You are on page 1of 19

Peer-to-Peer Netw. Appl.

DOI 10.1007/s12083-016-0533-5

CTMR-collaborative time-stamp based multicast routing


for delay tolerant networks in post disaster scenario
Amit Kumar Gupta 1 & Jyotsna Kumar Mandal 2 & Indrajit Bhattacharya 1 &
Tamal Mondal 1 & Sourav Sanu Shaw 1

Received: 15 March 2016 / Accepted: 26 October 2016


# Springer Science+Business Media New York 2016

Abstract Due to high chances of loss in connectivity, a Delay been adopted to suit a post disaster condition. Collaborative
Tolerant Network (DTN) can be used to communicate between bundle creation and selection mechanism has been utilized so
nodes without having any fixed connection between the source that localized redundant information flow is minimized.
and destination. A Post Disaster Scenario presents a very chal- Suitable time and space limits have been selected to further
lenging environment to communicate in and to analyze those reduce redundancy. The group forwarding strategy is based
situations is even a harder task to accomplish. It necessitates on probabilistic measures calculated using historical encounter
very efficient co-ordination to successfully accomplish situation records, and on a multiple parameter priority queue. This pro-
analysis and resource management. Efficient co-ordination be- tocol has been implemented in the ONE simulator and is com-
tween relief and rescue teams in such situations can be achieved pared with other existing unicast and multicast routing proto-
through multicasting, since it allows sending single packet to cols on important routing parameters like delivery ratio and
multiple destinations. Though multicasting in MANET has delivery delay. Comparison results show that CTMR can be a
been studied extensively, but the implementation of efficient novel efficient solution to multicasting in a DTN.
multicasting in DTN is a very challenging task due to its fre-
quent partitioning characteristic. In this work a Collaborative Keywords Opportunistic network . Delay tolerant networks .
Time-Stamp based Multicast Routing (CTMR) Protocol has Post disaster scenario . Multicast . Collaborative . ONE
been proposed. The messages have been implemented using simulator
customized bundles, where the destination of a multicast bundle
consists of a group of nodes. Node grouping mechanism has
1 Introduction

* Amit Kumar Gupta In the context of networking, multicasting is the process of


amitgupta.kgec@gmail.com communication between one to many as well as many to many
groups of entities. It differs from unicasting and broadcasting
Jyotsna Kumar Mandal with respect to the number of destinations a message has been
jkm.cse@gmail.com
assigned to. A unicast message has a single destination; a
Indrajit Bhattacharya broadcast message has all the possible entities as destinations
indra51276@gmail.com whereas a multicast message has multiple numbers of selective
Tamal Mondal destinations. Multicast offers many advantages over both
tamalkalyanigov@gmail.com unicast and broadcast mode of communication. It offers quick
Sourav Sanu Shaw and increased probability for message deliveries for well-
souravshw@gmail.com known multicast destinations as compared to unicast. On the
other hand, it reduces overhead by not overwhelming the sys-
1
Kalyani Government Engineering College, Kalyani, West Bengal, tem with messages as in a broadcast mode. DTN [1] is a wire-
India less networking approach that can be used in situations where
2
University of Kalyani, Kalyani, West Bengal, India end-to-end connectivity cannot be guaranteed, and where
Peer-to-Peer Netw. Appl.

networks suffer huge disruptions because of external factors DTN may consist of several paths in multiple hops. Hops are
like wireless radio range limitation, sparse and mobile nodes, the mobile nodes moving from one place to another. The topol-
energy constraints, etc. In any post-disaster scenario DTN net- ogy of the DTN layer often changes dynamically due to fre-
works are preferred. Working of any existing communication quent link ups and downs. As mentioned in [6], data transfer in
infrastructure such as Internet or any cellular network cannot be DTNs occurs in data units called messages or bundles. Hence,
guaranteed in those challenging situations. This results in im- the term message/ bundle can be used alternatively. There are
proper communication infrastructure, where it cannot be en- number of ways in which a node sends a bundle to single
sured when a message reaches the destination. Since TCP/IP destination, such as Epidemic [8], Prophet [9], Max-prop
and other communication protocols cannot work in DTN, [10], Spray and Wait [11], RAPID [12], MBRP [13], Bubble
hence the DTN Research Group has proposed architecture ca- Rap [3], Delegation Forwarding [14], etc. These strategies are
pable of communicating in challenging scenarios like DTN [2]. developed to communicate in DTN in one to one mode. These
DTN has wide applications in terrestrial environment, and post- protocols are not adequate in emergency response scenarios, as
disaster scenario that lack continuous connectivity. Previous there are multiple group of members (e.g. doctors, nurse, police,
studies have shown that in some DTN networks (e.g. Pocket and army) present in a relief/rescue team and they are scattered
Switched Networks, PSN) DTN routing approach in not an at different places. For such group communication, there can be
appropriate solution [3]. It has been suggested that some separate unicast operations to each members of a group. But
forwarding mechanisms can be an alternate approach in these due to limitations in path availability, power and storage restric-
situations. In Internet architecture routing and forwarding are tion in DTN environment, it is not a promising solution [15].
distinguishable. In DTN, routing algorithm is executed for com- Efficient routing using multicasting can be constructed by using
putation of paths for message delivery to achieve routing ob- partial knowledge of unicast routing protocols [16]. In the pro-
jectives. On the other hand, forwarding signifies moving bun- posed work, as mentioned earlier we have taken such partial
dles between two DTN nodes. Some DTN routing algorithms knowledge i.e. forwarding strategy from Prophet unicast
provide forwarding mechanisms, but these approaches are not routing protocol and on top of that developed a novel multicast
appropriate for most DTN scenarios as much of this mobility routing strategy applicable for post disaster scenarios.
pattern is unpredictable [4]. Hence, bundle forwarding based on Multicast helps disconnected rural areas to utilize the oppor-
some network behavior aspects can be an alternate solution. tunities in exchanging information effectively and efficiently,
Again, only forwarding cannot be a good solution when there for applications like data circulation, file sharing, etc. Many
are multiple receivers of a bundle. In DTN, the choice of wheth- multicast protocols have been proposed in Adhoc wireless net-
er to select a routing or a forwarding strategy depends on the works. These include Multicast Extensions to Open Shortest-
implementers [5]. Path First (MOSPF) [17], Distance Vector Multicast Routing
It has been observed that some forwarding strategies like the Protocol (DVMRP) [18], Ad hoc Multicast Routing protocol
Bubble Rap [3] perform better as compared to some unicast (AMRoute) [19] and the On-Demand Multicast Routing
DTN routing protocols in terms of certain network parameters. Protocol in multi hop wireless networks (ODMRP) [20].
But, these forwarding strategies are applicable only when there These MANET routing protocols are not suitable for DTN,
is a single destination for a message. In real DTN scenarios due to the challenge of frequent network partitions. The main
there can be multiple receivers of any message. As multicasting objective of this work is to implement a routing protocol that
supports efficient distribution of data to group members [6], transmits information to multiple destinations via multiple paths
there should be a focus on group formation. As mentioned to achieve better delivery of messages over a DTN.
earlier, multicasting in MANET is not a challenging issue as A Post-disaster recovery area is divided into number of dif-
the receivers or group members of a multicast packet are well ferent sites, as the affected area can be huge and scattered in a
defined. As the data transfer delay in a MANET is too short, large scale disaster. In each of these sites, relief work is per-
group membership changes during data transfer can be ignored formed. Government organizations work together with a num-
[6]. But due to frequent partitioning in DTN, large amount of ber of NGO’s. It has been observed that non-government relief
delay in data transfer can happen and membership changes can organizations choose a site for their relief operations for a par-
be frequent rather than exception. So an efficient multicasting ticular period of time [21]. The main purpose of such relief
and group formation mechanism is needed so that multicast work is to provide help to the needy by providing immediate
packets can be delivered to the members. rescue and emergency resources. The term ‘resource’ in this
The bundle layer protocol [7] defines a series of adjoining context refers to the emergency requirements of any affected
data blocks as bundles, where each bundle has enough infor- area after any large scale disaster. These emergency require-
mation to allow applications to make decisive progress. ments can be generally categorized as Rescuers, Medical,
Bundles are generally routed in a store-carry-and-forward man- Food, Water, Clothes. As it is a very challenging task to com-
ner between participating nodes over the network. Two nodes municate and analyze the requirement of particular needs and
are called neighbors if they are within their connection range. A the volume, hence, there is necessity to device an efficient
Peer-to-Peer Netw. Appl.

routing algorithm to deliver packets to different members of a 2 Related work


relief/rescue team in a post-disaster situation. Without knowing
the actual needs of various sites, it is not possible to coordinate A number of works have been reported in the field of routing
the relief operation efficiently. There should be an efficient in DTN. Most of the existing DTN routing protocols focus on
mechanism of information exchange that takes care of the relief unicast mode of transmission, such as Epidemic [8], Prophet
operation. This information exchange is performed using smart [9], Max-prop [10], Spray and Wait [11], RAPID [12], MBRP
phones carried by the relief workers. Every relief worker or any [13] that transmit a bundle to a single destination. As men-
affected person in the disaster area is considered as a node in the tioned earlier, these protocols are not suitable for emergency
proposed protocol. Smart phones are used to create bundles that response scenarios including a post-disaster scenario. When
incorporate messages created by the relief workers or affected multiple destination nodes are present, a multicast routing
people. These smart phones are capable of receiving and send- protocol is preferable over others. Very limited work has been
ing bundles. Multicasting is a preferred communication mode performed in the area of multicasting and group membership
in these cases as the destination is not a single node in most of mechanisms that define constraints applied on group member-
the cases. A single bundle is sent to multiple nodes which are ships in multicasting in DTN. In this section, different seman-
assigned as the destination group or DTN endpoints. A bundle tic models that explicitly specify constraints on group mem-
is considered to have been successfully delivered to a destina- bership are discussed first. Then it has been shown that the
tion group when at least a predefined subset of the nodes of proposed multicast routing technique supports one of these
destination group has received the bundle successfully. These semantic models. Recent multicast routing strategies that in-
destination groups are statically fixed before the field clude Unicast Based Multicast, Static Tree Based Multicast,
deployment. Dynamic Tree Based Multicast, OS-Multicast, Non-Custodial
The number of groups constructed in a disaster area depends Multicast over Prophet, Scalable Multicasting, Mobility based
on their responsibilities. We have considered various groups like routing algorithm in DTNs, Directional routing and schedul-
a group of Police, Doctors, Nurses, Ambulances, etc. There might ing for green vehicular DTNs, Quota Based Multicast Routing
besub-groupsunderagroup.Asforexample,underDoctorgroup in DTNs, Social Profile based Multicast Routing for DTNs,
there may be many sub-groups depending upon their specializa- and A Novel Mobility Based Routing Protocol for Semi
tions or expertise. Information about destination groups includes Predictable DTNs are discussed and critically reviewed after-
the number of members in the group, member identification, wards. We also discuss two of the most recent forwarding
name of each member, etc. This information helps the sender strategies in DTN that include Bubble Rap (BR) and
node to decide what bundles need to be sent to which group. Delegation Forwarding (DF). A comparative study of existing
In the present work a collaborative time-stamp based recent DTN multicast routing methods and DTN forwarding
multicast routing protocol is designed for a DTN. Each node, strategies has been presented in Table 1. A brief discussion of
before sending data to a particular destination group, scans its these is presented next.
neighborhood to find out the neighbors with similar type of
interest. Nodes are said to have similar types of interest if they 2.1 Multicast semantics model
have similar messages destined to the same group. Among all
such nodes, the most promising node takes the responsibility to In the multicast semantics model [6], authors have described
create the bundle and forward it towards the destination. All three semantic models to implement group formation in DTN
necessary information regarding the destination group members multicasting. Different semantic models [6] include- a)
like the number of resource instances (e.g. doctors, ambulance, Temporal Membership model (TM), b) Temporal Delivery
nurses, and food) required for the sites, etc. are padded with the model (TD) and c) Current Member Delivery model (CMD).
message. As the bundle is received at the destination group, it is In TM model the intended receivers of the bundle are
decoded to get necessary content like the need of a site from defined at the time of bundle creation. The actual receivers
where the bundle was sent. This eases need assessment and can be a subset of the set of intended receivers. Similarly,
resource management for entities in the destination groups. they define the group id of the receiver. Here, a member-
The remaining paper is organized as follows. Section 2 ship interval specifies that the group members are defined
shows the related work in the field of routing protocols in within that time period. In TD model the bundle specifies
DTN. Section 3 presents the proposed DTN multicast routing an extra constraint of delivery interval which indicates the
protocol followed by details about the bundle creation tech- time period within which the bundle should be delivered to
nique, forwarding techniques, and bundle priority. Section 4 the intended receivers. In CMD Model, they have specified
describes the experimental setup required for performing our a CMD flag which shows whether the receiver is the mem-
simulations in the ONE simulator. Simulation results are pre- ber of the group at the time of receiving the bundle. When
sented and discussed in Section 5. Section 6 ends the paper the CMD flag is set the intended receivers of the bundle
with a conclusion. should be the group members at the time of delivery. If the
Peer-to-Peer Netw. Appl.

Table 1 DTN multicast routing and forwarding strategies

DTN multicast routing strategies


1. Unicast based multicast [4]
Protocol Strategy Remarks
It is based on the concept of doing a one-to-many data communication by This technique is not adequate for DTN where efficient data transfer is very
multiple one-to-one source-to-destination unicast data transfers. If a challenging due to frequent network partitioning, limited power, limited
node creates a multicast bundle for n destinations then that bundle is bandwidth, limited storage. It does not provide promising solution in
replicated n times, as shown in Fig. 1. The sender queries the underlying emergency situations like post disaster scenario, military battlefield, etc.
network to collect all the discovered historical end-to-end paths to that where efficient data transfer is of utmost need.
destination. If the source finds any established link, then a copy of the
multicast bundle is forwarded to the next hop.
2. Static tree based routing (STBR) [4]
Here, nodes construct a tree starting from source to intended receivers. The If the tree is not available, sender just keeps the bundle in buffer and waits
nodes between sender and the receiver are selected as intermediate hops for next availability of the tree to send the bundle towards the
for the bundle using Dijkstra’s algorithm. As the route is static, if a node destination. This protocol is suitable for scenarios where network
misses the contact then it has to wait for the next contact. Use of partitions occur in a scheduled manner (like satellite communication).
multicast tree reduces redundant traffic. When communication starts, the But in an emergency situation where network partitions occur frequently,
source queries its underlying network to find out historical end to end this protocol is not promising. There should not be full dependence on a
paths to all group members, as presented in Fig. 2. It then builds up a multicast tree. STBR also suffers from large buffer overheads.
shortest path tree based on this information and keeps sending bundles
along the tree.
3. Dynamic tree based routing (DTBR) [4]
Here, routes are dynamic and use explicit addressing approach. Nodes This protocol suffers from lack of coordination between multicast group
determine next hop through current or newly derived contact members as it is not the responsibility of any multicast group member to
information. Each bundle has a unique multicast tree that may change transfer bundles to other members. In a DTN scenario due to non- esti-
hop-by-hop to adopt changes in link up and down. Each node has the mated delay in data transfer the group members may frequently join or
capability to extend its own tree, as has been shown in Fig. 3. It com- leave the group. Some mechanism is needed where the destination itself
putes the smallest cost tree based on its current local view of DTN layer. can send the bundle to other destinations, if opportunity arises.
It forwards the bundle using its own discovered multicast tree. If there is
a newly available path to a destination which is not discovered by the
upstream nodes, this node may immediately take advantage of that in-
formation for sending bundles.
4. On demand situation aware multicast protocol (OS-multicast) [22]
It is based on the demand of path discovery and situation awareness of It does not eliminate redundant bundle created from same proximity for
connection availability. Situation request and reply system bundles gain same destination. A strategy is required that shares bundle’s contents of
availability of the network. DTN nodes create a situation request bundle neighboring nodes. Also, this protocol does not rely on global
into entire neighborhood to know the paths to destinations. The knowledge of the network. A post disaster scenario requires some past
neighbors create a situation reply bundle based on historic encounters to knowledge about neighboring nodes initially. An efficient method must
the receivers of the multicast bundles. The node combines all paths to take care of this from initial level by calculating delivery probability of
form a static mesh. When bundle is forwarded, the hop generates a new each neighbor who wants to create similar bundles for same destination
dynamic mesh over which the bundle is transferred to next hop. group.
5. Non-custodial multicast over the DTN-prophet protocol [23]
Here, PRoPHET model is used for probabilistic calculations along with This protocol may be reasonable for DTN scenario but it is not efficient in
little mobility information. Nodes share delivery predictions along with post disaster scenario because the duration for handling critical situations
the node’s moving directions. A pseudo multicast tree is implemented lasts for hours or days. Authors have tested for a maximum of 4600 s
that comprises of one source, a set of destinations and a set of which is not confident enough for implementing in a real scenario.
intermediate nodes. Using these information, nodes transfer bundles via
multiple hop to the destination.
6. Scalable multicasting [24]
Here, a DTN multicast protocol called RelayCast is presented whose A relay node plays a vital role in transferring packets, yet no mechanism
operation is based on 2-hop relay DTN routing which is performed in has been proposed to select a relay node. The multicast source simply
two phases. In phase 1, the multicast source sends a packet to a relay sends the packet to a relay node if it is in the vicinity of that source.
node (could be multicast receiver). In phase 2, if there remains any
multicast receiver who has not yet received the packet, the relay node
delivers the packet.
7. Mobility based routing algorithm in DTNs [25]
It is based on the information about the mobility of nodes and their contacts.
Authors have considered the node mobility as schematic where mobility of
Authors in this work assume that greater mobility of nodes results in nodes can be predicted by getting information from daily activity of
better contacts. This is basically a unicast routing protocol that achieves
people who carry smart phones. But in a post disaster scenario the
high delivery probability with low message overhead compared to other mobility pattern is not so easily predictable. Also, lack of proper support
unicast routing scheme. for multicasting renders it inefficient.
8. Directional routing and scheduling for green vehicular DTNs (DRSS) [26]
Peer-to-Peer Netw. Appl.

Table 1 (continued)
It presents a direction based routing scheme for green vehicles that can Authors have designed this technique for varying mobility patterns (traffic
optimize energy efficiency with the considerations of congestion, buffer patterns). The proposed technique can be applicable in a post disaster
and delay. The proposed technique learns from traffic patterns and selects situation where vehicle routes are intact after any disaster. It fails to perform
optimal route for packet delivery towards the destination. otherwise.
9. Quota based multicast routing in DTNs (QBMR) [27]
Here, a multicast routing scheme has been designed by giving extra This approach is highly prone to unnecessary bandwidth utilization.
attention on group membership maintenance. Authors in this work adapt a However, in a DTN efficient utilization of network parameters (bandwidth,
quota replication scheme to increase delivery rate and decrease message power, etc.) is of utmost importance. Unbounded replication of messages
relay cost. It uses a quota based multicasting where a quota >1 means that can’t be afforded in a resource scarce DTN environment.
the message can be duplicated and forwarded to another node (may or may
not belonging to the multicast group).
10. Social profile based multicast routing for DTNs (SPM) [28]
This paper proposes a multicast scheme based on exploration of some The dataset is not equally suitable for a post disaster situation as in a
social features of DTN. The social features have been obtained from disaster scenario, node mobility is much more unstable and unpredictable.
infocom6 data trace. Authors have adapted efficient affiliation and So, its applicability in a post-disaster scenario is uncertain.
language based group formation technique for message delivery purpose in
a multicast scenario.
11. A novel mobility based routing protocol for semi predictable DTNs (MBPR) [29]
This protocol proposes mobility based routing using space-time graph from Such a routing strategy is not suitable for emergency response scenario as
known mobility patterns of nodes. When the space-time graph is incom- there might be multiple group members in a relief/ rescue team and they are
plete, this technique relies on previous encounters to estimate future con- scattered at different places. No test for support of multicasting has been
tacts. It achieves better performance in terms of delivery probability, delay performed.
and hop count compared to other unicast routing schemes.
DTN forwarding techniques
1. Bubble rap [3]
It is a social based forwarding strategy that exploits two social and It used datasets gathered by Haggle Project over 2 years from Infocom05,
structural metrics namely community and centrality. It is designed for Hong Kong, etc. These datasets are not applicable in post disaster situations
pocket switched networks, a type of DTN where repeated disconnection where node mobility is unpredictable. We require techniques that suit
and re-wiring of network can be possible. This technique evaluates the specific requirements of a post-disaster situation.
impact of above mentioned metrics on forwarding and proposes a hybrid
algorithm that selects high centrality (within a community some node are
more popular and interact with more nodes than others) nodes and com-
munity members of destination as relays. In this work authors have used
utility of two community detection algorithms together for getting rich
information about human social clustering for useful design of a forwarding
strategy.
2. Delegation forwarding [14]
It is another novel forwarding approach where authors have shown that the Authors here assume buffer space of each node to be unlimited and that the
network cost is proportional to the square root of population size. Authors buffer will hold all replicated messages till delivery. In real DTN
assume that a set of mobile nodes Ni € M where |M| = N. Each generated environment there are limitations on storage, bandwidth, etc. Strategies are
message has a particular source S € M and a destination D € M. Generated required that meet real scenarios more precisely.
messages are transmitted from one node to another based on quality metric
associated with each node and message. Main goal of this work is to reduce
the number of replicas generated for each message by choosing the best
node or highest quality nodes for eventual delivery to the destination.

CMD flag is not set the bundle would be delivered during (Collaborative Time-Stamp based Multicast Routing,
the delivery interval like the Temporal Delivery Model. CTMR), TM model for group formation has been taken
When the flag is set the intended receivers must belong as the group members in TM model are defined during
to that group at the time of delivery. These three models bundle creation. The grouping mechanism has been
provide efficient grouping mechanism in DTN for discussed later in Section 3.1.
multicast routing. In our proposed routing protocol

K R1
S S M

S R2 R3
R1, R2, R3 R1 R1 R1
J
Fig. 1 Unicast based multicast replicating a multicast bundle to multiple
unicast bundles Fig. 2 Tree formation in STBR
Peer-to-Peer Netw. Appl.

M creating a bundle CTMR forwards it to the right node based on


S K the proposed forwarding strategy used to prepare bundle pri-
M ority queue. Important features in CTMR are discussed in the
following sections.

R1
R3 3.1 Grouping mechanism in CTMR
K
R2 R2
As mentioned in Section 2.1, several semantic models have
Fig. 3 Route discovery as in DTBR been proposed which explicitly put constraints on the group-
ing mechanism. The models unambiguously define the re-
Table 1 presents a brief discussion and critical review of the ceivers of a multicast packet. In the proposed scenario, we
existing DTN multicast routing strategies and some DTN have considered the members of the relief organization
forwarding strategies. The discussions about DTN multicast (Doctors, Nurse, Ambulance etc.) and the team members of
routing strategies in Table 1 show that none of the existing rescue teams (Police, Army etc.) as the receivers of the
routing strategies have been built by taking into account the multicast bundles. The source and intermediate nodes of the
specific constraints posed by a DTN network that include multicast bundle can be normal people of affected sites or any
sparse resources, limited buffer size, very rare communication member of the above mentioned teams. It has been considered
opportunities, precious bandwidth, etc. We require a DTN that relief organizations and rescue teams are divided into
routing strategy that overcomes these constraints, at the same groups. For example, in any relief or rescue team there are
time perform efficiently for important parameters. several doctors, nurses, volunteers or police. We can think of
In the next section we discuss our proposed DTN multicast these as separate groups as they are distinguishable based on
routing protocol CTMR in details. their role within any organization. In CTMR, these groups are
named using specific identification. For example, D, N, RW
are the group names of Doctors, Nurses, Relief Workers re-
3 Collaborative time-stamp based multicast routing spectively. Again, a composite group involving users of dif-
protocol (CTMR) ferent types are named using combined group names of all
such types. For example, a group involving some doctors as
Collaborative Time-Stamp based Multicast Routing (CTMR) well as some nurses is called as DN group. As there might be
is the proposed DTN multicast routing protocol designed es- several such group combinations the number of groups to
pecially for a post-disaster scenario. In a large scale disaster form and their members has been customized before deploy-
there can be multiple affected sites. Number of governmental ment in the proposed work. Each and every node is aware of
or non-governmental relief /rescue organizations work in this group information. Group members decide the final des-
these affected areas. As timely information about those affect- tinations of multicast bundles. In our work, TM semantic
ed areas are necessary for efficient serving of affected people, model has been used to draw constraints on the groups due
there is a requirement of efficient coordination between the to frequent joining or leaving of the group members. At the
team members of rescue and relief organizations. In proposed time of bundle creation, along with other parameters the
CTMR technique, nodes belonging to any affected site create group-id and the membership interval [t1, t2] has been pro-
bundles specifying the requirement of that particular area. It vided to the bundles, so that they become aware of the nodes
may so happen that in an area a number of nodes create same belonging to the destination group. A membership interval is
type of bundle due to similar requirements. This creates re- the interval included in the bundle at the time of bundle crea-
dundancy in the network. The CTMR technique takes care of tion. For a bundle with group id G and membership interval
this by minimizing bundle redundancy during bundle crea- [t1, t2], the intended receivers of the bundle are the nodes that
tion. Nodes have the facility to create a multicast bundle in are members of group G at time during the period [t1, t2]. The
collaboration with their neighboring nodes which helps to intended receivers of a group are defined at the time of bundle
improve the performance of CTMR technique along with re- creation by including the membership interval to the bundle.
ducing the hop count. The most promising node among its Due to frequent join or leave, there is a subset of intended
neighborhood is given the responsibility to create and deliver receivers, called the actual receivers, which actually receive
the bundle. CTMR also employs efficient grouping mecha- the bundle. A node may belong to multiple groups but would
nism to bind nodes into destination groups judicially that helps only receive bundles if it belongs to the destination group
the nodes to know correct destinations for their bundles. assigned to that bundle. If a node is acting as an intermediate
CTMR defines proper bundle format that has been customized hop, then it temporarily receives all bundles to transmit to the
to suit requirements in the post-disaster DTN scenario. After next forwarding node.
Peer-to-Peer Netw. Appl.

3.2 Bundle format Therefore, a patience limit has been introduced. It defines a
time limit that a node must wait before creating a new bundle.
CTMR follows a customized bundle format, as shown in If any node wants to create a bundle then it defers its decision
Table 2, to create effective bundles prior to sending them. It until patience limit is elapsed and waits for reply from its
defines the header information regarding bundle. neighboring nodes. Suppose at a time instance t, node A wants
Fields used in the bundle format are explained in the fol- to create a bundle and B, C are its neighbors, as shown in
lowing part. Fig. 4.
Node A broadcasts this information as a temporary mes-
Bdl_ID: The unique id of the bundle to uniquely identify sage constituting its bundle type named as BR_Bdl_Type to
it among all the bundles. nodes B and C and waits for reply till patience limit (Fig. 5).
TTL (Time to Live): The time limit for which the bundle The purpose of this temporary message is to inquire its
would survive after creation. neighborhood if any other node wants to create similar type
Bdl Criticality: It is the criticality factor provided by the of bundle or not. After receiving the broadcast message, the
node itself that ranges from 1 to 5. Nodes assign a criti- neighboring nodes are prohibited from sending any broadcast
cality factor to their created bundles, based on their message till that patience limit expires. They can send or re-
choice. A criticality factor of 1 means least critical while ceive the bundles but they cannot broadcast any other tempo-
5 means highly critical. rary message. B and C, after getting the broadcast message,
Sen ID: The id of the creator who creates the bundle. decide whether they want to create that type of bundle or not.
Grp Id: The destination group ID of the intended If one or more nodes want to create same type of bundle, they
receivers. incorporate this information along with maximum delivery
Bdl Type: It specifies the type of bundle based on the type probability as the reply of the broadcast message and send it
of node this bundle belongs to. It helps the nodes to de- back to A.
termine which bundle is destined to which destination The delivery for a single multicast bundle is calculated on
nodes. the basis of the number of receivers who received the bundle
BdlLoc: It is the location of the node that created the divided by the number of nodes belonging to that group of
bundle at the time of creating it. At the time of bundle receivers (the destination group size). The delivery probability
creation GPS co-ordinates are included into the bundles of bundle i, termed BDPi, is calculated as in formula (1).
as bundle locations. These co-ordinates comprise of lati-
Total Group Members who received the Bundle
tude and longitude of the location of the sender node at BDPi ¼ ð1Þ
the time of creating the bundle. Total Number of Group Members
Bdl Created Time: It is the time at which the bundle has Using (1), the delivery probability of CTMR routing pro-
been created. tocol (of all the bundles), termed BDP, is calculated as in
formula (2).

Xn
3.3 Creating a bundle BDP ¼ 1 BDPi ð2Þ
n j ¼1
Bundle creation is an important phase of the proposed CTMR
protocol, where collaboration is used extensively to reduce
redundancy for creating multicast bundles. Since the protocol Here n represents the total number of bundles created and i
is defined over DTN, special emphasis is given for reduction denotes each bundle.
in redundancy and hop count. As affected people of any par- Suppose C wishes to create a similar bundle of same type
ticular region or relief/rescue workers of any affected region as that of A. So it replies affirmatively along with its maxi-
may report similar messages regarding the requirement of that mum delivery probability and its bundle contents to A.
region, so it is obvious that more than one user lying in very Figure 6 shows that C replies back to A along with its maxi-
close proximity may create bundles of same type. This is mum delivery probability with an affirmative answer.
because nodes in a small area continuously monitor their Node A incorporates all information from the neighboring
neighborhood. Same type of bundles from neighboring nodes nodes to its message content and decides, based on the max-
thus increases the redundancy and overhead thereafter. imum delivery probability, which node must create the bundle

Table 2 Bundle format for bundles in CTMR

Bdl_ ID TTL Bdl Criticality Sen ID Grp ID Bdl Type Bdl Loc Bdl Created Time
Peer-to-Peer Netw. Appl.

B C B
C
Yes + 0.73 +
Bundle Content
A
A 0.64
Fig. 4 Nodes B, C are in range of A
Fig. 6 Node C replies with a BYes^ along with its Max DP

for further forwarding. If Node A itself has the highest deliv- for similar bundles created from nearby locations. Nodes col-
ery probability, then A simply creates the bundle. If any other laborate among themselves to create a multicast bundle which
node within A’s neighborhood has maximum delivery proba- is delivered in a more efficient way in the disaster affected
bility, then that node is instructed by node A to create the area. Figure 9 shows the step by step procedure to create a
bundle. Node A sends the entire appended message to that bundle collaboratively at each node as used in CTMR.
node. The patience limit has been used here to reduce dupli- As presented in Fig. 9, the nodes first select the type of
cate bundles and hence redundancy. It must be noted here that bundle such as Doctor or Police type. Nodes then write the
the patience limit not only reduces bundle replication but it message content and initiate send message. Rest of the work is
also helps to reduce the number of bundles for same type of carried out with the CTMR routing protocol. Initializing a
destination groups. This contributes in saving of storage, min- send operation does not mean that the node has created the
imizing congestion and bandwidth utilization which are very bundle. It means that the node has just participated in bundle
important issues in DTN. Suppose Node A finds that C has creation. The bundle is created by the most probable node
better delivery probability for the destination, hence A in- among the neighborhood of the initiator node. This strategy
structs C to create the bundle along with final appended bun- is responsible for collaborating among neighboring nodes to
dle contents of all other nodes as shown in Fig. 7. decide which node creates the bundle. Rest of the processes,
To verify the efficiency obtained by incorporating such as shown in the form of an algorithm in Fig. 9, is self-
collaboration, we use an example to explain how this strategy explanatory.
efficiently decreases hop count and data redundancy. Suppose
two nodes C and A are lying at a close proximity to each other.
We imagine that there are few injured persons and they need 3.4 Metadata of the bundles
immediate medical attention. Both nodes C and A create a
doctor type bundle. The Doctor type bundle is destined to Once a bundle is created the sender node stores it in its buffer.
nodes R1, R2, and R3. It can be observed from Fig. 8a that Nodes maintain a metadata of every bundle at the time of its
node C creates the bundle and passes it to M, node M passes it creation or after receiving other bundles. The fields defined in
to J, node J delivers it to R1and R3. Finally, R3 delivers it to the metadata are used to identify the bundles that need to be
R2. forwarded in the next available opportunity.
Similarly, as can be seen from Fig. 8b, node A also creates The metadata of the bundles contain the following fields:
the bundle and passes it to Z, node Z passes it to L, node L
delivers it to R1, and finally R1 delivers it to R3 and R2. In the 1. Bdl Location
above case, total hop count is 8 for two bundles. Following the 2. Bdl Type
strategy as used in CTMR let us assume that node C has a 3. Bdl ID
better delivery probability than node A for the selected desti- 4. Group ID
nations. Node C creates the bundle utilizing the initial collab-
oration. The created bundle follows the same path from C to Details about these fields are similar to that explained in
all the destination nodes of the bundle as in Fig. 8a. It results in Section 3.2. This metadata is used by the nodes to select bun-
a hop count of 4 for the bundle. This shows how CTMR helps dles in appropriate order, to be forwarded to next hop on
to minimize message redundancy and hop count substantially available opportunity.

C B C B

Create Bundle +
BR_Bdl_Type BR_Bdl_Type
A Bundle Content
A
Fig. 5 Node A is broadcasting temporary message Fig. 7 Node A instructs C to create bundle
Peer-to-Peer Netw. Appl.

Step1: Start
M J R3 Step2: [Creating Multicast Bundle]
M 2.1: Select BdlType
C J R1 R2 2.2: Assign BR_Bdl_Type = BdlType
2.3: Assign Patience Limit
(a) [Check If any neighboring node is present in the node’s range]
2.4: If any Node is Present in its range or patience limit does not
L R3 expire
A R1 2.5: Broadcast temporary message BR_Bdl_Type to all Nodes in its
Z Z L R2
range.
[Wait for reply against BR_Bdl_Type]
(b) 2.6: If (any Broadcast Bundle arrives) and patience limit does not
expire.
Fig. 8 Analysis of nodes A and C of creating the bundle separately
2.6.1: Check BdlType from BR_Bdl_Type
2.6.2: If (BR_Bdl_Type = BdlType)
2.6.2.1: Send My Max Delivery Probability according to the
3.5 Time-stamp based next hop communication Destination List with "Yes” as the reply.
[Wait for the reply]
2.7: If (reply = Create Bundle)
Selection of appropriate forwarding node is performed 2.7.1: Create Bundle (Bdl ID) incorporating all bundle contents
through a proposed Time-stamp based criteria in CTMR. 2.8: If any reply arrives and patience limit has not expired.
2.8.1: If Reply is Yes with message content, max delivery
This check is performed whenever any node comes in the
predictions and patience limit has not expired.
range of another node. CTMR employs a strategy that is 2.8.2: Store all received message content.
used to determine that if a pair of nodes has communicat- 2.8.3: Select the host whose Delivery probability is highest.
2.8.4: If the Selected host is other than this host
ed recently, then for a certain period of time those two Then
nodes would be refrained from communication. This pe- 2.8.5: Reply to that host with "Create Bundle” instruction with all
the bundle contents.
riod of time is the proposed time-stamp. The idea behind [Abort Multicast Bundle creation]
this is that it is very unlikely that these two nodes will 2.9: Go to Step 3
2.10: Else if Selected Host is this host
have fresh bundles to exchange within that time period, 2.10.1: Select Bdl ID
since they have already communicated recently. In a di- 2.10.2: Select TTL
2.10.3: Assign Group Id
saster affected area, relief workers work at a particular site 2.10.4: Assign Sender = This Host
2.10.5: Add (This node) to List of Hops Travelled
for a long period of time. Nodes belonging to a particular 2.10.6: BdlType = MyBdlType
site are within communication ranges for long periods. In 2.10.7: BdlLoc = My Current GPS co-ordinates.
2.10.8: CreateBundle (Bdl ID) incorporating all bundle
the absence of a time-stamp, nodes keep trying to com- content;
2.10.9: Calculate the Size of the Bundle
municate with each other. Considering energy as a price- 2.10.10: If (Size <= Current Buffer Space)
less constraint, it is important to focus on saving the en- 2.10.11: Decrease MyBufferSpace by the size of the bundle
2.10.12 Else if (Size > Current Buffer Space)
ergy. The proposed time-stamp based strategy effectively Drop bundles in FIFO to make space for newly
created bundles.
reduces energy consumption of the nodes by putting a Step 3: Exit
limit on communication attempts. The time interval for Fig. 9 Collaborative strategy to create a bundle at a node
the time-stamp is fixed before deployment. A node always
maintains a list where the id of the node and its last con-
tact time is stored. Whenever a node comes in contact 3.6 Bundle forwarding
with another node, this list is checked to find out whether
that node had contacted within the time-stamp period. If Once appropriate next hop nodes are selected, the bundles
the condition is true, then the connection is no more en- need to be forwarded to them. The metadata is first exchanged
couraged. For example, if node A has made a contact with whenever any node comes in contact with another node that is
node B at time t. If time stamp period is 10 min, within ready to accept the bundles. The sender, after receiving the
10 min, node B is supposed not to have much fresh bun- metadata from the receiver, decides upon bundles to be sent.
dles to send to node A and vice-versa. Therefore, they are The receiver, ready to accept the bundles, would reply affir-
not allowed to communicate within the next 10 min. The matively along with all of the delivery predictions of the nodes
threshold for the time-stamp period is defined before field it contacted. This prediction technique is based on a well-
deployment so that all nodes are aware of this value. The known DTN routing protocol namely Probabilistic Routing
timestamp threshold has been set to 10 min in our simu- Protocol using History of Encounters and Transitivity
lations as it has been observed that for the threshold value (PRoPHET) [9] protocol. This delivery probability calculation
of 10 min, CTMR achieves optimal value for average employs an algorithm that attempts to take advantage of actual
bundle latency, increased average bundle delivery proba- encounters by maintaining a set of probabilities for successful
bility and reduced average bundle overhead as explained deliveries to known destinations. Whenever two nodes meet,
later in Section 5. The message generation interval has they exchange additional information called delivery predict-
been kept between 5 and 10 s. The corresponding results ability after expiry of the time-stamp period. The sender, after
have been shown in Section 5. receiving the delivery predictions for all encountered nodes by
Peer-to-Peer Netw. Appl.

Desnaon/ Table 3 Bundle priority queue of node S communicating with node R


Sender Next Hop Bdl ID μ Age δ Max DP Priority values
Sending Beacon
M1 5 423 0 0.31 1.81
Sending Metadata M2 4 45 −378 0.42 1.6515
M3 2 36 −387 0.65 5
ACK
M1: A bundle with its destination as the Group of Doctors, D1
Delivery Predicons
M2: A bundle with its destination as the Group of Supplier, S1
ACK M3: A bundle with its destination as the Group of Police, P1

Time Time
Aging: CTMR considers most recent bundles to be more
Sender preparing
Final Priority Queue
important as new bundles are more likely to bring recent
and valued information that need to be transmitted.
Sending Bundles according
CTMR gives higher preference to the bundles that are
to Final Priority Queue
recently created. The time-to-live (in minutes) for a bun-
dle in a DTN varies from hours to days. When a bundle
Fig. 10 Time activity diagram of communication between sender and reaches the destination then also it remains alive inside
receiver nodes
the buffer of many intermediate nodes till its TTL expires.
Longer the age of a bundle, more it is replicated thus
the receiver, creates a bundle priority queue. Bundles are occupying more buffer space at nodes. This results in
forwarded to next hop according to the bundle priority queue. fewer chances for delivering newly created bundles that
The task of intermediate hops is to forward bundles opportu- contain important and relevant information. Aging pa-
nistically to other hops or to the destination. The selection of rameter is used as a solution, in the preparation of
the next forwarding node, and also the order of the bundles to Priority Queue of bundles, as explained in section 3.7.1.
be transferred, depends on certain forwarding parameters like Criticality of Bundle: Criticality factor has been intro-
location, aging, criticality etc. These are discussed in details duced in CTMR through which a node determines the
next. importance of its bundles. If a node is carrying a bundle
with higher criticality, then it is treated with higher im-
Location: Whenever two nodes want to communicate, portance. The criticality of bundles varies from 1 to 5. For
they exchange their bundle metadata with each other. example, in a post-disaster scenario emergency bundles
CTMR protocol checks the location difference of two or are given the highest priority i.e., 5 and any update or
more similar type of bundles from these two nodes. status bundle is given the lowest priority i.e., 1. All inter-
Nodes are encouraged to send only those bundles that mediate criticality values fall within this range. The crit-
have been created from different locations rather than icality is used in the preparation of the priority queue as
the bundles from same or nearby close locations. The explained next in section 3.7.1.
Sender calculates the location difference of similar type
of bundles from within itself and the receiver node. It is
checked whether the location difference between two 3.6.1 Construction of priority queue
bundles from the two communicating node is greater than
the threshold limit defined for location difference or not. CTMR employs a bundle priority queue for arranging the
The threshold is a portion of the max size of the disaster bundles in order by using an additional queue. It is created
area region. If the location difference is less than the by the sender for all targeted receivers separately. In the pri-
threshold limit, then it is enquired whether the destina- ority queue those bundles are placed first whose destination is
tions of the two bundles are same or not. If yes, then that the current receiver. The remaining bundles are ordered
bundle is aborted. The location difference between two through formula (3).
GPS co-ordinates is calculated using the Haversine for-  . 
mula. This formula has been used to calculate the dis- PBi ¼ μ*0:3‐ δ 60 *0:005 þ DPðMax ðGroupId ÞÞ ð3Þ
tance between two GPS co-ordinate locations as in [30].
Bundles are transferred from one node to another only if Here, PBi represents the priority value of the ith bundle.
their location difference is above the threshold location In (3), μ is the criticality of the bundle which was assigned
difference that has been calculated empirically, as pre- by the creator node at the time of its creation. δ is the
sented in Section 5. aging factor, calculated as the difference between current
Peer-to-Peer Netw. Appl.

Step1: Send Beacon Messages Table 4 The ONE simulator parameters


1.1 Wait for Reply
1.2 If the Reply is affirmative Simulation time 12 Hours
Step 2: [Exchange Metadata of Bundles] Transmit range 60 Meters to 90 Meters
2.1 Send this node’s Metadata Buffer size 200 MB
2.2 After getting the Meta Data Of the other Node
Wait time 10 Seconds to 240 Seconds
Step 3: Select bundles first whose one of the destination is that
Speed 0.5 to 1.5 m/sec
neighboring node, with which this communication has been
established World size 4500 m × 3400 m
3.1 Assign priority value to 5 to those bundles. Bundle interval 5 sec to 10 Sec
3.2 [Check the bundles of the same type from both nodes] Bundle size 5 KB to 2 MB
If (Same_Type) Then 1st Doctor group (D1) 25,7,29,11,13,68,95,99
3.3 If (location Difference is more than or equal to predefined 1st Police group (P1) 4,28,13,22,1,86,55,47
threshold limit)
1st Relief worker group (RW1) 8,6,17,20,25,7,107,88
Then
1st Army group (A1) 2,12,14,27,3,112,42,63
3.4 [Add those bundles to the next index of the priority list
according to the final priority formula.] 1st Nurse group (N1) 9,1,10,24,19,97,113,110
PBi = μ * 0.3 + (δ / 60) * 0.005 + DP(Max(Group Id)) Management group (M1) 0,19,23,26,11,13,83,109
Else if location difference is less than the threshold Supplier group (S1) 29,13,17,19,2,118,53,56
3.5 Ignore those bundles 2nd Doctor group (D2) 28,35,20,4,9,100,93,115
Step 4: [Add bundles that do not belong to same BdlType according 2nd Police group (P2) 7,13,8,23,17,48,58,84
to the priority queue formula.]
2nd Relief worker group (RW2) 14,25,29,0,9,22,54,102
PBi = μ * 0.3 + (δ / 60) * 0.005 + DP(Max(Group Id))
Step 5: [After The Bundle List is generated]
2nd Army group (A2) 57,50,82,103,66,46,44,92
5.1 Send the Bundles one by one from the priority queue 2nd Nurse group (N2) 49,117,11,98,101,85,18,90
5.2 End Connection Fire fighter group (F1) 67,59,60,94,64,80,69,52
Step 6: Simultaneously while receiving bundles
6.1 Send affirmative acknowledgement (ready to receive
bundles)
Table 3 presents an example showing the calculation of the
Step 7: Reply my delivery probability along with Metadata to the
sending node priority queue of bundles prepared by a node S (sender) for
Step 8: Receive & Store bundles in buffer another node R (receiver) belonging to the Police group. Once
Step 9: Decrease the priority of the bundles which have this node as the priority values are calculated, the bundles are ordered ac-
the destination. cordingly to be transferred to the next hop.
Step 10: Exit According to the strategy to construct the priority queue, M3
Fig. 11 Bundle forwarding and priority queue construction at nodes is given the highest priority 5 as node R is its immediate desti-
nation. M2’s age is 45 s. The eldest bundle is M1 with an age of
423 s. Hence δ value for M2 is (45–423 = −378). Similarly M1
bundle age and the age of the eldest bundle at the sender. has a δ value of 0. For M1, node R gives the Max DP of 0.31
δ is divided by 60 to convert seconds into minutes and which means that node R has contacted members of Doctors
then multiplied by 0.005 to normalize the final priority Group and the maximum delivery probability among them for
value to the range of [0,5]. δ is subtracted because greater node R was 0.31. Also the criticality value μ, as assigned by the
the age of the bundle, the lesser is its priority value. It creator node to the bundles are 5, 4, and 2 respectively. Final
helps to increase the priority of those bundles which are priority values of these bundles are calculated using (3). Based
recently received. This is how the aging constraint is tak- on these priority values, the bundles are forwarded to the re-
en care of in CTMR. The queue is constructed whenever a ceiver node in the order {M3, M1, M2}. Figure 11 shows the
sender node comes in contact with a receiver. DP in (3) bundle forwarding strategy presented as an algorithm.
represents the delivery probability calculated by receiver
for the bundles’ destinations, using PRoPHET. If the
sender has a bundle with its destination as the current
receiver then that bundle is given the highest priority 4 Experimental setup
5 at the priority queue. This is because it is important to
send those bundles first whose destination is the current To evaluate the performance of our proposed strategy and to
receiver. Otherwise unnecessary delay creeps in. The validate our claims, we have performed simulations. Based on
steps involved in construction of final priority queue have the parameters described in Table 4, simulations have been
been shown in Fig. 10 as a time activity diagram of com- perf ormed in the ONE (Opp ortunistic Networks
munication between the sender and the receiver. Environment) simulator [31]. The ONE is a DTN specific
Peer-to-Peer Netw. Appl.

Fig. 12 Snapshot of ONE simulator executing CTMR routing protocol, after 2458 s

simulation environment capable of routing bundles between to that group. A node can belong to more than one group. It
various nodes that move by following certain mobility rules. may so happen that some nodes do not belong to any group
Each simulation has been executed for 12 h. The transmit (affected people, say). Such nodes can generate as well as can
range denotes the communication range of the wireless de- be a forwarder of a bundle. However, nodes that belong to a
vices. The range has been varied from 60 m to 90 m. Wait group can forward a bundle to another node only if that bundle
Time denotes waiting period when the node arrives at a par- does not belong to that group. In Each Simulation, 210 nodes
ticular location. The node may choose it randomly from 0 to are considered which are numbered from node 0 to node 210.
240 s. The bundle interval denotes the interval in which new Out of these, 0 to 118 nodes belong to some group or groups
bundles are created. At every 5 to 10 s, a new bundle is cre- and the rest do not belong to any group. A snapshot of the
ated. Bundle size varies from 5 KB to 2 MB. We have created ONE simulator executing the proposed CTMR routing proto-
the destination groups in our Simulation by grouping certain col has been obtained and reproduced here, as shown in
similar nodes together. The groups in this simulation have Fig. 12. Figure 12 presents the status of the simulation after
been selected as Doctor, Police, Relief Worker, Army, 2458 s of continuous execution of the CTMR routing based on
Nurse, Management and Supplier. Each node in ONE simu- simulation parameters as described in Table 4. The results
lator is identified using a unique node address. The numbers obtained, as presented in section 5, are the results of average
after each group denote the node numbers of nodes belonging runs of ten simulations each.

0.676

0.674 3150
Bundle Delivery Probability

0.672 3100
Average Bundle Delivery Delay

0.670 3050
(in seconds)

0.668 3000

0.666 2950

0.664 2900

0.662 2850

0.660 2800
60 70 80 90 60 70 80 90
Wireless Communication Range (in meters) Wireless Communication Range (in meters)

Fig. 13 Bundle delivery probability of CTMR with varying wireless Fig. 14 Average bundle delivery delay of CTMR with varying wireless
communication range for each node range for each node
Peer-to-Peer Netw. Appl.

120 different multicast strategies and forwarding strategies has


been shown in Figs. 23, 24, 25, 26 and 27.
100
Bundle Overhead Ratio

Multiple results have been obtained by varying wireless


80 communication ranges for the nodes as shown in Figs. 13,
14 and 15. In real scenario the outdoor communication range
60
of IEEE 802.11, i.e. Wi-Fi technology is 90 m (approx.).
40
Beyond that range nodes are not able to communicate. In
simulation we have varied the range between 60 and 90 m.
20 It is found that the delivery probability of CTMR is highest
when the wireless range for the nodes is set at 70 m. At 60 m
0
60 70 80 90 wireless range, bundles are delivered at a satisfactory rate. As
Wireless Communication Range (in meters)
the contact opportunity increases at the range of 70 m, bundle
Fig. 15 Bundle overhead ratio of CTMR with varying wireless range for
delivery probability also increases. It can be observed from
each node
Fig. 13 that as the communication range increased to 80 m the
delivery probability had decreased because of an increased
5 Results and discussion interference of other wireless signals and the probable chance
of buffer overflow. When the wireless range is increased to
The proposed multicast routing protocol CTMR has been 90 m the buffer overflow problem is increased which results in
measured on the basis of three performance metrics. These increased dropping of bundles thus reducing delivery proba-
are bundle delivery probability, average bundle delivery delay bility. At 90 m the delivery probability improves slightly but is
and average hop count. To obtain graphs for the same, average found best at 70 m. Connection between nodes (smart phones)
result of ten runs of simulation with a 95 % confidence inter- ceases after 90 m. Figure 14 shows the Average Delay of the
val has been considered. For our simulations, we have consid- bundles that have been delivered. Here it is observed that the
ered a real DTN scenario where DTN links may be up for a delay at 70 m range is the minimum among all other ranges. It
very short period of time [32]. We have also compared the has increased when the communication range of the nodes
forwarding strategy used in CTMR which is based on increased above 70 m. This is due to the interference of wire-
PRoPHET with Bubble Rap and Delegation Forwarding, less links of nearby nodes, and there is wastage of time re-
two popular forwarding techniques in terms of hop count, quired for switching from one node to other. Due to this it
bundle delivery ratio and average delay. Empirical test has results in higher delays. Figure 15 demonstrates bundle over-
been performed to obtain the optimal wireless communication head ratio as the wireless range is increased. It can be observed
range for the communicating nodes (Figs. 13, 14 and 15), that wireless range of 70 m for the nodes has more overhead
since it shows a large impact on overall performance. than 60 m but lesser than 80 and 90 m. Basically, bundle
Similar empirical tests have been performed to yield optimal overhead is gradually increasing because each simulation that
values for other forwarding parameter values such as the lo- runs on the particular wireless range suffers from number of
cation difference (Figs. 16, 17 and 18) and patience limit time dropped bundles. This is due to the overflow of buffer space
period (Figs. 19, 20, 21 and 22). The comparison with with the increase of wireless range of nodes as more nodes

Fig. 16 Bundle delivery 0.681


probability with varying location
differences and time-stamp 0.680
Bundle Delivery Probability

periods
0.679

0.678
Time Stamp
0.677 Period (mins)
0
0.676 10
20
0.675 30
40
0.674

0 1 2 3 4

Location Difference (in %)


Peer-to-Peer Netw. Appl.

Fig. 17 Average bundle delivery 2950

latency with varying location


differences and time-stamp 2940
periods

Bundle Latency
2930
Time Stamp
Period (mins)
2920
0
10
2910 20
30
40
2900

2890
0 1 2 3 4
Location Difference (in %)

(and thus more bundles) get involved in communication. of delivery probability. It has been found that bundles under a
Wireless range of 70 m has best delivery probability and la- time-stamp period of 10 min and location difference of 2 to
tency is least among all wireless ranges. Thus, it has been 3 % are more probable to reach the destination than other
observed that a wireless range of 70 m is optimum for each values for these parameters. This claim is further supported
node to deliver the bundles as delivery probability is notably from Fig. 17 where it is observed that the delivery latency of a
the most important parameter that needs to be considered in bundle with a time-stamp of 10 min and location difference of
post disaster scenarios. Hence the value of 70 m has been set 3 % is good but that with a location difference of 2 % is having
as a standard wireless communication range for testing the highest latency for the same time-stamp period. Though mes-
CTMR routing protocol’s performance in the considered sim- sage bundles with a location difference of 3 and having time
ulation environment. stamp of 30 min give the smallest latency, such bundles suffer
Empirical test has been performed using multiple simula- from low delivery ratio as in Fig. 16. Figure 18 shows the
tion to calculate optimum values for location difference and bundle overhead ratio, where it can be observed that location
time-stamp periods by varying these values, as presented in difference of 2 % and time stamp period of 10 min has lowest
Figs. 16, 17 and 18. This test is performed because these overhead among all. Location difference of 3 % is having high
parameters need to be set before deployment, as depending overhead for time-stamp period of 10 min. Message bundles
on these parameters the relief management team sets its pref- with time-stamp period of 10 min and location difference of
erences. Since the most important parameter in a post-disaster 2 % have the best performance as compared to other values for
scenario is the successful delivery of bundles, the delivery these parameter values. This suggests the use these values as
probability has been given highest importance. Figure 16 standard in our experiments and so the time stamp period has
shows the performance of proposed CTMR protocol in terms been fixed as 10 min and the location difference has been set

Fig. 18 Bundle overhead ratio


90.4
with varying location differences
and time-stamp periods
90.2

90.0
Bundle Overhead

Time Stamp
Period (mins)
89.8
0
89.6 10
20
30
89.4
40

89.2

89.0
0 1 2 3 4

Location Difference (in %)


Peer-to-Peer Netw. Appl.

0.696

0.694

Average Bundle Delivery Delay


2960
Bundle Delivery Probability

0.692

0.690

(in seconds)
0.688 2950

0.686

0.684 2940
0.682

0.680
2930
0.678

0.676
60 120 180 240 300 60 120 180 240 300
Patience Limit (in seconds) Patience Limit (in seconds)
Fig. 19 Bundle delivery ratio with different patience limit Fig. 21 Average bundle delay with different patience limits

to 2 %. After getting optimum location difference and time- Figure 23 shows the comparison between the proposed
stamp values, simulations for different values of patience limit multicast routing protocols CTMR with other existing DTN
have been performed to obtain optimum value as in Figs. 19, multicast routing protocols in terms of bundle delivery prob-
20, 21 and 22. ability. Here we have set the parameters values such that the
From Fig. 19 it can be observed that the bundle delivery nodes carry a wireless communication range of 70 m. System
probability gradually increases till the patience limit is in- values contain the location difference of 2 %, time-stamp pe-
creased to 240 s. Bundle delivery probability is highest at riod of 10 min and a patience limit of 240 s (these values have
the patience limit of 240 s. Beyond this, it has decreased. been empirically measured, as mentioned previously). The
Figure 20 suggests the use of 240 s as the desired patience graphs in Fig. 23 prove that the performance of CTMR is
limit. This is because at this limit the bundle overhead ratio is better than other existing multicast routing protocols. The
the least. From Fig. 21 it is further observed that the average main novelty in CTMR is that it does not allow creating re-
bundle delivery delay increases gradually with increasing pa- dundant bundles from neighboring nodes lying in close prox-
tience limit. This is attributed to the fact that the more we imity. Also, whenever two nodes meet for communication,
invest time for collaboration (required by the patience limit) CTMR ensures that multiple copies of similar types of bundles
the more a bundle is waited, thus increasing the average de- are filtered out and not further transmitted into the network for
livery delay. The discarded bundles’ contents are added to the same destination group. These improvements and advan-
those bundles which have been created after collaboration. tages are ensured by our judicious selection of important pa-
This in turn increases the size of bundles. As the patience limit rameter values. CTMR also encourages a node to transfer
increases the number of created bundles decreases as has been bundles that are recently created with higher priority.
shown in Fig. 22. From these observations it is suggested to Unicast based multicast has minimum bundle delivery delay
set a patience limit of 240 s as standard value for our further compared to other DTN multicast routing protocols including
experiments. CTMR as shown in Fig. 24. But this routing protocol in not
6200

91.0
6150
Bundle Overhead Ratio

Total Bundles Created

90.5 6100

6050
90.0

6000

89.5
5950

5900
60 120 180 240 300 0 60 120 180 240 300
Patience Limit (in seconds) Patience Limit (in seconds)
Fig. 20 Bundle overhead ratio using different patience limit Fig. 22 Number of created bundles with different patience limits
Peer-to-Peer Netw. Appl.

Fig. 23 CTMR verses other


multicast routing protocols for 0.7
bundle delivery probability

Bundle Delivery Probability


0.6

0.5

0.4

0.3

0.2

0.1

0.0
Non-Custodial Prophet CTMR OS Multicast STBR DTBR U-Multicasting

Different Multicast Routing Strategies including CTMR

appropriate for a post disaster scenario as there remains other it might be observed that we are considering fittest nodes for
network issues with it, as mentioned earlier. Except unicast message/bundle transfer. Besides, in all three cases the main
based multicasting, CTMR has lowest bundle delivery delay goal is to reduce the number of replication in the network.
among all other multicast routing protocols, as can be seen Though in above mentioned two forwarding techniques, the
from Fig. 24. This is because it gives higher priority to those authors claimed that the mobile nodes are not using any prior
bundles which are recently created, and two bundles from knowledge of the network but while considering the best or
nearby locations having same destinations are also ignored. fittest nodes they are fetching information from mobility
CTMR encourages relaying those bundles which are created traces offline or quality metric associated with each node. In
from locations having large enough location difference, to DTN scenario, to judge whether a node is a good candidate or
ensure no redundancy is present, as far as bundles are not, we must have to analyze some priory knowledge of that
concerned. node. In CTMR we have considered that by calculating the
Next, we have compared the forwarding strategy based on delivery probability (DP) of each node with respect to the
time-stamp constrained Prophet as used in our proposed destination group for each bundle.
scheme CTMR with the Bubble Rap and Delegation As can be seen from Fig. 25, we have compared CTMR
Forwarding schemes, as shown in Figs. 25, 26 and 27. The forwarding based on time-stamp constrained Prophet with
proposed algorithm CTMR is conceptually identical to the Bubble Rap and Delegation Forwarding in terms of average
above mentioned forwarding techniques as in all of the cases bundle delivery probability. CTMR forwarding performs

Fig. 24 CTMR verses other 4500


multicast routing protocols for
average bundle delivery delay (in 4000
Average Bundle Delivery Delay

seconds)
3500

3000
(in seconds)

2500

2000

1500

1000

500

0
Non-Custodial Prophet CTMR OS Multicast STBR DTBR U-Multicasting

Different Multicast Routing Strategies including CTMR


Peer-to-Peer Netw. Appl.

7.8
0.40
Average Bundle Delivery Probability

7.6
0.35
7.4
0.30

Average Hop Count


7.2
0.25
7.0
0.20
6.8

0.15 6.6

0.10 6.4

0.05 6.2

0.00 6.0
CTMR Forwarding Bubble Rap Delegation Forwarding
5.8
Fig. 25 CTMR forwarding based on time-stamp constrained Prophet vs. CTMR Forwarding Bubble Rap Delegation Forwarding
Bubble Rap and delegation forwarding in terms of average bundle deliv- Fig. 27 CTMR forwarding based on time-stamp constrained Prophet vs.
ery probability Bubble Rap and delegation forwarding in terms of average hop count

much better than both bubble rap and delegation forwarding from source to destination. This is attributed to the fact that
as at each step in bundle forwarding it is selecting the most Prophet calculations in CTMR force a carrier to transfer bun-
eligible candidate i.e. the fittest node for data transfer. The lack dles to immediate neighbors even if they appear to only mi-
of far-sightedness capability in bubble rap and delegation nutely improve delivery probability. This results in a large
forwarding reduce their capability to estimate good future number of intermediate nodes in the path from a source to a
values. This takes its toll in the long run, thus reducing their destination. From the above results and discussions, we con-
delivery probability to a large extent. Forwarding based on clude that CTMR is a good routing scheme for considered
Prophet provides good future estimates based on past perfor- DTN scenario.
mances. As far as average bundle delivery delay is concerned,
as presented in Fig. 26, CTMR performs almost at par with
bubble rap and much better than delegation forwarding. The
introduction of time-stamp based criteria in forwarding mech- 6 Conclusion
anism in CTMR, though helps in increasing delivery ratio, it
also increases delivery delay to some extent. This is because The area of Post Disaster Situation Analysis and Resource
forwarding to next hops is halted temporarily until the pre-set Management needs more focus because of increased severity
time-stamp period is exhausted. This increases the delivery and losses accompanied with disasters. A DTN presents a good
delay of CTMR forwarding slightly. Though BubbleRap alternative to create fast network architecture in case the existing
scheme takes minimum number of hops for bundle delivery, communication network is crippled after a disaster strikes. In a
as in Fig. 27, it is not a good candidate in terms of Delivery DTN, communication channels may connect and disconnect
Ratio as shown in Fig. 25. CTMR forwarding takes the max- continuously. Longer periods of disconnection are not rare in
imum number of hops, on an average, to deliver its messages DTN. Specialized routing protocols need to be developed for
communication to be made possible in a DTN. Unicast DTN
routing protocols have limitations in the context of performance
14000
in a post-disaster situation where most of the tasks are performed
Average Bundle Delivery Delay

12000 using collaboration and are group-based. In the current work, we


have proposed a multicast DTN routing protocol named
10000
Collaborative Time-stamp based Multicast routing (CTMR) pro-
(in seconds)

8000 tocol that is developed for efficiently handling post-disaster relief


work and resource management. A multicast protocol is capable
6000
of reducing bundle redundancy and overhead to a large extent. In
4000 the proposed strategy, bundles are created collaboratively among
2000
a certain neighborhood of a node such that redundant bundles of
similar type and same destination are avoided. Several important
0 parameters have been introduced in CTMR like optimal wireless
CTMR Forwarding Bubble Rap Delegation Forwarding

Fig. 26 CTMR forwarding based on time-stamp constrained Prophet vs.


communication range for each communicating node, optimal
Bubble Rap and delegation forwarding in terms of average bundle deliv- location difference & time-stamp period values, and optimal pa-
ery delay tience limit values. These parameter values, along with the
Peer-to-Peer Netw. Appl.

probabilistic calculations of potentiality for a node to deliver to 9. Lindgren A, Doria A (2006) Probabilistic routing protocol for in-
termittently connected networks. IETF, draft-lindgren-dtnrg-proph-
another node based on the well-known PRoPHET model, have
et-02.txt, March 2006
been used in combination to generate a bundle priority queue. 10. Burgess J, Gallagher B, Jensen D, Levine BN (2006) MaxProp:
This queue decides the order in which the bundles at a node are routing for vehicle-based disruption- tolerant networks. In Proc.
to be transmitted to the next hop in communication. CTMR has IEEE Infocom, April 2006
been implemented in the ONE simulator and its performance has 11. Spyropoulos T, Psounis K, Raghavendra CS (2005) Spray and wait:
an efficient routing scheme for intermittently connected mobile net-
been compared with other existing multicast routing protocols in works. In Proc. ACM SIGCOMM Workshop Delay-tolerant net-
DTN. Simulation results, with a 95 % confidence interval, have working, pp 252–259
shown that under the considered scenario of a post-disaster en- 12. Balasubramanian A, Levine BN, Venkataramani A (2007) DTN
vironment, CTMR performs best as far as the overall bundle routing as a resource allocation problem. In: Proc. ACM
SIGCOMM, August 2007
delivery probability and the average bundle delivery delays are 13. Erramilli V, Crovella M, Chaintreau A, Diot C (2008) Delegation
concerned. forwarding. In Proceedings of the 9th ACM International
Symposium on Mobile Ad Hoc Networking and Computing 2008
May 26, ACM, pp 251–260
14. Iranmanesh S, Wu-Chin K (2015) A novel mobility-based routing
7 Future work protocol for semi predictable disruption tolerant network. Int J
Wireless Inf Networks 138–146
There are different kinds of nodes in a disaster scenario which 15. Abdulla M, Simon R (2006) A simulation analysis of multicasting
in delay tolerant networks. In: Simulation Conference, 2006. WSC
could be selfish node or malicious nodes. It is an important
06. Proceedings of the Winter. IEEE, pp. 2234–2241
concern that if a node is malicious or selfish then it can destroy 16. Del Duca Almeida V, Oliveira AB, Macedo DF, Nogueira J (2012)
the entire team work. Detecting and handling such malicious Performance evaluation of MANET and DTN routing protocols. In:
nodes would be beneficial for an effective communication. In Wireless Days (WD), 2012 IFIP. IEEE, pp 1–6
our future work, we would like to incorporate measures to detect 17. Moy J (1994) Multicast Extensions to OSPF. RFC 1584,
March 1994
and handle such malicious nodes that might be present in the 18. Waitzman D, Partridge C, Deering S (1988) Distance vector
network. multicast routing protocol. RFC 1075, November 1988
19. Xie J, Talpade RR, Mcauley A, Liu MY (2002) AMRoute: ad hoc
multicast routing protocol. Mobile Netw Appl 7(6):429–439
Acknowledgments This research work is an outcome of the 20. Bae SH, Lee SJ, Wouldiam S, Gerla M (2000) The design, imple-
Government of India Project titled DiSARM funded by Information mentation, and performance evaluation of the on-demand multicast
Technology Research Academy, Media Lab. Asia, Department of routing protocol in multihop wireless networks. IEEE Network,
Electronics & Information Technology, Ministry of Communications January 2000, pp 70–77
and Information Technology. 21. http://www.ndma.gov.in/en/. Accessed 12/01/16
22. Ye Q, Cheng L, Chuah MC, Davison BD (2006) OS-multicast: on-
demand situation-aware multicasting in disruption tolerant net-
works. In Vehicular Technology Conference, 2006. VTC 2006-
References Spring. IEEE 63rd, vol 1, pp 96–100. IEEE
23. Santiago J, Casaca A, Pereira PR (2008) Non-custodial multicast
over the dtn-prophet protocol. In: Wireless Sensor and Actor
1. Farrell S, Cahill V (2006) Delay and disruption tolerant networking. Networks II 2008. Springer US, pp 197–208
Artech House, Inc 24. Lee U, Oh SY, Lee KW, Gerla M (2008) Relaycast: scalable
2. Cerf V, Burleigh S, Hooke A, Torgerson L, Durst R, Scott K, Fall K, multicast routing in delay tolerant networks. InNetwork Protocols,
Weiss H (2007) Delay-tolerant networking architecture. RFC4838 2008. ICNP 2008. IEEE International Conference on 2008 Oct 19,
3. Hui P, Crowcroft J, Yoneki E (2011) Bubble rap: social-based pp 218–227. IEEE
forwarding in delay-tolerant networks. IEEE Trans Mob Comput 25. Kawecki M, Schoeneich RO (2016) Mobility-based routing algo-
10(11):1576–1589 rithm in delay tolerant networks. EURASIP J Wirel Commun Netw
4. Ye Q, Cheng L, Chuah MC, Davison BD (2009) Performance com- 2016(1):1–9
parison of different multicast routing strategies in disruption toler- 26. Zeng Y et al (2013) Directional routing and scheduling for green
ant networks. Comput Commun 32(16):1731–1741 vehicular delay tolerant networks. Wirel Netw 19(2):161–173
5. Fall K, Scott KL, Burleigh SC, Torgerson L, Hooke AJ, Weiss HS, 27. Lo SC, Luo NW, Gao JS, Tseng CC (2014) Quota-based multicast
Durst RC, Cerf V (2007) Delay-tolerant networking architecture. routing in delay-tolerant networks. Wirel Pers Commun 74(4):
http://tools.ietf.org/html/rfc4838 1329–1344
6. Zhao W, Ammar M, Zegura E (2005) Multicasting in delay tolerant 28. Deng X, Chang L, Tao J, Pan J, Wang J (2013) Social profile-based
networks: semantic models and routing algorithms. Proceedings of multicast routing scheme for delay-tolerant networks. In: 2013 I.E.
the 2005 ACM SIGCOMM Workshop on Delay-Tolerant International Conference on Communications (ICC), pp 1857–
Networking. ACM. 1861. IEEE
7. Scott K, Burleigh S (2007) Bundle protocol specification. IETF, 29. Iranmanesh S, Chin KW (2015) A novel mobility-based routing
RFC 5050 protocol for semi-predictable disruption tolerant networks. Int J
8. Vahdat A, Becker D (2000) Epidemic routing for partially connect- Wireless Inf Networks 22(2):138–146
ed Ad hoc network. Duke University Technical Report Cs-200006, 30. Gupta AK, Bhattacharya I, Banerjee PS, Mandal JK, Mukherjee A
April 2000 (2015) DirMove: direction of movement based routing in DTN
Peer-to-Peer Netw. Appl.

Indrajit Bhattacharya is an
architecture for post-disaster scenario. Wireless Networks (Springer
Assistant Professor at Kalyani
US), June 2015
Government Engineering
31. Kernen A, Ott J, Karkkainen T (2009) The ONE simulator for DTN College (KGEC) in West Bengal,
protocol evaluation. Proc. Int. Conf. Simul. Tools Tech., pp 55 India. He has completed his PhD
32. Lindgren A, Phase KS (2006) Evaluation of queuing policies and from Jadavpur University in West
forwarding strategies for routing in intermittently connected net- Bengal, India in 2014. He obtain-
works. In: Communication System Software and Middleware, ed his Master in Computer
2006. Comsware 2006. First International Conference on 2006 Science from the University of
Jan 6, pp 1–10. IEEE Calcutta, West Bengal. He has a
teaching and research experience
of more than 14 years in different
institutes of repute. He is the
Amit Kumar Gupta is a Ph.D.
Principal Investigator of the
research scholar from the
Project titled DiSARM (Post
University of Kalyani, West
Disaster Situation Analysis and Resource Management using Delay
Bengal, India. His areas of interest
Tolerant Peer-to-Peer Opportunistic Networks) at KGEC, funded by
include delay tolerant networks,
Information Technology Research Academy, Media Lab. Asia, Govt. of
wireless networks, sensor net-
India. His research interests include delay tolerant networks, wireless
works, security in wireless net-
networks, sensor networks, and radio frequency identification.
works, social networks data anal-
ysis, and data management in
wireless transmission. He has
c o m p l e t e d h i s M a s t e r ’s i n
Computer Applications and in
Information Technology from Tamal Mondal is a research
theUniversity of Calcutta. He scholar from the Kalyani
carries a teaching experience of Government Engineering
four years of teaching Master’s students. College (KGEC), West Bengal,
India. His areas of interest include
delay tolerant networks, wireless
networks, sensor networks, social
networks data analysis, and other
related fields. He has completed
Jyotsna Kumar Mandal is a h i s M a s t e r ’s i n C o m p u t e r
Professor at the Computer Applications from KGEC.
Science and Engineering
Department of The University of
Kalyani, West Bengal, India. He
has been an ex-Dean to the
Faculty of Engineering,
Technology & Management at
the University of Kalyani. He
has an active academic carrier
with more than 30 years of teach- Sourav Sanu Shaw is a master’s
ing and research activities in vari- student at Kalyani Government
ous reputed institutes and Engineering College (KGEC),
Universities. He is an active mem- West Bengal, India. His areas of
ber of more than ten professional interest include delay tolerant net-
bodies like IEEE, ACM, to name a few. He has frequent involvement with works, wireless networks, social
organizing international and national seminars in different areas of com- networks data analysis, and other
puter science and allied subjects. His fields of interests include coding related fields.
theory, data and network security, remote sensing and GIS based appli-
cations, data compression error corrections, watermarking, steganogra-
phy and document authentication, image processing, visual cryptography,
MANET, wireless and mobile computing, unified computing, chaos the-
ory and applications. He leads various government and non-government
projects in his computing laboratory.

You might also like