HARP - HYBRID AD HOC ROUTING PROTOCOL

Navid Nikaein, Christian Bonnet and Neda NikaeinInstitut Eurecom

2229, Route des CrˆetesB.P. 193

06904 Sophia Antipolis, FranceFirstname.Name@eurecom.fr

Abstract— This paper presents a bandwidth-efficient low-delay routing protocol for mobile adhoc networks called HARP - hybrid ad hoc rout-ing protocol. HARP is a hybrid scheme combin-ing reactive and proactive approaches. The routingis performed on two levels: intra-zone and inter-zone, depending on whether the destination be-longs to the same zone as the forwarding node. Wepropose a new architecture that separates topologycreation from route determination. This architec-ture optimizes routing performance according totwo criteria: network properties and applicationrequirements. Topology creation generates a log-ical structure with respect to network properties,and the routing protocol discovers and maintainspaths to satisfy application requirements.

Keywords—Ad hoc networks, routing, zone, sta-bility, graph terminology.

I. I NTRODUCTION

A mobile ad hoc networkmanet is a set ofwireless mobile nodes forming a dynamic au-tonomous network through afully mobile in-frastructure [1]. Nodes communicate with eachother without the intervention of centralized ac-cess points or base stations. In such a network,each node acts both as a router and as a host.Due to the limited transmission range of wirelessnetwork interfaces, multiplehops may be neededto exchange data between nodes in the network,which is why the literature sometimes uses theterm multi-hop network for a manet. Manet wasfirst referred to as apacket radio network in themid-1960 [2][3]. A mobile ad hoc network in-cludes several advantages over traditional wire-less networks, including: ease of deployment,speed of deployment, and decreased dependenceon a fixed infrastructure. Manet is attractive be-cause it provides an instant network formationwithout the presence of fixed base stations andsystem administrators. Many critical issues haveto be addressed in manet such as unicast and mul-ticast routing, QoS support, power control, secu-rity, etc. This paper deals with the problem ofunicast routing in the mobile ad hoc network.

Designing routing protocols in a mobile ad hocnetwork is different from wireless networks dueto its fully mobile infrastructure, which affectsmobility management. In the literature relatedto routing protocols used in manets, there existsthree main routing mechanisms:proactive, reac-tive andhybrid. In the proactive or table-drivenapproach, each node keeps up-to-date routes toevery other node in the network in its routing ta-bles. Routing information is periodically trans-mitted throughout the network in order to main-tain routing table consistency. In the reactiveor on-demand approach, a node initiates a routediscovery procedure only when it wants to com-municate with its destination. Once a route isestablished, it is maintained by a route main-tenance process until either the destination be-comes inaccessible or until the route is no longerdesired. In the hybrid approach, each node main-tains only routing information for those nodesthat are within its zone and its neighboring zones.That is, it exhibits proactive behavior within azone, and reactive behavior between zones. Thesize and dynamics of a zone differ from proto-col to protocol. Consequently in hybrid schemes,a route to each destination within a zone is es-tablished without delay, while a route discoveryand a route maintenance procedure are requiredfor every other destination. We propose a hybridrouting protocol as its name indicates: HARP -hybrid ad hoc routing protocol.

The paper is organized as follows. Section IIprovides both related works and our observationsabout different routing mechanisms. Section IIIlocates our contributions through an architecturethat separates topology creation from route deter-mination. Section IV reviews an algorithm that isused in HARP to build a logical zone hierarchyof nodes; and it is called DDR - distributed dy-namic routing [4]. Then section V presents dif-ferent phases of HARP, followed by some math-ematical analysises. Finally, section VII providesconcluding remarks and highlights future works.

II. RELATED WORK AND OUR

OBSERVATIONS

Several routing protocols have been proposedwith the goal of achieving efficiency. Certaintable-driven or proactive routing protocols are[5], [6], [7], [8], [9]. The proactive protocolsdecrease the delay of route determination to adestination, but they waste a significant amountof scarce wireless resources in order to main-tain up-to-date routing tables. Such protocolsare scalable in relation to the frequency of end-to-end connection. Although proactive protocolsare not scalable in relation to the total numberof nodes, they can be made scalable if a hier-archical architecture is used. Finally, proactiveprotocols are not scalable in relation to the fre-quency of topology change. On the other handamong the on-demand or reactive routing proto-cols, we can find [10], [11], [12], [13], [14], [15],[16], [17]. The reactive protocols decrease thecommunication overhead at the expense of an ex-tra delay for route determination; and they arenot optimal in terms of bandwidth utilization be-cause of the flooding nature of the route discov-ery. Reactive protocols remain scalable in rela-tion to the frequency of topology change. Suchprotocols are not scalable in relation to the totalnumber of nodes, however, similarly to proacticeapproaches they can be made scalable if a hierar-chical architecture is used. Finally, reactive pro-tocols are not scalable in relation to the frequencyof end-to-end connection. The hybrid protocolscombine proactive and reactive features; and wecan find zone routing protocol (ZRP) [18] andzone-based hierarchical link state (ZHLS) rout-ing protocol [19]. The hybrid protocols can pro-vide a better trade-off between communicationoverhead and delay, but this trade-off is sub-jected to the size of a zone and the dynamics ofa zone. Furthermore, hybrid approaches providea compromise on scalability issue in relation tothe frequency of end-to-end connection, the totalnumber of nodes, and the frequency of topologychange.

We propose a zone level hierarchical routingprotocol, denoted by HARP - hybrid ad hoc rout-ing protocol. In HARP, each node maintainsonly routing information of those nodes that arewithin its zone, and its neighboring zones. Therouting is performed on two levels: intra-zoneand inter-zone, depending on whether the desti-nation belongs to the same zone as the forward-

ing node. Intra-zone routing relies on an exist-ing proactive mechanism, and HARP includesreactive mechanism for the inter-zone routing.Zone creation and proactive behavior in relationto network properties are provided by DDR - dis-tributed dynamic routing [4]. On the other hand,HARP is responsible for discovering and main-taining paths to satisfy application requirements.HARP generates and selects path(s) according tothe notion ofzone level stability, which is an ex-tension of node level stability previously used in[14]. The zone level stability is defined by theconnection stability of a zone regarding its neigh-boring zone. HARP appliesearly route mainte-nance regarding the degree of zone stability. Thatis, HARP avoids the extra delay caused by pathfailure during data transmission, and refresh thepath before instability period.

Similar to ZRP and ZHLS, HARP is a hybridapproach based on the notion of zone. Differentfrom ZRP and ZHLS, HARP only concerns withfinding and maintaining a path between sourceand destination, and leaves topology generationto DDR - distributed dynamic routing [4]. Thisseparation simplifies the routing protocol, andmakes the design modular. Different from ZHLSand ZRP, HARP limits the flooding to subset offorwarding nodes in each zone. This reduces bothbandwidth utilization, and energy consumptionof non-forwarding nodes. HARP applies zonelevel stability as a metrics of route determinationwhich is not the case in ZRP and ZHLS. Unlikeprevious routing protocol, HARP appliesearlypath maintenance which is more suitable for pri-ority classes.

III. HARP AND DDR

HARP finds and maintains a path in orderto route user data regarding application require-ments while DDR generates a logical structurewith respect to network properties. The appli-cation requirements include delay, loss rate, sta-bility, jitter, etc; and the network properties arenumber of nodes in the network, frequency ofend-to-end connection (i.e. number of communi-cation), and frequency of topology change. Thenetwork varies from a group of sensors to a groupof cars , i.e. from no mobility at all to high mobil-ity; and the application differs in types of the traf-fic, e.g. data, audio, video. HARP and DDR com-municate with each other to satisfy both applica-tion requirements and network properties. Fig. 1shows a layered view of HARP and DDR.

HARP RoutingProtocol ...

Application

Network

DDR

Fig. 1. Layered View of HARP and DDR

IV. D ISTRIBUTED DYNAMIC ROUTING

ALGORITHM

A. Basic Idea

The main idea of the DDR algorithm is toconstruct a forest from a network topology ina distributed way by using only periodic mes-sage exchanges between nodes and their neigh-bors. Each tree of the constructed forest forms azone.1Then, the network is partitioned into a setof non-overlapping dynamic zones. Each zone isconnected via the nodes that are not in the sametree but they are in the direct transmission rangeof each other. So the whole network can be seenas a set of connected zones. Each node is as-sumed to maintain routing informationonly tothose nodes that are within its zone, and infor-mation regardingonly its neighboring zones.

B. General Description

DDR combines two classical notionsforestand zone. Forest is previously used in DST -distributed spanning tress for routing in mobilead hoc networks [17]. Zone is also used in zonerouting protocol (ZRP) [18], and zone-based hier-archical link state (ZHLS) routing protocol [19].The combination of these two classical notions,zone and forest, provides us with an appropriatestructure which in turn can give us a better trade-off between delay and communication overhead.Although DDR benefits from classical conceptslike zone and forest, unlike previous solutions itachieves several goals at the same time. Firstly,it provides different mechanisms to reduce rout-ing complexity and improve delay performance.Secondly, it is a fully mobile infrastructure in astrong sense: it does not even require a physicalor global location information. Finally broadcastis reduced noticeably.

The DDR - algorithm consists of six cyclictime-ordered phases:preferred neighbor election,forest construction, intra-tree clustering, inter-tree clustering, zone naming and zone partition-

�We will use the terms tree and zone interchangeably.

4

2

e

q

i p

t

yx

vu

z

m1 3

1

1

2

4

4

3

2

3

5

1

3

22

3

2 3

4

34

3

4

3 3

2

n

k

b

l

o

hc

r

a

w

T

dj

gG

fs

(a) Network Topology

z

z4

z

z

em

x

u

n

o

l

1

3

z67

2

5

h

a

z

f

z

d

y

b

sk

c

r

gG

v

T

j

wz

pi

qt

(b) Network Topology under DDR

xy

Name of zone i Z

Bridge

i

Edge of graph G

Edge of Forest F

Gateway node g

Node x with degree y

g

(c) Legend

Fig. 2. DDR Infrastructure

ing, which are executed based on the informa-tion provided by beacons. Abeacon is a periodicmessage exchangedonly between a node and itsneighboring nodes. The content of a beacon isprimitive at the beginning, and it will be enrichedduring each phase of the algorithm. At the be-ginning, each node in the network topology car-ries out the preferred neighbor election algorithmto choose a preferred neighbor. The preferredneighbor of a node is the node that owns max-imum neighborhood degree among neighboringnodes. Then, a forest is constructed by connect-ing each node to its preferred neighbor and vice

versa. It has been proven that whatever is the net-work topology, connecting each node to its pre-ferred neighbor always yields a forest (i.e. wehave no cycle) [4]. Next, the intra-tree cluster-ing algorithm is carried out in order to give anappropriate structure within a zone, and build theintra-zone routing table. After that, the inter-treeclustering algorithm provides a natural structureamong zones which is kept in the inter-zone rout-ing table of every gateway node. Gateway nodesare the nodes that are not in the same zone, but inthe direct transmission range of each other.2Eachtree is assigned with a name by executing thezone naming algorithm. Since the constructedforest contains a set of trees where each tree isassigned with a name, then the network is parti-tioned to a set of non-overlapping dynamic zones.Note that DDR only uses beacons to perform ev-ery phase of the algorithm, e.g. the forest con-struction, the intra-tree clustering, the inter-treeclustering, etc. Therefore, it avoids global broad-cast throughout the network, thus causing a moreefficient use of radio resources.

Fig. 2(a) represents an arbitrary network topol-ogy. Once DDR algorithm is executed on eachmobile node, the network is partitioned into a setof non over-lapping dynamic zones, as it is il-lustrated in Fig. 2(b). Each node in the networkmaintains two tables: the intra-zone table and theinter-zone table. The intra-zone table keeps theinformation within a zone, and it is filled dur-ing the intra-tree clustering algorithm. It containstwo fields: node ID numberNID, and learned pre-ferred neighborsLEARNED PN. The field NID

represents the ID number of a node that formsan edge of the forest with the owner of the tabledirectly. The fieldLEARNED PN represents thenodes that are reachable indirectly via their as-sociatedNID in the intra-zone table. Thereforethe NID indicates the next hop for the nodes intheLEARNED PN. Table I (a) and I (b) depict theintra-zone table of nodek ands belonging to thezonez� regarding Fig. 2(b), and they are denotedby Intra ZTk andIntra ZTs respectively. Theintra-zone table gives the current view of a nodeconcerning its zone, and it is updated upon re-ceiving beacons.

In contrast to the intra-zone table, the inter-zone table keeps the information concerning

�There exist two kinds of gatewaynodes: out-gateway andin-gateway, regarding whether a packet leaves a zone or en-ters to a zone.

TABLE IINTRA-ZONE TABLE OF NODESk AND s

REGARDING FIG. 2(B)

NID LEARNED PN

s f� b� q� y� t� xc -g -

(a) Intra-zone table ofnodek � Intra ZTk

NID LEARNED PN

y x� tk c� g

b� f� q -

(b) Intra-zone table ofnodes � Intra ZTs

neighboring zones. This table represents thebridges3, which are detected during the executionof the inter-tree clustering algorithm. Table IIshows the inter-zone table of nodeg, which is de-noted byInter ZTg. Each entry inInter ZTgcontains the ID number of a gateway nodeGNID,the zone ID of this gateway node, i.e. neighbor-ing ZID NZID, and the stability of this neighbor-ing zone regarding nodex Z STABILITY . Thezone stability is defined by the connection sta-bility of a zone regarding its neighboring zone.For each beacon received, the zone level stabil-ity of the current zone with respect to the bea-coning zone is incremented if the euclidean dis-tance of the current ZID and the old ZID becomessmaller than a critical distance; otherwise it is re-set. The zone stability is directly related to theZID which is assigned during the zone namingphase. Indeed, the ZID determination is basedon randomly chosen NIDs in a zone. It thereforeidentifies the zone and it can simply reflect thezone stability.

TABLE IIINTER-ZONE TABLE OF NODEg � Inter ZTg

GNID NZID Z STABILITY

r z� ++G z� ++

Therefore, as it is shown in Fig. 2(b) the wholenetwork can be seen as a set of connected zoneswhere each node can communicate with anothernode in the network.

�A bridge is an edge that connects two gateway nodes.

V. HYBRID AD HOC ROUTING PROTOCOL

A. Basic Idea

HARP aims at establishing the most stablepath from a source to a destination in order toimprove delay performance due to path failure.HARP applies the path discovery mechanism be-tween zones that intends to limit flooding in thenetwork, and that filters the candidate paths assoon as possible according to the stability cri-teria. As stability is the most desired parame-ter, HARP offers different mechanisms to antic-ipate path failure along with path maintenanceprocedure whose complexity is reduced by theproactive nature of the routing algorithm withina zone. These procedures reduce the delay thatstems from a path failure during data transmis-sion.

B. Routing Mechanism

The routing mechanism in HARP is performedon two levels: intra-zone and inter-zone, depend-ing on whether the destination belongs to thesame zone as the forwarding node. The intra-zone routing involves onlyforwarding becauseHARP applies proactive approach within a zone(inherited from DDR), which means that routegeneration and selection are performed during theintra-tree clustering phase of DDR. Therefore theonly task of a node within a zone is toforwardthe data traffic along a pre-computed path. Theinter-zone routing impliesrouting since HARPuses a reactive approach between zones to gen-erate routes, and selects the most stable one tothe destination. The inter-zone routing includes:path discovery andpath maintenance phases inorder to discover and maintain a path. The pathdiscovery phase consists of two parts: path re-questPREP and path replyPREP. The path re-quest propagates from zone to zone via the gate-way nodes thanks to the both intra-zone table andinter-zone table, while the path reply is unicastedback from the destination to the source. Inside azone, the path request follows the tree structureprovided by DDR. As a consequence, the pathrequest propagation is limited to a subset of for-warding nodes. After this limited flooding, sev-eral paths may candidate for a given destination.The destination chooses the most stable path andsends a path reply back to the source. Path main-tenance provides different mechanisms to ensurethat packets can be safely transmitted from source

to destination. Each path is associated with a re-fresh time after which a new path discovery phaseis triggered. This is done to avoid path failure asthe network topology may change after a certaintime. Nodes may also have unanticipated behav-ior that may cause path failure. In this case a re-active path recovery procedure is triggered in ad-dition to the previous mechanism, which can beseen as a proactive path recovery.

B.1 Path Discovery Algorithm

The main objective of the path discovery al-gorithm is to generate and select the most sta-ble path between source and destination. Notethat the stability is a concave function, and it isdefined by the connection stability of a zone re-garding its neighboring zone. Assume that a nodesays wants to send data to its destinationd. Be-fore sending data to the noded, nodes checksif node d exists in its intra-zone table. If so,nodes forwards the data towards noded accord-ing to its intra-zone table without any delay. Wedenote this case intra-zone routing. Otherwise,noded belongs to a different zone as nodes, sothat nodes sends a path requestPREQ to everyother neighboring zonez via gateway nodesg.We denote this caseinter-zone routing. The sta-bility is unknown at the beginning, but theout-gateway nodes will change the stability value asthey propagatePREQmessage from zone to zone.Each intermediate node routesPREQ through itszone up to out-gateway nodes according to itsintra-zone table. In contrast, each out-gatewaynode forwardsPREQ from its zone to the nextzone(s) according to its inter-zone table, and up-dates the current stability of the path. Upon re-ceiving thePREQ by an in-gateway nodeg, thisnode verifies whether destination noded belongsto its intra-zone table or not. If so,g forwards thisPREQaccording to its intra-zone table; otherwiseg routes thisPREQ throughout its zone, and pos-sibly to every other neighboring zones. In orderto assure that the path replyPREP message tra-verse exactly the same path back to the source,each gateway node keeps somerouting state in-formation found inPREQandPREPfor each traf-fic. This routing state information includes: gate-way ID from which a path request is received (orcorresponding gateway), and path stability. Thedestination noded will eventually receivePREQ

if it is reachable by the source. Then nodedchooses the most stable path among the received

path requests. If more than one path with thesame stability is found, noded chooses one ran-domly. Afterward, noded creates a path replyPREPcorresponding to the most stable path, androutes thisPREPback to the gateway from whichit received thePREQ. Route reply includes thediscovered stability which helps the source nodeduring the path maintenance procedure (c.f. V-B.2). Each node routes back thePREP towardsthe corresponding gateway node according to theeither zone tables and the routing state informa-tion. Once thePREParrives at the source nodes,s creates data packets, and sends them throughthe discovered path. Consequently, each nodeforwards the data packet to the next zone via thecorresponding gateway towards the destinationaccording to the both intra-zone table and inter-zone table. Furthermore, a gateway node filtersa path request if it receives a path request with alower degree of stability than the earlier one forthe same connection; and it updates the routingstate information if it receives a path request witha higher degree of stability. This filtering is calledselective filtering, and guarantees the uniquenessof the most stable pair of gateway nodes.

B.2 Path Maintenance Algorithm

The goal of the path maintenance algorithm isto improve the delay performance based on thepath stability. The path maintenance algorithmincludes: path refreshment, path waiting time,path error, andnew path discovery phases. Thepath refreshment phase constructs anew path be-fore a period of time calledpath discovery up-date time, and then switches the traffic to this newpath at the beginning of the update time. Thispath discovery update time is estimated based onthe discovered path stability. Although the sourcecan keep communicating with the destination af-ter the update time, it may confront high prob-ability of link failures. Therefore, source noderenews the path discovery phase if it approachesto the path discovery update time. If a link failureoccurs in the meantime, the corresponding nodeholds the traffic for a duration ofwaiting time ex-pecting to receive some routing information cor-responding to its target gateway node, and at thesame time it sends apath error back to the source.The rationale for waiting time is that HARP ap-plies the proactive approach inside a zone, andthere is a chance of receiving new routing infor-mation embedded in the periodical beacon. Fur-

thermore, source node benefits from the discov-ered path stability to get the actual stability of ev-ery zone during data transmission. Note that thestability is a concave function. For this purpose,the source node puts the discovered stability indata packets so that each out-gateway node canverify whether this stability satisfies the actualstability of the next zone. If not, this out-gatewaynode sends back apath error with the actual sta-bility to the source node so that the source can up-date the path discovery update time accordingly.The path error is unicasted back to the source ifthe return path still exists; otherwise it is broad-casted to the network with an appropriate time-to-live (TTL). Once the source receives a path error,it initiates anew path discovery.

B.3 Example

Fig. 3 depicts the format of bothpathrequest PREQ and path reply PREP mes-sages; they include a source addresssrc@,a destination addressdest@, a port numberport#, a gateway node IDGID, and the es-timated stability stability. The stability isa concave function, and it is calculated bymeans ofPREQ.stability = minf PREQ.stability,Inter ZTout gateway.Z STABILITY .

src@ dest@ port# GID stability

Fig. 3. Fields of a path requestPREQor a path replyPREP

Fig. 4 shows the fields of adata packet DPKT.The fields of a data packet includes a source ad-dresssrc@, a destination addressdest@, a portnumberport#, a gateway node ID numberGID,the discovered stabilitystability, and the datadata.

src@ dest@ port# GID stability data

Fig. 4. Fields of a data packetDPKT

For example in Fig. 5, consider the scenarioof intra-zone routing where nodes wants tocommunicate with one of the nodes within itszone z�, e.g. f� b� q� y� k� c� g� x� t. Accordingto its intra-zone table (see table I(b)), nodescan reach the nodesx� t via y, and the nodesc� g throughk, while other nodesf� b� k are di-rectly reachable. Therefore intra-zone routing ta-ble always indicates the next hop for each des-tination within the zone. So if nodes wants

to send some data to nodec, it firstly for-wards DPKT�s�� c�� p�� k���� data� includ-ing the date to the nodek and thenk passesDPKT�s�� c�� p�� c�� data� to the nodec (seedash-dot-dot line betweens andc in the figure).

z

z4

z

z

em

r

bu

n

o

l

1

3

z67

2

5

h

a

G

z

f

z y

k

d

s

preq

preq preq

preq

preq

prep

prepprep

preq

prep

data

preq

preq

cg

x

tq

zw

j

T

v

i p

(a) Routing under HARP

Path Request

Path Reply

preq

prep

Date

(b) Legend

Fig. 5. Routing and forwarding in HARP : path dis-covery, path maintenance, and data traffic delivery

Inter-zone routing occurs if the destinationnode belongs to a different zone as the sourcenode. For instance in Fig. 5, source nodesis a member of zonez� while destination noded belongs to zonez�. Therefore nodes willsend a path requestPREQ�s�� d�� p�� s����through its zone. Note that at the beginningthe source node sets the field ofGID to its ad-dresss�, and thestability to �. This PREQ

crosses from the zonez� to the zonesz�� z�� z�via gateway nodesc� g� x� q� t respectively; as itis illustrated in the figure (follow the solid ar-row lines froms to d). These gateway nodes up-date the field ofGID to their ID, and store theold value ofGID by means of correspondinggate= s@. Also, these nodes forward thePREQ totheir neighboring zones according to their inter-zone table: i.e. via gatewaysr� G� v� i� p respec-tively. Now thePREQ traverses the zonesz�� z�to the zonez� via z� T . Note that nodeT ap-

plies selective filtering to one of the path re-quests received fromG andv. Assume that nodeT chooses nodeG as the most stable gatewaynode. Finally thePREQenters the zonez� wherenoded belongs to viaw� j. The nodesw� j for-wards thisPREQ to d according to their intra-zone routing table. Upon receiving thisPREQ

by destination noded, d will choose the moststable one. Assume the path viad� j� T�G� g� sas the most stable path, so that noded createsthe path reply corresponding to the chosen pathPREP�d�� s�� p�� j�� stability�; where thej�indicates the address of the gateway in whichthe chosen path has just been received, and thestability points to the whole path stability (in thefigure follow the dashed arrow line fromd to s).Nodej updates the fields ofsrc@ andGID in thepath reply toPREP�j�� s�� p�� T�� stability�,and saves the old value ofsrc@ by means of cor-respondinggate=@d. The routing state informa-tion at each gateway node points to the next gate-way towards the destination at the time of datapacket transmission. NodeT also updates thepath reply and pass thePREP to nodeG insteadof nodev since nodev has been filtered out be-cause of the stability priority. Each intermediatenode routes thePREPto the next gateway foundin PREP according to its intra-zone table. OncePREParrives at the source nodes, s creates datapacketsDPKT�s�� d�� p�� g�� stability� data�;whereg� is the gateway node through which thepath reply has been received and thestability in-dicates the discovered stability during the pathdiscovery. Then, nodes sends theDPKT to thedestination for the duration of the path discoveryupdate time (follow the dash-dot-dot line in thefigure). This path is refreshed ifs still wants tosend packets tod.

VI. M ATHEMATICAL ANALYSIS

We compare the communication overhead fortopology creation among DDR, ZHLS and ZRP.Assume that all the nodes are uniformly dis-tributed in the network. Consider a network withN nodes andM zones. Let� be the routing zoneradius in term of number of hops, therefore�� isthe amount of intrazone control traffic producedby each node. The total amounts of communica-tion overhead generated by ZHLS, ZRP and DDRare summarized in TABLE III.

It can be shown that the communication over-head (C.O.) ofC�O�DDR andC�O�ZRP are close,

TABLE IIICOMMUNICATION OVERHEAD COMPARISON

Protocol Topology creationZHLS N��M �NMZRP N��

DDR N��M

but it is subjected to the choice of� and M .ZHLS generatesNM messages more than DDRfor creating topology, when the number of zonesare the same. The communication overhead gen-erated to perform route discovery operation forZHLS, ZRP, and HARP is:C�O�RD � P �N �Y �, whereP � � � ��M is the probabilitythat destination belongs to a different zone as thesource, andY is the total number of nodes af-fected by the directed path where the reply packettransits. For the same number of zones, rout-ing overhead generated by ZHLS and HARP areclose. It can be shown that HARP generates lessrouting overhead than ZRP, because zones (Nzones) in ZRP are highly overlapped. There-fore, the overall overhead generated by HARPand DDR is less than both protocols.

VII. C ONCLUSION

This paper presents a hybrid routing protocol,called HARP, which benefits from the separationof logical topology creation/maintenance tailoredto the routing protocol. This separation leadsto a clear and modular design of routing pro-tocols according to specific application require-ments. HARP establishes and maintains the moststable path from source to the destination in or-der to enhance the reliability of data transmissionin mobile ad hoc networks. HARP uses a hier-archical topology provided by DDR in order toreduce the control message overhead. The mathe-matical analysis shows that the overall amount ofoverhead induced by HARP and DDR is smallerthan ZRP and ZHLS. In our future work, we willevaluate the performance of HARP under variouscondition of traffic and mobility with the relatedprotocols.

ACKNOWLEDGMENT

The authors wish to thank Shiyi Wu, DavidTurner and Sergio Loureiro for their useful dis-cussions.

REFERENCES

[1] “Mobile ad hoc network(MANET),” Web Site at:http � ��www�ietf�org�html�charters�manet �charter�html, Work in Progress.

[2] �, “Special issue on packet radio networks,”Proceed-ing of IEEE, vol. 75, no. 1, 1987.

[3] “Packet radio networks,” Web Site: http �

��www�tapr�org�tapr�html�pktf�html.[4] Na. Nikaein, H. Labiod, and C. Bonnet, “DDR-

distributed dynamic routing algorithm for mobile adhoc networks,” inMobiHOC 2000. IEEE, Augest2000, pp. 19–27.

[5] C. E. Perkins and P. Bhagwat, “Highly dynamic desti-nation sequenceddistance-vector routing(DSDV) formobile computers,” inSIGCOMM’94.

[6] S. Murthy and J.J. Garcia-Luna-Aceves, “An efficientrouting protocol for wireless networks,”ACM MobileNetworks and Applications Journal, 1996.

[7] T. Chen and M. Gerla, “Global state routing: A newrouting scheme for ad-hoc wireless networks,” inProc. IEEE ICC’98.

[8] C-C. Chiang, “Routing in clustered multihop, mo-bile wireless networks with fading channel,” inIEEESICON’97, pp. 197–211.

[9] A. Iwata, C-C. Chiang, G. Peiand M. Gerla, and T.-W.Chen, “Scalable routing strategies for ad hoc wirelessnetworks,” IEEE Journal on Selected Areas in Com-munications, Special Issue on Ad-Hoc Networks, vol.17, no. 8, pp. 1369–1379, 1999.

[10] M. Jiang, J. Li, and Y. C. Tay, “Cluster based routingprotocol,” IETF Draft, 1999.

[11] C. E. Perkins, E. M. Royer, and S. R. Das, “Ad hoc on-demand distance vector routing,” IETF Draft, 1999.

[12] D. B. Johnsonand D. A. Maltz,Dynamic Source Rout-ing in Ad Hoc Wireless Network, Kluwer AcademicPublishers, 1996.

[13] V. D. Park and M. S. Corson, “A highly adaptivedistributed routing algorithm for mobile wireless net-works,” in INFOCOM’97.

[14] C-K. Toh, “A novel distributed routing protocol to sup-port ad-hoc mobile computing,”IEEE InternationalPhoenix Conf. on Computers and Communications,IPCCC’96.

[15] R. Dube, C. D. Rais, K. Wang, and S. K. Tripathi,“Signal stability based adaptive routing(SSR) for ad-hoc mobile networks,” IEEE Personal Communica-tions, vol. 4, no. 1, pp. 36–45, 1997.

[16] Y-B. Ko and N. H. Vaidya, “Location-aided rout-ing in mobile ad hoc networks,”4th Annual Interna-tional Conferenceon Mobile Computing and Network-ing, MOBICOM’98.

[17] S. Radhakrishnan, N. S. V. Rao G. Racherla, C.N. Sekharan, and S. G. Batsell, “DST - a routing pro-tocol for ad hoc networks using distributed spanningtrees,” IEEE Wireless Communications and Network-ing Conference, pp. 100–104, 1999.

[18] M. R. Pearlman and Z. J. Hass, “Determining the op-timal configuration for zone routing protocol,”IEEEJournal on Selected Areas in Communications, vol. 17,no. 8, pp. 1395–1414, 1999.

[19] M. Joa-Ng and I-Tai Lu, “A peer-to-peer zone-basedtwo-level link state routing for mobile ad hoc net-works,” IEEE on Selected Areas in Communications,vol. 17, no. 8, pp. 1415–1425, 1999.