Analysis of Routing Protocol Performance on Multi-Hop ...nitin/TRM Documents/Nagar... · Analysis of Routing Protocol Performance on Multi-Hop Wireless Ad Hoc Networks Nitin Nagar

1884 PDPTA '01 International ConferetrJ.

Analysis of Routing Protocol Performance on Multi-HopWireless Ad Hoc Networks

NitinNagarDepartment of Computer Science

University of Nevada, Reno, NY 89577

Abstract- Multi-Hop Wireless Ad-Hoc Networks arecharacterized by dynamic and frequently changingtopologies, bandwidth constrained variable capacity ~

links and energy-constrained operation. Nodes insuch a network are usually laptops or hand heldpersonal digital assistants (PDA), which havelimited resources like battery power, CPU power,storage capacity, and transmission range. InternetProtocol (IP) based routing protocols are notsuitable for these types of networks as they relyeither on periodic updates or require stabletopologiesfor efficient routing. The objectivefor thispaper is to evaluate proposed routing protocols forMH-WANET based on peiformance. This evaluationshould be done theoretically and throughsimulation. The goal of this paper is to presentuseful insight in proposed routing protocols forMulti-Hop Wireless Ad-Hoc Networks through themeans of theoretical and simulation based study.The simulation environment is based on a Network

Simulator 2 from Berkeley with wireless extensi0!lprovided by Monarch group at Carnegie MellonUniversity (CMU).

Terms: QoS, Multi-Hop Wireless Networks,Routing

l.Introduction: Wireless communication

between mobile users is becoming morepopular than ever before. This is due to recenttechnological advances in laptop computers andwireless data communication device, such aswireless modems and wireless LANs. This haslead to lower prices and higher data rates. Thereare three distinct approaches for enablingwireless communication between two hosts.The first approach is to let the existing cellularnetwork infrastructure carry data as well asvoice. The second approach is to form a SimpleWireless Ad-Hoc Network (SWANET), whichallows a mobile node related to its fixed station

Angkul KongmunvattanaDepartment of Computer Science

University of Nevada, Reno, NV 89577

to roam in its neighborhood even if it is outsideof its host station [7]. The third approach is toallow communication among all users wantingto communicate with each other utilizing morethan one hop (Multi-Hop). These networks areknown as Multi-Hop Wireless Ad-HocNetworks (MH-WANET). MH-WANET hasseveral advantages compared to traditionalcellular systems and SWANET, which includeson demand setup, fault tolerance andunconstrained connectivity. MH-WANET doesnot rely on any pre-established infrastructuresand can therefore be deployed in places with noinfrastructure [2]. Nodes act as a host as well asrouter; therefore a routing protocol is necessaryto make the routing decisions. Currently astandard routing protocol for MH-WANETdoes not exist; however; establishments of thesame protocol are in progress. Many problemsremain to be solved before any standards can bedetermined. This paper looks at some of theseproblems and tries to evaluate some of thecurrently proposed protocols.

Many routing protocols have beenproposed, but few comparisons betweendifferent protocols have been made. Of thework that has been done in this field, only thework done by monarch project at CarnegieMellon University (CMU) has compared someof the different proposed routing protocols andevaluated them based on some quantitativemetrics [2]. Other simulation results do exist[3] that have been done on individual protocols.However, these simulations have not used the

same metrics and therefore are not comparablewith each other. Our simulations haveconfirmed that Dynamic Source Routingprotocol (DSR) [4] and Ad-Hoc Distance

~-,r=i '01Internatwnal Conference 1885

,eclor (AODV) [5] protocol for MH-WANET~orms better on less mobile conditions and: SR outperforms AODV on less source.:;.:--=sity.AODV protocol demonstrated a stable-..jng overhead; network routing load and rate_ hich link breakages are reported at high~~ and high source density. It; however,:-. e a better packet delivery ratio in sharp

-~ra.stto DSR whose performance was 30%r ~'tanAODV.

':~Iulti-Hop Wireless Ad-Hoc'tt\vorks: A new kind of wireless network

:oroposedby Internet Engineering Task Force- a"" which is known as "Multi-Hop Wireless_-Hoc Networks" (MH-WANET). In MH-..,;Tf each node is capable of dynamically_~ -,'italinformation with respect to changing- .:..Uons,and variable topology on energy and

. .:.:.~ldth constrained links. Each node has a~ ess interface of limited range and they~icate with each other over radio or

- ~ media. Nodes in the MH-WANET are- mobile, but can also consist of stationary

such as access points to the Internet. In'":9Ulationswe have used Lucent Wave

~el, which gives a range o,f 250 m to-::Dbilenode. A Multi-Hop Wireless Ad-

:\etwork (MH-WANET) uses no__~ administration. This is to ensure that-~ork won't collapse when one of the:=~es moves out of transmitter range of-as. Nodes should be able to enter/leave

~rk at any time. Since the range of the:=3I1smitteris limited, multiple hops may-_0:)00 to reach other nodes. Every node

~~ng in an MH-WANET must be able to.:~ packets for other nodes. Hence each

_ ... _ as a host and as a router. A router is an

!:ich, among other functions runs a.~ f!"otocol. MH-WANET is also capable

_-~.::'g topologychanges and malfunctions'.:6. It is fixed through network- .f"_-;!tion.For instance, if a node leaves

- .::!k and causes link breakages, affected~ easily request new routes and the

. - ":','ill be solved. This will slightly_, ~ delay but the networkwill still be

u.-_ -~ The radioenvironmentas an access

media has special properties that must beconsidered when designing protocols for MH-WANET. Multi-Hop in a radio environmentmay result in an overall transmit capacity gainand power gain due to the squared relationbetween' coverage and required output power.By using Multi-Hops, nodes can transmit thepackets with a much lower output power.However, usage of Multi-Hop Wireless Ad-HocNetworks also introduces problems unique toitself. For example the nodes are always energyconstrained and links are bandwidth constrainedwhere the messaging is usually performed usingbroadcast messages, hence there is emphasis onto find ways to limit the exchange ofinformation between nodes.

3.Routing Protocols for Ad Hocmobile networks: Since routing requiresmultiple hops through the network, a routingprotocol is needed. A routing protocol is aservice protocol that is'used (by routers, but notby hosts) to maintain routing tables. The linkstate and distance vector protocols are avoidedin Multi-Hop Wireless Ad-Hoc Networks, due tofrequently changing topologies and requirementto make dynamic routing decisions. Anothercharacteristic of conventional protocol is thatthey assume bi-directional links; in the wirelessradio environment this is not always true.Routing algorithms in a MH-WANET can beclassified as either proactive or reactive.Proactive protocols attempt to continuouslyevaluate the routes within the network so thatwhen a packet needs to be forwarded, the routeis already known and can be immediately used.Proactive schemes have the advantage that whena route is needed, the delay before actual packetscan be sent is very small. On the other handproactive schemes needs time to converge to asteady state. This can cause problems if thetopology is changing frequently. Since theconventional routing protocols do not meet thedemands for Multi-Hop Wireless Networks, newprotocols like table driven protocols or ondemand driven protocols have been proposed.Table driven routing protocols attempt tomaintain consistent, up-to-date routinginformation from each node to every other node

1886 PDPTA '01 International Conference

in the network whereas Source initiated on-

demand routing protocol takes a differentapproach by creating a route only when desiredby the source node. An ideal routing protocolhas distributed operation, loop free, demandbased operation, multiple routes, unidirectionallink support, security, power Conservation andQoS support.

Dynamic Source Routing protocol: TheDynamic source routing [4] is a source initiatedon-demand routing protocol based on theconcept of source routing. DSR uses no periodicrouting messages, thereby reducing networkbandwidth overhead, conserving battery powerand avoiding large routing updates throughoutthe MH-WANET. Instead, DSR relies onsupport from the Media Access Layer (MAC),which should inform the routing protocol aboutlink failures. The basic advantages of DSRprotocol are that nodes do not requiremaintaining routing information, and routes areoften predetermined even before transmissiontakes place. Secondly, DSR is on demand innature, so there is minimal routing overheadover the mobile nodes and it also works insymmetric as well as in asymmetric conditions.However, each packet carries a slight overheadcontaining the source route of the packet. Th,isoverhead grows when the packet has to gothrough more hops to reach the destination. Sothe packets sent will be slightly bigger becauseof the overhead. The DSR protocol is composedof two mechanisms that work together to allowthe discovery and maintenance of source routesin the Multi-Hop Wireless Ad-Hoc Network. Inparticular, unlike other protocols, DSR requiresno periodic packets of any kind at any levelwithin the network.

Ad hoc on Demand Distance Vector

(AODV) Routing protocol: AODV [5] isactually an on-demand version of DSDVprotocol, which is a table-dri ven routingprotocol. The main difference between theDSDV and the AODV routing protocol is thatthe routing table in ,AODV protocol is requiredto be updated only when desired by the sourcenode. In AODV protocol, when a source node

wants to send information to the destinationnode, it first cheeks the routing to see theavailability to the route destination. If the routeis not available in its routing table, or theprevious valid route has expired then it initiatesa route discovery process. The AODV protocoluses hello messages that are broadcastedperiodically to the intermediate neighbors. TheDSR protocol has the advantage of supportingboth symmetric and asymmetric links, where asAODV protocol only support symmetric links.

4. Simulation Setup: We have usednetwork simulator 2(ns-2) [I], which is adiscrete event simulator, targeted at networkingresearch. It provides substantial support forsimulation of TCP, routing, and multicastprotocols. The current version of the networksimulator does not support wireless networks.The network simulator alone is only intended forstationary networks with wired links. Themonarch group at CMU has developed a modelto simulate multi-hop wireless network,complete with physical, data link and MAClayer models on ns-2. Lucent Wave LAN is usedas a radio propagation model in the simulator.Wave LAN is a shared media radio with anominal bit-rate of 2Mb/see and a nominal radiorange of 250 meters.

,Traffic and mobility models: The trafficsource is CBR (continuous bit-rate). The source-destination pairs are usually spread randomlyover the network. The size of data packet is 512bytes. The packets are sent at the rate of 4packets per second. The mobility uses randomway point model [2], which states that, a groupof nodes, which pause for a specific duration oftime during their motion. We use theconfigurations used by many other researchersi.e. 1500m x 300m field with 50 Nodes. In thisfield a node begins its journey from a randomsource to a random destination with a randomlychosen speed (uniformly distributed between 0-20m/sec). The pause time is varied which alsoaffects the relative speeds of the mobiles.Simulations are done for 900 sec. Each datapoint represents an average of two runs withidentical traffic models, but different randomly

"~! '01 International Conference 1887

;merated mobility scenarios. For maintainingTJ1esswe use identical mobility and traffic

.:enarios for DSR and AODV Protocols. A1'ical simulation with ns and the mobility.:.:nsion consists a scenario file, which

.xscribes the movement pattern of the nodes,_-.1 a communication file that describes the

ffic in the network.

Ietries: Four key performance metrics:-~ket delivery ratio, routing overhead. error=~l-ets transmitted and normalized routing=-f'. These metrics are evaluated to examineAIlOUSaspects of the routing protocols. Packet

.ery ratio is the ratio of the data packets..c ered to the destination to those generated by

~ Constant Bit Rate (CBR) sources. Routingrlzeadis the total number of routing packets

:..::.smittedduring the simulation where each=smission of the packet (each hop) counts as

-~ mmsmission. Error packets transmitted ist: wtal number of error packets transmitted~g a simulation. During a simulation, each

-..enlink is reported by a single transmissionerror packet. Normalized routing load

~res the total number of routing packets_-.:>nUttedfor each data packet deliyered at the

_"'1ation.

-Simulation Results: As noted in

_,;),'14.1, simulations were performed using~ different node speeds: a maximum speed

~ _ of 20m/s, 30m/s and 40m/s respectively.a comparison was conducted for the

"'Colwith 10 sources and max set speeds of- r..30m/s and 40m/s. These speed values are""1:'.~,maximum speed limits set in the~ion scenario files. The speed by which

_~ move is generated randomly by the~ which is usually falls in the range of 0:s: max speed limit set by the user during::..rio generation. For all the simulations, the-:mication pattern is peer-to-peer, with

. ~.lI1having either 10 or 20 sources sending.,;".:,...etsper second. In my simulations I not

~d pause time as a base to calculate the=:: of my performance metrics but also;ued the performance metrics with base as

changing speed of mobile node. The maximumspeed limit of the mobile nodes is changed from20m/s to 40m/s. The performance of the DSRand AODV protocols can also be measured onthe basis of each individual pause time. Wewould like to clearly state that we used theperformance metrics and parameters similar tothat of the Johnson's [2] paper at Mobicom'98to verify accuracy of our simulations. I havetaken in consideration some new parameters toexplore new aspects of routing protocolsperformance.

Routing Overhead Details: Figures 6,8 and9 show the performance of DSR and AODVprotocols respectively, over multi hop wirelessad hoc networks. The performance metric inconsideration is "routing overhead" occurredduring the simulation. "Routing Overhead' isthe total number of routing packets transmittedduring the simulation. [2]. Each curve representsthe routing overhead with various sources andvariable speeds. The results clearly show that forlower speeds, DSR protocol performs better thanAODV by a factor of 6. However as the speed isincreased the DSR continues to out performAODV by at factor of 10 at 40m/s speed. Figure9 shows the routing overhead for DSR protocolwith various sources as a function of speed inMulti-Hop Wireless Ad-Hoc Networks (MH-WANET). The results show that for DSRprotocol with 20 sources and a pause time of 0(constant mobility), the routing overheadincreases by 900% when the speed of mobilenode is doubled to 40m/s. In similar situation theRouting Overhead in AODV protocol increasesby only 200%. However, the Routing Overheadby AODV protocol is at least two times to thatof DSR protocol. But it does manage tohighlight that change in Routing Overhead inAODV protocol is at least 50% less drastic asDSR protocol. Hence overall, DSR protocol hasless routing overhead compared to AODVprotocol. The routing overhead in DSR protocolcould be kept low if a higher pause time ischosen for the mobile nodes, when movement athigh speed becomes necessary

1888 PDPTA '01 International Conferena

~ ~ m ~ ~ ~ ~ ~ ~Pause feme (sees)

Fig 1. DSR: 10 Sources

0.99 --~---

.

---~~

I

."

.~. .~. - .. .0.96

0.97

::,~'~~ :.....................~.. p.ause1

-()_ p8UW

222426283J3234Speed (fMl~'SIsec)

36 Je ..


. ".,. "I.:;'~D":~'~'~::;~~'~':':'::':';;:""':'~~":';~""!'u,;~'U

0.95 t.' C> , 3).., I

. ;.,- -G, 3JmIs09~ . 40mIs

o65f L?:~~

o.e~: i

O.7S~:

01

0.66

0.60 100 200 ]'JJ .Q) 500 600 700 9D 9]JPause11mC1 (secs)


Packet Delivery Ratio: The Figure I showsthe packet delivery ratio for DSR protocol withvarious sources in relation to pause time in amulti hop wireless ad hoc network. The resultsclearly show that when the 50 nodes with 10sources are in constant mobile condition andwhen the maximum speed limit is increasedfrom 20m/s to 30m/s, there is a drop of at least50% in packet delivery ratio. Increase in the

'I ~-f096'

[' !,- .096 _ _ _ ". ",

'.

'.'.. .,"

086

0...

O~"

~ ~ ~ ~ ~ n ~ ~ ~ ~Speed (meters/see)


012

0

0

j

H ro.W ; ;::'

o ::-.'!066; ~: ;.;

i 064< t. .'

~Oli2t

~+ 2QmI,".3- 30mIs

otOmf.

......

.......'

....

06 G..:!

058

0560 100 :&XI XI) .4(1) 9)) 9J)P,US' timt (secs)

700~~

Fig S. AODV: 10 Sources

10

DSR 10.4OmIs

0 OSR 2O.:2Om/sNJOV 10. 20mfs

,A/:JrN20.20mfs

'.

o r~~~ '.~, .-.:; ::~~.: ~:~ ~'::::::::: :"::.::::::.: ~....-.o ~ ~ m m ~ ~ ~ ~ ~

Pause Itme (secs)

Fig 6. Routing Overhead

number of sources with the higher speed limitresults in the decline in the packet delivery ratiofor lower pause times. In Figure 2, we haveplotted packet delivery ratio for DSR protocol inrelation to variable speed as a base in MH-WANET. Fig I presents a suspicious result. ThePacket delivery ratio for pause time 0(continuous mobility) improves when comparedto other pause times when the

1889

.....s,

i )5,

! 3r~2S~

~If~"

L' ..,,/1 ......." ....- - ~ .,-

'~::::i::~~~ ~.~.~ 28-:..~~-.. j,Sp..d(mlfef$l$eC)

36 36 '"

Fig 7: Error Packets Transmitted

110"8

7.. £]...

u:..:'.

ISA.c~: 3t ,1:1

0' ~+"~:::::'::::'.~::::::~..:::.::~:::~::...::~: ~.:.:. .'" _o 100 m m m ~ ~ ~ ~ ~

PaIlS. time (sees)

Fig 8. Routing Overhead (30m/see)

16DSR 10 4Om/'s

Qm. DSR1O' 40mIs- . -+. - ADOV lij AOmfs

AOrN 10: <IOm/s"

700 £OJ 9Xt

Fig 9. Routing Overhead (40m/sec)

Speed limit is incremented to 40mls, which isalmost two times the initial speed. Figure 5shows packet delivery ratio for AODV protocolwith 10 sources as a function of pause time in amulti hop wireless ad-hoc networks. The resultsclearly demonstrate that packet delivery ratiodrops significantly for lower pause times wheneither the speed is increased or the number ofsources is increased. The drop is sharper with 20sources and speed of 40mls. From Figures1,2,3,4 and 5 it is observed that packet deliveryratio for less number of sources and moderatespeed limit (20mls) as in the case of DSRprotocol, is at least 10-15% more than the

7aJJ.. +... DSR10.20mls"'-Du' OSR20.20m/s

AOOV 10. 20mIsiAOOV 3). 20mls

Fig 10. Error Packets Transmitted

09 OSR 10.2Omfs

0 OSR20.2OmIsNJfN10.2OmhN:JfN 3), 2Om/s

0811

~ 0.7i-O.6~

~a,St -;,

~ 0..~~ 0.3

02-. -..

O.:t~Iil..'ti"...:::::::~::::::::::::::;.::'...c..hU.. -.. _o 100 m D @ ~ ~ ~ ~ ~Pause tirne (ucs)

Fig.II Normalized Routing Load

.'

.....

0.' '..'

02~ .:::>..,::"'::: :::.:....................O.,'j-'~ 2' ;i; 28 ; -n 34 3; 3J 40

Speed (meiers/see)

Fig 12. Normalized Routing Load

AODV protocol. However, if extensive runs formultiple scenarios is performed and with a largernumber of sources AODV out performs DSRprotocol in terms of packet delivery ratio.Similarly if we observe Figures 4 which mappacket delivery ratio for AODV protocol inrelation to variable speed limit in MH-WANETthe results demonstrate a better packet deliveryratio for the pause time of 30 seconds when thespeed reaches 40mls. Comparing it with the Fig5, a interesting observation can be made wherepacket delivery ratio for DSR protocol is higherin comparison to AODV protocol with 10sources and continuous mobile conditions.

1890 PDPTA 'Ollnter1U1iionalConferenu

Error Packets Transmitted Details: Toinvestigate the impact of broken links in DSRand AODV protocol, I computed the errorpackets transmitted by the intermediate nodes inrelation to pause time and speed. Figure 7 showthe frequency of broken links for DSR andAODV protocols for various sources in relationto pause time. The nodes are assumed to bemoving with a fixed speed of 20m/s. The resultsclearly confirm the fact that as speed increasesthe rate at which the broken links are reportedincreases. In this case, DSR protocol performsbetter as there are at least half the number ofbroken links reported compared to AODVprotocol. This makes DSR protocol performbetter to maintain a network topology in pseudo-stable condition. Figure 10 shows the errorpackets transmitted for DSR and AODVprotocols in relation to speed in multi hopwireless ad hoc networks. I compared the rate atwhich the broken links are reported when thenodes were in state of constant motion (pausetime 0). As the results indicate that, with 20sources, DSR protocol reports at least 8 timesmore broken link errors than AODV protocolwith similar number of sources. This resultconfirms the key weakness of a multi hpwireless ad-hoc network, in which performanceoften degrades when there is a frequent changein network topology.

Normalized Routing Overhead Details: FigureII and 12, shows the Normalized routing loadfor DSR and AODV protocol as a function ofpause time and speed, in multi-hop wireless adhoc networks. According to the results obtainedin Fig 4.9, DSR protocol always has a lessNormalized routing load compared to AODV,by a factor of almost 8-9, as AODV makesextensive usage of hello packets which ittransmits periodically to sense any change intopology of the network. This gives an edge toAODV protocol in packet delivery ratio overDSR protocol for large number of sources (>30).It will be a challenge for AODV routingprotocol to maintain low routing overhead, errortransmission error and normalized routing load,in case of lower pause times and frequentlychanging topology. Hence, our results confirmsthat even though AODV protocol manages tooutperform DSR protocol, in terms of packets

successfully delivered, but the otherperformance metrics show AODV performancedecreased by at least 50%, compared to DSRprotocol.

Comparison Summary: DSR and AODVprotocol deliver a greater percentage of theoriginated data-packets when the mobility isminimum. The DSR protocol performs better athigh mobile condition (pause 0) only with lessnumber of sources. If the speed is doubled from20m/s to 40m/s, for 10 sources the performanceimproves but when the same is repeated for 20sources, the packet delivery ratio drops at leastby 30%. AODV performs average at highmobility (pause 0) but posts an increase inpacket delivery ratio as number of sources isincreased (Source Density) or if speed isincreased. However performance is improvedonly for a certain speed (In this case 30m/s).When speed is increased. AODV protocol in thissituation continues to have a stable rate of linkbreakages reported. This confirms that AODVprotocol is bound to give a good performance ifthe Source density is high and nodes are movingwith high speed.

6.Conclusion: The simulations have shown

that the need of the hour is a special routingprotocol for MH-WANET. One that is loop free,supports demand based & distributed operations,multiple routes, unidirectional links, security,power Conservation and provides high QoS tothe data. Till date there is no such ideal routingprotocol for the same. This paper is based on myresearch done for my Masters Degree. We havetried to duplicate the results obtained by theMonarch group at CMU and also tried toevaluate the performance of two basic protocolsi.e. DSR and AODV w.r.t two new performanceparameters. This has given us useful insight intoworking and problems in MH-WANET.Solutions to the problems might be manyincluding developing hybrid protocols, makinguse of best features of both the routing protocols.However, we would like to conclude this paperby stating that developing routing protocols for adynamically changing environment like MH-WANET is a challenging task and we would like

JDPTA '01 International Conference 1891

to continue our efforts to continue our efforts inthis field.

7.References:

[I] Nitin Nagar and Carl G. Looney, "ActiveProgrammable Networks and QoS inWired/Wireless Networks: A Survey", PDPTA,Las Vegas, May 2001.

:2] David B. Johnson et aI., "A PerformanceComparison of Multi-Hop Wireless Ad-Hoc~etwork Routing Protocols", Mobicomm'98,Jallas, Texas October 1998.

:3] Charles E. Perkins, Samir R. Das andElizabeth M. Royer, " Performance Comparison~f Two On-demand Routing Protocols for Ad-Hoc Networks", Infocomm 99

:~] David B. Johnson, David A. Maltz et ai,The Dynamic Source Routing Protocol for:.Iobile Ad Hoc Networks", IETF MANET

orking group, 17th November 2000.'::.HP://www.ietf.org/intemet -drafts/draft -ietf-c:1aTIet-dsr-04.txt.

)] Charles E. Perkins et ai, "Ad hoc On-:-emand Distance Vector (AODV) Routing'\'.~obileAd-Hoc Networking working group, 241h'~ovember 2000. http://www.ietf.orglintemet-rafts/draft-ietf-manet-aodv-07.txt

[6] The CMU Monarch Project. "The CMUMonarch Projects Wireless and MobilityExtensions to ns".http://www.monarch.cs.cmu.edu/ (1998-11-29).

[7] Mobile Ad-Hoc Networks (MANET),http://www.ietf.orglhtml.charters/manet-charter.html[8] Chang and Tassiulas. "Energy ConservingRouting in Wireless Ad-hoc Networks."INFOCOMM 2000, March 2000.

[9] MacDonald and Znati. "A Mobility BasedFramework for Adaptive Clustering in WirelessAd-Hoc Networks." IEEE Journal on SelectedAreas in Communications, Vol. 17, No.8,August 1999.

[10] Royer. Elizabeth M. and Chai-Keong Toh,"A Review of Current Routing Protocols for AdHoc Mobile Wireless Networks," IEEE PersonalCommunications, April 1999.