A Survey and Performance Evaluation of Ad-Hoc Multi-Hop ... Survey and... · street lighting applications. The suitability of the protocols - AODV, DSDV, OLSR and DSR, which represent

2015 International Siberian Conference on Control and Communications (SIBCON)

A Survey and Performance Evaluation of Ad-Hoc Multi-Hop Routing Protocols for Static Outdoor

Networks

D. Dugaev, S. Zinov, E. Siemens, V. Shuvalov Siberian State University of Telecommunication and Information Sciences

Novosibirsk, Russia Faculty of Electrical, Mechanical and Industrial Engineering

Anhalt University of Applied Sciences Koethen, Germany

[email protected] [email protected] [email protected] [email protected]

Abstract-This paper shows a survey of routing algorithms for

ad-hoc wireless networks, which can be used in intelligent

street lighting applications. The suitability of the protocols -

AODV, DSDV, OLSR and DSR, which represent two main

routing approaches, employed in wireless mobile ad-hoc

networks (MANET): proactive and reactive, has been

analyzed, based on an NS-3 simulation.

In section 1, the main goals of this research are stated. In

sections 2 and 3, the general routing agendas are described, as

well as a short explanation of the four protocols is given.

Further, we concentrate on analyzing the routing performance

and predicting the "weak points" of each protocol, where a

major performance degradation is assumed to be reached

under a given network conditions (topology, number of nodes,

mobility and nodes' density).

In sections 4 to 6, our assumptions are tested and confirmed

with the help of a network scenario modeling, implemented

inside NS-3 discrete-event network simulator, where the

network performance has been evaluated in a form of average

throughput, one-way delay, jitter and packet loss rate (PLR)

per UDP connection.

Finally, we conclude, that the DSR routing protocol has best

performance characteristics among the others and the

particular routing approach used in DSR (the source routing

with route caching [1], [2]) overcomes the rest in the test

network with the given parameters. Nevertheless, for the other

routing protocols (DSDV and OLSR), appropriate

modifications can be made for achieving performance

optimization.

Keywords-multi-hop routing protocols, AODV, DSDV,

OLSR, DSR, wireless ad-hoc networks, NS-3 simulation.

I. INTRODUCTION

A problem of efficient data routing in wireless ad-hoc and mobile networks remains an important topic among researches of network protocols and the industry. For the

moment, a huge amount of various algorithms and routing schemes have been developed in order to solve specific routing issues in specific wireless networks, such as various types of WSNs - Wireless Sensor Networks, as well as the algorithms which aim at solving generic data routing problems in mobile ad-hoc networks, which is the main focus of the given paper.

In a context of this article, four MANET routing protocols have been selected for further investigation and performance measurements by means of simulation in NS-3 discrete-event network simulator [3] - AODV (Ad-Hoc On demand Distance Vector) [4] , DSDV (DestinationSequenced Distance Vector) [5] , OLSR (Optimized LinkState Routing) [6] and DSR (Dynamic Source Routing) [I] .

The wireless ad-hoc networks are used in many application scenarios, but here the research will be more focused on outdoor applications, in particular, on intelligent lighting control networks, where the network delay and throughput factors are crucial. Therefore, as for performance parameters, a few metrics have been chosen - one way delay (OWD), jitter, throughput and packet loss rate (PLR). Moreover, those kinds of networks have a number of specific features, such as static topology, big coverage and, in our case, a wireless connectivity. Those features have been simulated in NS-3 in order to provide more accurate results.

Due to a huge variety of different ad-hoc routing implementations, a task of collecting information and researching every one of them becomes non-trivial. However, since the overwhelming majority of them are based on two main routing principles with a few variations, which will be explained below, we have shortened an amount of the protocols being investigated to just the aforementioned four - AODV, DSDV, OLSR and DSR. Those selected protocols represent main concepts of multihop routing in ad-hoc networks and are most popular and widely used in research as well as in various industrial applications.

978-1-4799-7103-9/15/$3l .00 ©2015 IEEE


Figure I. A classification of the chosen ad-hoc routing protocols

II. RELATED WORK

In the work of Josh Broch et al. [7] , a performance comparison of multi-hop wireless ad-hoc routing protocols has been made. In particular, four protocols were selected as well - DSDV, AODV, DSR and TORA [8] , and the simulation has been conducted in ns-2 network simulator [9] with an emphasis on implementation of accurate MAC layer and physical IEEE 802.11 models. Their research has revealed, that, in conditions of nodes' movement, DSR and AODV protocols show better performance in terms of packet losses (the ratio between a number of received and a number of sent packets) and the routing overhead - a number of service packets generated by the routing protocol. However, their work was more focused on investigation of the protocols' behavior under different network mobility load, and did not consider the effect of increase of data packet intensity, since they used constant bit rate (CBR) generators on the source nodes. Moreover, the results of their work do not show the delay/jitter values of the data packets, as well as overall data throughput generated by CBRs and affected by the routing protocols.

Richard Draves et al. [10] have conducted a set of experiments on a real test-bed consisting of 23 nodes placed inside an office building in a static topology. The main idea of their research was to find out the effectiveness of the routing metrics, alternative to minimum hop count - ETX (expected transmission count), per-hop RTT and per-hop packet-pair. For this purpose, a modified version of DSR has been used, called as LQSR (Link-Quality Source Routing). This study has shown, that, under static topology conditions, the ETX metric outperforms minimum hop count, however, in a case of mobile nodes, the minimum hop count remains optimal metric due to its better flexibility against quick topology changes. Unlike in our research, the TCP connections with pre-defined HTTP traffic have been used there to measure TCP throughput and the latency.

Although the last work does not contain a direct comparison of several multi-hop routing protocols, it gives a valuable advice to consider alternative routing metrics (especially ETX and SNR [10] ) as an additional factor which can influence the routing performance in certain conditions.

Authors of [11] compare DSR and AODV routing protocols using OPNET simulator [12] in networks with rapidly changing topology. DSR is shown to be not suitable

TABLE I. MAIN FEATURES OF THE FOUR TESTED PROTOCOLS Protocol AODV DSDV OLSR DSR Routing Reactive

Proactive Proactive Reactive

approach (on-demand) (on-demand) Routes are Route table

Route Route Route cache stored in table table

Multiple route No No No Yes

selection Route Hop count Hop count Hop count Hop count metric

Underlying Distance- Distance-Link-State

Source routing Vector Vector Routing type Support of asymmetric No No Yes Yes

links for such conditions, while AODV is. Paper [13] compares DSR, AODV, and DSDV performance in static WSN networks using NS-2 simulations. DSR showed best performance results: packet delivery rate, average delay, routing overhead, average energy per packet. Authors of [14] gives broad survey of performance of 6 routing protocols (DSDV, FSR, OLSR, AODV, DSR, DYMO) with respect to mobility of nodes in simulated (NS-2) MANET scenario.

Many works dedicated to routing protocols comparison uses scenarios with mobile nodes, which is quite common case for wireless networks. All of above mentioned works used simulation (NS-2, OPNET) environments for tests. Works that used mobile scenarios showed protocol performance in respect to mobility of nodes, while the works dedicated to stationary networks showed the protocols' performance in respect to network size. This paper gives an

analysis of performance of four routing protocols (OLSR, DSDV, AODV, DSR) in relation to traffic load in network for stationary sensor networks using NS-3 simulator.

III. MAIN MULTI-HOP ROUTING CONCEPTS

Considering a wide topic of wireless multi-hop networks, there are 2 main routing concepts which can be highlighted (Figure I): • reactive (on demand) scheme; • proactive scheme.

The majority of protocols based on the proactive scheme are often referred as, so called, "table-driven" class of routing schemes, whereas the concept of reactive or on demand methods can belong to the category of "non tabledriven" (e.g., DSR), as well as to "table-driven" methods (e.g., AODV).

The reactive protocols are based on the on-demand strategies, i.e. the network path is created only when the source node requests a data transmission to some destination node in the network. For this purpose, a route discovery procedure must be invoked each time a data entity has to be transmitted. When the data transmission has been finished, the established route becomes inactive again [15] .

In the proactive protocols, information about the routes from each node to all the possible destinations is gathered "on-the fly" during service message transmissions. In this case, each node has its own routing table containing


s

�� 5

.. � 2 ......

....... . ....... ...

1 3 ......... .... �

. ......... .

... ,.,. ..........

.......

4

............ � ... . �� ...

. ........... ... \

. .... �

, . .... 6 .. · .. :·�

Figure 2. AODV route discovery procedure. Figure 3. DSDV route update procedure

The source "S" broadcasts RRIo'Q to the destination "f)" through the Io'very node here broadcasts its forwarding table to the others. For simplicity intermediate nodes. The destination "f)" replies with RREP on a chosen here, only partial update messages are shown, which contain the infiJrmation shortest path {D, 6, 5, S}. about the routesfrom "S" to "f)".

information about the paths from it to all the other nodes in the network. Such information is continuously being updated by sending control packets, e.g. route discovery, throughout the network. This route update mechanism becomes beneficial when there are strict transmission delay limitations in the conditions of wireless ad-hoc networks due to their specific features, such as power consumption restrictions, dynamic topology, noisy wireless environment, etc. [15] .

IV. OVERVIEW OF THE CHOSEN ROUTING PROTOCOLS

In the following subsections, an overview of these four protocols will be made. In Table 1, a generalized collection of the protocols' features is shown.

AODV

The Ad-Hoc On demand Distance Vector algorithm is designed for performing routing in mobile ad-hoc networks (MANET) with multiple hops. The main objective of the algorithm is to establish a route from a node to any other node in the network, using special control messages. The AODV algorithm is capable of setting up unicast (one-toone), as well as multicast (one-to-many) routes, and of maintaining them in current up-to-date state.

The feature of the algorithm lies in the way how the routes are calculated. Namely, each node in the network does not have a "full" routing table containing the information about every path to every node, instead, only current mostrecent information is used. In other words, AODV algorithm do not need to know a global network map and, therefore, do not persistently flood the network with global control information. Instead, the appropriate routes are constructed only when a corresponding data transmission is required, i.e. reactively, also commonly known as on-demand routing mode.

In order to successfully perform a routing function, AODV algorithm introduces 3 types of control messages used for establishing and maintaining active routes: RREQ (Route Request), RREP (Route Reply) and RERR (Route Error). Each node, in tum, uses this control messages to

create and update its own routing table, which consists of the following fields: <Destination IP address><Next-Hop address><Destination Sequence Number (DSN»<Route lifetime>. It is important to note, that a node stores and maintains only up-to-date information about the routes, which is determined by the <Route lifetime> field - when it expires, the respective record is deleted.

A. Route discovery procedure

When a node has to send data to a destination, it will perform a simple logics: lookup its routing table, if the destination IP address is found there ...... send data to the next hop; if there is no such record ...... initiate a route discovery procedure.

Route discovery procedure consists of three main phases (Figure 2): • generate RREQ and broadcast it through the network; • on reception of RREQ each node, looks up its own

routing table. If there is no route to given destination or there is one, but its DSN is less than DSN in RREQ ...... broadcast RREQ further. If there is such route with fresh DSN (DSN is equal or more than DSN in RREQ) ...... go to next step;

• when the destination node or any neighbor with the appropriate actual route to it received RREQ, it sends RREP message back to the source using the path with minimal amount of hops. During this phase, all intermediate nodes which took part in RREP forwarding, update their routing tables;

• additionally, the RREP-ACK acknowledgement message can be sent back if the appropriate "active acknowledgement" option is active.

B. Route maintenance phase

Once a successful route to the destination is established, it shall be properly maintained considering the nodes' mobility and unreliable transmission medium. This is done by using the following scheme.


Figure 4. OLSR traffic control (TC) messages transmission

� .. 2 .... � ..... . . .....

Y..-..- 1 ·····.�.b <c,).�..- 3 ..

s� ;;��r�\ ......... 5

'( 5 \\� � . • • • •

.

�

� ... � "" �·······4 � .. 6 ·····��S ,S'l ,-D ,\) '

Figure 5. DSR route request broadcast

In OLSR, TC messages are used to update topology infimnation throughout A source node "5" broadcasts a route request (RREQ) towards a destination the network. Here, node "f)" sends TC through its chosen MPR (node 4), node "D". While traversing, the RREQ is being updated by intermediate which broadcasts it further. nodes, which add their route information.

First of all, movements of the nodes within the ad-hoc network affect only the nodes participating in construction of the routes, which means that the movement/disappearing of "passive" nodes (the ones currently not contributing to any active paths) will not trigger any further broadcast protocol actions. There are three general scenarios where the change in the active route can take place: a movement of the source node, movement of the intermediate node/nodes or the destination node's move. Depending on the node's role in the path, they will react differently on the route changing. If the source node moves from its current position and tears the path, it can reinitiate the route discovery procedure in order to establish a new route. If the destination or any intermediate node goes out of the path's range, a Route Error (RERR) message is sent to affected nodes along the path. The RERR message is generated by the node which is closest to the source (i.e. upstream from the break). When the rest of the nodes upstream get this message, they mark the corresponding route as invalidated and set the distance to destination value to infinity. Finally, when the source node receives RERR, it may reinitiate route discovery if it's still needed.

C. Summary

The AODV protocol is designed specifically for highly mobile wireless ad-hoc networks, where it shows good performance in terms of low control overhead and shortest path utilization. The protocol is purely on-demand, therefore, it does not flood the network with big amount of service packets every time a topology change has happened, which make a significant effect on overall network performance.

The main concern about AODV is that it is optimized to operate in highly mobile networks, unlike our static topology scenario, which will be described in the next section. To prove this statement, appropriate simulation tests have been conducted further.

DSDV

The Destination-Sequenced Distance Vector protocol is a successor of the traditional Distance Vector routing algorithm used in wired networks (for instance, by the RIP protocol [16] ), it is specially improved for operation in mobile ad-hoc networks. In particular, it can handle looproutes as well as count-to-infinity problem which are quite usual in MANETs.

The key attribute of DSDV is - it is a proactive routing protocol. It presumes that each node (routing device) in the ad-hoc network possesses entire network topology information, i.e. it has a complete routing table which contains all possible paths from any source to any destination. Each entry in such a routing table consists of the following fields: <Destination Node> <Next Hop> <Metric Value> <Destination SN>. The <Destination SN> field is used to identify the freshness of the specific route (for avoiding loops). A route exchange mechanism of DSDV is illustrated on Figure 3.

Since the DSDV is a purely proactive protocol, there is no route discovery procedure preceding nodes' communication as it takes place in on-demand routing algorithms, because each ad-hoc node is assumed to have a global and up-to-date network topology map at any given time. However, in order to maintain the freshness of the routing table, the DSDV has to use special Routing Management Procedure every time there is a change in MANET's topology, which might create additional delays.

When an ad-hoc node moves across the network, it causes a topology change, since it breaks old neighbor's links and creates new ones. When it happens, all nodes which have detected the link change (some of its neighbors got disappeared) update their routing tables (set hop count to infinity, increment SN) and generate a special service message containing either an entire updated routing table (so called, full updates) or only the table changes (so called, incremental updates). This message is sent to their corresponding direct neighbors, which, after having received


it, perform the same actions (update routing table, generate the message). In this way, the network updates are propagated throughout the entire network topology.

A. DSDV properties

Since DSDV is specially developed to operate in MANETs, it has some advantages against reactive protocols. In particular, a low route-request latency and better performance in networks with low to moderate mobility, as well as in network with a few nodes and many data sessions.

However, there are some crucial drawbacks in DSDV and in proactive routing in general. Firstly, they tend to generate high control overhead which can flood an entire network and, therefore, significantly decrease overall network performance (lower throughput, higher delays). Secondly, in case of big ad-hoc networks (100+ nodes) with burst (non-intensive) type of traffic, there is no sense to maintain huge routing tables especially considering usually limited computational resources of mobile ad-hoc nodes.

B. Summary

The DSDV protocol certainly has advantages against reactive routing protocols in low- and middle- sized networks with low or static mobility, and where it is important to send data packets as fast as possible (with delay optimization). However, due to general properties of proactive routing concept, the DSDV can hardly perform the routing function in big networks, especially considering unreliable and noisy transmission media.

In any case, an appropriate simulation has to be conducted in order to test the protocol specifically for our network scenario. It is possible, that some way could be found towards improving DSDV performance.

OLSR

Optimized Link-State Routing is a routing protocol developed for MANETs. It is an improvement of the classical link state protocol. As a link-state protocol, it is a proactive protocol that scans the network to discover its current topology and to build routing tables, even if there is no activity. Its feature, in comparison to simple link state protocols, is low amount control traffic due to a reduced number of nodes, which broadcast and forward control traffic. These nodes are called multipoint relays (MPRs). MPRs are chosen in a way that each node has at least one 1-hop neighbor MPR for each of 2-hops neighbors. The denser a network is, the less MPRs are required, hence, less control traffic is flooding the network.

A. Algorithm

Basic algorithm (without using multi interface nodes, and non-olsr networks gateways) is the following: 1. At the first stage nodes discover their I-hop and 2-hop

neighbor nodes. This discovery occurs through HELLO messages. Each node broadcasts its HELLO messages which contain information on current node and information about its I-hop-neighbors that are discovered so far.

2. When each node knows its I-hop and 2-hop neighbors, it can choose a set of its MPRs. This set is chosen from its set of I-hop neighbors, so each 2-hop neighbor have at least one I-hop neighbor that is MPR of this node.

3. After a node has selected its MPRs it notifies these MPRs via HELLO messages.

4. MPRs send TC - traffic control - messages that are being retransmitted by each MPR, so each TC message is flooded over the entire network. TC message contains information about all neighbors of a node that initially has sent that message. So in the end all nodes know neighbors of each other (Figure 4).

5. Having information about links in the network, each node calculates its routing table, e.g. using Dijkstra algorithm [17] .

6. At this point, the traffic can be routed through the network.

7. Each node periodically issues HELLO and TC messages, so new or lost links can be detected. If node does not hear HELLO-messages from another node early discovered as I-hop neighbor, link is declared as lost. Information about lost links propagated to other nodes via subsequent HELLO- and TC-messages, as described above.

B. Summary

OLSR is an improvement of the link state protocol that is intended to reduce control traffic overhead in the network, which is especially important in MANET. Given that, it is also a proactive protocol ensuring better results in dense networks with random traffic between different nodes. In such a case, control traffic overhead will be effectively minimized by using MPRs, and the latency will not be affected by discovery, since actual routes are constantly maintained.

DSR

Dynamic Source Routing is another reactive routing protocol where the routes between two nodes are discovered only at the moment when an actual data transmission is required. It is a non-table routing method where the source node provides the whole route that packet should traverse inside of packet and routing decisions on each intermediate node are made based on this data, instead of routing tables as it works in many other routing protocols (LSR, OLSR, AODV, DSDV). Thereby, this protocol requires additional service data overhead, which contains routing information in each packet. However, it does not send any control traffic when there is no data transmission.

A. Algorithm

When node A wants to send data to node B, the node A lookups its route cache for a route to node B. If node A doesn' t have a route to node B, node A starts route discovery procedure. For this, node A broadcasts a Route Request, which contains information about initiator, target, request identification and list of nodes that forwarded current request so far. Each node that receives this Request adds itself to route contained in packet and broadcasts it further. As node


Helper class

Routing IPv4 Devices (AODV, OLSR, stack (Ad Hoc WiFi) DSR, etc,) Node class Mobility model

Common module Simulator module

Core module

Figure 6, NS-3 simulator programming architecture

B, to which the packet is destined to, gets the Request with entire route from A to B, it lookups its route cache for route to A. If it finds the route, then it sends RouteReply packet to A via this route with route from A to B discovered by the request. If there is no backward route in cache, the node sends new RouteRequest (B,A), but also packs RouteReply into request, so node A will know the route to node B, and will not try to make a new request. Such algorithm allows asymmetric links in the network, since there are no constraints that force requests and replies to traverse along the same path. This mechanism is illustrated on Figure 5.

During data transmission, a whole route is included into the data packet and each node on that route is responsible for transmitting the packet to the next node. If a node fails to send packet further, it sends RouteError message to originator of the data packet. When node receives RouteError, it marks route as invalid in its cache, and uses another route from the cache. If there is no suitable route in the cache, the node initiates route discovery as it described above.

For increasing efficiency, overhearing can be used. That means, that each node in the network inspects all packets it sees, regardless to whom these packets are sent and uses this information to fill its route cache.

B. Summary

One significant advantage of DSR is the possibility to route traffic relying on asymmetric links. Also, the absence of control traffic when there is no activity in network is beneficiary for networks with low traffic loads. The downside would be possible latencies for new connections, which would lead to low network performance in scenarios where network activity is random.

IV. NS-3 SIMULATION SCENARIO

In order to evaluate network performance under different MANET routing protocols, and since the software implementation of some of them is outdated and, it is hard to deploy a real test-bed with comparable network conditions. So, the NS-3 simulation tool [3] has been chosen to assess the effectiveness of the protocols in the investigated scenario. The NS-3 is a discrete-event network simulator and represents a set of different library components, see Figure 6 - core, simulation, node libraries, physical and channel models, network routing protocols implementations, written in C++, which together comprises a flexible simulation

On-Off application

UDP socket

AODV,DSD�DSR,OLSR implementations

IPv4 forwarding

Wi-Fi ad-hoc channel

Physical layer

Figure 7. The structure of the layers in NS-3 simulation script

environment, convenient for various protocol tests under defined network scenarios. For our purposes, we have used the latest release of NS-3 (version 3.20) with open source implementations of all four routing protocols which are being tested (AODV, DSDV, OLSR and DSR).

Since the aforementioned protocols have already been implemented in NS-3, our task in the simulation part of this research work was focused on creating a wireless ad-hoc network topology, choosing a mobility model, signal propagation and interference models in the wireless medium as well as defining network traffic parameters, such as traffic intensity, transport protocol, randomization of source/destination pairs, etc.

For our simulation purposes, a static grid network topology with a fixed size (100 nodes) has been chosen. The distance between the nodes in the grid has been set to 100 meters in all experiments. For the physical and channel layer simulation, we used Y ans Wifi [18] physical interference model to simulate signal's interference and the NS-3 AdhocWifiMac [18] implementation which determined the employed modulation scheme, bit rate and MAC-layer frame format. The entire layered structure, used in the simulation, is illustrated on.

As for a chosen L3 implementation, a traditional static IPv4 addressing scheme has been used along with one of the four ad-hoc routing protocol implementation. Further, in order to simulate the packet traffic inside a network, we used NS-3 OnOfApplication module [18] , which was modified to generate and to send the so-called On-Off packet traffic, the properties of which will be described below.

The On-Off traffic profile represents one of the numerous network traffic models, used in simulation scenarios with the main concept of "ON" and "OFF" time intervals, within which the traffic source is either active (sending packets with predefined format and data rate) or inactive (not sending any packets). This model is commonly used in the simulation for capturing network traffic scaling behaviors [19] . Those ON/OFF intervals can be deterministic or random, depending on the simulation needs. In our experiments, the ON-OFF traffic profile has been used with corresponding ON and OFF intervals being randomly and uniformly distributed in the following ranges: [1, 3] seconds for ON


Figure 8. An example of the wireless network topology. used on the simulation. with 3 simultaneous ON-OFF UDP flows

time and [3, 5] seconds for OFF time. The sending rate was set to 512 kbps in all experiments.

For a network with the fixed size of 100 nodes (in a shape of 1Ox1O grid), a number of simultaneous UDP ON/OFF connections has been created in a range from 10 to 100 connections with a step of 10, where each connection corresponds to a unique source-destination pair of nodes, randomly chosen as well, with a fixed total duration of 20 seconds. Each simulation experiment with the given network size (100 nodes) and the number of simultaneous UDP connections has been iterated 103 times. These results then have been processed and evaluated in the next section.

V. EVALUATED NETWORK PARAMETERS

Before explaining the evaluation of network parameters which will be used for routing protocols' performance comparison, a summary of the simulation test-bed with highlighting its main features is given: • each node represents a Wi-Fi device in the ad-hoc mode

with specified parameters (modulation scheme, error model, interference model, signal propagation model);

• nodes form a static square grid topology with a fixed link distance of 100 meters;

• network size is constant and equals to 100 nodes mapped as 1Ox1O static grid;

• IPv4 static addressing scheme has been used; • ON/OFF UDP connections have been generated in order

to create required traffic intensity inside the ad-hoc network.

In order to conduct the routing protocol comparison, some metrics had to be chosen. In this paper, there are 5 such metrics that have been calculated for each simulation scenario, representing an average value per 1 connection (UDP flow): • average throughput; • average one-way delay; • average jitter; • average packet loss rate (PLR); • average hop count.

Throughput per connection can be described as an amount of received information during a transmission

Figure 9. Average throughput per connection vs number of simultaneous UDPtlows

interval and represents an important performance metric which is basically showing the upper bound of possible connection capacity that the network with the given L2, L3 and L4 protocols is able to provide. The routing algorithm here plays an essential role since the selection of a set of links (i.e., route or path) towards the destination should be optimal, like it is done in all four protocols that are being tested, where a number of hops are chosen as a minimization parameter.

In the simulation script, the average throughput per connection was calculated as follows:

Throughput = Nrec.bits avg T .. transmlsswn (I)

Where: Nrec. bits - number of received bits; Ttransmission - a total "ON" transmission time during a UDP

connection with ON-OFF traffic. One-way delay (OWD) metric shows a time difference

between the instants of receiving and sending the packet and plays an important role in Quality-of-Service (QoS) evaluation, especially in delay-sensitive applications. Since UDP datagrams are used for transmission, it becomes possible to evaluate "pure" one-way delay without the effect of various L4 retransmission mechanisms, used in the reliable transport protocols (such as TCP, UDT or SCTP). For calculating OWD, the following formula has been used:

Where:

OWD = TTOTAL avg

N rec. packets (2)

TTOTAL - a total time of received packets' delay in a UDP flow operation;

Nrec. packets - number of received packets. Average jitter gives a representation of delay variation

during a data transmission. This metric is important in terms of assessing a "delay stability" of a connection, which plays a key role in many delay-sensitive applications. The average jitter has been automatically calculated by means of NS-3 simulator module "FlowMonitor" [20] .


Figure 1 O. Average OWD per connection vs number of simultaneous UDP flows

Packet loss rate (PLR) shows a fraction of lost packets among the total transmitted packets per UDP flow. As it was mentioned above, the UDP transport protocol is non-reliable, which allows evaluating more accurate PLR values. The PLR metric has been calculated as:

Where:

N dropped packets • 100

Nsent packets

Ndroppedpackets - number of dropped packets; Nsent packets - total number of transmitted packets.

(3)

Average hop count represents a very important performance characteristic of the chosen routing protocol, since its value directly affects all three previous metrics (throughput, OWD and PLR) and shows how the protocol handles an increasing traffic intensity (number of UDP flows) by optimally distributing the routes along the network. The average hop count has been calculated as:

Where:

HOPcount avg

N LhopSi (4)

N

N - total number of generated connections in one simulation run;

hops i - number of hops in i-th connection.

VI. SIMULATION RESULTS

In this section, the results of conducted network simulations are presented. There are 5 plots, where each corresponds to the performance characteristics, which have been defined in a previous section - average throughput, one-way delay, jitter, packet loss rate and hop count per UDP connection. The x-axis of every plot represents a number of simultaneous connections, taking place in the simulation at that moment.

Figure 9 shows a behavior of average throughput per connection. It is clearly seen, that, as the overall traffic intensity grows, the throughput smoothly decreases because of an increase of inter-node interference, which causes more

Figure 11. Average jitter per connection vs number of simultaneous UDP flows

frequent packet losses. Moreover, the three protocols (AODV, DSDV and OLSR) show very similar results, which can be explained by the fact, that these protocols always forward data traffic along a single (chosen by the shortest hop count metric) path, and are not able to see and consider alternative paths, unlike DSR.

The DSR, instead, shows much better performance in terms of throughput, because, among all the four tested protocols, it is the only one that actually uses alternative paths to destination by caching multiple routes. This caching allows a node to use alternative routes without the need to send new route discovery messages every time a route has been broken, which, in fact, significantly decreases an overall one-way delay, see , lowers the probability of a packet loss and, therefore, increases the chances for a packet to reach the destination. Moreover, when DSR ". pcap" traces, generated by NS-3, have been investigated, it turned out that DSR were able to send packets via different routes to the same destination, which also increased the throughput and lowered the PLR. This route caching mechanism becomes especially important on high traffic loads, where the maximum number of connections was set to 100 - that is 100 uniformly distributed random source/destination pairs, and even in this case, the DSR outperformed the rest of the protocols with about 150 kbps.

Figure 10 also represents a relation between number of connections and the average delay per flow. In this, quite unexpectedly, DSDV shows the worst result, in spite of the fact that DSDV is a proactive protocol, which does not need a mechanism of constant route requests, since it already knows the whole network topology. However, this behavior can be explained by the fact, that the packet loss probability in the network is quite high, and, in order to restore the damaged route, DSDV relies on periodic topology updates, which can take up to a few seconds (5 seconds of PeriodicUpdatelnterval in our setup). OLSR uses the proactive scheme as well, but in this simulation scenario, it uses smaller update interval value (l second) than DSDV. AODV performs a little bit better than DSDV, since it does not depend on periodic updates, but, instead, uses a reactive


Figure 12. Average PLR per connection vs number of simultaneous UDP flows

routing scheme with dynamic route requests and route replies, which, due to unreliable transmission medium, tend to get lost as well, prolonging an overall packet transmission time. In terms of one-way delay, DSR outperforms the other three protocols again, due to aforementioned route caching mechanism, which decreases the delay, but, as a "side effect" (due to the use of alternative routes with different number of hops), creates an additional jitter which will be clearly seen on a . All in all, looking at the one-way delay plot, only DSR shows acceptable result - even at the maximum traffic load the OWD value is about 0.8 seconds, however, we should also consider the jitter, where DSR has the highest value - 0.425 seconds (Figure 11). In contrary, AODV, DSDV, and OSLR have 1 second OWD even at the lowest load (10 connections) and up to 6 seconds in the worst-case scenario (DSDV, 100 connections), which is unusable in many networking applications, especially the ones operating in the real time conditions.

In Figure 11, a behavior of average jitter per connection is shown. Here, DSR shows the best result (42 milliseconds under 10 flows), until it reaches 50 connections. After that, DSR performs worse, showing a 425 millisecond jitter under 100 connections. The other three protocols behave very similar to each other and maintain a relatively stable jitter up to 200 milliseconds (DSDV, 100 connections). Such different behavior of DSR comparing to the rest of the protocols, as it was mentioned above, can be explained by the route caching mechanism, which allows DSR to use alternative routes for the same connection, without relying on a route discovery procedure. This causes an additional jitter to appear, since different routes have different delay values, so the jitter between the two packets, which have reached the same destination along different paths, increases. The similarity in the behavior of AODV, DSDV and OLSR is explained by the fact that these protocols always try to use the same, shortest path route, which maintains more or less stable delay variation.

Figure 12 shows the relation between packet loss rate (PLR) and the amount of connections. AODV here shows the worst performance where PLR ranges from 28 to 45 percent, which makes this protocol hardly usable in the

Figure 13. Average hop count per connection vs number of simultaneous UDP flows

chosen simulation scenario - square grid static topology, wireless medium and UDP random ON-OFF packet traffic. DSDV and OLSR, which both are proactive protocols, perform better, showing l 3% packet loss in the worst case. DSR has no competitors here with PLR equals to 0.04% and 0.87% for 10 and 100 simultaneous UDP flows, correspondingly. For such network topology and traffic conditions, the result shown by DSR is close to optimal. Advantage DSR over proactive protocols(OLSR, DSDV) in this metric can be explained by the fact that, although proactive protocol sense the link faults all the time, and can fix the routes before transmission started, link fault sensing requires some time (e.g. relatively big HELLO messages timeout for OLSR), so ongoing connection fault link would cause losses during this time. DSR on the other hand would not find out about link breakage on known routes before transmission, but it will quickly detects route error, since each nodes tracks if retransmission were successful and informs source about broken routes.

The last plot, shown in Figure 13, depicts the dependency of the average hop count versus the connection count. It can be clearly seen, that under any tested protocol, except DSR, the average hop count value decreases as the number of connections grows. This behavior can be explained in the following way - as the traffic load increases, the probability of a connection loss grows as well, and, since the more "continuous" routes, with large amount of hops, have more chances to get broken, they gradually start disappearing under elevating traffic intensity, which is illustrated on the . This thesis especially holds true for AODV, DSDV and OLSR, because their packet loss rate (Figure 12) is rather significant. As for DSR, this behavior is not observable and, more or less, remains constant around 7.5 hops, since its PLR does not exceed 0.87% even under maximum load throughout the simulation.

VII. CONCLUSION

In this paper, a simulation analysis of four widely used wireless multi-hop routing protocols has been made with a survey describing main principles of their operation. It has been shown, that in the established simulation scenario -


with 10 by 10 static grid topology and ON-OFF random packet traffic - DSR routing protocol demonstrates the best results in a majority of performance characteristics (throughput, OWD, PLR and hop count). Under such conditions, OLSR also performs well, unlike DSDV and, especially, AODV, which has shown intolerable packet loss rate, up to 45%.

The main reason of DSR' s well and stable performance lies in its route caching mechanism, which allows DSR to immediately choose various routes to the same destination without the constant need in route discovery procedure. This fact becomes especially important in such unreliable transmission environment as wireless multi-hop network, where the packet loss probability is high and, therefore, a route gets often broken and the repeated route discovery procedure must be repeated often. The AODV case is a direct evidence of that, where the packet loss are mostly caused by an increasing amount of the route discovery procedures, while the data packets had remained queued and eventually got dropped due to timeouts.

One of the unexpected results is the DSDV highest average delay, in spite of the protocol' s proactive nature. This fact can be explained by high packet loss probability and relatively big DSDV route update interval to be waited in order to establish a new path to the destination. OLSR has the same problem, but handles it more effectively than DSDV.

All in all, a following conclusion can be made. Under the proposed network model, a classical on-demand (reactive) approach, represented by AODV, is inapplicable due to low tolerance to intensive route breakages. On the other hand, the proactive routing approach in a form of DSDV and OLSR can be employed with applications, where the delay metric is less important. Finally, the DSR protocol is able to effectively use resources of the network and has showed the overall result, close to the optimum. The only drawback of DSR, which has been observed in this research, is a higher jitter, which could have negative effect on some network applications.

VIII. FUTURE WORK

In a future research, most efforts will be focused on the in-depth analysis of the chosen protocols, especially on such characteristics as a percentage of service traffic in a network traffic profile, generated by UDP On-Off application, and throughputJdelaylPLR relation from a number of hops per connection.

Furthermore, taking into account main conclusions of this work, a development of a routing algorithm, based on reinforcement learning [21] methods and, in particular, on a Q-learning [22] approach will begin. The developed algorithm should consider the advantages and drawbacks of protocols from the two routing domain principles - reactive and proactive, as well as from source-routing approach with route caching, and perform better in a network scenario, which has been given in this paper - a static grid topology with On-Off UDP traffic.

REFERENCES

[1] David B. Johnson, David A, Maltz, and Josh Broch, DSR: The Dynamic Source Routing Protocol for Multi-Hop Wireless Ad Hoc Networks, in Ad Hoc Networking, edited by Charles E. Perkins, Chapter 5, pp. 139-172, Addison-Wesley, 2001.

[2] [RFC 4728] http://www6.ietforg/rfclrfc4728.txt

[3] NS-3 network simulator official website, www.nsnam.org

[4] c. Perkins, E. Belding-Royer, and S. Das, "Ad hoc On-Demand Distance Vector (AODV) Routing," IETF RFC 3561, Jul. 2002.

[5] Perkins Charles E., Bhagwat Pravin: Highly Dynamic DestinationSequenced Distance-Vector Routing (DSDV) for Mobile Computers, London England UK, SIGCOMM 94-8/94.

[6] T. Clausen and P. Jacquet. Optimized Link State Routing Protocol (OLSR). RFC 3626 (Experimental) , October 2003.

[7] J. Broch, D. A. Maltz, D. B. Johnson, y-c. Hu, and J.Jetcheva,"A performance comparison of multi-hop wireless ad hoc network routing protocols", Mobicom' 98, October 1998 , pages 85-97.

[8] Vincent D. Park and M. Scott Corson. Temporally-Ordered Routing Algorithm (TORA) version 1: Functional specification. IntemetDraft, draft-ietf-manet-tora-spec-OO.txt, November 1997. Work in progress.

[9] Kevin Fall and Kannan Varadhan, editors. ns notes and documentation. The VINT Project, UC Berkeley, LBL, USC/ISI, and Xerox PARC, November 1997. Available from http://wwwmash.cs.berkeley.edu/nsl

[10] R. Draves, et aI., "Comparison of Routing Metrics for Static MultiHop Wireless Networks," ACM SIGCOMM, Portland, OR, August 2004.

[11] Yinpeng Yu; Yuhuai Peng; Lei Guo; Xingwei Wang, "Performance evaluation for routing protocols in wireless mesh networks, " Educational and Information Technology (ICEIT) , 2010 International Conference on , vol.2, no., pp.V2-107,Y2-110, 17-19 Sept. 2010.

[12] Hasan, M.S.; Hongnian Yu; Griffiths, A. ; Yang, T.-C., "Simulation of Distributed Wireless Networked Control Systems over MANET using OPNET, " Networking, Sensing and Control, 2007 IEEE International Conference on , vol. , no. , pp.699,704, 15-17 April 2007.

[13] Ihbeel, A.A.S.; Sigiuk, H.I. ; Alhnsh, A.A. , "Simulation based evaluation of MANET routing protocols for static WSN," Innovative Computing Technology (INTECH) , 2012 Second International Conference on , vol. , no. , pp.63,68, 18-20 Sept. 2012.

[14] Javaid, N. ; Yousaf, M. ; Ahmad, A. ; Naveed, A. ; Djouani, K., "Evaluating impact of mobility on wireless routing protocols , " Wireless Technology and Applications (ISWTA), 20 I I IEEE Symposium on , vol. , no. , pp.84,89, 25-28 Sept. 2011.

[ I S] D. Dugaev, S. Zinov, E. Siemens A Survey of Multi-Hop Routing Schemes in Wireless Networks applied for the Smartlighting Scenario II V international science conference 'Technologies and equipment for information measurement", Tomsk, Russia, May 2014.

[16] [RFC 2453] G. Malkin, "RIP Version 2," RFC 2453, Nov. 1998. http://www.rfc-editor.org/rfclrfc2453.txt

[17] Skiena, S. "Dijkstra's Algorithm. " §6.1.1 in Implementing Discrete Mathematics : Combinatorics and Graph Theory with Mathematica. Reading, MA: Addison-Wesley, pp. 225-227, 1990.

[18] NS-3 doxygen documentation. [Online]. Available: www.nsnarn.org/doxygen-release/index.html

[19] Victor S. Frost and Benjamin Melamed, "Traffic Modeling for Telecommunications Networks" , IEEE Conununications, March 1994.

[20] G. Carneiro, P. Fortuna, and M. Ricardo, "Flowmonitor - a network monitoring framework for the network simulator 3," NSTOOLS , 2009.

[21] R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction. The MIT Press, March 1998.


[22] Chang, Y.-H., Ho, T. : Mobilized ad-hoc networks: A reinforcement learning approach. In: ICAC 2004: Proceedings of the First International Conference on Autonomic Computing, pp. 240-247. IEEE Computer Society, USA 2004.

Documents

A Survey and Performance Evaluation of Ad-Hoc Multi-Hop ... Survey and... · street lighting applications. The suitability of the protocols - AODV, DSDV, OLSR and DSR, which represent