protocol - Motivation, RFC 5325, September

2008 [13] Bundle Protocol Specification, RFC 5050, November 2007 [14] Delay-Tolerant Network Architecture, RFC 4838, April 2007 [15] NASA. “Delay Tolerant Networking (DTN)”. [Online]. Available:

http://www.nasa.gov/mission_pages/station/research/experiments/DTN.html [Feb. 18, 2011]

[16] Vinod Venkkaraman, Hrishikesh Bhatt Acharya, Harsh Shah. (2009, May 6). Delay Tolerant Networking - A Tutorial. [Online]. Available: www.cs.utexas.edu/~vinodv/files/dtn-tutorial.pdf [Feb. 17, 2011]

[17] Peng Yang, Mooi Choo Chuah. “Context-Aware Multicast Routing Scheme for Disruption Tolerant Networks” [Online]. Available: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.94.338&rep=rep1&type=pdf [March. 4, 2011]

[18] Jon Crowcroft, Eiko Yoneki, Pan Hui., “Promoting Tolerance for Delay Tolerant Network Research,” ACM SIGCOMM Computer Communication Review 38, pp.63-68. [Online]. Available: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.158.9124&rep=rep1&type=pdf [Feb. 17, 2011]

[19] Dave Wick. “Delay Tolerant Networks in a Nutshell.” Bachelor Thesis. University of Bern, Switzerland, 2007. [Online]. Available: http://rvs.unibe.ch/research/pub_files/Wi07.pdf [Feb. 17, 2011]

[20] Michael Freedman. Class Lecture, Topic: “Delay Tolerant Networks (and email)”. COS 461: Computer Networks, Princeton University, Feb. 3, 2010. [Online] Available: http://www.cs.princeton.edu/courses/archive/spr09/cos461/docs/lec22-dtns.pdf [Feb. 26, 2011]

[21] S. Keshav. Class Lecture, Topic: “Naming, Addressing and Routing in Delay-Tolerant Networks”. Future Internet Workshop, Lisbon, Dec. 2006. [Online]. Available: http://www.net.t-labs.tu-berlin.de/arcadia/resources/Keshav-DTN.pdf [Marc. 4, 2011]

[22] Michael Demmer, Eric Brewer, Kevin Fall, Sushant Jain, Melissa Ho, Rabin Patra. “Implementing Delay Tolerant Networking”. [Online]. Available: http://www.dtnrg.org/docs/papers/demmer-irb-tr-04-020.pdf [March. 2, 2011]

[23] Stephen Farrell, Vinny Cahill. “Security Consideration in Space and Delay Tolerant Networks,” [Online]. Available: http://smc-it.jpl.nasa.gov/docs/Abstracts/M47.pdf [March. 2, 2011]

[24] S.Guo, M.H. Falaki, U.Ismail, E.A. Oliver, S.Ur Rahman, A. Seth, M.A. Zaharia, and S.Keshav. “Design and Implementation of the KioskNet System (Extended Version)”. [Online]. Available: http://www.cs.berkeley.edu/~matei/papers/2007/tr_ictd_kiosknet.pdf [March. 2, 2011]

[25] Robert A. Nichols, A. Roger Hammons Jr., Daniel J. Tebben and Anurag Dwivedi. “Delay Tolerant Networking for Free-Space Optical Communication Systems”. Appl. Phys. Lab., Johns Hopkins Univ., Baltimore, MD. [Online]. Available: http://ieeexplore.ieee.org/xpl/freeabs_all.jsp?arnumber=4567371 [March. 2, 2011]

[26] Wenrui Zhao. “Routing and Network Design in Delay Tolerant Networks”. Ph.D. Dissertation. College of Computing, Georgia Institute of Technology, 2006. [Online]. Available: Available: http://smartech.gatech.edu/bitstream/handle/1853/14085/zhao_wenrui__200612_phd.pdf;jsessionid=9193BFA7D09066B5A3F41B6D8544877D.smart2?sequence=1 [March. 2, 2011]

[27] Hemal Shah, Yogeshwar P. Kosta. “Evolution of Routing Techniques, Routing Protocols and Routing Efficiencies for Delay Tolerant Network”. [Online]. Available: http://www.ijcaonline.org/manets/number1/SPE49T.pdf [March. 2, 2011]

[28] Sumair Ur Rahman, Urs Hengartner, Usman Ismail and S.Keshav. “Securing KioskNet: A Systems Approach”. [Online]. Available: https://www.cs.uwaterloo.ca/research/tr/2007/CS-2007-43.pdf [March. 3, 2011]

[29] Bundle Security Protocol Specification, March 2011. [Online]. Available: http://tools.ietf.org/html/draft-irtf-dtnrg-bundle-security-17 [March. 3, 2011]

[30] Yili Gong, Yongqiang Xiong, Qian Zhang, Zhensheng Zhang, Wenjie Wang, Zhiwei Xu. “Anycast Routing in Delay Tolerant Networks”. [Online]. Available: http://www.techrepublic.com/whitepapers/anycast-routing-in-delay-tolerant-networks/2556719/post [March. 3, 2011]

[31] Samuel C. Nelson and Robin Kravets. “Achieving Anycast in DTNs by Enhancing Existing Unicast Protocols”. [Online]. Available: http://www.winlab.rutgers.edu/~snelson/achievinganycast.pdf [March. 3, 2011]

[32] Ederson Rosa da Silva and Paulo Guardieiro. “Anycast routing in delay tolerant networks using genetic algorithms for route decision”. In Proceedings of IDCS, 2008.

[33] Sushant Jain, Kevin Fall, Rabin Patra. “Routing in a Delay Tolerant Network”. [Online]. Available: http://conferences.sigcomm.org/sigcomm/2004/papers/p299-jain111111.pdf [March. 3, 2011]

[34] Wenrui Zhao, Mostafa Ammar and Ellen Zegura. “Multicasting in Delay Tolerant Networks: Semantic Models and Routing Algorithms”. [Online]. Available: http://conferences.sigcomm.org/sigcomm/2005/paper-ZhaAmm.pdf [March. 4, 2011]

18

Peer review from Guojun Wang ([email protected]): The authors have given an excellent description to the delay tolerant network, including the protocols, the overall architecture, the technologies, implementations and so on. The levels between each part are very clear and all the formats are qualified. However, it will be better to give an overview of what you have presented at the beginning of this paper instead of putting that in the conclusion part. You also give one sentence to the origin of DTN, which is followed with tag [7] in the conclusion paragraph. Actually, it reduces the role of conclusion. In addition, you give a new opinion at the end, "Moreover, we think that the implementation of DTN is a good solution to supply people in remote areas... ". New idea should be avoided when you are giving an end to this paper. Author’s Comment: We thanks for deep review from reviewer. The overview of this work appears on the Abstract section, where it was written, "we summarize the overview of Delay Tolerant Network and introduce a case study about the implementation of Delay Tolerant Network." We used "summarize" word instead of "explain", because we know that there are still so much detail of the concept that we cannot cover in this paper, due to the page and word limit. Later we paraphrase again the sentence so it can be more clearly by become, “In this paper, we present comprehensive overview of Delay Tolerant Network and introduce a study case about the implementation of this network”. At the conclusion, we think that, it is important to sum up and deliver the global idea that is appear in the entire paper in simple sentence, which is "the implementation of DTN is a good solution to supply people in remote areas with the Internet service for their communication to the world even though DTN still has many open issues." Thus we think, this last sentence is important in order to conclude the core idea of the technology that has been talked about. However, after carefully think about the point that reviewer mentioned in Conclusion, we decided to remove the citation in the reference, because it will lead reader to think that we still present new idea in this conclusion, however that point is already presented at the main section of the paper including its citation. Peer review from Merabi Kechkhoshvili ([email protected]) First I will make my point about overall feeling from this report. It is obvious that authors had wide range of references and they did a deep research in order to present this document. It is written in really professional manner and is appealing to person who is familiar with particular topic. I would also add that report is done in appropriate style: • Well-divided paragraph structure. • Good knowledge of language and broad vocabulary. • Although the topic is huge, it is written without redundant information.

However there are some minor details that are worth to pay more attention: • Firstly, I want to say about introduction part: it is not describing how the document is organized. I

would like you to put more information about chapters and what are you going to describe in following sections.

• Secondly, while reading the document, several figures are too difficult to understand. I think you need to put more information and some explanation about figures.

• Finally, from my perspective this paper will be difficult to read be people, who are not really familiar with the topic.

But once again I must stress exceptional quality of presented report. Author’s Comment: We thanks for deep review from reviewer. After reading the points of your review carefully, we consider to make some changes in the Abstract part to make the document organization more clearly, by putting sentences about what the reader can expect from this work. We agree that the organization of the document nor should be mentioned in the Introduction part, because that part should be as introductory for the topic

19

itself, not to the document organization or chapter explanation. Moreover, the figures that is presented in this paper have been explained by paragraph above them, and due to limited page and word, we decide not to explain more about that. We’ve also realized this work digs little bit deeper and detail of DTN concept, thus it will not just give basic general overview of this technology.