Manet Study

8/2/2019 Manet Study

1/18

Copyright 2005 John Wiley & Sons, Ltd.

Study of MANET routing protocols by GloMoSim

simulator

By Ashwini K. Pandey and Hiroshi Fujinoki*,

This paper compares ad hoc on-demand distance vector (AODV), dynamicsource routing (DSR) and wireless routing protocol (WRP) for MANETs todistance vector protocol to better understand the major characteristics ofthe three routing protocols, using a parallel discrete event-drivensimulator, GloMoSim. MANET (mobile ad hoc network) is a multi-hopwireless network without a fixed infrastructure. Following are some ofour key findings: (1) AODV is most sensitive to changes in traffic load inthe messaging overhead for routing. The number of control packets

generated by AODV became 36 times larger when the traffic load wasincreased. For distance vector, WRP and DSR, their increase wasapproximately 1.3 times, 1.1 times and 7.6 times, respectively. (2) Twoadvantages common in the three MANET routing protocols compared toclassical distance vector protocol were identified to be scalability for nodemobility in end-to-end delay and scalability for node density inmessaging overhead. (3) WRP resulted in the shortest delay and highestpacket delivery rate, implying that WRP will be the best for real-timeapplications in the four protocols compared. WRP demonstrated the besttraffic scalability; control overhead will not increase much when trafficload increases. Copyright 2005 John Wiley & Sons, Ltd.

Ashwini K. Pandey received his Bachelor of Science from Utkal University, India, in 1994 and his Masters of Science in Physics and in Com-puter and Information Science (CIS) in 2001 and 2004 respectively, from Southern Illinois University Edwardsville (Edwardsville, IL). He cur-

rently works at Stereotaxis Inc., as an Engineering Physicist. His research interests include communication networks, computational physics,and materials physics.

Hiroshi Fujinoki received his Bachelor of Art in Economics from Meiji University, Tokyo, Japan, in 1991; his Masters of Science in ComputerScience from Illinois Institute of Technology, Chicago, IL, in 1995 and his Ph.D. in Computer Science from the University of South Florida,Tampa, FL, in 2001. He has been working for the Department of Computer Science at Southern Illinois University Edwardsville as anAssistant Professor since 2001. His research interests include routing, transport and network protocols, operating system support for networkapplications, server performance optimization and network security.

*Correspondence to: H. Fujinoki, Department of Computer Science, Engineering Building, EB 2067, Southern Illinois University Edwardsville,Edwardsville, IL 62026-1656, USA.E-mail: [email protected]

Introduction

With the growing popularity and fallingprices of mobile hand-held comput-ing and information exchange de-

vices, the need and capability of these devices arealso growing. This is creating new problems andchallenges. Some examples of recent and not sorecent wireless devices are cellular phones, per-

sonal digital assistants, tablet PCs and laptop PCs.All of these have the capability and need toexchange information over a wireless medium ina network. Currently, the wireless networks thatallow communication between mobile devices can

be classified into the following two categories:

1. Networks having a fixed infrastructure: anexample of such a network is a cellular phone

INTERNATIONAL JOURNAL OF NETWORK MANAGEMENTInt. J. Network Mgmt 2005; 15: 393410Published online in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/nem.579


2/18

network. Amobile cellular phone depends ona fixed infrastructure of base stations thatcover fixed areas. A mobile phone communi-cates with the nearest base station and the

base station in turn transmits the informationto another base station, wired network or

another mobile phone. When a mobile phoneis at an intersection of the coverage areas oftwo base stations, it is switched to the basestation with the stronger signal without any

break in communication and without the userbeing aware of it.

2. Networks that do not have a fixed infra-structure: this is an emerging but highlyuseful and promising type of network com-munication method. There are several situa-tions where such a network would beindispensable; mostly, in unplanned eventslike natural disasters and wars, but also in aplanned event. For example, a meeting of

business people scattered over a large areahaving no fixed infrastructure will be bestsupported by this kind of network. This typeof network can be described as a network ofmobile devices that is created or destroyed asneeded and hence it is named a mobile ad hocnetwork or MANET.

The distinguishing feature of MANET is the lackof any fixed infrastructure and any central con-

trolling authority. When there is no centralcontrolling authority, the devices comprising anetwork are all equal and in such a situation anydecision needed to maintain a network becomesdistributed. This creates the need for distributedrouting algorithms, resource allocation schemes,network entry and exit rules and network security.Moreover, as it is quite possible that a majority ofthe mobile devices in such a network will be hand-held devices, the need to conserve battery powerwill drive down the transmission power of theindividual devices. Consequently, communication

between two devices would often require relay byintermediate devices, which introduces the prob-lem of multi-hop routing.

The distinguishing feature of MANET isthe lack of any fixed infrastructure andany central controlling authority.

In wireless networks, physical links do not existand a single transmission of a packet will transfera packet to multiple nodes within the communi-cation range of a transmitting node at the sametime. We call this inherent broadcast of MANETslocal broadcast to distinguish it from global

broadcast. It is guaranteed that at least a copy of apacket will reach a destination node if every inter-mediate node, except the destination, repeats local

broadcast without any explicit routing, as long assuch a path exists. However, routing is still neededfor MANETs for the following reasons. If packetsare transmitted by global broadcasts, excess copiesof each packet will be transmitted in the networkand to the destination. Thus, global broadcastswill entail unnecessary transmissions of packets,which wastes battery power of intermediate nodesfor transmitting duplicated copies of packets at thesame time as wasting transmission bandwidth.

Several routing protocols have been proposedfor mobile ad hoc networks.14 These can be cate-gorized as proactive (also known as table-driven)protocols, reactive (known as source-initiated ordemand-driven) protocols or the hybrid of thereactive and proactive protocols. A categorizationof the prominent ad hoc routing protocols isshown in Figure 1.

Proactive ProtocolsIn proactive routing protocols, routing informa-

tion to reach all the other nodes in a network isalways maintained in the format of the routingtable at every node. When the network topologychanges (i.e., existing nodes have moved, somenew links have been created or existing ones aredropped), such changes in link states are an-nounced to all the nodes in a network. Thus, routesto all possible destinations are discovered inadvance of packet transmissions.

If a proactive protocol is used for MANETs,an immediate problem is that rapid changes innetwork topology might overwhelm the networkwith control messages (messages for updating therouting table at every node) and the excess mes-saging overhead will compromise the throughputof actual data transmissions. Examples of proac-tive protocols are DV (distance vector) protocol,5

DSDV (destination sequenced distance vector)protocol,6 WRP (wireless routing protocol),7 and

394 A.K. PANDEY AND H. FUJINOKI

Copyright 2005 John Wiley & Sons, Ltd. Int. J. Network Mgmt 2005; 15: 393410


3/18

FSR (fisheye state routing) protocol.8 The fourprotocols are also called table-driven protocolssince the routing table will be updated for eachchange in link states in a network and routes arediscovered using information stored in routingtables.

Reactive Protocols

As its name suggests, this type of protocol dis-covers a route only when actual data transmissiontakes place. When a node wants to send informa-tion to another node in a network, a source nodeinitiates a route discovery process. Once a routeis discovered, it is maintained in the temporarycache at a source node unless it expires or someevent occurs (e.g., a link failure) that requires

another route discovery to start over again. Reac-tive protocols require less routing information ateach node compared to proactive protocols, asthere is no need to obtain and maintain the routinginformation for all the nodes in a network.Another advantage in reactive protocols is thatintermediate nodes do not have to make routingdecisions.

An obvious disadvantage in reactive protocolsis delay due to route discovery, called route acqui-

sition delay. Furthermore, if routing informationchanges frequently, as is the case in MANETs, andif route discoveries are needed for those changedroutes, reactive protocols may result in a largevolume of messaging overhead, since route recov-eries require global broadcasts. Currently popular

reactive protocols are DSR (dynamic sourcerouting) protocol,3 AODV (ad hoc on demand dis-tance vector) protocol2 and ABR (associativity-

based routing) protocol.9

Hybrid (Combination of Proactiveand Reactive) Protocols

Because of the initial delay due to route discov-ery and high control overhead in reactive proto-cols, a pure reactive protocol may not be the best

solution for routing in MANETs. On the otherhand, a pure proactive protocol used for a largenetwork may not be feasible because of the needto keep a large routing table at all times. A proto-col that uses the best features of both reactive andproactive protocols may be a better solution forMANETs. An example of such an approach is theZRP (zone routing protocol),10 although it isnot the panacea for all the limitations of otherprotocols.

STUDY OF MANET ROUTING PROTOCOLS 395


DSDV FSRZRP

AODV DSR ABRWRP

Ad-hoc RoutingProtocols

Proactive

Protocols

Hybrid

Protocol

Reactive

Protocols

AODV: Ad hoc On Demand Distance VectorDSR: Dynamic Source Routing

ABR: Associativity Based Routing

DSDV: Destination Sequenced Distance VectorWRP: Wireless Routing Protocol

FSR: Fisheye State Routing

ZRP: Zone Routing Protocol

Figure 1. Categorization of ad hoc routing protocols


4/18

Aprotocol that uses the best features ofboth reactive and proactive protocolsmay be a better solution for MANETs.

Performance comparisons for ad hoc routingprotocols have been reported in the recent past.1116

A comparison of DSR and AODV with some otherprotocols showed that the performance of DSRwas superior to AODV when node mobility washigh, although DSR had higher routing overheadas compared to AODV.11 In a similar work byDas,12 it was observed that, for metrics like delayand throughput that have real-life applicationimplications, DSR performed better than AODV inconditions where the node density and/or nodemobility were low. According to Das, DSR alwaysgenerated fewer control messages for routing thanAODV. However, Das argued that AODV resultedin fewer control messages than DSR under hightraffic load and high node mobility.

We are not aware of any previously publishedwork that measures how much better WRP, DSRand AODV protocols are than a classical distancevector protocol in ad hoc networks. For example,how much better those MANET routing protocols

will be than a classical distance vector protocol, inwhat aspects they are better than a distance vectorprotocol and in what conditions those MANETrouting protocols will be better than a distancevector protocol, surprisingly, have not beenanswered. In this paper, we try to find answersfor such unanswered but significant questionsin order to understand the advantages of theMANET routing protocols. In addition to thosegoals, we try to understand the major propertiesin the existing MANET routing protocols.

The rest of this paper is organized as follows. In

the second section, the four routing protocols com-pared in this paper are described with regard totheir procedure and data structure in order toclarify their design motivations and characteris-tics. The third section describes our simulationexperiments; experimental modelling and experi-mental designs are described, followed by the keyobservations obtained from our experiments. Thefourth Section 4 presents conclusions of this paperand proposals for future work.

Four Routing ProtocolsCompared

In this section, the four existing routing proto-cols we compared are described in terms of theirimplementation, design motivations and the major

known characteristics, to provide a foundation forcomparisons.

Distance Vector (DV) Protocol

The DV protocol is a classic routing protocolwhose refined versions are used in the currentwired networks. It is a proactive protocol andis based on the concept of DV: every node ina network maintains a distance table (a one-dimensional array or vector, called a distancevector), where each entry in a distance tablecontains the shortest distance and the address ofthe next-hop router on the shortest path to everydestination in a network.

In the DV protocol, each node knows only thedistance to its directly connected neighbours at the

beginning. The DV initially contains only the dis-tance to the direct neighbours (the distance to allother nodes is initialized to be infinity). Everynode exchanges its DV with all its directly con-nected neighbours. After a node receives a DV

from a neighbour node, the node updates its ownDV to reflect the least-cost path to other nodes thatare not immediate neighbours. This processrepeats until there is no further update in the DVat all nodes in a network. When this process iscompleted, each node will have a DV that containsthe least-cost path to all the other nodes in thenetwork. When routing information changes atany node (e.g., link failures), a node sends its newDV to all of its immediate neighbours. The newDV will be propagated to all the other nodes in anetwork using the same procedure to propagate

the DV.The DV protocol has several advantages for

MANET wireless networks. First of all, the proto-col does not require a global broadcast, which isthe property most essential for a routing protocolfor large networks. Another advantage is the shortroute acquisition delay. Since this protocol isproactive, routing information for every destina-tion should be available in the routing table at eachnode. The lack of need for route discovery on




5/18

demand results in short route acquisition delay.The above two advantages also imply traffic loadscalability, since the messaging overhead of theprotocol will be constant irrespective of traffic loadas long as there is no change in link states in anetwork.

The major known disadvantages of the DV pro-tocol are as follows. The convergence time forpropagating routing information will be long,especially when the link cost is increased.17 Be-cause of the long convergence time, it is possiblethat another change in link states occurs while theinformation for the previous change in link costhas not been completely propagated to the entirenetwork. This could cause an erroneous routingdecision, the well-known counting-to-infinityproblem, which can result in temporary routingloops.

Another disadvantage of the DV protocol is thenon-availability of alternative paths. Since the pro-tocol uses a distributed approach, each node doesnot maintain complete information about linkstates in a network. Lack of complete knowledge ofthe link states for all links in a network means thateach node is unaware of alternative paths to reacha destination. The unavailability of the informationfor multiple alternative paths to reach a destination(if they exist) will make the process of finding analternative path during a sudden link failure a time-

consuming process, if not impossible.The third problem is the large routing table. Forad hoc networks as MANET, the contents of therouting tables will be short-lived. Maintaininglarge routing tables, while their contents dynami-cally change in a short time, will result in a high

but unnecessary maintenance overhead. Finally,the DV protocol assumes symmetric links (e.g.,costs of links are the same for the two directionson a link), which is not necessarily the case forwireless networks. This is because each transmit-ting host usually uses a different signal frequency

in wireless networks even when two hosts com-municate with each other. Because of these prob-lems, the DV protocol is seldom used in its originalform.

Wireless Routing Protocol (WRP)

WRP was proposed by Murthy.7 WRP is anextension of the DV protocol that eliminates the

possibility of routing loops. Nodes in a networkusing WRP maintain a set of four tables:

1. Link cost table. This table contains the cost ofthe link to each immediate neighbour nodeand information about the status of the link

to each immediate neighbour.2. Distance table. The distance table of a nodecontains a list of all the possible destinationnodes and their distances beyond the imme-diate neighbours.

3. Routing table. The routing table contains a listof paths to a destination via different neigh-

bours. If a valid path exists between a sourceand a destination node, its distance is re-corded in the routing table along withinformation about the next-hop node to reachthe destination node.

4. Message retransmission list (MRL). The MRL ofa node contains information about acknowl-edgement (ACK) messages from its neigh-

bours. If a neighbour does not reply with anACK to a hello message within a certain time,then this information is kept in its MRL andan update is sent only to the non-respondingneighbours.

WRP works by requiring each node to send anupdate message periodically. This update messagecould be new routing information or a simple

hello if the routing information has not changedfrom the previous update. After sending an updatemessage to its all neighbours, a node expects toreceive an ACK from all of them. If an ACKmessage does not come back from a particularneighbour, the node will record the non-responding neighbour in MRL and will sendanother update to the neighbour node later.

The nodes receiving the update messages lookat the new information in the update message andthen update their own routing table and link costtable by finding the best path to a destination. This

best-path information is then relayed to all theother nodes so that they can update their routingtables. WRP avoids routing loops by checking thestatus of all the direct links of a node with its directneighbours each time a node updates any of itsrouting information.




6/18

Dynamic Source Routing (DSR)Protocol3

DSR protocol, as its name implies, is a sourcerouting protocol: a complete sequence of interme-diate nodes from a source to a destination will be

determined at a source node and all packets trans-mitted by a source node to a destination follow thesame path. Every packet header contains the com-plete sequence of nodes to reach a destination.DSR protocol is a reactive protocol and its primarymotivations are: (1) to avoid periodic announce-ments of link states required in proactive proto-cols, by separating route discovery from routemaintenance; (2) to avoid long convergence timeof routing information; and (3) to eliminate a largerouting table for forwarding packets at intermedi-ate nodes. The routing table, in a sense that it is thedata structure to always hold routing informationto reach every possible destination in a network,is not used in the DSR protocol. In DSR, routes arediscovered on demand and a route cache is usedto hold routes that are currently in use.

As with most of the reactive protocols, DSR con-sists of two procedures: route discovery and routemaintenance.

Route discoveryEvery node in a networkmaintains a route cache that contains a list of hop-

by-hop routes to each destination node currentlyactive and its expiration time (i.e., the time afterwhich a route is considered stale and discarded).Before a source node starts transmitting data to adestination node, it first looks up its route cache tosee if a valid route to that destination exists. If sucha route exists, then the route discovery processends and the source starts transmitting data usingthe route found in its route cache. Otherwise, asource node broadcasts a route request packet(RRP) to find a route to reach the destination. This

broadcast is a global broadcast, which floods an

RRP in a network through all alternative paths toreach a destination. While an RRP is being broad-cast and propagated to the destination, it adds theaddress of every node it encounters to its list. If thesame address appears more than once in the list,an RRP drops itself to avoid a routing loop. Whenan RRP reaches the destination node, the destina-tion returns the learned path extracted from theRRP to the source node. For wireless networks thatconsist of asymmetric links, the destination node

can send that path information back to the sourcenode as a global broadcast, which allows the DSRto work for asymmetric links.

Route maintenanceRoute maintenance inDSR is a mechanism to inform network failures to

all nodes in a network. Its primary motivation isto expedite detection of network failures by explic-itly announcing them to every node in a networkusing global broadcasts. No matter if it is a link ornode failure, a node that is connected to the otherend of a failed link is responsible for detecting afailure in DSR. On detecting a network failure, thedetecting node broadcasts an error message, calledan error packet, to all the other nodes in a networkto inform the failure. When other nodes receive anerror packet, they will disable the paths that gothrough the failed link in their route cache.

An obvious advantage of DSR is that sourcenodes are aware of the existence of alternativepaths, which implies that recovery from a linkfailure will be easy and quick. Another advantageis that there is no chance of a routing loop (or it iseasy to detect and avoid one). Furthermore, nodesdo not have to maintain a routing table, which isan advantage especially for a large network wherenodes continue to move.

Being a reactive protocol also means that a route

is stored in the route cache only when one isneeded, which implies low maintenance overhead.Since most routes are short-lived and networktopology frequently changes in MANETs, on-demand routing will make more sense than proac-tive protocols in terms of maintenance overheadfor routing information at each node (this is

because a node does not have to modify anythingif a failed and/or changed link is not a part of anyactive path from this node).

The disadvantage in DSR is long route acquisi-tion delay due to route discovery if short trans-

mission delay is a significant factor. Long routeacquisition delay may not be acceptable in certainsituations, such as mobile communication in a bat-tlefield. It is also quite possible that the path be-tween a source and a destination may not be theshortest path (this is because resumed links willnot be explicitly advertised in DSR), resulting inpaths with suboptimal end-to-end delay. Anotherdisadvantage is that messaging overhead of theprotocol will be high during busy times, when




7/18

many connections must be established in a shorttime since broadcast is used in route discovery. Alarge packet header will also cause low payloadutilization, since each packet has to contain alist of all the intermediate routers to reach adestination.

Ad Hoc On Demand DistanceVector (AODV) Routing2

The AODV protocol is a reactive routing proto-col that provides a good compromise betweenreactive source routing protocols and proactiveprotocols. The trade-off problem AODV addressesis the one between high messaging overhead dueto periodic announcements of link states in proac-tive protocols and the large packet header neededto contain the entire route information to reach adestination in source routing protocols. Unlikepure DV protocols, routes are discovered andmaintained on demand in AODV. Unlike DSR,AODV uses a distributed approach, meaning thatsource nodes do not maintain a complete sequenceof intermediate nodes to reach a destination.Unlike DV and WRP, each path is established as apair of two streams of pointers chained between asource and a destination node (details of this aredescribed in a later section), which eliminates the

need for broadcasting error packets on a linkfailure. Like DSR, AODV uses route discovery androute reply mechanism to create and maintain aroute on demand.

Route discoveryWhen a source node wantsto send information to a destination node, it firstlooks up its own routing table to see if a valid routeexists. If a valid route does not exist, a source node

broadcasts a route request message that containsthe source address, source sequence number, des-tination address, destination sequence number,

broadcast ID and hop count. The combination ofthe source address and the broadcast ID is used touniquely identify each route request messagewhile a route request message is globally broad-cast. Any node that has a valid route to the desti-nation or the destination node is supposed torespond to route request messages by sending aroute reply message.

During a route discovery, two pointers are setup at every intermediate node between the source

and the destination nodes. The two pointers arethe back pointer and the forward pointer. A chainof the forward pointers is set up between a sourceand destination node while a route request mes-sage propagates from the source node to a desti-nation. Similarly, a chain of back pointers is set up

while a route reply message propagates back fromthe destination (or from a node that already has avalid route to the destination) to the source. As aresult, all the intermediate nodes on a routemaintain a pair of the forward pointer and the

back pointer for every connection that goesthrough them.

Every route request contains a list of intermedi-ate nodes that have been encountered. If the sameintermediate node appears more than once in thelist, the route request message will be dropped (thechain of forward pointers must be deleted for aroute request message to be deleted). This guar-antees loop-free routing.

Route maintenanceThe route maintenanceis performed using three different types of mes-sages: route-error message, hello message androute time-out message. The purpose of the time-out message is obvious: if there is no activity on aroute for a certain amount of time, the route point-ers at the intermediate nodes will time out and thelink will be deleted at the intermediate nodes. The

periodic hello messages between immediateneighbours are required to prevent the forwardand backward pointers from expiration. If one ofthe links in a route fails, a route-error message isgenerated by the node upstream (i.e., from anintermediate node to source nodes on the link andthe message is propagated to every source node inits upstream that uses the failed link. Thus, theerror packets will not be globally broadcast inAODV. Then, the source nodes in the upstreamwill initiate the route discovery process.

The primary advantages of the AODV protocolare as follows. Route caches are small in AODV,

because of its on-demand routing. Routes areguaranteed to be loop-free and valid. Convergencetime is short for propagating changes in link states

because link failure information will be propa-gated only to the nodes that are using a failed link(i.e., no broadcast for error packets). Informationof a link failure will be propagated following the

back pointers to reach such nodes. This implies




8/18

that messaging overhead to announce link failureswill be less than for DSR, where link failure infor-mation is broadcast. As another advantage, eachdata packet does not contain the complete list ofall the nodes on a route in AODV, which reducesthe size of message packet. Like DSR, a source

node is aware of multiple alternative paths.One of the disadvantages of the AODV protocol

is that nodes cannot perform routing (forwarding)packets as aggregate (at least in the latest existingimplementation of AODV). This is because a set ofpointers is used to maintain a route and each flowrequires its own pair of back and forward point-ers. For nodes where a large number of connec-tions exist, overhead for maintaining pairs of twopointers will be significant and may not be traffic-load scalable. Another disadvantage is longerroute acquisition delay compared to that for proac-tive protocols since route discovery must still takeplace on demand. Unlike DSR, AODV requiresperiodic hello messages to maintain pointers setup at every node on a path. Use of broadcastduring route discovery, which contributes to high

messaging overhead, is still the major overhead.Table 1 summarizes the discussions regarding thefour routing protocols in this section.

Simulation ExperimentsTo compare the performance of the four routing

protocols described in the previous sections, simu-lation experiments were performed. In thissection, experimental modelling, design and keyobservations from our simulation experiments aredescribed.

Experimental Modelling

To compare the four routing protocols, a paral-lel discrete event-driven simulator, GloMoSim,was used. GloMoSim (global mobile informationsystem simulator) is a simulation tool for largewireless and wired networks.18 Our simulationexperiments were executed on two desktop PCs:



Properties DV WRP DSR AODV

Type of routing Proactive Proactive Reactive Reactive

Distributed Yes (hop-by-hop) Yes (hop-by-hop) No (source Yes (hop-by-hop)routing)

Routing loops Possible Not possible Not possible Not possible

Use of broadcast No No Yes Yes

Control overhead Constant to the Constant to the Affected by the Affected by thenumber of sessions number of number of sessions number of

sessions sessions

Routing entries All destinations All destinations Destinations in use Destinations inuse

Alternative paths Not available Not available Available Available

Request response Short Short Long (if not Long (if notcached) cached)

Advantages Short response Short response Quick path Small routing

time time recovery tableLow message OH Quick recovery

Disadvantages Routing loops Large routing Long response Long responseLarge routing table table time timeLong convergence Large packet Aggregatetime header routing is not

possible atintermediatenodes

Table 1. Major properties of the four protocols compared


9/18

one with a Pentium 3 processor, 850 MHz with 512MB RAM; and the other with a Pentium 4 proces-sor, 3.2GHz with 2GB RAM.

To compare the four routing protocols, aparallel discrete event-driven simulator,GloMoSim, was used.

We focused on three performance measure-ments to compare the four routing protocols: meanend-to-end delay, packet delivery rate and routingoverhead as measured by the number of controlpackets generated for routing. The three measure-ments in our experiments were defined as follows:

1. End-to-end delay. The average time from thebeginning of a packet transmission (includ-ing route acquisition delay) at a source nodeuntil packet delivery to a destination.

2. Packet delivery rate. The ratio of the number ofuser packets successfully delivered to a des-tination to the total number of user packetstransmitted by source nodes.

3. Messaging overhead. The number of controlpackets generated for routing by each routingprotocol.

The control parameters we used in our simulationexperiments were traffic load, node density andnode mobility. Mean end-to-end delay, packetdelivery rate and routing overhead were thenmeasured for node mobility (Experiment 1) andnode density (Experiment 2) for three differentlevels of traffic load.

Traffic load generated by each source node wasmodelled by a constant bit rate data stream, whosetransmission rate was defined by packet transmis-sion interval for fixed-size packets. Three differentlevels of traffic load defined by the packet trans-

mission intervals are: (1) low traffic loadonepacket transmitted every 10s; (2) medium trafficloadone packet every second; and (3) high trafficload: one packet every 0.1s.

Movement of each node was modelled using therandom waypoint model.3 In this model, eachnode remains stationary for the duration of itspause time. At the end of a pause time, a nodestarts moving in a randomly selected direction inthe network terrain at a fixed speed. Once a node

reaches its new location, it remains stationaryduring its next pause time. At the end of the newpause time, a node again starts moving in anotherrandomly selected direction in the network. Thismovement process was continued during a simu-lation experiment.

The network terrain size was fixed for 2000 2000m. The radio signal transmission range wasfixed at 175m (radius of 175m). The transmission

bandwidth of each link was fixed at 2Mbps (= 2 106 bits per second) and the simulation time was500s for all the experiments. In every experiment,there were 25 randomly selected pairs of a senderand a receiver node. Traffic load was simulated insuch a way that each of the 25 sources transmitteda constant bit rate data stream to one of the ran-domly selected 25 receivers at approximately thesame time. Data packet size was fixed at 1460

bytes. The above parameters were used for all ofour simulation experiments.

To avoid the experimental artefact of having anetwork that was totally devoid of any traffic

before a simulation experiment began, a randompattern of FTP and HTTP network transmissionwas simulated for 10s. This involved sending datapackets of size 1460 bytes from randomly selectednodes to randomly selected destinations for arandom duration using either FTP or HTTP. Thissimulated a real network that was already in use

before a simulation experiment began.In our experiments, TCP was not used as thetransport-level protocol, because the main objec-tive in this project was to observe the net char-acteristics of each routing protocol. Since TCPperforms flow control and packet retransmission,it would have prevented us from observing essen-tial characteristics (such as the number of packetsactually transmitted and dropped) of the fourrouting protocols. To directly observe the majorcharacteristics of the routing protocols, we se-lected UDP for the transport layer in our experi-

ments. IEEE 802.11 was used for the MAC layerprotocol. For the radio layer protocol, no captureand free space were used for the propagationmode.

Experimental Design

In our simulation, the following two experi-ments, mobility tests (Experiment 1) and node




10/18

density tests (Experiment 2), were developed. Thetwo experiments are defined in the followingsections.

Experiment 1 (mobility tests)The effect ofnode mobility was studied. In this experiment,

there were 50 nodes moving using the randomwaypoint model described before. The speed ofnode movements was 45km/h, with the followingfive levels of node mobility: (1) perpetual mobil-itymove pause time 0s; (2) high mobilitymovepause time 120s; (3) medium mobilitymovepause time 300s; (4) low mobilitymove pausetime 400s; and (5) zero mobilitymove pausetime 500s.

Performance of the four routing protocols wascompared for end-to-end delay, packet deliveryrate and messaging overhead for various combi-nations of traffic load and node mobility.

In Experiment 1, 10 experimental runs wererepeated for each configuration, with a differentstarting random seed value for each run. Differentinitial seed values ensured that the networkstarted out with different node placements anddifferent starting directions for their movement.The results were averaged. The most extremeresults (highest and lowest values) werediscarded.

For the table-driven protocols (DV and WRP), it

was expected that messaging overhead would notchange significantly when traffic load wasincreased. This is because the number of routingupdates for these protocols is dependent only onchanges in link states in a network, which is inde-pendent of traffic load. Since WRP uses periodiccontrol packets to maintain routes, it was expectedthat WRP would have a higher messaging over-head than DV. For AODV, we expected that mes-saging overhead would increase with increasedtraffic load as AODV uses periodic messages tocheck the consistency of the network, which

would increase when the network traffic loadincreased. For DSR, since there is no periodic routemaintenance required, it was expected that thecontrol overhead in DSR would be less than thatof AODV.

Experiment 2 (node density tests)In thisexperiment, the effects of node density on meanend-to-end delay, the packet delivery rate androuting overhead were measured for the four

routing protocols. Effect of node density wasstudied while all the nodes in a network weremoving at a velocity of 45km/h with no pause(i.e., perpetual node mobility). The following threelevels of node density were defined: (1) low-density networks50 nodes in a network; (2)

medium-density networks75 nodes in a net-work; and (3) high-density networks100 nodesin a network. Similar to Experiment 1, 10 experi-mental runs were repeated for each experimentalconfiguration. The same randomization techniqueand statistics methods as Experiment 1 were used.Simulations were performed with 100, 75 and 50nodes, while 25 nodes were acting as senderswith another 25 randomly selected receivernodes.

For the two reactive routing protocols (DSR andAODV) where minimum-path routing is not guar-anteed due to lack of explicit announcements ofnewly added nodes and links, increase in nodedensity was expected to result in longer end-to-end delay than the proactive protocols. This is

because the network diameter will be increased ifthe number of nodes in a network is increasedwithout increase in the number of links. Increasein the network diameter implies increase in pathlengths. Thus, we expected that the end-to-enddelay in the two reactive protocols would be moreaffected by increase in the node density than the

proactive protocols.Another possible impact from the increase in thenumber of nodes is that messaging overhead forproactive protocols will increase, since the tablesize will increase to account for extra nodes. Forreactive protocols, increase in the number of nodesin a network will also increase the messaging over-head in global broadcast during route discovery,while this is not the case for proactive protocols.The node density tests attempted to measure theimpact of node density on proactive and reactiveprotocols.

Experimental Results and Analysis

Experiment 1 (mobility tests)The left halfof Table 2 (under the column heading Nodemobility) shows the number of control packetsobserved for the five different levels of nodemobility. The left-most column, Load, indicates




11/18

three traffic load levels. Figure 2 shows thenumber of the control packets when the traffic loadwas low. The graphs show the minimum, averageand maximum values (after discarding theextremes) with error bars indicating approximatepercentage difference.

Table 2 shows that DSR resulted in the smallestnumber of control packets for all the experimentsin the mobility tests. Under low and mediumtraffic load, WRP generated the largest number of

control packets (1276 and 1285 packets for low andmedium traffic load, when the node mobility waszero). Under high traffic load, AODV generatedthe largest number of control packets (3899packets), followed by WRP (1246 packets). One ofthe possible causes for high messaging overheadin WRP for low and medium traffic load is thatWRP generated control packets to prevent loops inrouting, which is not the case for the other threeprotocols.



Node mobility Node density

Load Protocols Perpetual High Medium Low Zero High Medium Low

Low DV 1,577 1,121 967 750 624 4,456 2,598 1,438WRP 3,159 2,037 2,013 1,687 1,276 2,512 2,073 1,800AODV 281 157 156 157 156 307 303 262

DSR 108 145 132 69 52 95 46 48Medium DV 1,552 1,086 947 766 647 4,434 2,600 1,254

WRP 3,112 2,020 1,730 1,404 1,285 2,432 2,197 1,967AODV 1,805 751 627 482 365 1,241 1,201 1,156DSR 452 357 408 383 351 284 318 187

High DV 1,568 1,166 979 927 799 4,742 2,723 1,444WRP 2,875 1,943 1,749 1,757 1,246 2,300 1,995 1,980AODV 10,123 5,097 4,901 4,797 3,899 11,768 12,333 9,431DSR 822 741 821 789 522 828 543 393

Table 2. Number of control packets observed for the five different levels of node mobility and threedifferent levels of node density

0

2000

4000

6000

8000

10000

12000

0 100 200 300 400 500

Pause Time (seconds)

Num

berofControlPackets

WRP

DSR

Distance Vector

AODV

Figure 2. Control overhead for various node mobility levels at three levels of traffic load


12/18

It was observed that the impact of traffic load torouting overhead was more significant for AODVand DSR than for WRP and DV protocols. ForAODV, the number of control packets increasedsteeply from 281 to 10,123 packets (which wasapproximately a 36 increase) when traffic loadwas increased from low to high at perpetualmobility (Table 2). The increase was from 108 to822 packets for DSR (approximately a 7.6 increase). For DV and WRP, the difference was

minor. For DV, it was decreased from 1577 to 1568(0.6% decrease). In WRP, it was decreased from3159 to 2875 (9% decrease). These results implythat WRP and DV are more scalable to increase intraffic load in terms of messaging overhead thanAODV and SDR.

To study the effect of node mobility to mes-saging overhead, the ratio of the control packetsgenerated at perpetual mobility to that of zeromobility was calculated for each of the four proto-cols. The ratio of increase in the number of controlpackets in DSR was 207.7% (= 108 packets/52

packets), 128.8% (= 452/351) and 157.5% (822/522)for low, medium and high traffic load (Table 2),when the node mobility was increased fromzero mobility to perpetual mobility. For the sameexperiment, DV resulted in 252.6%, 239.8% and196.2% increase, while WRP resulted in 247.7%,242.2% and 230.7%, and AODV in 180.7%, 494.5%and 259.7% increase. DSR demonstrated thelowest increase rate in the number of controlpackets when mobility was increased from zero to

perpetual for medium and high traffic load(209.7%, 128.8% and 157.3% increase). This impliesthat DSR is least vulnerable to node mobility interms of routing overhead in the four protocols.The increase rate of control packets for WRPwas least affected by the traffic load (only 17%difference between low and high traffic load),which implies that WRP will be traffic-loadscalable.

The left half of Table 3 shows the packet deliv-

ery rate as a percentage for the four protocols forthe five different levels of node mobility. Figure 3shows the packet delivery rate of the four proto-cols observed for the low traffic load. Note that thepacket delivery rate may not reach 100% evenwhen any node in a network is not moving. InUDP, the packets dropped due to buffer overfloware not retransmitted, resulting in undeliveredpackets.

From Table 3, it can be seen that the packet deliv-ery rate was low (12.017.1%) for the four proto-cols when the node mobility was perpetual, even

at low traffic load. When the node mobility waszero and traffic load was high, WRP resulted inhigher packet delivery rate than DV protocol by4.7% (19.1% for WRP, while it was 14.4% for DV),presumably due to WRPs better route conver-gence and maintenance than those used in the DVprotocol. AODV and DSR demonstrated similarresults in the packet delivery rate, when trafficload level was low. The packet delivery rate forAODV was 13.8%, 30.3% and 57.9%, while it was





Low DV 12.3% 16.4% 34.6% 47.4% 86.9% 24.2% 16.8% 16.3%WRP 17.1 18.4 38.7 64.9 89.1 26.3 22.2 22.4AODV 13.8 16.3 30.3 38.4 57.9 21.8 17.1 19.1

DSR 12.0 20.1 28.8 41.2 60.3 20.0 14.6 16.2Medium DV 14.6% 16.6% 29.5% 44.2% 69.4% 23.2% 15.0% 11.3%

WRP 23.0 36.6 52.7 55.0 76.6 27.1 18.2 21.0AODV 19.6 32.9 38.5 45.2 71.6 26.0 21.1 13.3DSR 19.5 21.8 28.8 35.3 62.2 25.0 18.3 12.7

High DV 10.4% 11.2% 14.0% 12.7% 14.4% 13.8% 11.3% 11.4%WRP 8.9 13.4 15.8 18.0 19.1 19.9 13.3 13.0AODV 5.4 12.9 16.8 19.8 26.0 9.3 8.3 8.1DSR 6.2 9.6 13.4 12.5 18.1 15.5 8.0 10.2

Table 3. Packet delivery rate (%) for the five different levels of node mobility and three levels of nodedensity


13/18

12.0%, 28.8% and 60.3% for DSR for perpetual,medium and zero node mobility.

A possible reason for the reduced packet deliv-ery rate in the two reactive protocols (i.e., AODVand DSR) was route fluctuation due to the delayneeded for route recoveries. Paths detected by

route discovery might change again before sourcenodes transmit packets if nodes were activelymoving in a network. Detected changes in linkstates will be notified to source nodes, which willtrigger the source nodes to initiate another routediscovery. Thus, data packets that are transmitted

before rediscovery of routes would be lost. For thetwo proactive protocols (i.e., DV and WRP), con-vergence delay in propagating updated routinginformation can be the primary cause of reducedpacket delivery rate when node mobility wasincreased. Data packets transmitted while the

latest routing information is being propagated ina network will be lost due to possible routingloops caused by temporary inconsistency in therouting tables.

Figure 4 shows the ratio of packet delivery rateobserved at zero mobility to that of perpetualmobility as a percentage. We observed that thepacket delivery rate of AODV was least affected bychanges in traffic load. For AODV, the ratio was419.6%, 365.3% and 481.5% for low, medium and

high traffic load (for low traffic load, the ratio wascalculated as 57.9 (the delivery rate of AODV atzero mobility)/13.8 (the delivery rate of AODV atperpetual mobility), which is 4.1956). The absolutedifference between the maximum and minimumwas 116.2% (481.5365.3%). The increase was

706.5%, 475.3% and 138.5% for DV (the differencewas 568.0%), while it was 502.5%, 319.0% and291.9% for DSR (the difference was 210.6%) and521.1%, 333.0% and 214.6% for WRP (the dif-ference was 306.5%). These results suggest thatAODV will be one of the protocols whose packetdelivery rate is least vulnerable to increase intraffic load.

The left half of Table 4 shows the mean end-to-end delay (in seconds) observed for the four pro-tocols for the five different levels of node mobility.DSR resulted in the largest end-to-end delay, while

WRP was the shortest. DSR resulted in the longestend-to-end delay except when traffic load washigh at perpetual node mobility (for this case only,DV resulted in the longest delay). The delay fromDV was 32.42s, which was approximately twotimes longer than DSR. DSR resulted in 2.79, 2.10and 1.52s for perpetual, medium and zero mobil-ity when traffic load was low. It was 0.28, 0.15 and0.12s for DV, while it was 0.15, 0.10 and 0.10s forWRP, and 0.56, 0.53 and 0.43s for AODV for the



0

10

20

30

40

50

60

70

80

90

100

0 100 200 300 400 500

Pause Time (seconds)

PacketDeliveryRated(%)

WRP

Distance Vector

DSR

AODV

Figure 3. Packet delivery rate for five different levels of node mobility at low traffic load


14/18

same experiments. For medium traffic load,similar results were observed.

The end-to-end delay of WRP was significantlyshorter than the other three protocols at the per-petual node mobility when traffic load was high.The mean end-to-end delay of WRP was 3.83s,which was only 11.8%, 21.2% and 24.7% of DV(32.42s), AODV (18.04s) and DSR (15.50s) proto-cols, respectively.

Possible explanations for the short end-to-enddelay in WRP are: (1) WRP is a proactive protocol;(2) WRP uses an efficient route maintenance

method (no loops and constant check of linkstatus), hence the average end-to-end delay would

be less as the likelihood of sending information oninvalid paths would be less; and (3) in the event oflink failures, WRP recovers much quickly.7 Theseadvantages are not present in DV protocol, exceptfor point (1); hence DV resulted in longer end-to-end delay than WRP, especially at high traffic loadand high mobility.

We also observed that DV constantly resulted inthe highest increase rate in end-to-end delay whennode mobility was increased from zero to perpet-



0%

200%

400%

600%

800%

Low Medium High

WRP

DSR

AODV

Distance Vector

RelativeIncreaseinpack

etdeliveryrate

Traffic Load

Figure 4. The relative increase in packet delivery rate when mobility was decreased to zero mobility



Low DV 0.28 0.28 0.15 0.13 0.12 0.12 0.19 0.17WRP 0.15 0.13 0.10 0.10 0.10 0.11 0.15 0.13AODV 0.56 0.47 0.53 0.46 0.43 0.40 0.36 0.41DSR 2.79 2.89 2.10 1.44 1.52 1.72 1.54 1.40

Medium DV 0.28 0.25 0.19 0.14 0.12 0.12 0.18 0.14WRP 0.21 0.17 0.15 0.12 0.11 0.11 0.19 0.14

AODV 0.26 0.20 0.19 0.20 0.20 0.25 0.26 0.22DSR 1.16 1.82 1.15 1.03 1.10 0.28 0.49 0.44High DV 32.42 5.02 4.54 4.72 4.58 2.58 4.60 5.17

WRP 3.83 2.59 2.47 2.31 2.56 3.12 5.59 3.40AODV 18.04 5.34 5.05 5.09 6.77 8.08 8.64 5.88DSR 15.50 7.75 7.12 6.03 4.65 7.02 10.83 7.48

Table 4. Mean end-to-end delay (in seconds) for the five different levels of node mobility and the threedifferent levels of node density


15/18

ual. Especially when traffic load was high, DVshowed the worst scalability for node mobility. InDV, the end-to-end delay increased at the highestrate when node mobility was increased from zeromobility to perpetual mobility. In other words, theother three routing protocols demonstrated that

they had better scalability for node mobility inend-to-end delay, compared to DV. As a result, itis concluded that one of the advantages in thethree exiting routing protocols for routingMANETs is their good scalability for node mobil-ity in end-to-end delay.

Experiment 2 (node density tests)The righthalf of Table 2 (under the column heading Nodedensity) shows the number of control packetsobserved for the three different levels of nodedensity. Figure 5 shows the number of controlpackets for the three levels of node density whentraffic load was high. Figure 6 shows the ratio ofthe control packets generated at high node densityto that generated at low density.

At low and medium traffic load, the table-basedprotocols (DV and WRP) generated more controlpackets than the demand-based protocols (AODV



0

2000

4000

6000

8000

10000

12000

14000

100 75 50

Node Density (nodes per network)

NumberofControlPackets

WRP DSR

AODV Distance Vector

Figure 5. Messaging overhead for three different levels of node density at high traffic load

0%

100%

200%

300%

400%

Low Medium High

Traffic Load

WRPDSR

AODV

Distance Vector

Relativeincreaseincontrolpackets

Figure 6. The ratio of the number of control packets observed at high node density to that observed atlow node density


16/18

and DSR). For low traffic load, DV generated 4456,2598 and 1438 packets for low, medium and highnode density, while for WRP the figures were 2512,2073 and 1800 packets. AODV generated 307, 303and 262 packets, while DSR resulted in only 95, 46and 48 packets. Similar results were observed for

medium traffic load.The messaging overhead in AODV sharply

increased when traffic load was increased. Thenumber of control packets increased from 307 atlow traffic load to 11,768 packets at high trafficload (at high node density for both), which was a38.3 increase from low traffic load. A similarincrease was observed for medium and low nodedensity. These results imply that AODV is nottraffic load scalable in terms of control overhead.

The ratio of control messages generated at highdensity to that at low density was calculated tomeasure the scalability in messaging overhead fornode mobility. We defined good scalability in mes-saging overhead for node density to mean that thenumber of control packets would not rapidlyincrease when node density was increased. DVresulted in the worst node density scalability,while DSR resulted in the best. When node densitywas increased from low to high, DV resulted in a309.9% (4456 divided by 1438), 353.6% and 328.4%increase in the number of control packets for low,medium and high traffic load. DSR resulted in

an increase of only 197.9%, 151.9% and 210.7%.Results for AODV and WRP were similar. ForAODV, the increase was 117.2%, 107.4% and124.8%, while for WRP it was 139.6%, 123.6% and116.2%. The same pattern was observed for thethree different levels of traffic load. These resultssuggest that an advantage in using the threerouting protocols designed for MANETs (com-pared to a classic DV) is in the good scalability incontrol overhead for increased node density.

The right half of Table 3 shows the observedpacket delivery rate for the four protocols for the

three different levels of node density. The packetdelivery rate increased as the node densityincreased for all four protocols at any traffic loadlevel. When both node density and traffic loadwere low, the packet delivery rate was 16.3%,22.4%, 19.1% and 16.2% for DV, WRP, AODV andDSR, respectively, while the rate was 24.2%, 26.3%,21.8% and 20.0% when the node density wasincreased to high. Similar results were observedfor medium and high traffic load. For the effect of

node density to packet delivery rate, we did notobserve any significant difference in the four pro-tocols.

The right half of Table 4 shows the mean end-to-end delay observed in Experiment 2. For lowtraffic load, the increase in the end-to-end delay

was minor for changes in node density for all fourprotocols. The mean end-to-end delay for DV was0.12, 0.19 and 0.17s for high, medium and lownode density, when traffic load was low. It was0.11, 0.15 and 0.13s for WRP. For AODV, it was0.40, 0.36 and 0.41s, while it was 1.72, 1.54 and 1.40s for DSR. These results imply that the nodedensity will not significantly affect end-to-enddelay if traffic load is low. However, the long end-to-end delay for DSR is conspicuous at low trafficload.

At medium traffic load, the higher node densityresulted in lower end-to-end delay. For example,mean end-to-end delay for DV was 0.12, 0.18 and0.14s for high, medium and low density, while itwas 0.28, 0.49 and 0.44s in DSR (similar resultswere observed for WRP and AODV for the sameexperiments). One possible explanation for thisresult is the presence of a larger number of alter-native paths to a destination node due to theincrease in the number of nodes in a network.

At high traffic load, the end-to-end delay signif-icantly improved (i.e., became shorter) as the node

density increased, except for AODV. Mean end-to-end delay for DV was 2.58, 4.60 and 5.17s for high,medium and low node density. It was 3.12, 5.59and 3.40s for WRP. For DSR the figures were 7.02,10.83 and 7.48s.

The ratio of the mean end-to-end delayobserved at low node density to that at high nodedensity for DV was 141.7%, 116.7% and 200.4%increase for low, medium and high traffic load. Theratio was 102.5%, 88.0% and 72.8% for AODV. Itwas 81.4%, 157.1% and 106.6% for DSR, and113.3%, 127.3% and 109.0% for WRP. Especially for

low and medium traffic load, the increase rate inend-to-end delay was not significant (less than20% increase), except for DV and DSR protocol atlow and medium traffic load (141.7% for DV at lowtraffic load and 157.1% for DSR at medium trafficload). DV resulted in a high increase rate in end-to-end delay when traffic load was high.




17/18

Conclusions and Future Work

This paper attempts to compare three represen-tative routing protocols for MANETs (WRP, AODVand DSR) to DV protocol in their packet deliveryrate, end-to-end delay and messaging overhead to

understand the advantages of the three protocolsdeveloped for MANETs. We developed two sets ofexperiments in this project. The first experimentcompared the four routing protocols with respectto node mobility in an ad hoc network for differ-ent levels of traffic load. In the second experiment,the four protocols were compared for differentnode density for different levels of traffic load. Thefollowing is a list of key findings obtained fromour experiments:

1. Although DSR and AODV are both reactiveprotocols, DSR resulted in the best (i.e., theleast) messaging overhead for all the experi-ments in both Experiments 1 and 2, andAODV generated a higher volume of controlpackets than even the two proactive protocols(DV and WRP). Since the major difference

between DSR and AODV for control over-head is the lack of periodic route mainte-nance in DSR, the periodic hello messagesused in AODV to maintain routes was mostprobably responsible for DSRs high control

overhead.2. Contrary to our prediction, DV performedmuch better than expected. DV resulted innearly similar results in packet delivery rate(Figure 3) and was even better in end-to-enddelay, especially compared to the two reac-tive protocols (AODV and DSR). The cate-gory where DV showed mainly poorerperformance than the other three MANETprotocols was messaging overhead (in thenumber of control packets). DSR, against ourexpectation, showed a worse performance in

messaging overhead when traffic load washigh. End-to-end delay for DSR was con-stantly longer than those of the three otherprotocols.

3. The impact of traffic load on the amount ofmessaging overhead for routing was high inAODV. For AODV, the number of controlpackets increased to 36 larger when trafficload was increased from low to high in ourexperiments (Table 2). For DV, WRP and DSR,

the increase was approximately 1.3 , 1.1 and 7.6 , respectively. The experimentalresults show that proactive protocols (DVand WRP) were less vulnerable to increase intraffic load than reactive protocols.

4. Experiment results suggest that DSR has the

best scalability for node mobility in messag-ing overhead, meaning that the number ofcontrol packets in DSR will not increasesharply even when node mobility increases.

5. The packet delivery rate was quite low(between 12.2% to 17.1%) for all four routingprotocols when the node mobility was per-petual (i.e., when nodes continued to move).Since the packet delivery rate was low whentraffic load was low and node mobility wasperpetual, this low packet delivery ratewas most likely because routing informationwas easily outdated and might not beupdated quickly enough during extremelyhigh node mobility. Thus, data packets weretransmitted to non-existing paths, causing ahigh rate of lost packets.

6. We found that the DSR had the longest end-to-end delay and that WRP had the shortestend-to-end delay for all traffic loads. Theend-to-end delay of WRP was significantlyshort compared to the other three protocols,when the traffic load was high at the perpet-

ual node mobility. Possible explanations forthe short end-to-end delay in WRP are: WRPmaintains the best-path information to a des-tination, avoids routing loops during routediscovery process and converges quicklyafter a link failure.

7. Results shown in Figure 6 suggest that one ofthe advantages in using the three routing pro-tocols designed for MANETs is the goodscalability for node density in messagingoverhead. The results show that increase inthe number of control packets was minor for

the three protocols designed for MANETs(AODV, DSR and MRP) when node densitywas increased. The largest increase inthe number of control packets when nodedensity was increased from low to high was124.8% for AODV, 210.7% for DSR and139.6% for WRP, while it was 353.6% for DV.

8. We observed that WRP resulted in a goodpacket delivery rate (WRP resulted in the bestpacket delivery rate expect when traffic load




18/18

and node density were both medium). Thisresult suggests that WRP will be a good pro-tocol if high reliability and throughput arethe priority.

9. DV resulted in the worst scalability for nodemobility in the end-to-end delay. When the

node mobility was increased from zero mo-bility to perpetual mobility, DV resulted in thehighest increase rate in end-to-end delay.This inversely implies that one of the primaryadvantages in the three routing protocolsdesigned for MANETs is the scalability fornode mobility in end-to-end delay.

Future study will include measuring the actualnumber of bytes transmitted for control messages,which would be useful to better differentiate thetwo on-demand protocols. Another future study

will be to perform the experiments for various dif-ferent node migration speeds. We used a nodemobility of 45km/h in our experiments this time.However, this is just one of the possible velocities.Keeping the migration speed lower may lessen orrule out cases of packets getting dropped even

before routing information is updated. This mayaffect the simulation results and perhaps will bringout the strengths and weaknesses of different pro-tocols unambiguously.

References1. Perkins C, Bhagwat P. Highly dynamic destination

sequenced distance vector routing for mobile com-puters. In Proceedings of the SIGCOMM 94 Conferenceon Communication Architectures, Protocols and Appli-cations 1994; 234244.

2. Perkins C, Royer EM. Ad hoc on demand distancevector (AODV) routing. In Proceedings of the SecondIEEE Workshop on Mobile Computing Systems andApplications 1999; 90100.

3. Johnson DB, Maltz DA. Dynamic Source Routingin Ad-Hoc Wireless Networking, Mobile Computing.

Kluwer: New York, 1996.4. Corson MS, Ephremides A. A distributed routingalgorithm for mobile wireless networks. ACMBaltzer Wireless Networks Journal 1995; 1(1): 6181.

5. Ford LR, Fulkerson DR. Flows in Network. PrincetonUniversity Press: Princeton, NJ, 1962.

6. Murthy S, Garcia-Luna-Aceves JJ. Loop-free internetrouting using hierarchical routing trees. In Proceed-ings of the IEEE INFOCOM 1997; 101108.

7. Murthy S, Gracia-Luna-Aceves JJ. An efficientrouting protocol for wireless networks.ACM Mobile

Networks and Applications Journal, Special Issue onRouting in Mobile Communications Networks 1996;183197.

8. Iwata A, Chiang CC, Pei G, Gerla M, Chen TW. Scal-able routing strategies for ad hoc wireless networks.IEEE Journal on Selected Areas in Communications,

Special Issue on Wireless Ad hoc Networks 1999;17(8): 13691379.9. Toh CK. A novel distributed routing protocol to

support ad hoc mobile computing. In Proceedings ofthe 1996 IEEE Fifteenth Annual International PhoenixConference on Computers and Communications 1997;136.

10. Peralman MR, Haas ZJ. Determining the optimalconfiguration for the zone routing protocol. IEEEJournal on Selected Areas in Communications, SpecialIssue on Wireless Ad hoc Networks 1999; 17(8):13951414.

11. Broch J, Maltz DA, Johnson DB, Hu Y, Jetcheva J.

A performance comparison of multi-hop wirelessnetwork routing protocols. In Proceedings of theFourth Annual ACM/IEEE international Conference onMobile Computing and Networking 1998; 2530.

12. Das SR, Perkins CE, Royer EM. Performance com-parison of two on-demand routing protocols for adhoc networks. In Proceedings of the IEEE INFOCOM2000; 312.

13. Raju J, Garcia-Luna-Aceves JJ. A comparison of ondemand and table driven routing for ad hoc wire-less networks. In Proceedings of IEEE InternationalConference on Communications 2000; 17021706.

14. Royer E, Toh CK. A review of current routing pro-

tocols for ad hoc mobile wireless networks. IEEEPersonal Communications 1999; 4655.

15. Hong X, Xu K, Gerla M. Scalable routing protocolsfor mobile ad hoc networks. IEEE Network 2002;1121.

16. Cordeiro CM, Agrawal DP. Mobile ad hoc network-ing. In Proceedings of the 20th Brazilian Symposium onComputer Networks 2002; 125186.

17. Jaffe JM, Moss FM. A responsive routing algorithmfor computer networks. IEEE Transactions on Com-munications 1982; 30(7): 17581762.

18. Zeng X, Bagrodia R, Geria M. Glomosim: a libraryfor parallel simulation of large scale wireless net-

works. In Proceedings of the 12th Workshop on Paralleland Distributed Simulations 1998; 154161.


If you wish to order reprints of this or anyother articles in the International Journalof Network Management, please [email protected].

Manet Study

Documents