DTIC ADA461662: A Self-Correcting Neighbor Protocol for Mobile Ad-Hoc Wireless Networks

6 Pages·0.11 MB·English

by  Defense Technical Information Center

A self-correcting neighbor protocol for mobile ad-hoc wireless networks
MarcMoskoand J.J.Garcia-Luna-Aceves
UniversityofCaliforniaatSantaCruz
BaskinEngineering
SantaCruz,CA95064,USA
{mmosko,jj}@cse.ucsc.edu
Abstract—Mobile wireless ad-hoc networks lack some basic The present work does not address the differences be-
abilitiestakenforgranted in wirednetworks, such astheability tween periodic topology protocols and location-based proto-
toknowadjacentnodes. Wepresentaneighbordiscoveryproto-
cols. Location-based protocols have a significantly different
col,withparticularapplicationtobroadcastflooding. TheNeigh-
approach to connectivity and routing. NXP is suited towards
bor Exchange Protocol (NXP) has two main improvementsover
simpleperiodicbroadcastschemes:(1)itonlysendsHellopackets distance-vectorroutingprotocolsandconnectivity-baseddom-
whennecessarytomaintaintopologyand(2)usessequencenum- inatingsetbroadcastdistribution.
bersinredistributedinformationtoaidinconvergence.Insimula- Broadcast distribution control generally follows a type of
tion,wecompareNXPtoaperiodicprotocolandsimpleflooding
dominating-setrouting or re-broadcastingbased on geometry.
forall-nodepacketbroadcastsandtwodisseminationtechniques.
Neighborinformationcontrolshownodespropagatepackets–
Weshowthatwemaintainsimilardeliveryrateswhileusingfewer
controlpacketsinmostconfigurations. eitherrepeatingapacketorsuppressingapacket.Theprotocols
assume full duplex links. If node A can hear node B’s Hello
packet,thennodeBmayhearA.Infact,thisissometimesnot
the case in common-channelwireless networks, where asym-
I. INTRODUCTION
metric fading and location-dependent noise levels may favor
hearingonenodeoranother.
Wirelessad-hocnetworksarecharacterizedbyaradiochan-
A problem arises when a node makes forwardingdecisions
nelandarbitrarytopologywithoutfixedinfrastructure,suchas
basedonincorrectinformation. Ifintheaboveexample,Aac-
cellsitesorbasestations. Allsuchnetworksneedamethodto
tually cannot hear B, but B hears and advertises A as a one-
detectandorganizenodes,whetheritistocomputeaschedule
hopneighbor,some third nodeC may choose B to forwardto
for TDMA-style MAC layers, determine adjacenciesfor rout-
A when in fact there may be a better forwarder. Section II
ing, or perform what is known as dominating-set routing [1],
describes the Neighbor Exchange Protocol (NXP), which ad-
whichuseslocalizedneighborhoodinformationtocreatepacket
dressesthisweaknessbyincludingper-nodesequencenumbers
distributionbackbones.
in topology broadcasts and using a three-state machine – Up,
Neighborprotocolsaredesignedtoexchangenodeinforma-
Hold,andDown–wherenodesintheHoldstatehavenotbeen
tion for determining which nodes are “alive” and reachable.
full-duplexverified.
Theygenerallyfallin to two groups: periodic or event-based.
NXP convergesto correct topology information and main-
PeriodicprotocolsbroadcastHellopacketswithsome,possibly
tainslivelinessinamobileenvironmentwithoutrequiringHello
variable,frequency.Thefrequencymayvarybasedonnetwork
packetsbesentonafixedschedule. Italsoallowsusingsmall
load, node mobility, or relative group mobility. Event-based
KeepAlivepackets rather than full topology Hello datagrams.
schemes send Hello packets based on eventssuch as distance
ThekeyfeatureofNXPistoadvertiseneighbors’ids,state(up
movedordetectedtopologychanges.
orhold),andsequencenumber.Byrepeatingthemostrecently
Manyprotocols use neighbor informationto controlbroad-
heardsequencenumbersforeachone-hopneighbor,NXP can
cast distribution over ad-hoc networks. Some use periodic
detectinconsistenciesandpro-activelytrytocorrectthem. In-
Hellomessages[1],[2],[3],[4],[5]andothersuseevent-based
cludinganodestateallowsadvertisingunverifiednodes(nodes
updates[6],[7],[8].Periodicschemesusuallyexchangelimited
insimplexstate) andalsoservestoindicateto one-hopneigh-
information– such as onlynodeids – overa limiteddistance.
borsif theyare goingto be aged-outbecause of lost neighbor
Hello packets in location-based schemes broadcast the send-
packets.
ingnode’slocation,suchasdeterminedbyGPS.Thelocation-
TopologyBroadcast ReversePath Forwarding(TBRPF) [9]
based protocols generally serve as full routing protocols, and
usestheTBRPFNeighborDiscovery(TND)mechanism.TND
arenotlimitedtotwo-hopneighborinformation. Thelocation
has several similarities to NXP. As of draft 3 (current draft
packetsmayfloodthenetwork.
is 5), TND is modular and independent of TBRPF, similar
to how we have used an external NXP process. TND uses
This workwas supported in part by the Office of NavalResearch (ONR)
undergrantN00014-99-1-0167. differential Hello messages whereas NXP uses either com-
Report Documentation Page Form Approved
OMB No. 0704-0188
Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and
maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,
including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington
VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it

does not display a currently valid OMB control number.
1. REPORT DATE 3. DATES COVERED
2002 2. REPORT TYPE 00-00-2002 to 00-00-2002
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER
A self-correcting neighbor protocol for mobile ad-hoc wireless networks
5b. GRANT NUMBER
5c. PROGRAM ELEMENT NUMBER
6. AUTHOR(S) 5d. PROJECT NUMBER
5e. TASK NUMBER
5f. WORK UNIT NUMBER
7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION
University of California at Santa Cruz,Department of Computer REPORT NUMBER
Engineering,Santa Cruz,CA,95064
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITOR’S ACRONYM(S)
11. SPONSOR/MONITOR’S REPORT
NUMBER(S)
12. DISTRIBUTION/AVAILABILITY STATEMENT
Approved for public release; distribution unlimited
13. SUPPLEMENTARY NOTES
14. ABSTRACT
15. SUBJECT TERMS
16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF 18. NUMBER 19a. NAME OF
ABSTRACT OF PAGES RESPONSIBLE PERSON

a. REPORT b. ABSTRACT c. THIS PAGE 5
unclassified unclassified unclassified
Standard Form 298 (Rev. 8-98)
Prescribed by ANSI Std Z39-18
pleteone-hopneighborinformationHellosorKeepaliveswhich variablefrequencyperiodsorevent-basedtriggers,suchasmo-
only carry a sequence number. TND sends a Hello every bility rate or distance moved, in other instances of NXP. The
HELLO INTERVAL,whereasNXPwillnotsendaHelloifthe presentworkfocusesontheredistributionofsequencenumbers
MAClayerhasotherwisesentapacketandtherehasnotbeen and indirect detection of topology changes using those num-
someothereventthatmandatesaHellopacket. NXPrelieson bers.
promiscuousmodeoperation. NXPusesthreetypesofpacketsandfourtimers.Thepackets
TNDmaintainssymmetriclinksbyrepeatingrecentlyheard are Hello, KeepAlive,andPoll. The timersareSendMessage,
Hellos in the NEIGHBOR REQUEST field. A receiverscans Purge,Recompute,andPoll. Werefertothelengthofthemean
thislisttoseeifitislistedandifnot,itsendsHellosuntila2- SendMessagetimerasPeriod. A nodesetstimerswitha jitter
waylinkisestablished,whichissimilartoNXP’s“forcehello” variationchosenuniformlyoveraninterval.NXPalsoresponds
mechanism, but NXP also considers sequence number differ- totheeventMessage: thereceptionofapacketfromtheMAC
ences.NXPextendsthisideatoallcommonone-hopneighbors layer. Receiving a packet may promote neighbor status from
andusesPollpacketstoconvergetopologyinformation. UnknowntoUp,DowntoHold,orHoldtoUp. PeriodicPurge
TND makes uses of sequence number differences to detect timersmaydemotestatusfromUptoHoldorHoldtoDown.
thenumberoflostHellosandifitisgreaterthanathresholda NXP maintains two tables. The NbrTable stores a row for
pair ofnodesmayagree to ignoreeach other by declaringthe each one-hopneighbor,describedbelow. NXP passes this in-
link“lost”.NXPdoesnothaveanysuchmechanism,butitdoes formation to a topology algorithm on topology changes, rate
detectintermittentlinksbyredistributingallsequencenumbers limitedbytheRecomputetimer. TheNbrPolltablelistsnodes
of neighboringnodes and comparing the resulting topologies, inaninconsistentsequence-numberstate. Anodewillperiodi-
andmaytrytoconvergethetopologywithPollpackets. callysendaPollpackettothesenodes,gatedbythePolltimer.
SectionIIIpresentstheresultsofsimulationexperimentsus- Each node has a HelloSeqno, which is monotonically in-
ing three broadcast distribution protocols over two neighbor creasingforeachHellosent. AKeepAlivepacketincludesthe
protocols and flooding. The broadcast distribution protocols currentHelloSeqno,butdoesnotincrementthevalue.Pollmes-
areAHBPandAHBP-EX[5],andDominantPruning(DP)[3]. sagesdo notincludea sequencenumber. We donotcurrently
Theanalysisconsideredthreegraphdensities,fourtrafficrates, considerissuesofsequencenumberwrap-aroundandnodefail-
twomobilityrates,andfiveHelloperiods.Exceptforonecom- uresresultinginreusingsequencenumbers.
bination, NXP performed better than a periodic protocol for A neighbor node may be in one of three states: Up, Hold,
dominating-setbroadcastdistribution.NXPeitherhadahigher Down. Hello packets advertise both Up and Hold nodes, in-
deliveryratioandahigherefficiency,orahigherdeliveryratio. dicating the state. A receiveronly uses Up nodes in topology
Flooding had the highest overall delivery ratio in all simula- computations. AnodeintheDownstateisconsidereddiscon-
tions, but generally was very inefficient in terms of the ratio nected and all topology information from it is discarded. A
of packets transmitted to packets received. The exception is nodeintheHoldstateisinahysteresisbetweenUpandDown.
AHBP-EXusingaperiodicprotocolwithalongHelloperiod.
Itappearsthattheout-datedneighborinformationgivesAHBP- A. DataStructures
EX a boost: stale neighbor information allows AHBP-EX to
The NbrPoll table is a simple list of node addresses. All
floodmoreoftenandresultsinhigherdeliveryrates.
nodes listed are candidates for a Poll message when the Poll
timer expires. The NbrTable stores records consisting of {id,
lastHeard, lastHelloSeqno, state, heardCount, nbr}. last-
II. PROTOCOL
Heard tracks the arrivaltime of the last packetfrom a given
We first present a general description of NXP then provide one-hop neighbor. lastHelloSeqno tracks the last valid
pseudo-codeoftheprincipleroutines. Thissectionfocuseson HelloSeqno from the node. state = {Up,Hold,Down} and
theuseofsequencenumberstoaidindirectandindirecttopol- heardCounttracksthenumberofpacketsheardsincethelast
ogyconvergenceandthestatechangesbetweenUp,Hold,and Purge.
Downforneighboringnodes. A Hello packetsent to the MAC layer has the fields {type,
The protocolworks by combining pro-activeHello packets seqno, size, nodeList, stateList, seqnoList}. The field type
that contain one-hop topology information with MAC layer is {Hello,KeepAlive,Poll},indicating the type of packet. The
snooping.NodesonlysendHellopacketsiftheMAClayerhas field seqNois the sendingnode’scurrentHelloSeqno. When
nottransmittedanyotherpacketwithinatimeoutperiod. Inan sending a Hello packet, the source increments HelloSeqno
802.11-styleMAClayer, allpacketsincludingcontrolpackets before sending the packet. For KeepAlive packets, a source
containthesourceaddressandmayserveaskeep-aliveindica- usesitscurrentHelloSeqno.Pollpacketsdonotuseseqno.
torstoneighboringnodes. ThefirstHellosentbyanodehasseqNo=1. sizeisthenum-
Our currentimplementationuses a fixed beaconinginterval ber of entries in the lists nodeList, stateList, se-
inwhichatleastonepacket(userdata,NXPHello,NXPPoll, qnoList.nodeListlistsallone-hopneighborsintheUpor
or NXP Keepalive)must be sent. It would be possible to use Holdstate. ForeachaddressA∈nodeList,seqnoList[A]
containsthecorrespondingHelloSeqnoknownatSforA. This newer,thennodewillqueueaPollformsg.source.Inlines
entry maybe 0, which indicatesthe sender has notreceiveda 14–17,areceivingnodewillqueueaPollpacketifitreceives
Hello from A, but only snooped a packet. stateList[A] apacketfromanunknownnodeandthepacketisnotaHello.
likewiseisthestateforA. ProcessHello() performs indirect detection of topol-
A KeepAlivepackethas only the type and seqNo fields. ogy changes in lines 14 – 16. For each node j ∈ msg.data
It does not repeat topology information. A Poll packet is the listed in a received Hello packet, the receiver checks to see
sameasaHellopacket,butthetypeindicatesitisaPolland if j is a known one-hop neighbor. If so, the receiver com-
itdoesnotuseseqnoListorstateList.Thenodeslisted pares the stored lastHeardSeqnoagainstthe redistributed
innodeListarerequestedtorespondwithaHellopacketat msg.seqnoList. If the received Hello has a more current
theexpirationoftheirnextSendMessagetimer. sequencenumber,thereceiverhasout-of-dateinformationand
willqueueaPollrequestfornodej.
B. NodeInitialization PROCESSMESSAGE(node,msg)
Atinitialization,anodecreatesemptyNbrTableandNbrPoll (1) isnew←false
(2) row←node.NbrTable[msg.source]
tables. It sets the three timers SendMessage, Purge, and Poll.
(3) if rowisNIL
The Recompute timer is only set by protocol events. There
(4) CreatenewrowinUpstate
are three global variables, set as: forceHello = true, re- (5) isnew←true
computePending = false, helloSeqno = 0. NXP also (6) if msgisHello
keepstrackofthenumberofpacketstransmittedbytheMAC. (7) PROCESSHELLO(node,msg,row)
(8) elseif msgisKeepalive
We initializemacUnicast, macBroadcast, macMulti-
(9) if msg.seqno>row.lastHeardSeqno
cast,andmacControltozero.
(10) Addmsg.sourcetonode.NbrPoll
NXPusestheconstantsHoldDownandAgeOuttodetermine (11) elseif msgisPoll
state changes. HoldDown controls when an Up neighbor be- (12) if node.id∈msg.data
comesaHoldneighbor.Weuse1*(Period+maxjitter). Age- (13) node.forceHello←true
(14) else
OutcontrolswhenaHoldneighborbecomesaDownneighbor.
(15) Promiscuoussnoopedpacket
We use 3 * (Period + max jitter). Using a longer HoldDown
(16) if isnew=true
results in significantly less overheadand somewhat lower de- (17) Addmsg.sourcetonode.NbrPoll
liveryrates. The constant HoldCnt determines the number of (18) if isnew=true
consecutivepackets needed to transition a node from Hold to (19) FORCERECOMPUTE(node)
(20) else
Up. WeuseHoldCnt=2.
(21) if row.state=Up
(22) row.lastHeard←now
C. Algorithms (23) elseif row.state=Hold
(24) if now−row.lastHeard≤period+jitter
NXP’sthreeprincipleroutinesareProcessMessage(), (25) row.heardCount++
Timer(), and ProcessHello(). The environment calls (26) else
ProcessMessage()on the arrivalof any packetfor NXP. (27) row.heardCount←1
(28) if row.heardCount≥holdcnt
These include packets addressed to the NXP SAPI and to
(29) row.state←Up
packets snooped by the MAC layer. The environment calls (30) row.lastHeard←now
Timer() at the expiration of any of the four times. Pro- (31) else
cesssHello() is called by ProcessMessage() for (32) Packetfromdownnode
HelloPDUs. (33) row.heardCount←1
(34) row.lastHeard←now
The salient features of Timer() is that a SendMessage
(35) row.state←Hold
timer will only send a packet if forceHello is true or if
theMAClayerstatisticsshowthattheMAClayerhasnotsent
anypacketssince thelast check. NXP compensatesforNXP-
originatedpacketsanddoesnotcountthose.APurgetimerwill III. SIMULATION
onlyforceatopologyrecomputewhenanodetransitionsfrom We conducted several types of simulations using Glo-
Hold to Down. Up to Hold transitions do not cause a recom- MoSim [10] with an 802.11 MAC layer with a 250m trans-
pute. missionrange. Thesimulationsmeasuredthedeliveryratioof
ProcessMessage() and ProcessHello() contain packets received to packets sent and the number of neighbor
the code to pro-actively use Poll and Hello messages in re- over-headpackets. Wesimulatedawidevarietyofconditions,
sponse to topology changes detected through sequence num- whichshowthatNXPhasthesamedeliveryratioasaperiodic
bers. Inlines8 – 10 of ProcessMessage(),the receiving protocol,butwithloweroverhead.
node checksthat the sequence number of a KeepAlivepacket Ineachscenario,therearefoursourcenodes,whichsend500
is the same as the stored node.lastHeardSeqno. If it is byteUDPpacketsfroma Poissonsourcewith meanrate of 2,
TIMER(node,timer) PROCESSHELLO(node,msg,row)
(1) if timerisRecompute (1) if msg.seqno>row.lastHeardSeqno
(2) Send all up node information in node.NbrTable to (2) oldnbrs←row.nbrs
topologyalgorithm. (3) row.nbrs←NIL
(3) elseif timerisSendMessage (4) foundme←false
(4) if (node.forceHello=true)or(MAClayerquiet) (5) row.lastHeardSeqno←msg.seqno
(5) hello.type=hello (6) foreach {n,state,seq}∈msg.data
(6) node.seqno++ (7) if state=Up
(7) hello.seqno=node.seqno (8) Addntorow.nbrslist
(8) For each {id,state,seqno} ∈ node.NbrTable (9) if n=node.id
wherestateisUporHold,addtohello.data. (10) foundme←true
(9) Transmithello. (11) if (seq<node.seqno)or(state=Hold)
(10) SetSendMessagetimer (12) node.forceHello←true
(11) elseif timerisPoll (13) else
(12) PollList←NIL (14) row2←node.NbrTable[n]
(13) foreach row∈node.NbrPoll (15) if (row2 notNIL) and(seqno >
(14) if row.state(cid:8)=Down row2.lastHeardSeqno)
(15) Addrow.idtoPollList (16) Addnodentonode.NbrPoll
(16) ifPollListnotNIL (17) if foundme=false
(17) Create Hello packet with hello.data ← (18) node.forcehello←true
{PollList,0,0} and hello.seqno ← 0. (19) if row.nbrs(cid:8)=oldnbrs
Transmit. (20) FORCERECOMPUTE(node)
(18) SetPolltimer
(19) elseif timerisPurge
(20) foreach row∈node.NbrTable Description Values
(21) if row.state=Up
(22) if now−row.lastHeard≥HoldDown Helloperiod(sec) 0.5,1,2,4,8
(23) row.state←Hold Distribution Flood,DP,AHBP,AHBP-EX
(24) row.heardCount←0 NeighborProto None,Periodic,NXP
(25) elseif row.state=Hold Sourcerate(pps) 2,10,20,40
(26) if now−row.lastHeard≥Ageout Mobility 1-10m/sor5-20m/s
(27) row.state←Down
Pattern randomwaypoint,0secpause
(28) row.heardCount←0
Simtime 300sec
(29) row.nbrs←NIL
(30) FORCERECOMPUTE(node) TABLEI
SIMULATIONVARIABLES
10, 20,and40packetspersecondpernode, dependingonthe
simulationscenario.Nodes1and2wereonfrom[0s-60s]and
except Recompute is ±1/5th the SendMessage length. The
[120s-180s].Nodes3and4wereonfrom[30s-90s]and[150s
Recompute timer is deterministic. The Distribution methods
-210s]. Repeaternodesaddarandomexponentialdelaywhen
are Distributed Pruning (DP) [3], Ad-hoc Broadcast Protocol
repeatingpackets,withmeandelayof5milliseconds.
(AHBP)[5], andblindfloodingwith packetcache(Flood). In
Thereare fivegeneralscenarios: base case (BC), high den-
thesimulationruns,wedenoteAHBPasAHandAHBP-EXas
sity(HD),highmobility(HM),groupmobility(GM),andlow
density (LD), all being variations on parameters listed in Ta- AX.TheNeighborProtocolsarePeriodic(PR)andNXP(NX).
Note that flooding does not use a neighbor protocol and does
ble I. The BC scenario uses 50 nodes with velocities 1 – 10
m/splacedatrandominthesimulationspace.TheHDscenario notdependonHelloperiods.
uses100nodes.TheHMscenariouses50nodeswithvelocities We have summarized the behavior of broadcast flooding in
5–20m/s.TheGMscenarioisBCwithacircularlysymmetric to two metrics with the independent variable being the Hello
nodedistributionoffourringswithradiiof50m,150m,400m, period. TheDeliveryRatioisthetotalnumberofuniquepack-
and 750m. There were 5, 10, 16, and 19 nodes in each ring, etsreceivedbyallnodesdividedbythetotalnumberofpackets
respectively. The GM scenarios model high noise/low noise originated by all nodes. The Inefficiency is the total number
environmentsbygroupingnodes. TheLD scenarioisBC, but ofpacketstransmittedbyallnodesdividedbythetotalnumber
with25nodes. ofuniquepacketsreceivedbyallnodesnormalizedby(divided
Eachscenarioisrunwith123configurations,basedonvari- by) the total number of packets originated by all sources. In
ations of Table I. The Hello period is the setting for the computingtheaverages,weconsiderasourcenodeoriginating
SendMessagetimer. The Purgetimer is set to the same mean packetstoreceiveallthosepackets
as the SendMessage timer. The Poll timer mean is 1.5 times Basedonoursimulationparameters,floodingalwayshadthe
longer than the SendMessage timer. The Recompute timer is bestdeliveryratio. Fig. 1 plotstheDeliveryRatio for the five
1/20th the SendMessage timer. The jitter in timer all timers Helloperiods. AstheHelloperiodincreases,thedeliveryratio
0.46 0.22
0.44 0.2
0.42
0.18
0.4
0.16
Delivery ratio 00..3368 Inefficiency 0.14
0.12
0.34
0.1
0.32 AH-PR AH-PR
AH-NX AH-NX
AX-PR AX-PR
0.3 DAPX--PNRX 0.08 DAPX--PNRX
DP-NX DP-NX
Flood Flood
0.28 0.06
0.5 1.0 2.0 4.0 8.0 0.5 1.0 2.0 4.0 8.0
Hello Period (sec) Hello Period (sec)
Fig.1. Deliveryratio Fig.2. Inefficiency
of the dominating-setdistributionprotocolsdecreases. Flood- REFERENCES
ing is invariantto Hello period as it does not use a neighbor [1] JieWu,MingGao,andI.Stojmenovic,“Oncalculatingpower-awarecon-
protocol. At long Hello periods (8 seconds), AHBP-EX with necteddominatingsetsforefficientroutinginadhocwirelessnetworks,”
inProc.InternationalConf.ParallelProcessing,Sept.2001,pp.346–54.
a Periodic protocolhas a higher deliveryratio. Stale one-hop
[2] Sze-YaoNi,Yu-CheeTseng,Yuh-ShyanChen,andJang-PingSheu,“The
informationallowsAHBP-EX to flood morepacketsresulting broadcaststormprobleminamobileadhocnetwork,”MobiCom’99,Aug
inahigherdeliveryratio. 1999.
[3] H.LimandC.Kim, “Floodinginwirelessadhocnetworks,” Computer
Intermsofefficiency(inefficiencyinourplots),Fig.2shows
Communications,vol.24,no.3-4,pp.353–363,Feb2001.
that for very short Hello periods, dominating-set distribution [4] WeiPengandXi-ChengLu, “Onthereductionofbroadcastredundancy
withneighborprotocolsislessefficientthanfloodingandfrom in mobile ad hoc networks,” in 2000 FirstAnnual Workshop on Mo-
bileandAdHocNetworkingandComputingMobiHOC.Sigmobile,Aug.
Fig.1hasalowerdeliveryrate. ForHelloperiodsof1second 2000,pp.129–30.
or longer, the adaptive protocols become more efficient than [5] W.PengandX.Lu, “AHBP:Anefficientbroadcastprotocolformobile
adhocnetworks,”JournalofComputerScienceandTechnology,vol.16,
Flooding.A1secondHelloperiodhasalmostthesamedelivery
no.2,pp.114–125,Mar2001.
ratioasfloodingatahigherefficiency. [6] S.Basagni,I.Chlamtac,V.R.Syrotiuk,andB.A.Woodward,“Adistance
The absolute valuesof results vary considerablydepending routingeffectalgorithmformobility(DREAM),” inMobiCom’98,New
York,NY,USA,Oct.1998,IEEE,pp.76–84,ACM.
onthescenarios(BC,GM,HD,HM,LD),butthegeneraltrends
[7] G.Karumanchi,S.Muralidharan,andR.Prakash, “Informationdissem-
are similar. The notable exception is that in the HD and GM ination in partitionable mobile ad hoc networks,” in Proc.18th IEEE
configurations,dominating-setdistributionhavehigherdelivery Symp.ReliableDistributedSystems,Oct.1999,pp.4–13.
[8] I.Stojmenovic,M.Seddigh,andJ.Zunic,“Dominatingsetsandneighbor
ratiosthanfloodingforHelloperiodsof2secondsandunderfor
elimination-based broadcastingalgorithmsinwirelessnetworks,” IEEE
highpacketrates(morethan10pps). TransactionsonParallelandDistributedSystems,vol.13,no.1,pp.14–
25,Jan.2002.
[9] R.G.Ogier,F.L.Templin,B.Bellur,andM.G.Lewis, “Topologybroad-
IV. CONCLUSION cast based onreverse-path forwarding (TBRPF),” IETFInternet draft,
WehavepresentedtheNeighborExchangeProtocol(NXP), draft-ietf-manet-tbrpf-05.txt,Mar2001.
[10] L. Bajaj, M. Takai, R. Ahuja, K. Tang, R. Bagrodia, and M. Gerla,
which uses variablefrequencyHello packetsto maintain two- “GloMoSim: Ascalable networksimulationenvironment,” Tech.Rep.
hoptopologyinformationinanad-hocnetwork.NXPalsouses 990027,UCLAComputerScienceDepartment,1999.
small KeepAlivepacketsrather thanrebroadcastfulltopology
Hellopacketsiftherehasbeennochangetotopologywithina
timeoutperiod. By redistributingsequencenumbersand node
statesinHellopackets,NXPmaydetecttopologyinconsisten-
cies.
NXPusesandredistributessequencenumbersandhasathree
state (up,hold, down)neighbormodel. This allowsdirectde-
tectionof inconsistencies. A Poll mechanism allowsnodesto
requestfull-topologyHellopacketswhenneeded.
SimulationexperimentsshowedthatNXPyieldsdeliveryra-
tiosthatareinmostcasesasgoodorbetterthanthoseattained
byaperiodicprotocol. Wherethedeliveryratioislower,NXP
also has a higher efficiency. An optimization to the protocol
wouldbeto piggybackPoll packetson Hello/Keepalivepack-
ets,whichwouldfurtherreducetheprotocoloverhead.