On Modeling Optimizations and Enhancing Routing Protocols for Wireless Multihop Networks

The contribution of this paper is two fold. It first analyses the flooding strategies for the Wireless Multihop Networks (WMNs) then it enhances the reactive and proactive routing protocols. For analysis purpose, we select four widely used flooding techniques for routing: i. traditional flooding, ii. Time-To-Live based Expanding Ring Search (TTL-based ERS) flooding scheme, iii. TTL-based Scope Routing (SR) flooding and iv. Multi-Point Relays (MPR) flooding. These techniques play a vital role and act as a backbone for routing protocols. Therefore, we compare efficiency of these techniques for six widely used routing protocols: Ad-hoc On-demand Distance Vector (AODV), Destination Sequenced Distance Vector (DSDV), Dynamic Source Routing (DSR), DYnamic MANET On-demand (DYMO), Fish-eye Scope Routing (FSR) and Optimized Link State Routing (OLSR). DSDV uses traditional flooding, AODV, DSR and DYMO use TTL-based ERS flooding, FSR uses TTL-based SR flooding and OLSR uses MPR flooding. This paper also presents mathematical models for flooding techniques and studies the affects of these techniques on their respective protocols in terms of energy and time consumption. This is done to measure the cost incurred by the routing protocols in the form of routing overhead and latencies. A novel contribution of this work is the enhancement in search set values and intervals of routing algorithms to improve the efficiency of selected existing protocols. A detailed comparison analysis of selected protocols with their default and enhanced routing algorithms in NS-2 is also a part of this work.


I. INTRODUCTION
Wireless Multihop Networks (WMNs) are the type of wireless networks that use multi-hop routing mechanism for communication. A mobile user with wireless node in WMNs can indirectly communicate on the move with other mobile users. Hence, routing protocols play a major part in providing and maintaining an indirect communication path between all mobile users that are part of this fully dynamic and distributed network. Different routing protocols are proposed that adapt varying techniques to search and maintain suitable path(s) between source(s) and destination(s). Moreover, routing strategies of these protocols are aimed to minimize The associate editor coordinating the review of this manuscript and approving it for publication was Ilsun You .
both the overhead and the convergence time for the dynamic topology of WMNs.
In WMNs, a route towards the destination is found by disseminating the route request messages. Upon reception, the destination sends the route reply message as a reply to the originator of route request. Various techniques have been proposed to propagate these responses throughout the network. Most widely used mechanism is the traditional/complete flooding or simply flooding. In traditional flooding strategy, the originator sends route request message to its nearest entities, which forward these requests further to their nearest entities until the message is successfully delivered to the destination node. The traditional flooding technique has very high route searching control overhead in WMNs. Due to high control overhead in traditional flooding strategy, several other VOLUME 8, 2020 This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ techniques have been proposed to minimize the overhead that include Time-To-Live (TTL)-based and Node Degree-based strategies [1].
In TTL-based strategy, flooding of route request message is confined within the specified region by assigning the value of TTL. The TTL value is fixed by the originator of request packet and is carried out in the route request packet. Each time when the request packet is relayed, a decrease in TTL value is observed. This process continues until value of TTL becomes zero and propagation of request packet stops. As a result, broadcast area of request message is governed by the TTL value. The destination node replies to the request only if it is within a specified region. Otherwise, source node restarts the search again with larger TTL value than the previous one. This process continues until the destination node is found or the originator gives up. On the other hand, in node degree based strategy, the source node selects a small subset of its neighboring nodes that are responsible to forward the request message. Similarly, each node within that subset selects another subset of forwarding nodes from its neighbors and continues relaying request message. This strategy also minimizes the flooding control overhead. However, computation of the optimal local subset is quite difficult in the case of WMNs.
This paper presents an empirical model of the above said routing strategies along with the routing protocols that employ these techniques. To evaluate these strategies, we select six widely used routing protocols, which are: Ad-hoc On-demand Distance Vector (AODV) [2], Destination Sequence Distance Vector (DSDV) [3], Dynamic Source Routing (DSR) [4], DYnamic MANET On-demand (DYMO) [5], Fish-eye Scope Routing (FSR) [6] and Optimized Link State Routing (OLSR) [7]. This work also contributes in terms of enhancing the efficiency of the selected routing protocols in the form of Normalized Routing Load (NRL) and reduced latencies. In addition, a mathematical model is presented to give critical analysis of the existing work. Further, the addition of NRL in existing schemes show significant improvements in terms of throughput and delay. It is worth mentioning that this study has been conducted by inspiring from [8]- [11] and [12]. The comparison of the performance of the selected routing protocols is done by simulating them in NS-2 environment. The efficiency is compared in terms of throughput, End-To-End Delay (E2ED) and (NRL).
Rest of the paper is arranged in the following manner. Section II provides insights about the previous work related to the topic. The empirical models of traditional flooding, routing protocols that employ this flooding strategy and their necessary enhancements are briefly discussed in section III. Similarly, the empirical models of TTL-based and degree based routing strategies along with the routing protocols that adapt these schemes are discussed in section IV and section V, respectively. Simulation results of the conventional routing protocol models with default parameters and the proposed empirical models with enhanced parameters are discussed in section VI. Finally, section VII concludes the paper.

II. RELATED WORK
Flooding provides multipath diversity and is considered a promising approach for routing. Authors in [13] presented a flooding based routing protocol (REALFLOW) for Industrial Wireless Sensor and Actuator Networks (IWSANs). The REALFLOW provides enhanced network stability and reliability when compared against four traditional flooding based protocols, which are: Normal flooding, Location flooding, Directional flooding and Single Gossiping with Directional Flooding (SGDF) in terms of packet delivery ratio within deadlines, consecutive packet loss and total packet transmissions.
Authors in [14], propose a protocol based on the selective-path priority table for the deployment of the mobile sink. Two short routes leading towards the cluster head or sink node are prioritised to make the priority table using the simple and effective rules. Rules are developed using the defined transmission and residual energy metrics. An energy-efficient routing protocol for prolonging network lifetime is presented in [15], which is referred to as Balancing Ant-based Routing Algorithm (EBAR). It uses the pseudo-random path searching algorithm along with the enhanced pheromone updating approach for balancing the energy consumption of the sensors. The results of this approach show that it has improved the lifetime and energy efficiency of the system as compared the previous schemes.
Radwan, A. A. et al. in [16] presented a detailed comparison of the major characteristics of AODV, FSR and Location-Aided Routing (LAR) protocols to analyze control overhead, throughput and E2ED using GloMoSim. They deduced from simulation results that FSR has low control overhead as compared to AODV and LAR, whereas AODV has high throughput in comparison with rest of the protocols. Moreover, LAR has low routing delay as compared to FSR and AODV protocols.
The work in [17] describes the multi-path routing scheme for underwater sensor networks. The purpose of this scheme is to monitor the nodes having the higher energy efficiency along with the packet delivery ratio, which is validated through the simulations. Simulation results also demonstrate that this scheme has improved the end to end delay, energy efficiency and packet delivery ratio as compared to the previous schemes. Various challenges, i.e., low throughput, short life cycle and other energy constraints of the nodes in a network are tackled by the K-means and fuzzy analytic hierarchy process [18]. Where, clustering is done by the K-means algorithm in the network and cluster head selection is done by the fuzzy analytic hierarchy process based on the considered factors in the system. This system was designed for improving the life cycle of the nodes while reducing the energy consumption.
In [19], route discovery using AODV routing protocol with directional antennas is performed. The proposed technique implements AODV routing protocol to transmit data from boarder node to the gateway node. When compared against existing CR-AODV (Cognitive Radio-AODV) routing protocol, the proposed technique performed better in utilizing the network energy consumption resulting in high data delivery.
In [20], performance of a reactive routing strategy is compared against Many-to-one Deliverability of Greedy Routing (MDGR). Link collision and congestion severely affect performance of greedy routing. In order to provide reliable characterization of data deliver-abilities, authors conducted a detailed analysis to measure the effects of link collision and network congestion in 2D wireless networks. Simulation results proved the correctness of the derived upper bound on critical transmission power.
A twofold problem is resolved in [21], where, traffic routing and security-based functions' activations are discussed in the multi-hop networking environment. Total energy of the nodes is reduced, and their security demand are also fulfilled. Authors in [22] discuss the movement of the mobile sink around the tracking region. It acts as an original source node in this region. There is an issue of flooding, when different nodes send the packets at the same time. So, this scheme addresses the issue related to the flooding and validated it to prove the system effectiveness. Its results show that flooding delay is minimized. In multi-hop networks, it is very difficult to compute the packet-level routing [23]. This information is very beneficial in the activities like managing the networks. Authors have addressed this path reconstruction issue using various algorithms and proved their effectiveness by performing simulations.
Authors in [12] proposed a detailed framework for reducing routing overhead. The control overhead for route maintenance operation of chosen protocols: plain flooding of DSDV, Multi-Point Relays (MPRs) flooding in OLSR and Fish-eye scope routing of FSR, is modeled according to appropriate computed probabilities. This further includes: change in MPRs, neighbor discovery error and number of link breakages.
Sinha et al. [24] proposed a dynamic multi hop broadcast routing protocol for emerging wireless networks. Tight upper and lower bounds on broadcast capacity are defined to improve the network throughput. The exponential growth of the spanning trees proved to be inefficient for the broadcast in topologies exhibiting the spanning trees. Therefore, this paper used adaptive broadcast techniques to improve the network throughput.
Kuang et al. [25] proposed a bandwidth based multihop WSN for real-time information sharing. When a node requests for the required content, there may exist multiple information provider nodes. So, in order to select an information provider, the distance and available path size is checked. An information provider with a wider path and the shortest distance is selected to send the required content to the requester.
Authors in [26] provided a comparative analysis of two routing protocols: RPL and thread. Form a detailed analysis of both protocols it is concluded that the thread routing protocol is better and it efficiently provides connectivity between low powered devices over a multihop WSN. However, in terms of scalability, the RPL routing protocol is better.
In paper [27], the authors focused on maximizing the throughput of the multicast traffic pattern in MWN using the Time Division Multiple Access (TDMA) approach. The authors generalized minimum frame length problem for multicast traffic from a single source to multiple destinations. This non-trivial characteristic of the proposed generalization mechanism makes it mathematically untrackable and computationally complex as compared to its unicast counterpart. In the proposed work, the concept of unicast compatible sets is extended to a multicast scenario using the Mixed Integer Programming (MIP) technique for an arbitrary network, where packets transmitted from each transmitter are received at different subsets of receivers. Authors used the linear relaxation of MIP, multicast tree optimization and compatible set generation to solve the optimization problem. Furthermore, different length packets, multiple modulation technique and usage of different coding schemes for delay minimization are also included in the proposed work.
Route discovery is an indispensable process for successful data sharing between leaf nodes and sink nodes. Therefore, broadcasting is widely used in literature for path finding. In [28], the authors proposed a customized proactive routing strategy to achieve energy efficiency in clustered Internet of Things Wireless Sensor Networks (IoT-WSNs). The simulation results of the proposed strategy proved that it outperformed existing strategies, in terms of maximized throughput and network lifetime.
In [29], the authors proposed a selective multipath routing protocol based on their first protocol named as Multipath-ChaMeLeon (M-CML) for ubiquitous network. The proposed protocol is named as M-CMLv2, in which authors considered three different Quality of Service (QoS) metrics and performed the network optimization analysis with statistical significance. To enhance the effectiveness of the M-CML, an intelligent algorithm is applied to reduce the improvident emission of packets. Different performance evaluation metrics like E2ED, NRL, packet delivery ratio, energy consumption and redundant packet suppression are compared for OLSR, AOMDV and M-CML using the NS-3 simulator. The simulation results showed that the M-CMLv2 reduced the redundant information and delivered packets successfully with acceptable E2ED. It also reduced NLR and the energy consumption of the nodes.
A cross technology communication routing protocol is proposed in [30]. This protocol uses Wi-Fi to transmit data over long distances and assists the ZigBee embedded devices. To establish a routing path, the Wi-Fi nodes send the bacon messages to the ZigBee embedded devices to awake them. The experimental results exhibit the effectiveness of VOLUME 8, 2020 the proposed routing protocol to establish communication between cross technology devices.
An optimized clustering and energy efficient routing protocol for mobile nodes in WSNs is proposed in [31]. A clustering algorithm is executed in a centralized manner and clusters of moving nodes are made. If a node is disconnected from a cluster it automatically connects with another nearby cluster head by calculating the distance. The objective of this routing protocol is to achieve the energy efficiency of nodes while sharing maximum data.
Authors in [32] provided a comparative analysis of ZigBee topologies for wireless lighting control system. These topologies are tested using two routing protocols: AODV and DSR. The comparison shows that end to end delay is minimum when DSR is implemented on start topology. However, it suffers from the problem of the center point of failure. On the other hand, mesh topology with the same routing protocol also emerged as a promising solution if the system is not time critical.
In [33] authors have proposed a mode selection mechanism for frequency division duplex nodes. These nodes are divided into two groups according to their frequency bands of data transmission and reception. In the proposed work, the nodes select their frequency band in such a way that each node can communicate with all other nodes, of the same network, present at its one hope. The bipartite graph coloring scheme is used for mode selection and simulation results depict the effectiveness of the proposed scheme.
Authors in [34] highlight the deficiency of an underlay network that it is designed without considering the importance of what a user of this network wants. Considering the importance of a user's preferences, authors propose a framework of a multihop network. They have also designed an interface, which transforms technical terms into non-technical terms so that an average user can easily understand it. The simulations depict that the proposed framework successfully achieves the objective of this work.
In [35], a lightweight countermeasure to cater the selective forwarding attached in WSNs is designed, analyzed and evaluated. A randomly deployed single checkpoint node is selected to detect the misbehaving forwarding node. Whereas, the proposed approach is used for both timeout and hop-by-hop based retransmission approaches, to recover the lost packets. The authors also considered the false detection and analyzed the false detection rate using the checkpoint node.
The authors in paper [36] proposed a computationally effective Receiver Based Flooding (RBF) algorithm for Mobile Adhoc Networks (MANETs). In the proposed work, the authors focused on reducing complexity and redundancy during transmission of packets. The computational complexity of the proposed RBF algorithm is O(nlogn) compared with the O(n 2 ) of the conventional RBF schemes. Further, the RBF scheme is extended to RBF-E1 and RBF-E2. The former shows good performance when the defer time is short; while, the latter reduces the complexity to O(n).
A dynamic topology resilient smart world routing protocol is proposed by Biswas, P. K. et al. [37]. The goal was to find an efficient route towards destination. It selects a data forwarder, which is closest to the final destination. Simulation results show minimum E2ED and packet loss ratio when compared against existing counterpart schemes.
In [38], the authors conducted a comprehensive survey about the integration of opportunistic routing and interflow network coding. Comparison of several joint routing protocols like Interflow NC with Opportunistic Routing (XCOR), Coding-Aware Opportunistic Routing (CAOR), Active Network Coding High-throughput Optimizing Routing (ANCHOR) etc. is provided along with their pros and cons.
Authors in [39] provided a comprehensive analysis of the stateless geographic routing protocol's performance in three dimensional Flying Ad-hoc Networks (FANETs). This routing strategy exploits the location information of nodes using greedy forwarding. Performance of the proposed work is compared against state-of-the-art in geographic routing.
During the past few decades the use of Unmanned Aerial Vehicles (UAVs) in MANETs has been emerged. However, the authors in [40] identified that the exiting protocols do not work well for UAVs. This article focuses on the route establishment for UAVs and identifies open research problems. According to the authors of this paper, the routing protocols for UAVs are categorized into two major categories i.e. Single hop routing and multi hop routing. Some of the proactive, reactive and hybrid protocols are also discussed under the heading of multi hop routing. Dijkstra's algorithm has been used by [41] to find optimal routing path in Adhoc networks. This work optimizes the transmit power and secure routes to derive the routing weights.
In [42], the authors proposed a bandwidth-satisfied and coding-aware multicast routing protocol (BCMRP). The scheme uses distributed hash table for reliable communication in partition based ad hoc networks. According to this study, mobility of the node and their limited transmission range can cause network partitioning in ad hoc networks. This partitioning may result in inaccessibility of certain nodes. This work intelligently avoids the problem of pre-partitioning.
Cooperative communication is appealing because of its flexibility in MANETs. A constructive relay based protocol is proposed by using the topological information and maintaining two types of tables (relay table and cooperative  table) [43]. Two types of routing problems: frequent link breakage and quick energy exhaustion are also taken into consideration.
Routing is a process of finding path(s) between source node(s) (S n ) and destination node(s) (D n ) to carry out communication. In this process, there may be n relay node(s) R n , while connecting S n with D n . Routing protocols are categorized based on the routing methods they employ, as well as the processes used to find and maintain paths in the network. In this work, we characterized routing protocols based on routing strategies they adopt to perform routing tasks; Traditional Flooding, TTL-based Expanding Ring Search (ERS) and SR and MPRs or node degree based routing technique, as shown in fig. 1. Before connecting the empirical model of the above said routing strategies along with their employing routing protocols, we make the generalized empirical model for Cost of Energy (CE) and Cost of Time (CT) parameters. The reason to model CE and CT first is that we will evaluate empirical models of routing strategies based on these two parameters. During evaluation, we use different search set values (k i ) to analyze their effect on CE and CT generated by the routing strategies. Following sections discuss the detailed model of the above said routing strategies; flooding-based routing, TTL-based routing, and degree-based routing for the chosen protocols.

III. FLOODING-BASED ROUTING STRATEGY
In flooding based routing strategy, route search which leads to D n , where n is the intermediate hop which is involved in delivering the data at the destination (from now on we use the term search instead of route search to D n ) is performed throughout the network, N . To find D n , S n initiates search based on search set value k i (the number of hops i.e., TTL), where i shows the broadcast region in which search is performed. Let, k N denotes the network diameter (predefined value of N be 255), and thus TTL value of k i be set up to N . In this case, whole network is flooded with route search control packets to find route(s) to D n .
In MANET, the nodes which are located in the range are capable to communicate directly with S n ; called adjacent or neighboring nodes. The total number of directly connected or neighboring nodes depicts the node degree (d). An isolated node has zero degree. Let, d avg be an average node degree of each node in the MANET, then the degree of connectivity for that node with its first hop neighbors, called expected forward degree, of that node is; d f . Here, nodes with higher degree d f forward received broadcast message with certain probability of success, P s [44]. In flooding based technique, S n initiates the search by broadcasting route request packet. This control packet is received by d and further disseminated by d f with P s in MANET until D n receives that packet.
The detailed steps of traditional flooding strategy are shown in algorithm 1 and well depicted in fig. 2. In fig. 2, S n is connected with 8 nodes, therefore, its d f = 8. For k 2 in fig. 2, each adjacent first hop neighbor of source node is connected with d of 4, therefore, d f for k 2 is 4 and so on. As TTL value of last k i is zero (as process of forwarding continues till k N −1 search set and stops at k N ). Therefore, d f of k N becomes zero. The routing control overhead of the above discussed traditional flooding in terms of CE, CE k N flooding , is computed using equation 1, taken from [12].
Algorithm 1 Traditional Flooding 1: begin 2: S n ← source node 3: msg ← requested message 4: TTL_VALUE of msg ← 255 hops 5: S n broadcast msg 6: R n ← receiver of msg 7: for all R n do 8: if already process msg then 9: do not broadcast msg 10: else 11: TTL_VALUE ← TTL_VALUE -1 12: if (TTL_VALUE=0) then 13: discard msg 14: else 15: R n broadcast msg 16: end if 17: end if 18: end for 19: end This flooding strategy has been adopted by different routing protocols e.g. DSDV, AODV, DSR and OLSR to perform route search in MANETs. However, in this work we have selected DSDV [3] to examine the behavior of flooding and section III-A is devoted to discuss it in detail.

A. ORIGINAL DSDV (DSDV-ORIG)
DSDV [3] is one of the proactive routing protocols that uses traditional flooding mechanism to perform search and maintain topological information in the Routing Table (  nodes present in the network. DSDV relies on the notification of beacon message from MAC layer to maintain connectivity information. This notification is based on the principle that if a node does not receive any beacon message from its neighbor for many successive beacon intervals, then the link is assumed to be broken. RU Per operation updates routing information across the network after a constant time interval and is known as full dump period. The changes of LS occur in an Active Route (AR) state and RU Tri are flooded throughout the network to disperse the updates across the network ( fig. 4(a)). Unlike RU Per , all link changes in a specified time interval are gathered by this function, before broadcasting the route updates in the network. The RU Tri operation may appear to be redundant along with the LSM Per operation. However, it has a certain benefit i.e., avoidance of routing loops. On the contrary, RU Per includes dispersion of destination sequence numbers throughout the network, which help in monitoring and maintaining liveliness of the routing information. Moreover, LSM Per operation is performed to eliminate the routing loops from the network using the latest sequence numbers.
To reduce routing overhead in flooding, DSDV uses Network Protocols Data Units (NPDUs) to transmit updates and carry distance vector information, as showon in Fig.3. RU Tri 's are transmitted using single NPDU after incremental dump period, which depends on the link breakage notification among ARs. Whereas, after full dump period, RU Per are transmitted through single or multiple NPDU(s) depending on the size of changes in RT entries from the last full dump period. A single InterScope_Interval is sufficient to broadcast the routing information throughout MANET, which is neither dense nor dynamic. Whereas, in a highly dynamic environment, RT information is broadcasted in multiple NPDUs for RU Per .
To achieve quality routing, a specific amount is incurred by the routing protocol (rp), denoted as: C (rp) total . This amount is in terms of energy consumption for transmitting routing control packets (CE) and time spent for routing (CT ) and is computed using equation 3 [12].
In case of DSDV, let CE (DSDV ) total be the total cost for DSDV on network layer. As, DSDV transmits RU Tri and RU Per on network layer; therefore, CE   . calculated using equation 4 [12].
Let τ NL denotes network life time and τ Per is the interval to generate RU Per (in the other words, full dump period), then CE (DSDV ) Per is estimated using equation 5.
The RU Tri updates are flooded when link breakages happen in AR that cause change in LS of AR and it costs C AR . The triggering cost of DSDV, C (DSDV ) Tri , is computed using 6.
If, at any instant, the value of |sgn(C AR )| becomes 1, then incremental update is flooded to notify this change.
Routing latency can increase end-to-end delay while delivering data to the destination. DSDV has the ability to maintain the data for a predefined time, termed as route settling time (τ rst ), which enables the selection of the best route. Holding time may also occur during the settling time of DSDV. Moreover, time is spent when updating the routing table with correct route(s) using RU Tri , after detecting the link breakages. It is denoted by τ TU . The total time cost or time spent by DSDV, CE (DSDV ) total , is determined using equation 7.
where, node and L denote the total number of nodes in AR and individual link delay in seconds, respectively.

B. ENHANCED DSDV (DSDV-MOD)
To reduce redundant Route Updates (RUs), τ rst and τ Per of DSDV-Orig have been modified in this work. We modified the values of τ rst and τ Per and changed them from 6s to 7s and 15s to 20s, respectively. RU Per intervals are mentioned in fig. 3(a). These modifications resulted in stabilized routes' advertisement and augmented the RU Per intervals, thus reducing the overhead of RUs. It is worth mentioning that the values of τ rst and τ Per are modified in such a way that there is rare trade-off in the performance parameters of DSDV-MOD.
That is none of the parameters is compromised to achieve any of the other parameters. On the other hand, suppose if we increase τ rst from 6s to 15s, the throughput is increased from 30 to 40% on the cost of delay, which is increased from 20 to 25%. The final selection of τ rst and τ Per is the result of in depth and comprehensive study of the protocol followed by the extensive simulations.

IV. TTL-BASED ROUTING APPROACHES
Instead of broadcasting control messages in entire network (as in traditional flooding), some techniques are used to limit k(i). ERS and SR work on the principle of setting TTL values for k(i), which limits the broadcast region.

A. ERS
The search set value, k i , in ERS is used to form broadcast levels (rings). Different k i 's are selected by ERS mechanism to perform search, whereas, each k i is associated with a specific waiting time. ERS strategy reduces the chances of flooding in the entire MANET by gradually increasing the broadcast rings during search process by setting different TTL values of k i . This results in limiting the search or Route Discovery (RD) coverage area to reduce energy cost, as depicted in fig. 3(b). The cost of ring i by setting k i , in terms of routing control packets per ring, CE ERS k (i) depends on the value of TTL. The average degree for individual k i in perspective of TTL in ERS, d k i avg , depends on d k i f . In fig. 2, d k i f is analogous to TTL values and waiting time in ERS. The average degree, d k i avg , is computed using equation 8, taken from [12].
However, it may be required to repeat these attempts by varying k i during ERS until D n is discovered. Let M is the maximum number of k i , then total CE consumed during ERS, CE total_k (M ) is given by equation 9.
where, k rrep denotes the search set (ring) that generates RREPs and is unicasted to S n . During RD, k i search varies with respect to particular protocol's strategy, k rrep denotes the ring that generates RREP (either Grat. RREP or Dest. RREP). The value of M varies according to success or failure of RD.
As k i for AODV, DSR and DYMO are different, therefore, CE of the respective protocol will also be different. In DSR, basic ERS mechanism is adopted, wherein every attempt after failure of previous attempt results in doubling of previous TTL along with waiting time in the next k i . The procedure of basic ERS is explained in algorithm 2. In AODV and DYMO, TTL_START value is assigned to TTL(k 1 ), and then is incremented by two for TTL(k 2 ) and TTL(k 3 ). When TTL reaches to its threshold value, TTL_THRESHOLD, TTL(k 4 ) is set to 35 in AODV and 10 in DYMO (according to TTL_NETDIAMETER [2], [5]), whereas, TTL(k 5 ) is set to 70 and 20 for AODV and DYMO, respectively (NET_TRAVERSAL in [2] and [5]). Algorithm 2 Basic ERS Mechanism 1: Preconditions: Mode ∈ {source_node(S n ), intermedi-ate_node (I n ), destnation_node (D n )} 2: while S n starts ERS do 3: set (TTL_VALUE ← 1 AND waiting_time ← τ ) 4: broadcasts RREQ (b_RREQ) 5: for all b_RREQ do 6: if (I n = D n ) then 7: D n generates RREP to S n , RDP successful, stop ERS 8: else 9: if no RREP receives to S AND τ expires then 10: if TTL_VALUE_TTL ≤ MAX_VALUE then 11: go to step 6 12: where, k llr represents the search set for Local Link Repair (LLR) activity, as shown in fig. 4(b). TTL value for k llr is calculated using LOCAL_ADD_TTL (value equal to 2) and MIN_REPAIR_TTL (the last known hop-count to the destination). It also depends upon the TTL value of k llr . In large networks, successful and accurate LLR calculation is required, because the chances of route re-discovery are reduced and more bandwidth is utilized. The TTL(k llr ) is calculated using equation 13.
where #hops is the number of hops to the sender of the currently undeliverable data packet. The local route repair attempts are often invisible to the originating node, and have TTL ≥ MIN_REPAIR_TTL + LOCAL_ADD_TTL.
The CT of reactive protocols depends on waiting time for k i . During RD, waiting time associated with each k(i) is different in AODV, DSR and DYMO. Let τ 1 is the time obtained as 2× NODE_TRAVERSAL_TIME, τ 2 is the value of BUFFER_TIME_OUT, and τ is the waiting time of DSR for NonPropagating RREQ [4]. Time cost of these routing protocols, CT (ERS) k−RD (k), is calculated using equation 14, taken from [12].
As k i based on TTL value for LLR is also performed during Route Maintenance (RM) process, therefore it has some time cost. Let CT CT AODV llr = τ 1 (TTL(k llr )) + τ 2 (15) In the next sub-section, we discuss the enhancements proposed for the selected reactive protocols to optimize routing overhead.

C. ERS IN ENHANCED AODV, DSR, AND DYMO (AODV-Mod, DSR-Mod, AND DYMO-Mod)
Routing latency as well as routing overhead solely depend on corresponding TTL values and waiting times [12]. To optimize the routing overhead, we set TTL(k 1 ) and TTL(k 2 ) to 3 and 6, respectively, in AODV-Mod, DSR-Mod and DYMO-Mod. Moreover, TTL(k 3 ) values in AODV-Mod and DYMO-Mod are taken to be 9, respectively. Whereas, TTL_NETDIAMETER and TTL_NETTRAVESAL of DYMO-Orig is changed to 35 and 70, respectively. On the other hand, TAP_CACHE_SIZE of DSR-Mod is adjusted to 256 and LOCAL_ADD_TTL of AODV-Mod is reduced to 1.

D. SR IN ORIGINAL FSR (FSR-ORIG)
In SR search strategy, different scopes based on TTL values, are selected to exchange routing information. To reduce overhead of routing updates, ''graded'' frequency technique is associated with each scope. The basic principle behind the working of SR is that entries corresponding to nodes within the smaller scopes, which have less TTL value, are propagated with highest frequency compared to the larger scopes. CE of any i th scope, CE S (i), with TTL value associated with TTL(S(i)) is calculated using equation 16, taken from [12].
where, d S avg (i) is the average degree of scope i. A protocol sets its scope levels, the overall energy cost of these scopes up to maximum number of scope levels M S in SR, VOLUME 8, 2020 FSR and Global State Routing (GSR) use SR strategy. We select FSR because its concept originates from GSR. In FSR, a moderate approach is taken as it uses only periodic updates (LSM Per and RU Per ), as shown in fig. 4(a), and uses two scope levels; IntraScope (IAS) and InterScope (IES) [6]. Energy cost for IAS is CE IAS with TTL value of 2 for k s (IAS) and for IES is CE IAS with TTL value of 255 for k s (IES). The periodic propagation in different scopes can be seen from fig. 2 and can be understood from algorithm 3. The energy cost for SR technique in FSR CE (FSR) SR is expressed using equation 18 where, τ IAS specifies the periodic interval for IAS and is called IntraScopeInterval [6], which is 5s and τ IES represents the periodic interval of IES and known as InterScopeInterval, which is by default 15s.  (20) To clearly understand the relationship between scope levels with their frequencies in SR, refer algorithm 3. The values used in this algorithm are taken from [6]. for all n 1 ∈ N 1 (S n ) do 8: if n 1 has link with N 2 (S n ) then 9: select n 1 as mpr 10: else 11: add n 1 as 1st-hop neighbor list 12: end if 13: for all mpr ∈ MPR(S n ) do 14: select mpr with highest degree 15: if (covers all N 2 (S n )) then 16: stop MPRs selection process In proactive protocols, routes must be updated quickly to maintain accuracy of routes in RT. As, FSR has no such mechanism to update routes after detecting link breakage, therefore, it needs periodic maintenance with less refresh intervals. In FSR-Orig, τ IAS is 5s and τ IES is 15s, which are too large intervals for mobile networks. Therefore, we set τ IAS to 2s and τ IAS to 5s in FSR-Mod to increase the frequency of topological information exchange.

V. DEGREE-BASED ROUTING STRATEGY
Degree-based strategy selects multi-point relays in the network based on node degree and reduces the routing control overhead in the network by minimizing duplicate retransmissions in the same region. In this mechanism, each node in the network selects a set of nodes in its 1 st -hop neighborhood, known as Multi Point Routing set (MPR mechanism is shown in fig. 2), that provides connectivity to the 2 nd -hop neighbors. Detailed functionality of MPR mechanism is depicted in algorithm 4. The energy cost CE of MPRs, CE k N mpr , is computed using equation 21.
Here, we define the MPR forwarding degree d mpr f as described in [12]: if x denotes a network node, then N (x) gives the set of 1 st -hop neighbors, whereas, N 2 (x) gives its 2 nd -hop neighbors. Similarly, N (x) lies in the neighboring region of x, such that x ∈ N (x). If y is the 1 st -hop neighbor of x, then it means that y is directly linked with x or in other words, they both are immediate neighbors. Furthermore, if both S and T denote collection of nodes, then x covers T iff T is either a subset or a superset of S. A set of Hence, MPR forwarding degree, d mpr f , is calculated using equation 22.
In fig. 2 (24) where, τ NS is the network start time and τ TC−def is TC default time period of OLSR.

B. DEGREE-BASED ROUTING IN ENHANCED OLSR (OLSR-Mod)
In enhanced version of OLSR, we enhanced the RU LSM and RU Per intervals. In OLSR-Mod, τ HELLO is set to 1s and τ TC is set to 3s, as RU Per interval for OLSR-Mod and OLSR-Orig are shown in fig. 3(a).
In the upcoming section, the simulation results of the original protocols and their enhanced versions are discussed. Thee different parameters are used for the performance evaluation of these protocols; throughput, E2ED and NRL.

VI. SIMULATION RESULTS
The performance of the modeled framework of original and enhanced routing protocols AODV, DSDV, DSR, DYMO, FSR and OLSR is measured through simulations in NS-2.
The values of different simulation parameters and results of simulations for reactive protocols are taken from [12]. The Constant Bit Rate (CBR) traffic sources having packet size of 512 bytes and Random Waypoint mobility model are considered in the simulations. Nodes are dispersed in square network area of 1000m × 1000m having 2Mbps bandwidth links to carry out communication. Three performance parameters: throughput, E2ED and NRL are chosen for the performance evaluation of the selected protocols.
It is worth mentioning that all of the selected protocols are enhanced in such a way that there is rare trade-off in the performance parameters of the modified protocols. That is none of the parameters is compromised to achieve any of the other parameters.
Also, the enhancements made in the protocols do not have any impact their overall complexity. Moreover, these modifications are not just the changes in the configuration parameters. Not only all of the six routing protocols have been studied in depth and compared but also after hundreds of experiments these configurations have been finalized.
For mobility analysis, two different cases are considered, which are: • 50 nodes having pause times ranging from 0s to 900s are simulated at a constant speed of 30m/s and • nodes having different densities, ranging from emph10 to 100, are simulated at a constant speed of 15m/s having a uniform pause time of 2s.
Moreover, different data flows with varying data rate (packets/s) e.g. 2,4,8,16 and 32 packets/s, are considered to examine efficiency of the protocols.

A. THROUGHPUT
It is defined as the amount of data exchanged successfully among source-destination pairs during specified simulation time (network life time). A protocol needs quick convergence to operate efficiently during high mobilities.

1) GENERAL BEHAVIOR OF REACTIVE PROTOCOLS REGARDING THROUGHPUT
As discussed earlier, different k i 's affect the efficiency of ERS in AODV, DSR and DYMO protocols appreciably. In AODV, k i s are reduced due to timely-based route checking in RT to make available correct route information. The routes are shortened and routing delays are reduced, using Grat. RREPs as compared to DSR, which has no such strategy as shown in fig. 5(a) at 0s, 100s and 200s pause times. Grat. RREPs provide quick paths and increase throughput of AODV in high mobilities.
In DSR, due to large TAP_CACHE_SIZE of 1024 and longer time intervals for routes' storage, RC contains stale routes. Moreover, the value of RouteCacheTimeout is 300s in DSR. Incorrect RCing in high mobility scenarios generates noisy and erroneous route information in DSR. The reason is the absence of such mechanisms, which are able to delete stale routes; except RERR messages. In result, VOLUME 8, 2020 it increases drop rates. Similarly, quick and efficient route repairing is required; once, the link breakage is detected, for re-establishment of paths by maintaining the broken links. Therefore, LLR (refer fig. 4(b)) makes AODV more adaptable to high mobility rate as compared to the other reactive protocols because it diminishes the chances of route re-discoveries.
DSR exhibits highest throughput than AODV and DYMO in moderate and immobile scenarios (as depicted in fig. 5(b)), because of RCing during RD process and PS for route maintenance makes end-to-end path calculation quick by reducing k i . Different mechanisms are used in various protocols like Grat. RREPs are used in AODV and DSR, LLR is used in AODV and RCing and PSing are used in DSR. However, DYMO lacks in providing such a mechanism; therefore, there exists lack of optimization to limit k rrep . AODV outperforms among all selected protocols in case of dense network; network with large number of nodes (refer fig. 5(b)) or with high data traffic (refer fig. 5(c)). AODV exhibits one major feature: LLR, which increases its suitability in scenarios having high scalabilities and high traffic rates. It is due to the reduction in routing latency, as depicted in fig. 5(c). On the other hand, DSR performs better in low scalable network scenario by reducing k i through PSing and RCing, however, due to dissemination of source routing information in dense networks, more Grat. RREPs increase the CE. This results in increased drop rates. Same behavior of DSR is also noticed in fig. 5(c), the case of varying data flow rates.

2) REACTIVE THROUGHPUT : ORIGINAL AND ENHANCED BEHAVIOR
In case of varying dynamicity, having pause times ranging from 0s to 200s (with step size of 100s), AODV-Orig attains increased throughput than DSR-Orig, as shown in fig. 6. The reason is the storage of stale routes during RC in DSR-Orig. On the contrary, both AODV-Orig and AODV-Mod check for the RTs having valid time and refrain from the usage of invalid routes. Moreover, an efficient repair mechanism for route re-establishment, known as LLR, is used. Furthermore, DSR-Mod results in high throughput as compared to AODV-Orig, due to 25% decrease in the actual size of RC. This also helps DSR-Mod to eliminate outdated routes and to make the initial ring size k 1 of DSR-Mod (having TTL value equal to 3) discover the routes faster as compared to DSR-Orig ( fig. 6).
Similarly, reduction in RING_TRAVERSAL_TIME of 25ms makes AODV-Mod more efficient in high mobilities and the DYMO-Orig, among reactive protocols, achieves overall less throughput. In DYMO, Grat. RREPs, LLR, PSing, RCing, and dissemination of source route information collectively result in increased drop rate as compared to other protocols. Same as that of AODV-Mod, low RING_TRAVERSAL_TIME in DYMO-Mod increases its throughput because of quick convergence as compared to DYMO-Orig for varying mobilities, as shown in fig. 6.
For varying network sizes, AODV-Mod produces higher throughput, as TTL_START (TTL (k 1 )), TTL_INCREMENT (TTL (k 2 )) and TTL_THRESHOLD (TTL (k 3 )) are set to 3, 3, and 9. By quick search through large value of TTL (k i ) along with reduction of RING_TRAVERSAL_TIME, AODV becomes more scalable for high densities as well as network loads, as depicted in fig. 6. Same thing happens in DYMO-Mod unlike DYMO-Orig, where two more k i s: (TTL (k 5 ) = 30) and TTL (k 6 )=60) are introduced. The addition of two more k i s increases the scalability of the system. The increase in initial ring size TTL (k 1 )=3 increases the scalability of DSR-Mod and enables it to perform better in scenarios having 80, 90, and 100 nodes, respectively. Fig. 5(b) and fig. 6 illustrate this thing. Fig. 7(a) shows that DSDV has the highest throughput in a network with high mobilities as compared to other proactive protocols. This is due to generation of RU Tri based on C AR S and delay, that advertise the routes which are in state of transition and cause damping irregularities in the RTs. The number of rebroadcasts of the routes having same sequence numbers is lessened by slowing down the advertising process of un-stabilized routes, which further improves the accuracy. Moreover, it also results in increase of throughput for DSDV for different mobility rates. On the other hand, a gradual decrease in throughput is observed due to low converging rate of OLSR in high mobility scenarios. The reason is that increasing mobility increases C mpr S and causes flooding. Although RU Tri 's are generated by both OLSR and DSDV, however, τ LSM is high for OLSR as compared to LSM Per interval of DSDV because of MAC layer notification. Moreover, 2 × HELLO_LOSS parameter decides generation of RU Tri in OLSR, which is approximately 4s. In result, OLSR has low convergence behavior in high mobility scenarios. As RU Tri 's are absent in FSR, therefore, it is less convergent to mobilities as compared to OLSR and DSDV, as portrayed in fig. 7(a).

3) GENERAL BEHAVIOR OF PROACTIVE ROUTING PROTOCOLS REGARDING THROUGHPUT
Among proactive protocols, FSR produces highest throughput for varying traffic rates, as shown in fig. 7(c). OLSR, on other hand, is suitable for networks having variations in their sizes, as depicted in fig. 7(b). In medium and high data traffic, FSR outperforms other protocols because it introduces SRs with graded frequencies, which minimize the routing overhead and works better even when available bandwidth is low. Consequently, it achieves more throughput in case of high data traffic rate and reduces routing overhead. Although, DSDV uses NPDUs to reduce routing control overhead of flooding; however, RU Per as well as RU Tri degrade its performance. MPRs in OLSR provide more optimizations in larger scalabilites, as shown in fig. 7(b), because it reduces VOLUME 8, 2020

4) PROACTIVE THROUGHPUT : ORIGINAL AND MODIFIED BEHAVIOR
Shortening the update intervals, τ HELLO and τ TC messages, make OLSR-Mod more convergent as compared to OLSR-Orig, as shown in fig. 3(a) and fig. 6. Therefore, increase in τ rst along with MAC layer notification of DSDV as compared to τ HELLO of OLSR (in OLSR-Orig and OLSR-Mod) collectively result the highest throughput attained by DSDV-Mod among all proactive protocols. Similarly to OLSR-Mod, we reduce topological information exchange intervals in FSR-Mod by decreasing the value of τ IAS and τ IES , which results high throughput as compared to FSR-Orig, refer fig. 6, whereas absence of RU Tri makes it less efficient for high mobilities.
Simulation of different scalabilities traffic rates, 2s pause time, also introduces some mobility, resulting frequent updates in MPRs. Therefore, OLSR-Mod outperforms OLSR-Orig in terms of scalability as well as traffic rates as depicted in fig. 7(b), 7(c) and in fig. 6. Same as that of OLSR, by increasing frequencies in exchanging information, FSR-Mod gives better results in selected scalabilities as well as network flows as shown in fig. 7(a), 7(b), 7(c) and in fig. 6.
On the other hand, changes in DSDV-Mod result less number of RU Tri due to more stabilizations and thus reduce the bandwidth consumption cost of flooding exchange. Thus, DSDV-Mod produces less drop rate by enhancing the protocol efficiency as compared to DSDV-Orig ( fig. 7(a), 7(b), 7(c) and in fig. 6).

B. E2ED
E2ED is the total time a packet takes to reach D n from S n . In this section, we simulate and analyze the E2ED performance of the above said routing protocols for varying pause times, network sizes and data traffic rates.

1) GENERAL BEHAVIOR OF PROTOCOLS REGARDING E2ED
For varying mobilities as well as varying network sizes, DYMO attains the lowest routing latency as demonstrated in fig. 8(a) and 8(b) because it only uses ERS (CT ERS ) to find suitable routes and results in less delay. Initial assessment of the values of RC in DSR and RT in AODV prior to RD through ERS causes delay in processing of both protocols. Although, CT of DYMO and AODV are calculated similarly as in eq. (14), however, LLR of AODV along with TTL (k 5 ) and TTL (k 6 ) are greater in AODV as compared to DYMO which results in less routing latency for DYMO. At higher mobilities, DSR suffers with higher E2ED (refer fig. 8(a)) because it looks for the desired route in RC first. Upon failure of searching the desired route, RD is initiates. Since, LLR is not implemented in DSR; therefore, it exhibits less E2ED as compared to AODV. Moreover, in moderate and high mobility scenarios, route search failure occurs frequently in RC, which leads to increased delay in DSR. Higher E2EDs are observed by reactive protocols, when considering medium to high network loads, as depicted in fig. 8(c). Diverse effects are observed in scenarios having different data packet rates and node densities in AODV and DSR, when Grat. RREPs are used. However, this strategy is not used for DYMO, which leads to reduction in delay for small data traffic rates and increases the latency in case of high data rates. Fig. 8(c) illustrates this thing. On the other hand, for different scalability cases, absence of Grat. RREPs keeps the value of E2ED to its lowest in DYMO protocol, refer fig. 8(b). Similarly, PS and Grat. RREPs help in reducing the system halts in different traffic scenarios for DSR, ranging from moderate to high. Initial validation of the RC values is encouraged in place of initiating simple RD process based upon ERS calculation, because it increases the delay in DSR for larger networks as compared to DYMO, refer fig. 8(b). Owing to the presence of LLR process, AODV possesses the maximum value of E2ED in all scalability scenarios. However, a decrease in latency is observed due to the decision based upon the congestion of LLR; 16 packs/s and 32 packs/s compared to 8 packs/s, refer fig. 8(c).

2) REACTIVE E2ED: ORIGINAL AND MODIFIED BEHAVIOR
DYMO-Mod produces the high E2EDs as compared to DYMO-Orig for higher mobilities and network flows. It uses two additional rings k 5 and k 6 , to achieve efficiency in high scalabilities, while these two rings introduce routing delay in case of highly mobile network with large data traffic rate. The quick search due to decrease in RING_TRAVERSAL_TIME along with increase in first three k i 's of both AODV-Mod and DYMO-Mod, minimizes delay in contrast to AODV-Orig and DYMO-Orig (refer fig. 8(a),8(b),8(c) and fig. 9). Presence of stale routes enhances the delay through production of Grat. RREPs and increase in route re-discoveries in DSR-Orig. In DSR-Mod, the large TTL values in k i affects other ring sizes drastically by reducing the routing latency. Moreover, they also avoid stale routes in small cache size and reduce the route re-discovery delay. Overall, DSR-Mod has less E2ED compared to DSR-Orig, as shown in fig. 9. In low scalabilities, due to the increase in path lengths, the delay increases via LLR. The route latency is shortened by decrementing the LOCAL_ADD_TTL value during repairing process in AODV-Mod, as obvious from fig. 8(b).

3) GENERAL BEHAVIOR OF PROACTIVE ROUTING PROTOCOL REGARDING E2ED
DSDV produces the highest E2ED among all proactive protocols working in scenarios having moderate to low mobilities VOLUME 8, 2020 because of τ rst , as evident from fig. 10(a). Furthermore, there is an intentional delay in advertisement of the unstabilized routes, in order to achieve reduction in the number of rebroadcasts of possible route entries. In FSR, LSM Per information is collected from MAC layer notification, increased link breakages in high mobilities increase round trip time of data packets due to absence of RU Tri . Thus, routes are only re-established through RU Per after τ IAS and τ IES as shown in fig. 10(a).
OLSR has the lowest routing latency as compared to other protocols for varying mobilities and network size, because of the highest frequency of topological information exchange, as depicted in fig. 10(a) and 10(b). In high mobilities, OLSR results more delay as compared to low mobilities, as in OLSR, RU Tri 's are generated after confirming link breakages from HELLO messages through expiration of HELLO_LOSS (2) whereas, HELLO_INTERVAL is also in seconds to check the connectivity. Thus, delay in detecting link breakage introduces latency in updates of RT entries, therefore causes delay in rapidly changing topological environment ( fig. 10(a) at 0s,  100s, 200s and 300s pause times). In all scalabilities, shortest interval of RU tri along more optimizations through MPRs incurs low latency, as exchange of TC messages through MPRs only, reduces retransmission latencies. As we consider 2s of pause times, therefore C mpr s increases routing update interval due to stabilization of MPRs from RU Tri , thus causing more E2ED in high densities, as portrayed in fig. 10(b).// DSDV attains the lowest E2ED for the high data traffic rate due to quick convergence and delayed advertisement provides stabilized and long lived routes ( fig. 10(c)). The highest E2ED is noticed for FSR in fig. 10(c), as broken routes are updated only through next RU Per after τ IAS and τ IES , therefore stable routes are not available for long time, thus RTT of a single data packet is much higher for more network loads.

4) PROACTIVE E2ED: ORIGINAL AND MODIFIED BEHAVIOR
Decreasing the value of update intervals of both OLSR-Mod and FSR-Mod reduces overall E2ED in all selected scenarios. Whereas, stabilized routes of DSDV-Mod and increased value of τ rst decrease its delay in all network scenarios, as can be seen from fig. 10(a), 10( fig. 9).

C. NRL
It is the number of routing packets transmitted for successful reception of data packets at destination(s).

1) GENERAL BEHAVIOR OF REACTIVE PROTOCOLS REGARDING NRL
In reactive protocols, CE (in terms of control packets) depends on number of k i 's in RD during ERS and on the mechanisms used to reduce CE of route re-discoveries. ERS is used in other routing protocols as well like AODV, DSR and DYMO. However, different TTL values of k i s in RD affect CE of the respective protocols. The absence of Grat. RREPs in DYMO leads to high TTL_VALUE of k rrep . Therefore, it incurs maximum routing overhead as compared to other reactive protocols in all mobilities and scalabilities (refer fig. 11(a) and 11(b)). On the contrary, less number of routing packets are produced in DSR RCing and PSing, (as can be seen from fig. 11(a)). The reason is the reduction in TTL_VALUE of k rrep in promiscuous listening mode. Moreover, DSR relies on MAC layer notifications for LSM after establishment of ARs unlike DYMO and AODV. In high mobilities with low pause times, reactive protocols generate more control packets as compared to moderate and low mobilities with more pause times, because CE of RM process in high link breakage augments control packets and disseminates of more RERR messages. Increase in number of Grat. RREPs due to high populations also increase routing overhead. Scenarios exhibiting moderate to high population and traffic rates possess less routing load for DYMO (refer fig. 11(b) and 11(c)) as compared to DSR and AODV. AODV attains the maximum routing load due to LLR and generation of HELLO messages to ensure proper connectivity of AR on network layer.

2) REACTIVE NRL: ORIGINAL AND MODIFIED BEHAVIOR
In AODV-Mod, we increase TTL_VALUES of k 1 , k 2 and k 3 . This change effects the number of control packets because CE of any protocol depends on TTL_VALUE of k i in ERS as depicted in fig. 12. Quick search in initial levels of ERS reduces routing load in mobilities as well as in traffic rate scenarios. We have also changed k 1 size of DSR along with reduction in TAP_CACHE_SIZE. Both of these changes in DSR-Mod reduce CE due to reduction in Grat. RREPs generation. In all mobilities, moderate and high scalabilities, and in all traffic loads in fig. 12, DSR-Mod generates less control packet as compared to DSR-Orig. Also modification in initial rings of DYMO-Mod along with TTL_NET_DIAMETER and TTL_NET_TRAVERSAL values lowers CE by providing quick search and less number of k i s for those D n which are at more distance, as compared to routing overhead of DYMO-Orig. One common noticeable behavior of reactive protocols after modifications are made is increase in the  routing load in scenarios having less scalability, where destination nodes are in smaller diameters of S n s. As we have modified TTL_VALUEs for k 1 and k 2 and assign 3 and 6 values, respectively for all protocols, thus, these changes increase routing load, in fig. 11(b) and fig. 12 at 10, 20 and 30 nodes. Fig. 13(a), 13(b) and 13(c), show that OLSR produces highest routing load as compared to DSDV and FSR. OLSR's update intervals for TC messages is more than other selected proactive protocols (given in fig. 3(a)); therefore, it consumes more CE. HELLO messages are also generated on network layer for neighbor sensing and for MPRs' computations unlike FSR and DSDV (rely on MAC layer ACK). Moreover, RU Tri are absent in FSR; whereas, they are used in DSDV and OLSR. In OLSR, C mpr s causes RU Tri through TC messages, whereas, RU Tri are generated only when link among AR breaks therefore OLSR's routing load is less when compared with DSDV. The lowest NRL is produced by DSDV in all mobilities as well as in low and medium scalabilities, because, RU Tri are produced in less number for small number of ARs (as shown in fig. 13(a)). Moreover, RU of incremental and periodic dumps through NPDUs help in reducing the routing overhead. As compared to full dump period of DSDV, FSR generates more routing load as compared to DSDV, as portrayed in fig. 12.

4) PROACTIVE NRL: ORIGINAL AND MODIFIED BEHAVIOR
After increasing τ rst value, more stabilization is achieved in DSDV-Mod by lowering the generation of RU Tri and producing less NRL as compared to DSDV-Orig, as shown in fig. 12 and 13.
Although in OLSR-Mod, RU and LSM exchange intervals are lessen, however, in all scalabilities and in all traffic loads OLSR-Mod generate less number of control packets as compared to OLSR-Orig. The reason for such behavior of OLSR-Mod in fig. 12 is because more stability of MPRs is achieved due to frequent updates, thus, reducing node degree in high scalabilities, and decreasing RU Tri in high data traffic. On the other hand, OLSR-Mod produces more routing overhead in low pause times, as compared to OLSR-Orig and vice versa for high mobilities, as visualized in fig. 12. In low mobilities, more stabilization of MPRs is achieved with lowest frequency of updates and in high mobilities high frequency is required to update MPRs to avoid redundant TC messages.
As FSR does not generate RU Tri , therefore, increase in τ IAS and τ IES , increase frequency of RU Per (refer fig. 3(a)). Thus in FSR-Mod, these frequent updates generate more routing load in terms of number of control packets in all scenarios as compared to FSR-Orig, as depicted in fig. 12.

VII. CONCLUSION
In WMNs, indirect communication is needed to exchange information among hosts. Routing protocols play a major role in the cases when two hosts want to exchange packets and they cannot communicate directly. Different routing techniques are employed by routing protocols to find a suitable path between the nodes. These routing strategies are VOLUME 8, 2020 designed to reduce the bandwidth overhead to enable efficient routing and also to reduce the time required for convergence after topological changes. In this work, we mainly focus on analyzing different routing techniques in different scenarios regarding node mobilities, network densities and data traffic rates. We characterize routing techniques in three categories: 1) traditional flooding, 2) TTL-based routing and 3) node degree-based routing. To study these techniques in different protocols, we evaluate and compare flooding in DSDV for traditional flooding, route discovery of AODV, DSR and DYMO to examine TTL-based ERS, fish-eye SR of FSR to evaluate TTL-based SR and topology control transmission of OLSR to compare MPRs flooding with traditional flooding. A novel contribution of this work is the enhancement in search set values and intervals of routing algorithms to improve their efficiencies. A comparison of selected protocols with their default and enhanced routing algorithms is conducted in NS-2 with the help of three performance metrics: throughput, E2ED and NRL. The results depict that NRL helps in reducing delay and overall overhead during the data communication.