Delay Tolerant Networks Routing Protocols

Published Date: 02 Nov 2017

The routing algorithms of DTN have the inbuilt storage management scheme such as Hop based TTL (Spray and Wait) or passive cure (Potential-based Entropy Adaptive Routing PEAR). There has been a significant amount of work in the past regarding buffer management policies. In this paper, we have proposed a new message deletion policy for multi-copy routing schemes. This scheme uses the acknowledgement method to remove the useless bundles from the network, which prevent the nodes from the buffer overflow problem and avoid transfer of useless message replicas thus relaxing the resources of the nodes. We evaluate our proposed method by simulating network, on four major DTNs routing algorithms: Epidemic, Spray and Wait, ProPHET and MaxProp. The simulation results clearly show significant improvement in the value of delivery probability and the overhead ratio for an Epidemic, Spray and Wait, and Prophet routing protocols.

Keywords: Delay Tolerant Networks, Epidemic, Spray & Wait, Prophet, MaxProp, Delivery Probability, Average Latency, Overhead Ratio.

Introduction

Conventionally, data networks are sculpted by connecting graphs whereby the presence of at least one end-to-end path among any source-destination duo is continuously assured. In these networks, any arbitrary link between two network nodes is supposed to be bidirectional supporting symmetric data rates with little error likelihood and latency (i.e. Round-trip time is in the order of milliseconds). In these networks, packets are not supposed to exist in a node’s buffer for a lengthy time period. On the basis of these basic suppositions, the Internet was designed and its most commonly used protocols, particularly the TCP/IP protocol suite, were designed.

However, these suppositions do not hold when designing existing and recently emerging wireless networks, especially those which are to be deployed in extreme environments (e.g. Battlefields, volcanic regions, deep oceans, deep space, developing regions, etc.). Under such challenging conditions, these networks suffer from excessive delays, acute bandwidth restrictions, extensive mobility of nodes, frequent power outages and repeated communication hindrances. Wireless networks working under these challenging conditions experiences connectivity which becomes considerably intermittent and no continuous end-to-end path(s) between any source-destination pair can be guaranteed [1].

Popular examples of such intermittently connected networks (ICNs) scenarios are satellites, deep space probes, Wireless Sensor Networks (WSNs), Mobile Wireless Sensor Networks (MWSNs) and Sensor/Actuator Networks (SANs) deployed in extreme regions, Mobile Ad-Hoc Networks (MANETs) typically consisting of nodes (e.g. GPSs, PDAs, Cellular Phones, Tracking devices, Laptops, etc.) mounted over continuously moving objects [1].

Many research interests focus on developing new approaches for routing in delay tolerant network environment. These routing schemes generally use the store-carry-and-forward approach, where intermediate nodes keep the message until encounter other nodes to set up new links in the path to the destination [2]. DTN Routing protocols can be broadly categorized on two bases: (1) on the basis of the number of copies and (2) on the basis of knowledge of future contact opportunities and message patterns. On the basis of the number of copies, we have Single-copy routing schemes which use only one copy per message and significantly reduce the resource requirements but suffer from long delays and low delivery ratios. Other one is Multi-copy routing schemes have a high probability of delivery and lower delays at the cost of buffer space and more message transfers.

Section 2 summarizes prior related work on routing in disrupted environment. From our summary of related work, it becomes apparent that there is a need of the buffer management technique to remove the useless replicas of the already delivered message.

Section 3 details the problem related to the existence of the replicas of delivered message. In this section, we have taken an example to show the replication process in DTN, and then we have carried out a hypothetical simulation to calculate the maximum possible improvement if this useless replicas is removed from the network.

Section 4 details the methodology used to remove useless replicas from the network. In this, we have explained the data structures required to implement the mechanism and also gave the detailed algorithms used for the implementation of the technique.

In section 5, performance evaluation is carried out. In this section, we have explained detailed simulation setup and different parameters used for the study. Also in this section we have analyzed the improvement in performance of base routing protocols when this extension mechanism is implemented.

Related Work

Basic DTN routing algorithms rely only on node movement, and no other information is used for the creation of the communication link. Examples of fundamental DTN routing are "Custody Transfer" and "Epidemic Routing",

Software of DTN research projects uses a random algorithm to simulate node movement whereas mobility in real life has a predictable pattern. Certain DTN routing algorithms are designed to utilize this predictable behavior of node mobility for predicting message delivery in a probabilistic manner [3].

PROPHET is a probabilistic routing algorithm in which when the source initializes the message, this message relies on the node mobility for its delivery to the destination. Jain et al in 2004 [4] carried out the study of DTN routing with predictable connectivity. These algorithms apply to DTN where there is a predictable intermittent connectivity (for example satellite movement in IPN) [5].

An optimal "oracle based" algorithm has been proposed in [6]. The algorithm has the full knowledge of future movements of the nodes and computes optimal forwarding decision (i.e. Time and next hop) which delivers a message to the destination in a minimum amount of time.

There are a number of other proactive approaches to routing which are made possible by stronger assumptions such as knowledge of connectivity patterns and control over peer movement [4, 7-12].

Data mule concept was introduced by Shah et al in 2003. No DataMule – DataMule forwarding is allowed, only direct transmission to the destination is carried out [13]. This idea of collecting data through a partitioned network has also been used in [14, 15]. In other works as mentioned earlier in literature, all the nodes are mobile, and algorithms are used to forward messages from any node to any other node [3, 16-21]. To reduce the overhead of flooding and improve the performance of DTN routing protocol, one simple approach is to forward the data packets with some probability p<1 [22]. Other intelligent approaches use the history based or utility based routing [3, 17, 19]. In these protocols, each node keeps track of a utility value for every other node in the network. Thus, these schemes reduce the extent of flooding and hence result in better performance than flooding based approaches [3, 6, 19].

Another approach is spray and wait in which there is a significant reduction in transmission overhead as compared to flooding based approach [23]. Also spray and wait have better performance w.r.t. Delivery delay as comparing to flooding. This approach does not make use of any knowledge of the network, not even the past encounters. It is also a flooding based approach but in this flooding is done in a controlled manner.

Apart from multi-copy routing in DTN, there are certain single-copy schemes which have been extensively explored in [6, 13]. Such scheme significantly reduces the number of transmissions by generating and forwarding only a single copy of the message in the network.

Generally the routing protocols of DTN manage their buffers as first-in-first-out (FIFO) queues [16]. Another approach is Drop Least Encountered (DLE) algorithm [24] which proposed dropping messages with the lowest likelihood of delivering and various work deployed this dropping technique [3, 8, 10, 25].

In 2006 Burges et al proposed MaxProp routing protocol. It is based on prioritizing both the schedule of packets transmitted to other peers and the schedule of packets to be dropped [26].

Resource Allocation Protocol for Intentional DTN routing (RAPID) was proposed by Balasubramanian et al. in 2007. RAPID is fabricated to explicitly optimize administrator specific routing metrics and to translate these metrics to per packets utilities [27].

In a prediction based scheme like SOLAR [3, 6, 10, 26, 28-30], the nodes with high delivery probability are used to forward the messages and in this way, these protocols try to reduce the message overhead and buffer contention. As the network is portioned in DTN, it takes very long time before every node receives a delivery probability of other nodes.

In 2007 Lei et al. proposed SMART (Selectively MAking pRogress Towards delivery), which combines the best features of prediction based scheme and opportunistically forwarding based schemes. SMART achieves a high delivery rate, small delivery latency while keeping message overhead low [31]. It is achieved by using node's mobility pattern and controlled flooding methodology.

MOBISPACE [28] and SOLAR [30] use location-visiting probabilities, which may raise privacy concerns and may be inaccurate in predicting the good forwarder.

Prioritized epidemic routing (PREP) was proposed by Ramanathan in 2007 which used four inputs (the current cost to destination, current cost from source, expiry time and generation time) for imposing partial ordering on the message on the deletion and transmission [32]. Yue et al. in 2011 used counter method to remove the message copies from the node. In this method, each node adds an encounter counter based on epidemic routing scheme.

Problem Definition

It is evident from the literature that the multi-copy routing schemes achieve higher delivery probability as compared to the single copy routing scheme. This improvement is achieved at the cost of higher resource utilization i.e. Multi-copy routing protocols require more buffer space, more bandwidth, incur more overheads and consume other vital network resources. In multi-copy routing scheme, multiple copies of the message are propagated on the network with the belief that one of the copies will reach the destination. When one copy reaches the destination the other copies become useless, but they still remain in the network and utilize the resources of the network. In this section, we have tried to analyze the impact of message replicas (which exists in the network after the delivery of the message in the multi-copy routing scheme) on the performance of the routing protocols.

The example below show how the replicas exits after the delivery of the bundle / message to the destination and then we have shown the hypothetical simulation in which it has been deduced that there is scope of improvement in the performance of these multi copy routing scheme if these replicas are removed from the network.

Example

In the figure 1, at time t1 source ‘S’ generates a message for the Destination ‘D’. At this point of time, there exists different disjoint regions of networks and S is connected to only 1, 2 and 3 nodes. At time t2, the source S forwards the message to the node 3.

At time t1 At time t2 At time t3

Figure 1: Network and data bundle status at time instances t1, t2 and t3

At time instance t2, the node 3 moves to the new location and thus the node 3 come in contact of a region with nodes 7, 8,9,10 and 11. At time t4, node 3 forwards the message to node 7 which in turn is forwarded to node 11. In this process, the node S, 3 and 7 retains a copy of the forwarded bundle.

At time t4 At time t5 At time t6

Figure 2: Network and data bundle status at time instances t4,t5 and t6

Then node 11 at time t6 moves to the new point in the network and thus comes in the communication range of node 14. A link is established between 11, and 14 and thus node becomes the member of a region with nodes 14, 15, 16 and D.

The message is forwarded to node 14 by node 11 retaining the copy of the original message. This message is forwarded to node D (Destination) by node 14 again retaining the copy of it. At time t8 the message is delivered to the intended destination. But if we see the figure and the state of the network we can identify that, in this process, there exists 5 replicas of this delivered message which are now useless and are utilizing the resources. These messages will keep on utilizing the resources and will participate in further replications as these nodes, which are holding these replicas, are not aware of the delivery of the message to the final destination. So there is a need of removing these replicas from the network so as to free up the held resources. This can be achieved if the delivery information is communicated back to these nodes.

At time t7 At time t8

Figure 3: Network and data bundle status at time instances t7 and t8

3.1 Problem Formulation: (Hypothetical Simulation)

To analyze the percentage scope of improvement in the performance of considered routing protocols by deleting the replicas after delivering the bundle to the destination, we have carried out a hypothetical simulation. In this simulation, at the time of delivery of the message to the destination we have dropped all the replicas from the network and evaluated the performance in this scenario. Then we compared this performance value to the base routing protocol’s performance values and calculated the difference in performance value. This difference is the maximum improvement scope in the performance.

To carry out this study we have used ONE simulator for Delay Tolerant Networks. The detailed simulation parameters for the considered scenario are summarized in table 1 below.

Table 1: Simulation Parameter for hypothetical simulation in which all replicas are dropped from the network when the message is delivered to the destination.

Parameter

Value

Simulation Time

43200 Sec (12Hours)

Interface Transmission Range

30m

No of Host Groups

1

No of Hosts Per Group/ Total number of hosts

50

Group Router

Epidemic; MaxProp; Spray And Wait; Prophet

Buffer Size

25 M

Movement Model

ShortestPathMapBasedMovement

Message generation Event Interval

One new message every 15 to 30 seconds

Message Size

250kB - 2MB

The results obtained are summarized in the table 2. We analyze that the maximum scope of improvement is there in the Prophet routing protocol. There we have seen 36%, 71%, and 31% improvement in delivery probability, overhead ratio and average latency respectively. Further in Epidemic routing protocol there is 31%, 14% and 26 % scope of improvement in delivery probability, overhead ratio and average latency respectively.

Table 2: Percentage improvement in the performance of considered routing protocols when all replicas are dropped at the delivery of the message

Protocol

Delivery Probability

Overhead Ratio

Average Latency

Without deleting replica

All Replicas deleted at delivery

% improvement

Without deleting replica

All Replicas deleted at delivery

% improvement

Without deleting replica

All Replicas deleted at delivery

% improvement

Epidemic

0.71

0.93

31.76

26.56

13.80

13.80

2734.06

2014.76

26.30

Spray And Wait

0.8105

0.93

15.91

15.94

5.99

5.99

2258.82

1570.71

30.46

MaxProp

0.9657

0.96

0.25

9.69

8.84

8.77

1468.96

1359.92

7.42

Prophet

0.6988

0.95

36.13

23.54

6.86

70.83

2585.92

1778.99

31.20

The results of the above simulation encourage us to work upon the mechanism to remove useless replicas of delivered messages from the network.

Methodology

Generally in the network with the multi-copy routing scheme there are a number of replicas of messages which exist in the network. These replicas utilize the network resources and could even exhaust them. Once the message is delivered to the intended destination, its replicas become useless, but they still keep on utilizing the resources and also participate in the further replication process. This replication and utilization of the resources can only be prevented if the delivery information of the message is propagated back to the other nodes in the network.

In this paper, we have proposed the mechanism to propagate this information back to the nodes in the network and to remove this useless replicas from the network.

In our mechanism, all the nodes in the network are required to maintain a list of known delivered messages ID. When the node comes in communication range of the other node, both nodes will exchange the knowledge of the known delivered messages. If a message in node buffer matches the ID of delivering messages in the list, then the node deletes corresponding message from its buffers.

To remove the replicas from the network, we need to convey the delivery information to the nodes in the network. To achieve this we have maintained two lists with each node namely deliveredMessage and del_msgs. Hash map deliveredMessage maintains the information of messages delivered at this node as a destination. So deliveredMessage hash map maintains node level delivery information. Del_msgs maintains the list of messages delivered throughout the network.

deliveredMessages: This is the hash map maintained by each router and contains the messages received by this router as a final recipient. DeliveredMessages are defined in MessageRouter.java module and is applied to each and every router/node defined in DTN scenario. It keeps track of delivered messages at the node level.

M1 M2 M3

D

Contact Occurs: Node S having message M5 meets Node D

M1 M2 M3 M5

D

Message Delivered: deliveredMessage is updated with Message M5

S

S

(a)

(b)

M5

M5

deliveredMessage data structure

deliveredMessage datastructure

Figure 4: DeliveredMessages Data Structure propagation

As shown in figure 4, S have a message M5 in its buffer, which is destined for D. When S meets the D, it delivers the message M5 to D and D makes the entry of this delivered message in its deliveredMessage hash map as shown in fig 4 (b).

del_msgs: This is the list maintained by each router and contains a collection of all known delivered messages in the network. Whenever two nodes come in contact with each other they exchange their del_msgs list with each other and from the information contained in other nodes del_msgs list the current node updates its own del_msgs list and adds the messages already delivered and not present in its del_msgs list. It keeps track of network wide delivered messages.

M1 M2 M3 M4 M5

N2

Contact Occurs: Node 1 meets N2

M1 M2 M3 M4 M5

N2

Contact Occurs: Del_msgs are exchanged to get the current status of delivered messages

M1 M3

N1

M1 M3 M2 M4 M5

N1

Updating del_msgs

(a)

(b)

Figure 5: del_msgs data structure propagation

When N1 meets N2 they exchange their previous knowledge of network wide delivered message information stored in del_msgs. As shown in fig 5 (a), N1 have knowledge of M1 and M3 delivered message information and N2 have knowledge of M1, M2, M3, M4 delivered messages. At the contact opportunity N1 and N2 exchange their del_msgs list and hence from the del_msgs list of N2, N1 learns that M2 and M4 have also been delivered. Thus, N2 updates its del_msgs list and add M2 and M4 in its del_msgs list (fig 5 (b)).

The complete procedure for extending base routing protocols so as to implement network wide propagation of message status information is shown below

Steps for the exchange of message delivery information

Step 1: Populating: Update the network wide delivered message information stored in del_msgs from the node level delivered message information stored in the deliveredMessage hash map.

Step 2: TTL Check: Check and remove the message ids from the del_msgs of those messages whose TTL has expired. This help to reduce the length of del_msgs list and hence reduce the processing burden on the node.

Step 3: Exchange: Two connected nodes exchange their del_msgs lists to update the network wide message delivery information.

Step 4: Replica Deletion: Drop those messages from the message collection of the node which have already been delivered to the intended destination. This frees up the resources held by these messages.

Figure 6: steps for the exchange of message delivery information

The overall procedure consists of four phases or steps as shown in figure 6. In step one, the del_msgs data structure is populated with the information stored in the deliveredMessage hash map. In the second phase, (TTL Check) the del_msgs list is shortened by removing the IDs of the messages whose TTL has already expired. In the third phase, del_msgs are exchanged for the two connected nodes, to share information of the delivered messages so that other node can also update its own del_msgs list. In the last phase, the replicas of the delivered messages whose IDs are present in del_msgs list are removed from the message collection of the router so as to free the space occupied and resources held by these messages. In the sections below, we will discuss in detail each step of the procedure defined.

Populating

The populating phase is executed when the two nodes come in contact of each other i.e. at connection establishment. It reads the messages stored in its deliveredMessage hash map and compares the IDs of these messages in deliveredMessage with the IDs of the messages stored in the del_msgs list. If there are some messages in the deliveredMessage hash map which are not present in the del_msgs list then those messages ID is added to the del_msgs list. This process is again repeated for the other node so as to populate the other node’s del_msgs list (fig 7 lines 6-10).

Algorithm: Populating del_msgs from Delivered Message hash map

Input: Node: n1, n2

DeliveredMessage hash map of n1 and n2

Del_msgs list of n1 and n2.

Output: del_msgs of n1 and n2 are updated with the delivered message information from the hash map of n1 and n2.

For all messages in deliveredMessage hash map of node n1 do

If not (n1 del_msgs contains deliveredMessage) then

Add deliveredMessage Id to del_msgs list of node1.

Endif

End

For all messages in deliveredMessage hash map of the node n2 do

If not (n2 del_msgs contains deliveredMessage) then

Add deliveredMessage Id to the del_msgs list of node2.

Endif

End

Figure 7: Populating Algorithm for adding message IDs from deliveredMessage to del_msgs

TTL Check

The messages whose IDs are present in the del_msgs list are checked for the TTL expiry. It starts with getting the list of messages stored in the del_msgs list of node. Then these messages TTL value are checked against the maximum TTL value configured for the network. If it has exceeded the maximum configured TTL value, those messages IDs are removed from the del_msgs list as those messages would have already dropped by the nodes due to TTL expiry (fig 8 lines 1-6). This complete process is again executed for the other node’s del_msgs list (fig 8 lines 7-12).

Algorithm: TTL Check for limiting the del_msgs list.

Input: Node: n1, n2

Del_msgs list of n1 and n2.

Output: del_msgs list of n1 and n2 with removed message Ids whose TTL has expired.

Obtain a del_msgs list for node n1.

For all message Ids in del_msgs list of n1 do

If (simulation_time - message_creation_time) > maximum TTL configured then

Remove message Id from the del_msg list of n1

Endif

End

Obtain a del_msgs list for node n2.

For all message Ids in del_msgs list of n2 do

If (simulation_time - message_creation_time) > maximum TTL configured then

Remove message Id from the del_msg list of n2.

Endif

End

Figure 8: TTL Check Algorithm for limiting the length of the list.

Exchange

In this phase, del_msgs lists are exchanged for the two connected nodes. The message IDs stored in their del_msgs list is compared with the message IDs stored in other node del_msgs list. If some message is found in other node’s del_msgs, which are not present in their del_msgs, those message ID is learned and added to their del_msgs list (fig 9 lines 1-11). Same is executed by the other node, to learn the delivered messages from the other node’s del_msgs list (fig 9 line 12-22).

Algorithm: Exchange of del_msgs between two connected nodes

Input: Node: n1, n2

Del_msgs list of n1 and n2.

Output: updated del_msgs list of n1 and n2 with network wide delivered message information.

For all messages in node n1’s del_msgs do

flag = false

For all messages in node n2’s del_msgs do

If (message Id of n2 del_msg = message Id of n1 del_msg) then

flag=true

Endif

End

If flag=false then

Add message Id of n1 del_msg to n2 del_msgs list

Endif

End

For all messages in node n2’s del_msgs do

flag = false

For all messages in node n1’s del_msgs do

If (message Id of n2 del_msg = message Id of n1 del_msg) then

flag=true

Endif

End

If flag=false then

Add message Id of n2 del_msg to n1 del_msgs list

Endif

End

Figure 9: Algorithm for the exchange of del_msgs between two connected nodes

Replica Deletion

The last step is replica deletion. In this step, the message collection of the node is compared against the updated del_msgs list which contains the list of message IDs of the known delivered messages in the network. If some messages exist in the message collection, which are already delivered to the intended destination and their ID is present in the del_msgs list those messages are dropped by the node and are removed from the message collection of the node (fig 10 lines 1-8). The same procedure is followed by other connected node, to drop the delivered messages (fig 10 lines 9-16).

Algorithm: Replica Deletion from the Message Collection of Node’s routers.

Input: Node: n1, n2

Del_msgs list of n1 and n2.

Output: updated message collection of n1 and n2 with replicas of delivered messages removed.

Obtain message collection of n1.

For all messages in del_msgs of n1 do

For all messages in message collection of n1 do

If message Id of message from message collection of n1 = message Id of del_msg of n1 then

remove the message from the message collection of n1

Endif

End

end

Obtain message collection of n2.

For all messages in del_msgs of n2 do

For all messages in message collection of n2 do

If message Id of message from message collection of n2 = message Id of del_msg of n2 then

remove the message from the message collection of n2

Endif

End

end

Figure 10: Algorithm for the replicas deletion from the message collection of node’s router

Performance Evaluation

The base routing protocols (Epidemic, Spray & Wait, ProPHET and MaxProp) are extended by adding the mentioned mechanism in the base protocols. These extended routing protocol performance is evaluated and compared with performance of base routing protocols through extensive simulations using ONE simulator. We employed a variety of scenarios by varying node buffer size.

Our results show that adding the procedure defined in section 4 fig 6 to the base routing protocol significantly improves performance as the procedure frees up the resources which can now be utilized to deliver other messages, which are not yet been delivered.

5.1 Simulation setup

The procedure defined in fig 7 to 10 is added to the base routing protocols namely Epidemic, Spray and Wait, Prophet, and MaxProp. These procedures are added to the Active Router class, hostConnected action listener and the messageTransfer action listener of the ONE simulator.

We simulated a 12 hour period and evaluated the different metrics used to compare the performance of routing protocols. Four routing protocols are considered in this study namely Epidemic, Spray & Wait, MaxProp, and Prophet. These routing protocols are implemented in a Java based ONE (Opportunistic Network Environment) simulator. Mobile nodes move according to the movement model selected. Once it reaches a destination, it randomly waits for 5 to 15 minutes (wait time). Then, it selects a new map location, and a random speed between 30 and 50 km/h. The mobile node moves to the new destination using the characteristics of movement model used. Each of the mobile node buffer size is varied from 5 to 35 Mbytes and the performance metrics are studied under different buffer size scenario. Node’s buffer works on FIFO strategy.

Messages are exchanged between random source and destination mobile nodes. It uses an inter message creation interval in the range [15, 30] (seconds) of uniformly distributed random values. Message size is in the range [500 KB, 2 MB] of uniformly distributed random values. All the messages exchanged have a time to live (TTL) of 2 hours when node buffer is varied.

We assume that the traffic matrix is not provided in advance, and the mobile node routes are not pre-assigned and fixed, so there isn’t any knowledge about the transfer opportunities.

Network nodes connect to each other using IEEE 802.11b with a data rate of 2 Mbit/s, and a transmission range of 30 meters. The simulation parameters are summarized in the table 3 below.

Table 3: Simulation Parameters

Parameter

Value

Simulation Time

43200 Sec (12Hours)

Interface Transmission Range

30m

No of Host Groups

1

No of Hosts Per Group

50

Group Router

Epidemic; MaxProp; Spray And Wait; Prophet

Buffer Size (Varied)

5 to 35 M

Message TTL

120 Min

Movement Model

ShortestPathMapBasedMovement

Message generation Event Interval

One new message every 15 to 30 seconds

Message Size

250kB - 2MB

Wait Time

300-900 Sec

5.2 Results

We evaluated the extended protocols under the varying node buffer size. The buffer size if varied from 5 M to 35 M in steps of 5. The performance metrics studied for the performance comparison is delivery probability, overhead ratio and average latency. The results corresponding to these performance metrics are discussed below.

5.2.1 The impact on Delivery Probability:

The fig 11 (a-d) shows the Delivery Probability of extended routing protocol compared to base routing protocol under the variable buffer size of the nodes. From the output shown in fig 11 (a-d), it is clear that the delivery probability of the extended routing protocol is significantly improved as now only the useful data packets are being handled and forwarded by the nodes. Thus in this case the eventual delivery of the data bundle contributes to the delivery probability of the nodes. Whereas, in the case of base routing protocols, it is uncertain that the data bundle being forwarded or delivered will contribute to the delivery probability as there is a high probability that it could be the replica of the delivered data bundle.

Further we see that there is about 25% - 66 % improvement (w.r.t. Varying buffer) in case of extended epidemic routing protocol. Similarly, for the Prophet routing protocol the percentage improvement is from 28% to 54% (w.r.t. Varying buffer size). In case of Spray and Wait routing protocol, there is a marginal improvement of 9% - 22% as in this case the replication level is based on the value of L.

Delivery Probability Comparison of With and without replicas deleted in Epidemic Routing

Delivery Probability Comparison of With and without replicas deleted in MaxProp Routing

Delivery Probability Comparison of With and without replicas deleted in Prophet Routing

Delivery Probability Comparison of With and without replicas deleted in Spray and Wait Routing

Fig 11 (a-d): Delivery Probability comparison of With and without replicas deleted DTN routing protocols individually under varying node buffer size.

5.2.2 Impact on Overhead Ratio:

Fig 12 (a-d) shows the Overhead Ratio comparison of the extended and base routing protocols with respect to the variable buffer size of the nodes. It is observed that the % improvement in the performance of extended Prophet routing is around 65%, in extended epidemic routing is about 54%, and in extended Spray and Wait routing is about 48%. This improvement in Overhead ratio is as because of the fact that, in case of extended routing protocols, the number of packets decreases as the useless replicas get deleted and hence only the useful packets are dealt with by the routing protocols.

It is also observed that the adding procedure for the propagation of delivery information has a higher impact on the performance when the nodes have small buffer size as compared to large buffer size. From fig, we can see that for the considered scenario, when the buffer size is less than 20M the performance improvement is high, but as it is increased more than 20M the performance remains the constant. This is because of the fact that now there is a significant amount of buffer space to accommodate the data packet, and there is no much need of shuffling them or to apply buffer management techniques.

Overhead Ratio Comparison of With and without replicas deleted in Epidemic Routing

Overhead Ratio Comparison of With and without replicas deleted in MaxProp Routing

Overhead Ratio Comparison of With and without replicas deleted in Prophet Routing

Overhead Ratio Comparison of With and without replicas deleted in Spray and Wait Routing

Fig 12 (a-d): Overhead Ratio comparison of With and without replicas deleted DTN routing protocols individually under varying node buffer size.

5.2.3 Impact on Average Latency:

The average latency of the extended and base routing protocols is compared and shown in fig 13 (a-d). It is observed that the average latency does not have much impact as compared to the improvement in delivery probability and overhead ratio. The average improvement in the extended Epidemic, Prophet and Spray and Wait routing protocols is about 20% - 30%. It is evident that this improvement in average latency is observed when the node has large buffer size. At the small buffer size there is a marginal improvement in average latency. MaxProp does not show any difference in average latency.

Average Latency of With and without replicas deleted in Epidemic Routing

Average Latency of With and without replicas deleted in MaxProp Routing

Average Latency of With and without replicas deleted in Prophet Routing

Average Latency of With and without replicas deleted in Spray and Wait Routing

Fig 13 (a-d): Average Latency comparison of with and without replicas deleted DTN routing protocols individually under varying node buffer size.

Conclusion

In this paper, we have proposed the scheme for the management of buffer so as to prevent the nodes from the excessive resource utilization. In this technique, the procedure for the propagation of message delivery information is implemented above the multi copy routing protocols so as to remove the useless replicas of the bundles from the network. We have simulated this extended routing protocol in the Java based ONE simulator and tested it on four multi copy routing protocols namely Epidemic, Spray and Wait, Prophet and MaxProp. From the simulation results, it can be analyzed that the extended routing protocols give better results than the base routing protocols. The improvement in delivery probability is about 66% for Epidemic routing protocol, 53% for Prophet and 22 % for Spray and Wait routing protocol as compared to their base counterparts. Similarly the improvement in overhead ratio is about 57%, 67%, and 48% for extended Epidemic, Prophet, Spray and Wait routing protocols respectively as compared to unacknowledged counterparts.

The proposed buffer management scheme gave better results as compared to the base protocol performance. In the future, we can work out on the other metrics also like energy requirements and CPU utilization under varying message TTL and other network parameters. Also, work can be carried out by implementing this extension scheme in other multi copy routing protocols.