Routing protocols in ad hoc networks: A surveycs752/papers/routing-001.pdfSurvey Paper Routing protocols in ad hoc networks: A survey Azzedine Boukerchea,1, Begumhan Turgutb,2, Nevin

Computer Networks 55 (2011) 3032–3080

Contents lists available at ScienceDirect

Computer Networks

journal homepage: www.elsevier .com/locate /comnet

Survey Paper

Routing protocols in ad hoc networks: A survey

Azzedine Boukerche a,1, Begumhan Turgut b,2, Nevin Aydin c,3, Mohammad Z. Ahmad d,4,Ladislau Bölöni d,5, Damla Turgut d,⇑a University of Ottawa, School of Information Technology and Engineering (SITE), 800 King Edward Avenue, Ottawa, Ontario, Canada K1N 6N5b Rutgers, The State University of New Jersey, Department of Computer Science, 110 Frelinghuysen Road, Piscataway, NJ 08854-8019, United Statesc Istanbul Arel University, Department of Industrial Engineering, Sefaköy, Küçükçekmece, Istanbul, Turkeyd University of Central Florida, Department of Electrical Engineering and Computer Science, P.O. Box 162362, Orlando, FL 32816-2362, United States

a r t i c l e i n f o a b s t r a c t

Article history:Received 19 February 2010Received in revised form 6 March 2011Accepted 12 May 2011Available online 25 May 2011Responsible Editor: A. Al-Dhelaan

Keywords:Ad hoc networksSensor networksRouting protocols

1389-1286/$ - see front matter � 2011 Elsevier B.Vdoi:10.1016/j.comnet.2011.05.010

⇑ Corresponding author. Tel.: +1 407 823 6171; faE-mail addresses: [email protected] (A. B

cs.rutgers.edu (B. Turgut), [email protected]@eecs.ucf.edu (M.Z. Ahmad), lboloni@[email protected] (D. Turgut).

1 Tel.: +1 613 562 5800x6712; fax: +1 613 562 562 Tel.: +1 732 445 3546; fax: +1 732 445 0537.3 Tel.: +90 0212 540 96 96; fax: +90 0212 540 974 Tel.: +1 407 823 3327; fax: +1 407 823 5835.5 Tel.: +1 407 823 2320; fax: +1 407 823 5835.

Ad hoc wireless networks perform the difficult task of multi-hop communication in anenvironment without a dedicated infrastructure, with mobile nodes and changing networktopology. Different deployments exhibit various constraints, such as energy limitations,opportunities, such as the knowledge of the physical location of the nodes in certain sce-narios, and requirements, such as real-time or multi-cast communication. In the last15 years, the wireless networking community designed hundreds of new routing protocolstargeting the various scenarios of this design space. The objective of this paper is to create ataxonomy of the ad hoc routing protocols, and to survey and compare representativeexamples for each class of protocols. We strive to uncover the requirements consideredby the different protocols, the resource limitations under which they operate, and thedesign decisions made by the authors.

� 2011 Elsevier B.V. All rights reserved.

1. Introduction

Wireless local area networks based on the 802.11a, b, gand n standards became one of the most ubiquitous waysof networking with mobile nodes. Most of these networks,however, are deployed in the configuration which can becalled ‘‘wired everywhere, except the first hop’’. If the goalof the user of the mobile computer is to connect to a web-site located halfway around the world, the best strategy isto escape as quickly as possible from the challenges ofwireless domain and enter the reliability of fiber optic

. All rights reserved.

x: +1 407 823 5835.oukerche), [email protected] (Nevin Aydin),

.ucf.edu (L. Bölöni),

64.

97.

networks and time-tested networking protocols. In suchnetworks, all the nodes connect to an access point whichusually has a wired connection to the Internet. From thepoint of view of the network and higher layers, this firsthop can be approximated as an Ethernet-type shared med-ium. In this scenario the nodes connected to the samewireless LAN communicate with each other only indirectly.

There are, however, many important applicationswhere this model is not applicable. First, even if the goalis Internet access, the access point might not be able tocover all the relevant mobile nodes due to limitations intransmission range, cost or access rights considerations.Another case is when Internet access is not desired (or issecondary importance), the main application being to com-municate locally among a group of (potentially mobile)nodes.

These scenarios can be serviced only if we allow some(possibly all) routing hops to be performed in the wirelessdomain. Such networks can be set up in any location in anad hoc manner, without the need of an existing wiredinfrastructure.

http://dx.doi.org/10.1016/j.comnet.2011.05.010

mailto:[email protected]







http://dx.doi.org/10.1016/j.comnet.2011.05.010

http://www.sciencedirect.com/science/journal/13891286

http://www.elsevier.com/locate/comnet

A. Boukerche et al. / Computer Networks 55 (2011) 3032–3080 3033

These networks are known as ad hoc wireless networks[92], other proposed names being infrastructureless wirelessnetworks, instant infrastructure [8] and mobile-mesh net-working [140].

One of the major technological challenges of such net-works is that they require new types of routing protocols.As opposed to the wired infrastructure, there are nodedicated router nodes: the task of routing needs to be per-formed by the user nodes, which can be mobile, unreliableand have limited energy and other resources.

The goal of this paper is to review the collection of tech-nologies which have been proposed for routing in ad hocnetworks. There are literally hundreds of different ad hocrouting protocols proposed. We strive not for a simple enu-meration of this extensive literature, but we try to uncoverthe design decisions behind the various protocols, theirinterrelationship, and the specific requirements taken intoconsideration by the designers.

This way, we hope to provide the student and research-er with a more clear description of the state of the art. Wehope that this systematic approach will help the researcherunderstand the open challenges of the field, as well asthose which have been satisfactorily solved. This can helpa researcher position its work in the context of the stateof the art, assess its originality and avoid duplication ofexisting work. The survey can also help the practitionerschoose the most adequate technology for a specificdeployment.

Early ad hoc routing protocols have been classified intoon-demand and table-driven protocols. Between these two,several hybrid approaches have been developed. Theincreasing size of the ad hoc networks considered madenecessary the use of techniques such as geographical rout-ing and hierarchical routing. Finally, in many deploymentsof ad hoc networks, the problem of energy conservationtakes precedence from all the other performance metrics,thus power aware routing protocols will be treated as aseparate class.

2. Applications of ad hoc networks

Ad hoc networks, defined in the broad sense by of theterm as wireless networking in the absence of a wiredinfrastructure, have a wide range of potential applications.Some of these applications have been already identified inearly ad hoc literature [92]. Other applications, however,were enabled only recently by the shifting landscape ofcomputing and communication. In the last decade, for in-stance, the default method for first hop internet accessshifted from the Ethernet line to the WiFi connection. Inthe last three years, smartphones and tablets emerged asa multi-purpose computing platform, relying exclusivelyon wireless connectivity and replacing personal digitalassistants and (in some cases) low-end mobile computers.These technological advances have enabled applicationssuch as urban sensing which could not have been predictedten years ago. On the other hand, some wireless technolo-gies progressed at a slower pace than predicted: vehicle tovehicle technologies are still in the prototype phase, homeautomation systems are rare, and sensing devices are still

too expensive to be considered disposable and too largeto be deployed by random spreading.

In the following we briefly survey some applicationareas of interest for ad hoc networks. There are someapplications where ad hoc networks are the only possiblesolution, for instance, networking in areas where noinfrastructure is available. Early work in ad hoc networksfrequently assumed such radical scenarios. Beyond theseapplications, however, there is a much larger field of po-tential applications where ad hoc networks compete withother possible technical solutions. Finally, there are appli-cation areas where ad hoc networks must be part of a com-bination of technologies.

Network extension: In this application area, the net-working infrastructure exists, but it has insufficient cover-age. The goal of the participants of the network is internetaccess, that is, their main communication partners are out-side the ad hoc network. The goal of the ad hoc network isto extend the internet connectivity beyond the reach of theaccess points. Most routes of the ad hoc network will con-nect the access points to the nodes.

Local interconnection networks: In this applicationarea, no infrastructure is available (or the nodes choosenot to use it). For instance, when networking in remote areas(such as in a scenario involving a camp of archaeologists inthe Central American forest), the infrastructure might nothave been there to begin with. In other applications, suchas disaster response, the previously existing infrastructurehas collapsed due to a natural disaster. In these applications,the communication partners of most nodes are within thenetwork. Example applications include point-to-point mes-saging and audio and video conferencing.

Ubiquitous computing: This area covers networkingbetween devices embedded in the environment. Commu-nication patterns in ubiquitous computing are stronglyinfluenced by the physical location and proximity – de-vices which are close to each other are more likely to com-municate then remote devices. In contrast, on the wiredinternet, physical location is almost irrelevant. Ad hoc net-works are a particularly good match for proximity basedcommunication. Note, however, that in areas where a per-vasive infrastructure is available, ad hoc networks competewith solutions which rely on the convenience of the defaultinfrastructure, even when technologically suboptimal. Arecent example involves solutions where a TV set-topbox is controlled from a smartphone, through an internetconnection traversing dozens of routers, even when thetwo devices are several feet from each other.

Urban sensing: This application area exploits the sens-ing and computation capabilities of smartphones, togetherwith the wide range of their deployment in urban areas. Ur-ban sensing is characterized by distributed sensing or datacollection, and, in many cases, by distributed data custom-ers. Smartphones can use both infrastructure based accessand as well as ad hoc connections. Ad hoc approaches havethe advantage of lower energy consumption, lower overallbandwidth consumption and improved privacy – but theyinevitably involve more complex interaction patterns.

Vehicular networking: This area covers applicationswhere one of the communication partners is a vehicle. Thisdefinition covers a very wide range of technologies.

3034 A. Boukerche et al. / Computer Networks 55 (2011) 3032–3080

One type of vehicular networking involves the shortrange communication between the vehicle and personaldevices carried by the passengers. The most widely de-ployed systems rely on Bluetooth, but other technologies,such as WiFi have also been suggested and implemented.Note that many devices, while connected to the local net-work of the vehicle, can also maintain long range wirelessconnections, for instance through 3G cellular telephony,and, increasingly, WiMAX and LTE.

Another area of vehicular networking is vehicle to vehi-cle (V2V) communication. Due to the high relative speed ofthe participants, the short duration of the encounters andthe peer-to-peer nature of the communication, V2V tech-nologies are a good match for ad hoc networking. V2Vhas applications in convoy driving, accident pre-emption,and lane changes with pre-negotiated lane clearance. Onesignificant obstacle of the deployment of V2V technologiesrelate to achieving a critical mass of deployment – a singleV2V-capable surrounded with vehicles which cannotunderstand it cannot achieve any of the benefits of thetechnology.

A final area of vehicular networking is vehicle-to-infra-structure (V2I) communication. Calling a V2I system an‘‘ad hoc network’’ appears to be self-contradictory – as wedefined the lack of infrastructure the defining characteristicof ad hoc networks. Yet, vehicles have a high relative speedwith respect to any specific access point in the infrastruc-ture – the communication needs to be either very brief, orthe system needs to handle very fast connection hand-offs,or it needs to be able to communicate through intermittentconnections. Thus, the infrastructure in V2I communica-tions has more in common with the typical mobile interac-tion partner in ad hoc networks, than the highly reliable,quasi-permanent connections characterizing infrastructurein traditional networking. The most popular applications ofV2I are currently toll collection systems. However, thesesystems will also offer significant support for the emergingintelligent transportation systems.

Personal area networks: This application area refers tonetworking among the portable devices carried by a singleuser. As long as these devices move with the user, this sys-tem can be considered as a local area network with theindividual components being in a fixed relative position.The most popular current PAN technology is based on theBluetooth standards. Quite often, one or more of these de-vices has its own internet connectivity through long rangecommunication. Very similarly with the vehicular net-works, aspects of ad hoc networking come into play whenthe personal networks of different users will need to inter-act, or when devices of one users’ PAN needs to establish ashort range communication with infrastructure elements(such as when performing payment processing throughnear-field communication).

3. Ad hoc routing protocols and comparisons

Mirroring the diversity of applications areas, research-ers have proposed a wide range of routing protocols forad hoc networks. The basic goals of these protocols arethe same: maximize throughput while minimizing packet

loss, control overhead and energy usage. However, the rel-ative priorities of these criteria differ among applicationareas. In addition, in some applications, ad hoc networkingis really the only feasible solution, while in other applica-tions, ad hoc networking competes with other technolo-gies. Thus, the performance expectations of the ad hocnetworks differ from application to application and thearchitecture of the ad hoc network, thus each applicationarea and ad hoc network type must be evaluated againsta different set of metrics (although some metrics can beapplied across several protocol categories).

In the reminder of this paper, we organize the discussedrouting protocols into nine categories based on theirunderlying architectural framework as follows (also shownin Fig. 1).

� Source-initiated (Reactive or on-demand) (Section 3.1).� Table-driven (Pro-active) (Section 3.2).� Hybrid (Section 3.3).� Location-aware (Geographical) (Section 3.4).� Multipath (Section 3.5).� Hierarchical (Section 3.6).� Multicast (Section 3.7).� Geographical Multicast (Section 3.8).� Power-aware (Section 3.9).

3.1. Source-initiated protocols

Source-initiated routing represents a class of routingprotocols where the route is created only when the sourcerequests a route to a destination. The route is createdthrough a route discovery procedure which involves flood-ing the network with route request packets are flooded tostarting with the immediate neighbors of the source. Oncea route is formed or multiple routes are obtained to thedestination, the route discovery process comes to an end.A route maintenance procedure maintains the continuityof the route for the timespan it is needed by the source.

Dynamic source routing (DSR) [57]: Johnson et al. pro-pose one of the most widely known routing algorithms,called Dynamic Source Routing which is an ‘‘on-demand’’algorithm and it has route discovery and route maintenancephases.

Route discovery contains both route request and routereply messages. In the route discovery phase, when a nodewishes to send a message, it first broadcasts a route re-quest packet to its neighbors. Every node within a broad-cast range adds their node id to the route request packetand rebroadcasts. Eventually, one of the broadcast mes-sages will reach either the destination or a node whichhas a recent route to the destination. Since each nodemaintains a route cache, it first checks its cache for a routethat matches the requested destination. Maintaining aroute cache in every node reduces the overhead generatedby a route discovery phase. If a route is found in the routecache, the node will return a route reply message to thesource node rather than forwarding the route request mes-sage further. The first packet that reaches the destinationnode will have a complete route. DSR assumes that thepath obtained is the shortest since it takes into consider-ation the first packet to arrive at the destination node. A

Routing protocols in ad hoc networks

Source-initiated

Location-aware

Multi-pathHybrid Multicast

Table-driven

DSR

AODV

TORA

ABR

ZRP

FSR

LANMAR GeographicalMulticast

Power-aware

Hierarchical

RDMAR

SLURP

ZHLS

DST

DDR

A4LP

DSDV

R-DSDV

OLSR

HOLSR

CGSR

WRP

GSR

STAR

SSBR

Goff et al.

AQOR

ARA

ROAM

FORP

LAR

DREAM

GPSR

Colagrosso et al.

ALARM

REGR

LAKER

Blazevic et al.

MORA

CHAMP

AOMDV

SMR

NTBR

Das et al.

DCMP

ADMR

AMRoute

DGR

GAMER

GeoGRID

GeoTORA

HSR

CEDAR

Eriksson et al.

H-LANMAR

DEAR

Singh et. al

Scott & Bombos

Chang & Tassiulas

CLUSTERPOW & MINPOW

SCaTR

DAR

Beraldi

QMRB

Yu et al.

LBAQ

LSR

SWORP

RPR

GRP

Mahmood & Comaniciu

MEHDSR

Liang & Liu

Karayiannis & Nadella

TMRP

Liu et al.

SMORT

REEF

MuSeQoR

Li et al.

QMRPCAH

Wu & Jia

Randaccio &Atzori

Fireworks

PPMA

ALMA

AQM

Mohanoor et al.

SLR

Beraldi et al.

LDR

DBR2P

Tam et al.

QOLSR

HOPNET

LRHR

FZRP

ANSI

OGRP

Song et al.

SOLAR

LBLSP

GLR

MER

CBM

DDM

ROMANT

EraMobileOD-PFS

Gong et al.

FDG

LLR

Abdulai et al.

AODV-ABR

RBR

Bamis et al.

Souihli et al.

DRM

TBR

SecMR

Ramasubramanian et al. Gui & Mohapatra

Sun & Li

Fig. 1. Categories of ad hoc routing protocols.


route reply packet is sent to the source which contains thecomplete route from the source to the destination. Thus,the source node knows its route to the destination nodeand can initiate the routing of the data packets. The sourcecaches this route in its route cache.

In the route maintenance phase, route error andacknowledgements packets are used. DSR ensures the valid-ity of the existing routes based on the acknowledgementsreceived from the neighboring nodes that data packetshave been transmitted to the next hop. Acknowledgementpackets also include passive acknowledgements as the nodeoverhears the next hop neighbor is forwarding thepacket along the route to the destination. A route errorpacket is generated when a node encounters a transmis-sion problem which means that a node has failed to receive

an acknowledgement. This route error packet is sent to thesource in order to initiate a new route discovery phase.Upon receiving the route error message, nodes removefrom their route caches the route entry using the brokenlink.

Ad hoc on-demand distance vector (AODV) [94]: TheAODV routing protocol was developed by Perkins andRoyer as an improvement to the Destination-SequencedDistance-Vector (DSDV) routing algorithm [93]. AODVaims to reduce the number of broadcast messages for-warded throughout the network by discovering routeson-demand instead of keeping a complete up-to-date routeinformation.

A source node seeking to send a data packet to a desti-nation node checks its route table to see if it has a valid


route to the destination node. If a route exists, it simplyforwards the packets to the next hop along the way tothe destination. On the other hand, if there is no route inthe table, the source node begins a route discovery process.It broadcasts a route request (RREQ) packet to its immediateneighbors and those nodes broadcast further to theirneighbors until the request either reaches an intermediatenode with a route to the destination or the destinationnode itself. The route request packet contains the IP ad-dress of the source node, current sequence number, theIP address of the destination node and the last known se-quence number. Fig. 2 illustrates the forward and reversepath formation in the AODV protocol. An intermediatenode can reply to the route request packet only if it has adestination sequence number that is greater than or equalto the number contained in the route request packet head-er. When the intermediate nodes forward route requestpackets to their neighbors, they record in their route tablesthe address of the neighbor from which the first copy of thepacket has arrived. This recorded information is later usedto construct the reverse path for the route reply (RREP)packet. If the same RREQ packets arrive later on, they arediscarded. When the RREP packet arrives from the destina-tion or the intermediate node, the nodes forward it alongthe established reverse path and store the forward routeentry in their route table by the use of symmetric links.Route maintenance is required if either the destination orthe intermediate node moves away and it is performedby sending a link failure notification message to each of itsupstream neighbors to ensure the deletion of that particu-lar part of the route. Once the message reaches to sourcenode, it then re-initiates the route discovery process.

Temporally ordered routing algorithm (TORA)[89,139]: Park and Corson proposed TORA, an adaptiveand scalable routing algorithm based on the concept of linkreversal. It finds multiple routes from a source to a destina-tion in a highly dynamic mobile networking environment.

D

S S

D

Timeout

(a) (b)Fig. 2. (a) Reverse path formation: the reverse path is set up from all thenodes back to the source through RREQ message traveling in the network.(b) Forward path formation: the forward path is constructed as the RREPmessage from the destination node D to source node S. The source nodestart sending data when it received the first RREP message to the sender[94].

In TORA control messages are localized to a small set ofnodes nearby the topological change. Nodes maintain rout-ing information about their immediate one-hop neighbors.The three basic functions of the protocol are route creation,route maintenance, and route erasure.

Nodes use a ‘‘height’’ metric to establish a directed cyc-lic graph (DAG) rooted at the destination during the routecreation and route maintenance phases. The link can beeither an upstream or downstream based on the relativeheight metric of the adjacent nodes. TORA’s metric con-tains the unique node ID, the logical time of a link failure,the unique ID of a node that defined the new referencelevel, a reflection indicator bit, and a propagation orderingparameter. Establishment of DAG resembles the query/re-ply process in Lightweight Mobile Routing (LMR) [34].Route maintenance is necessary when any of the links inDAG is broken. Fig. 3 describes the control flow of routemaintenance in TORA.

The main strength of the TORA protocol is its approachto handling the link failures. TORA’s reaction to link fail-ures is optimistic: it reverses the links to re-position theDAG for searching an alternate path. Effectively, each linkreversal sequence searches for alternative routes to thedestination. This search mechanism generally requires asingle-pass of the distributed algorithm since the routingtables are modified simultaneously during the outwardphase of the search mechanism. Other routing algorithmssuch as LMR use a two-passes search for the same task,while both DSR and AODV use a three-pass procedure.TORA achieves its single-pass procedure with the assump-tion that all the nodes have synchronized clocks (via GPS)to create a temporal order of topological change of events.The ‘‘height’’ metric is dependent on the logical time of alink failure.

Associativity-based routing (ABR) [116]: Toh proposesthe ABR algorithm which considers route stability as themost important factor in selecting a route. Routes are dis-covered by broadcasting a broadcast query request packet.Using these packets, the destination becomes aware of allpossible routes between itself and the source.

The ABR algorithm maintains a ‘‘degree of associativity’’by using a mechanism called associativity ticks. Each nodemaintains a tick value for each neighbors, which is increaseby one every time a periodic link layer HELLO message isreceived from the neighbor. Once the tick value reaches aspecified threshold value, it means that the route is stable.If the neighbor goes out of the range, then the tick value isreset to zero. Hence a tick level above the threshold valueis an indicator of a rather stable association between thesetwo nodes. Once a destination has received the broadcastquery packets, it has to decide which path to select bychecking the tick-associativity of the nodes. The route withthe highest degree of associativity is selected since it isconsidered the most stable of the available routes.

Signal stability-based adaptive routing (SSBR) [37]:Dube et al. propose the SSBR protocol in which the mainrouting criteria are the signal and location stability. As inother on-demand routing protocols, the route request isbroadcast throughout the network, the destination replieswith the route reply message and then the sender sendsdata through the selected route. Additionally, the signal

Node i loses its lastdownstream link

Was the linklost due to a

failure?

Case 1:Generate newreference level

Do all the neighborshave the same reference

level?

Case 2:Propagate the highest

neighbor's reference levelIs the reflection bit (r) in the

reference level set to 1 ?

Case 3:Reflect back a

higher level

Did this node originallydefine that reference level

(oid = i)?

Case 4:Partition detected,

erase invalid routes

Case 5:Generate newreference level

YES

YES

YES

YES

NO

NO

NO

NO

Fig. 3. Flow diagram of route maintenance in TORA [89]. The only routes with a height greater than NULL value are maintained. The figures shows all fivepossible paths in the form of a decision tree which is based on the state of the node and the preceding event.


strength (link quality) between neighboring nodes plays amajor role in the route selection process in this protocol.

SSBR consists of two sub-protocols: the Dynamic Rout-ing Protocol (DRP) and the Static Routing Protocol (SRP).DRP interacts with the network interface device driverthrough an API to determine the actual strength of a re-ceived signal. Using this signal information the DRP main-tains a signal stability table which categorizes each linkwith the neighboring nodes as strong or weak. This tableis updated with every new packet received. For instance,if a HELLO packet is received, the signal strength is moni-tored and the signal stability table is upgraded while forother packets such as route update packets, data packets,and so on, the packet is sent to the SRP for further process-ing. The SRP performs the routine tasks such as forwardingpackets according to the existing routing table, replying toroute requests, and so on.

The route request is given an option on the type of linkit requests, i.e., strong, weak or a combination of both. Ifthe route request specifies only strong links, all the routerequest packets coming from a perceived weaker link aredropped. Thus, the final discovered path consists of onlystrong links. If there are multiple paths from source to des-tination using strong links, the destination can chooseamong them (or it might simply choose the first path it re-ceives). If no strong links are found, the protocol could fallback on other available weaker links.

The paper also discusses two further enhancements tothe route selection process. The first selection concernsthe selection of alternative routes from a set which In thefirst case, the link strength is added for each hop into theroute request packet and then forwarded towards the des-tination. In this case, the destination does not select thefirst route request packet received, but waits for a periodof time to choose the best route among all the route re-quests within a set time interval. The second improvementsuggests that any intermediate node can make a graciousroute reply for a route it already has a prior informationabout.

Preemptive routing in ad hoc networks [47]: In con-ventional protocols, a path is considered broken only afterseveral retransmissions have timed out. The algorithmintroduced by Goff et al. attempts to initiate the discoveryprocess of an alternate route just before the probable routefailure. The algorithm generates a preemptive warningwhen the signal power of the packet received drops belowa predefined preemptive threshold. The correct setting ofthe preemptive threshold is the main challenge of the algo-rithm. If the value is too high, unnecessary warnings maybe generated which can lead to greater overhead, unneces-sary route discoveries, and switches to possibly lowerquality paths. On the other hand, if the value is too low,the path breaks much earlier than the alternate route isselected. This leaves a short time period for building an


alternate path. As temporary channel fading can oftencreate a weak reception without leading to route failure,the algorithm uses successive ‘‘query’’ packets to decidewhether the generated warnings are valid.

Ad hoc QoS on-demand routing (AQOR) [132]: Xueand Ganz propose AQOR, an on-demand routing protocolenabling QoS support in terms of bandwidth and end-to-end delay. The AQOR mechanism estimates the bandwidthand end-to-end delay requirements and use these metricsto make admission and resource reservation decisions.

The AQOR integrates on-demand route discovery, sig-naling functions for resource reservations, and hop-by-hop routing to provide QoS support in ad hoc networks.Since most QoS violations are detected at the destinationnode, the routing overhead generated can be reduced byinitiating the route recovery process at the destinationnode. Route maintenance is accomplished by sending peri-odic HELLO messages. Routes are discovered on-demandby a limited flooding mechanism. The requested band-width and end-to-end delay values are specified withinthe route request packet. The bandwidth requirementsare calculated based on the available link capacity andthe bandwidth used by the flow. If this request is accepted,the node updates its routing table with an explored statusand broadcasts it to all its neighbors. However, if no replyis received in a specified time, the route entry is removedfrom the node’s table and late-arriving replies are simplyignored. Route caching is not used since the route requestpackets are forwarded from the node to the destination todetermine the bandwidth and end-to-end delay require-ments. To further reduce control overhead, the packetshave a time-to-live (TTL) parameter which stops the pack-ets from traveling unnecessarily throughout the network.

ARA-The ant-colony based routing algorithms [50]:Gunes et al. present a novel technique for ad hoc routing

Nest Food

Maximum pheromoneconcentration on theshortest path

Path 1 Path 2

Path 3

Path 4

Fig. 4. Pheromone concentration along the shortest path used by ants todiscover food from their nest [50].

2

1

s

3

F

FF

F

Fig. 5. Forward ant route discovery phase. A forward ant (F) is send from the sendnodes, which initialize their routing table and the pheromone values [50].

by using concepts of swarm intelligence and the ant colonymeta-heuristic. This class of algorithms aims to solve thecomplex optimization and collaboration problems withoutdirect communication among the participants. Indirectcommunication is achieved by stigmergy, the process ofleaving traces in the environment, similar to the behaviorof ants leaving pheromone signals.

The route discovery phase uses two types of controlpackets: the forward ant (FANT) and the backward ant(BANT). The FANT establishes the pheromone track to thesource node while the BANT establishes the pheromonetrack to the destination. When the route is required, thesource broadcasts FANT packets to all its neighbors. A nodewhich receives a FANT for the first time creates a routingtable record which contains the destination address, nexthop, pheromone value. The source address of the FANT istaken as the destination address, the previous node ad-dress as the next hop, and the pheromone value is calcu-lated based on the total number of hops required by theFANT to reach a particular node. When the FANT reachesthe destination, the node updates its own informationand sends the BANT back. Once, the BANT reaches thesource, the path can be used. Figs. 5 and 6 show the for-ward and backward ants’ route discovery phases.

The ARA algorithm doesn’t need special route mainte-nance packets since it uses the transmitted data packetsto maintain the route. The pheromone value of the pathis increased by d/ each time a data packet is forwardedalong the path, but decreases in time when no packetsare transmitted.

Routing on-demand acyclic multipath (ROAM) [98]:ROAM algorithm by Raju and Garcia–Luna–Aceves coordi-nates among nodes in directed acyclic subgraphs. It is anextension of the DUAL [41] routing algorithm. The ROAMalgorithm guarantees that route search query will fail toreturn a destination path only is all the routers agree thatthe destination is unreachable.

Each router in ROAM maintains distance, routing, and linkcost tables. While the distance table maintains the distancesof nodes for each destination and neighbors from the respec-tive node, the routing table contains the distance to eachdestination, the feasible distance and the reported distance.The link cost table provides the link costs to each of the adja-cent neighbors of the router. A router updates its routing ta-ble for a destination when it needs to: (i) add an entry for aparticular destination; (ii) modify its distance to the desti-nation; and (iii) erase the entry for the destination.

The routers in ROAM are either in active or passive states.If a router has send queries to all its neighbors and awaiting

4

5

D

6

F

FF

F

er (S) toward the destination node (D). The forward ant is relayed by other

2

1

4

5

D

6

s

3

BB

B

B

BB

B

Fig. 6. Backward ant route discovery phase. The backward ant (B) has the same task as the forward ant. It is send by the destination node toward the sourcenode [50].


a reply, it is in active state, otherwise in passive state. Selec-tion of loop-free paths allows a router to select a neighboras its successor only if it is a feasible successor. This providesa shortest loop-free path to the destination. A when it re-quires a path to a destination, the source router starts a dif-fusion search, with the packet propagated through routerswhich have no entry of the node. The first router with anavailable route to the destination responds to the sourcewith the distance to the node. At the end of the search,either the source has a finite distance to the destinationor the destination is unreachable.

The flow oriented routing protocol (FORP) [113]: TheFORP protocol proposed by Su and Gerla aims to transmitreal-time data streams in ad hoc networks, which requirein-order delivery of packets with tight delivery bounds. Ifalternate routes are not available to immediately redirectthe data packets in case of route failures, real-time packetsmay be dropped. FORP introduces the ‘‘multi-hop handoff’’mechanism in which the nodes use their mobility informa-tion to determine future route changes resulting inrebuilding of an alternate routes much sooner.

Similarly to other on-demand schemes, FORP maintainsrouting information only for active source/destinationpairs. The protocol predicts the link expiration time (LET)for each hop on the route to calculate the route expirationtime (RET). During the route discovery phase, the sourcebroadcasts a Flow-REQ message containing a sequencenumber, source ID, and destination ID. Each node appendsits own ID and the LET of the last link in which the messagewas received before forwarding to the next hop. When theFlow-REQ arrives at the destination, it contains the list ofall the routes travelled and the LETs for each hop. Usingthis information, the destination calculates the RET byselecting the minimum LET value for the route. The nodesare assumed to have a common time reference from GPSfor instance. Once the route is selected, a Flow-SETUP

message travels to the source along the chosen path.While the connection is in progress, the intermediate

nodes continue adding the LETs to the forwarded packetsto enable the destination to keep track of the RET predic-tion. If the destination determines that a ‘‘critical time’’ isreached, i.e. the route is close to expire, a Flow-HANDOFF

message is generated and propagated throughout the net-work. These Flow-HANDOFF messages reach the sourceand based on their LETs and RET, the source determinesthe new route. The ‘‘critical time’’ is calculated as Tc = RET– Td where RET is the route expiration time and Td is thedelay experienced by the last packet arrived on the sameroute.

On-demand routing and channel assignment in mul-ti-channel mobile ad hoc networks [48]: Gong et al. con-centrates mainly on designing an efficient channelassignment algorithm at the MAC layer to be used withmost on-demand routing strategies at the network level.The authors state that there are intra-flow and inter-flowinterferences due to adjacent nodes on the same or differ-ent channels respectively. To mitigate the interferenceproblems, the authors implement two enhanced versionsof the AODV routing protocol: Enhanced 2-hop CA-AODVand Enhanced k-hop CA-AODV. In these algorithms, theRREQ and RREP messages are similar to AODV with the dif-ference of the neighbor table at each node consisting ofboth the route and the indices of the channels already cho-sen. If the node is not assigned to a channel, a channel israndomly chosen and assigned from the available list(see Fig. 7). RREP messages also carry channel informationwhich is updated by each node receiving the RREP. At theend of the RREQ-RREP cycle, every node in the route is as-signed a channel which is different from any of its k-hop(for the k-hop extension of the algorithm) neighbors onthe same route.

Space-content adaptive time routing (SCaTR) [17]:Boice et al. present a routing framework which takes intoconsideration the possibility of intermittent connectivityin a mobile ad hoc network. SCaTR uses past connectivityinformation by defining proxy nodes to route traffic to-wards the destination when no direct route is available.It is built upon the existing AODV protocol in such a waythat when the network is fully connected, it works identi-cal to AODV. When a network partition occurs, proxynodes are created based on the distance of a node to thedestination. A node closer to the destination advertizes it-self as the proxy destination and buffers messages on theroute until the final destination is discovered or anothernode is selected as a better proxy. This design feature en-ables the protocol to behave no worse than the standardAODV in most cases (when the network is entirely con-nected). During the route discovery phase if a route tothe destination is not established, a proxy request (PREQ)is forwarded to find the nearest proxy destination. PREQcould also request for multiple destinations and a proxy re-ply (PREP) is sent back by the responding node. Each proxynode has a finite buffer to store different messages fromvarying source–destination pairs. The PREP message con-tains updated information of the proxy’s contact valuefor the destination, its remaining buffer space, and numberof messages stored for the particular source–destinationpair from which it has received the PREQ. Each node also

Init

Wait

recvRREQ recvRREP

Keep my channel

Update set A Update set A

Am I an active node? Channel conflict?

Update my channel

seYseYPick a randomchannel

Know a route to destination? Am I the source?

No No

sendRREP sendRREP

No Routeestablished

Yes

YessendRREQ

No

Fig. 7. Algorithm flowchart of CA-AODV [48]. This figure shows that if the node does not have a pre-assigned channel, the channel will be chosen randomlyfrom the list of available channels.


has a separate buffer for its own messages. The sourcenode collects all the PREPs, compares the contact valuesobtained and updates its table with the highest value alongwith the route where the data packets are forwarded.

Distributed ant routing (DAR) [104]: Rosati et al. pro-pose a distributed routing algorithm based on ant behaviorin colonies. Ant colony optimization algorithms have beenwidely used in MANETs and the authors aim to design analgorithm incorporating the salient features of many exist-ing approaches. The main design goal of DAR is to mini-mize the computation complexity. Each node containsrouting tables which are stochastic with the next hopbeing selected based on weighted probabilites. These prob-abilities are calculated based on pheromone trails left byants. Forward ants are used to find new paths. If multiplepaths are available at a node, the next hop could be se-lected either randomly or the most optimal one. DARmostly uses hop-by-hop optimal forwarding with the for-ward ant routed at each node based on probabilites forthe next hop. Data packets are forwarded deterministicallyby selecting the highest probability at every node on thepath. Thus, a global route is created by using local hop-by-hop information.

Forwarding Dilemma game (FDG) [85]: Naserian andTepe propose a game theoretic approach to forwardingflooding packets in MANET with AODV as the underlyingrouting protocol. The game is played within the networkonly when a node receives a HELLO or any other floodingmessage since the nodes are the players. The game, called

the forwarding dilemma game (FDG), is composed of thenumber of players receiving the packet, the forwardingcost and the network gain factor and it offers primarilytwo strategies – forwarding or dropping the packet. Usinga mixed strategy Nash equilibrium, the probability of for-warding the flooding messages are calculated. The FDG isimplemented in AODV with the existing HELLO messagesused in neighbor discovery. Since the HELLO messagesare forwarded only to the winners of the game, the numberof nodes participating in the route discovery process is re-duced. The structures of the RREQ and RREP packets ofAODV have been modified for the calculation of the prob-ability of packet forwarding.

Long lifetime route (LLR) [28]: Cheng and Heinzelmanargue that many routes in ad hoc networks are short lived,triggering frequent route discovery processes, which inturn account for extra control overhead and packet latency.They propose two techniques which allow the network toselect long lifetime routes (LLR).

The g-LLR approach is a global approach where the opti-mal LLRs are computed in a centralized manner. As suchinformation is not normally available to the nodes in a net-work, its importance is mostly as benchmark. The d-LLRapproach selects long lifetime routes in a distributed man-ner, using only local information. The authors show thatthe performance of d-LLR closely matches that of g-LLR.

From a practical implementation point of view, the LLRapproach can be used to enhance most existing ad hocrouting algorithms. In addition, having longer lifetime

1

987

654

32 1

987

654

32

(a) (b)


routes at the network layer, also improves the performanceof the transport layer protocols built upon them: UDP, TCPas well as wireless-specific variations on TCP such as TCP-NB (no-backoff).

Polarized gossip protocol for path discovery [12]:Beraldi looks at gossip protocols for path discovery, wherea node forwards a packet with some predetermined prob-ability. In contrast to classical gossip algorithms which for-ward each message with the same probability, this workconsiders the probability dependent on node locationsand distances between each other. We have the polarizingnode n with two gossiping probabilities: pF (the forwardprobability) and pB (the backward probability). The mes-sage is forwarded with probabilty pF when the node re-ceives this message from at least one node which isfarther from the destination node than the node itself.Otherwise, it forwards the message with probability pB

(see Fig. 8). Distances between nodes can be found by sim-ple estimates of relative positions between neighbor nodeswhich can be determined by using periodic beaconingschemes.

On-demand packet forwarding scheme (OD-PFS) [4]:Al-Karaki and Kamal propose a clustering approach fol-lowed by a routing protocol exploiting the clusteringframework in MANETs. A fixed and scalable virtual wire-less backbone, called the virtual grid architecture (VGA),is created. The physical network topology is mapped ontoa virtual grid topology. The routing is then carried outusing a combination of hierarchical and virtual backbonerouting. Nodes are divided into fixed clusters (also calledzones) which require few virtual topology updates. Theauthors show that the virtual topology is stable as longas at least one mobile node is present in each cluster. Withclusters being fixed, the architecture becomes simple andscalable. Network zoning is accomplished by dividing thenetwork into disjoint but adjacent regular shape zones.Zone lengths in homogeneous MANETs are chosen suchthat two mobile nodes in adjacent zones can always com-municate directly with each other. In heterogeneous MAN-ETs, network topology can be changed by varying thenodes transmission range. Clusterheads (CHs) are selectedeither by periodic elections using an eligibility factor (EF)or dynamically when needed. CHs can also be elected on-demand. A simple on-demand packet forwarding scheme(OD-PFS) is implemented over the VGA. Four standarddirections (North, South, East and West) are used for sim-

p < p_c

p > p_cd

z

s

Fig. 8. A sketch of the partition of the node of a network. The forwardingprobability for the nodes belonging to Z is higher than the critical value pc.The nodes outside Z forward a packet with probability lower than pc. [12].

ple packet forwarding using the well-known route request(RREQ) and route reply (RREP) cycle. The path includesthe direction of the packet to be forwarded. For example,N-W-W-S-W-N-N-N-E is a sample path. A transitiveclosure approach is used where each CH maintains an adja-cency matrix of neighboring CHs with alive links. By calcu-lating successive transitive closures of the adjacencymatrix, path existence between a pair of nodes can be com-puted. The local path restoration in VGA is shown in Fig. 9.

QoS routing with traffic distribution (QMRB) [52]:Ivascu et al. use a mobile routing backbone to supportQoS in a MANET. The mobile routing backbone (MRB)dynamically distributes traffic within the network andselects the route with the best QoS between a source–des-tination pair. The proposed scheme classifies the nodes inthe network into QoS routing nodes (QRN), simple routingnodes (SRN) or transceiver nodes (TN). QRNs possess QoSguarantees, SRNs simply route packets through the net-work while TNs send and receive packets but cannot relaythem. The MRB is formed by these different types of nodeswhile it is not essential that all nodes in the network jointhe MRB. Nodes not joining the MRB may still communi-cate with it through a working link.

Node classification for the MRB is computed by the fourQoS support metrics (QSMs) for each pair of nodes.

� Static resources capacity (SRC): computed by theweighted sum of the size of the node packet queues,speed of the CPU, power of the battery and the maxi-mum available bandwidth.� Dynamic resources availability (DRA): indicates the cur-

rent load in the resource usage of a node. The usage rateof the static resources are used to calculate the availabledynamic resources.

1

987

654

32 1

987

654

32

Global

Local

clusterhead mobile node

(c) (d)

Fig. 9. The illustration of local path restoration in VGA [4].

N = 25

A B

C

D

E

n = 4

n = 10

n = 6

n = 5

n = 3

Fig. 10. Illustration of two logical groupings of 25 nodes located in sparseand dense regions of a network for the 2P-Scheme algorithm [1].

D S

Primary route

Alternate route

Destination Source

Fig. 11. A fish bone structure formed by the primary route and alternateroutes [65].


� Neighborhood quality (NQ): the number of nodes in theneighborhood of a node which can successfully forwardpackets.� Link quality and stability (LQS): the power of signal

received and the statistical stability of its links.

The node aptitude is computed based on the node clas-sification by following the formula:

MNaptitude ¼ lSRC þ gDRAþ rNQ þxLQSþ /BW;

where l, g, r, x and / are coefficients and BW is the avail-able bandwidth. Once the MRB is set up, route discovery isinitiated using RREQs along the MRB.

Adjusted probabilistic route discovery [1]: Abdulaiet al. observe that rebroadcasting route request packetsin a MANET leads to extensive control overhead and highlevels of channel contention. This work proposes two prob-abilistic methods aiming to reduce the number of RREQpackets using a predetermined fixed-value forwarding prob-ability. Unlike other similar algorithms, the proposedmechanism does not use GPS based devices for locationtracking but mainly relies on basic topology information.It also uses the minimum connected dominating set(MCDS) requiring global topological information of a net-work. In case of receiving duplicate packets at a node, theforwarding probability is adjusted. The first probabilisticroute discovery scheme is called the 2P-Scheme in whichthe nodes are categorized into two groups based on theircurrent neighborhood information. If the node is situatedin a sparse region, it gets assigned to Group-1 and if in adense region, it is assigned to Group-2 (see Fig. 10). Nodesin Group-1 are allocated a higher forwarding probabilitythan those in Group-2. In the second scheme, 3P-Scheme,the nodes are classified into three groups: Group-1 forsparse regions, Group-2 for medium dense regions andGroup-3 for dense regions. For this scheme, the forwardingprobabilities are assigned in non-decreasing order.

Adaptive backup routing (AODV-ABR) [65]: Lai et al.provide an extension to the AODV-BR scheme which usedthe concept of backup routes to AODV. It sets up a meshand multipath routing using RREP messages and aims toreduce control overhead. The mesh structure is createdby overhearing data packets transmitted from the nodesin the neighborhood (see Fig. 11). This helps in reducingcontrol messages and also react faster to the topologychanges. Whenever a link break is detected at a node, thenode itself initiates a handshake procedure with its neigh-bors and repair the broken route. Backup route request(BRRQ) and backup route reply (BRRP) messages are usedin the route repair process. The authors also suggest com-bining AODV-ABR with local route repair once the routebreaks by broadcasting a RREQ message from the nodewith the broken link.

Low overhead dynamic route repairing [135]: Yu et al.repair broken routes dynamically by using informationfrom overhearing the nodes. Once the route is down, theproposed protocol intelligently replaces the failed linkswith backup links along the main route. An initial mainroute is constructed between a source and destination pairalong which the data packets are transmitted. While thesepackets are forwarded, neighboring nodes overhear these

transmissions. These nodes are actually potential candi-dates for replacing a node on the main route in case of afailure. These candidate nodes are recorded in the packetheaders and used during the route reconstruction in the fu-ture. The route construction uses a main route request(MREQ) packet initially, similar to a RREQ packet in AODV.The MREQ is flooded throughout the network and once itarrives at the destination, the main route reply (MRRP) issent back. The MRRP keeps track of the hop count betweenthe source and destination. This hop count is used duringroute repair when a particular link on the main route goesdown. A repair query (REPQ) packet with the hop count ofthat node is forwarded in case of the main route failure. Anode receiving the REPQ checks to see if it has a savedroute to the destination. If so, it sends a repair reply(REPR); otherwise, the hop count value is compared withits own hop count. If its hop count is smaller, it propagatesthe REPQ packet towards the destination, else the packet isdropped. This enables nodes closer to the destination to re-ceive the REPQ and reply with a valid backup route.

Link availability-based QoS-aware (LBAQ) routing[136]: Yu et al. propose the LBAQ routing protocol basedon node mobility prediction and link quality measurement.While a node moves, it may experience varying capacity,reliability and bandwidth availability. Instead of trying topredict the mobility patterns of the mobile links, the linkavailability is incorporated into the routing metrics to helpchoose the link with the highest availability in the route.The expected transmission count (ETX) along with powerconsumption is also used as a route metric. The route met-rics used are:


� Link availability: probability that two nodes of a linkstay directly connected at a time t0 + t provided thatthey were connected at time t0.� Link quality: the number of retransmissions required to

send a packet on a link between two nodes (ETX). This iscomputed by measuring the loss rate of broadcast pack-ets between the node pairs.� Energy consumption: power consumption at each hop.

Using the metrics above, a combined cost function isdesigned based on which the route is constructed.

DiðTiÞ ¼ c0ðEiÞ þ c1ðQiÞ þ c2ð1� LiðtÞÞ;

where c0, c1 and c2 = 1 � c0 � c1 are the weighting coeffi-cients, Li and Qi are availability and quality of link i, Ei isthe energy consumed on link i, Ti is the traffic carried onpath i, and Di is the final cost of the route. A source nodesends a RREQ packet during route discovery. When anintermediate node receives this RREQ packet, it calculatesits own cost from the equation above and forwards thepacket with this new information. The total cost to the des-tination is computed when the RREQ reaches the destina-tion node. At the destination, the node waits for a fixednumber of RREQs before selecting the route with the leastcost. If a link breaks, then an alternate route is selected fordata transmission.

Labeled successor routing (LSR) [101]: Rangarajan andGarcia–Luna–Aceves notice that many modern on-demandprotocols are built on top of AODV, using the same destina-tion sequence numbers. Thus, they inherit the performanceproblems of AODV: (a) most route requests are answeredby the destination and (b) it can suffer from temporaryloops, de facto partition and count-to-infinity. The LSR ap-proach is an attempt to overcome these problems by usingthe information already needed in route requests to estab-lish and maintain loop-free routes and allows other nodesthan the destination to initiate route replies. LSR uses un-ique source-sequenced labels (SSLs) of flooded route re-quest (RREQ) messages to build loop free paths withinthe network. Directed acyclic graphs (DAGs) from sourceto destination are created and route replies are forwardedto the source through the reverse paths. In LSR, the desti-nation must answer every RREQ it receives since everynode relaying a RREQ associates a relay sequenced label(RSL) with the SSL of the forwarded RREQ. The RREPs takedifferent paths along the DAG built by the RREQs. Therecould be multiple DAGs through the same node due to dif-ferent RREQs forwarded through the node. To avoid pathloops, nodes use the RSL to select only a unique RREP withan SSL and drop the others. The LSR creates unique routerequest labels (RRLs) derived from the SSL of the RREQ for-warded. A source floods a RREQ SSL and the participatingnodes on this DAG store the SSL as the RRL for that partic-ular destination. The source can then forward packetsthrough any of these nodes towards the destination.

Stable weight based on demand routing protocol(SWORP) [123]: Wang et al. propose a weight based mech-anism for routing in MANETs. Weights are assigned to dif-ferent routes during route discovery using the routeexpiration time (RET), the error count (EC) and the hopcount (HC). Route discovery in SWORP is similar to DSR

with a source node initiating a RREQ message. The destina-tion node sends the RREP when it receives an RREQ for it-self. When multiple RREQs are received from differentpaths, the destination node calculates the RET, the ECand the HC for each path. With these metrics, the weightvalues for each path is calculated by the destination nodeand the route with the largest value is chosen as the pri-mary route. The RET is computed by the destination nodeusing the GPS signal strength to determine when the linkbetween a pair of moving nodes may be disconnected.The EC denotes the number of link failures caused by a mo-bile node while the hop count is the distance from the nodein terms of hops. Using these values, the weight functioncalculates the weight for each path using the followingequation:

Wi ¼ C1 �RETi

MaxRETþ C2 �

ECi

MaxECþ C3 �

HCi

MaxHC;

where C1, C2 and C3 are the weighing factors with their sumequal to one. Other important routing protocol functionssuch as route maintenance is carried out similar to DSRand AODV with a RERR packet generated when a link isbroken.

Recycled path routing (RPR) [38]: Eisbrener et al. pres-ent a new strategy towards broadcasting route request(RREQ) packets in MANETs during route discovery. It usesexpired routes stored in the route cache to make an edu-cated decision on forwarding RREQ packets towards thedestination. The authors implement controlled flooding inthe direction of the destination node but without any priorlocation information. RPR propagates RREQs withoutflooding every node and uses the hot or cold concept. Nodescloser to the destination are considered hotter than thenodes farther away. These hot or cold values are basedon the route request for a particular destination node.Hot nodes rebroadcast the RREQ message while cold nodesdiscard it. The reasoning behind this forwarding strategy isthe fact that a broken or expired route can be used towardsfinding the new route. Essentially, expired route caches arereused unlike other on demand protocols. If the destina-tion node is within the expired route cache of a nodereceiving the RREQ, that node is also considered hot. Onthe other hand, if the node is colder than a pre-definedthreshold, the RREQ is not forwarded at all. Initially, thenodes have empty route caches, meaning no hot or coldsettings. During the network initiation, all nodes broadcastpackets using traditional flooding and setup the routes.Timers are used to maintain the expired routes. Once thistimer expires, the expired routes are purged from the routecache to free up memory and remove redundancy.

Gathering based routing protocol (GRP) [3]: Ahn pre-sents the gathering based routing protocol which collectsnetwork information during route discovery to be used la-ter by the source node. Initially, the source node broadcastsa destination query (DQ) packet which is continuously for-warded towards the destination. This process is similar tothe route discovery process in DSR or AODV using RREQs.When the DQ reaches the destination node, a networkinformation gathering (NIG) packet is forwarded by thedestination. This NIG packet is then propagated over theentire network collecting information. Every NIG packet


is assigned a sequence number. Nodes receiving a NIG

packet with a new sequence number attach network infor-mation to this packet and forward it along effective outgo-ing links (EOLs). EOLs are those links in which the NIG

packets are not received. Once the NIG packet arrives atthe source node, the optimal path is calculated by thesource node based on the collected network information.Data packets are then transmitted through this calculatedroute.

Source routing with local recovery (SLR) [108]: Senguland Kravets start from the observation that although on-demand routing reduces the cost of routing in high mobil-ity environments, the route discovery process, which istypically done through network-wide flooding, consumesa significant amount of bandwidth. This is especiallyexpensive if the route discovery must be repeated due tolinks broken due to node mobility. To alleviate this prob-lem, the authors propose bypass routing, a process whichpatches a route using local information acquired on-de-mand, without the need of network-wide flooding. TheSLR protocol is an implementation example of the bypassrouting approach.

Bypass routing in SLR localizes the reaction to a routefailure and initiates only a local recovery procedure. Usinglink state information from the MAC layer, a local patch onthe route is created bypassing the broken link. By using lo-cal recovery with link state information, the chances ofrecovering from a broken route increases. When a routefails, first the local route cache is searched for an alternateroute. If no route is available, bypass recovery is initiatedby querying the neighbors. Based on replies from thesenodes, the path is again reconstructed. MAC caches providethe connectivity information to immediate one-hop neigh-bors and the state is updated by each node whenever anycommunication is heard in the neighborhood. MAC cachesalso help determine the validity of the routes in the routecaches. Error recovery is initiated by salvaging used routecaches, bypass recovery and error reporting (see Fig. 12).The node looks for alternative routes to the destinationin its cache to salvage a packet. If the salvaged packetreaches its destination, the source receives an enhancedroute error message from the destination stating that thesalvaged route is in fact an alternate to the broken route.

Hint based probabilistic protocol [13]: Beraldi et al.propose a probabilistic forwarding framework which usesmeta-information to forward packets towards the generaldirection of the destination. The meta-information is

S

N

U

T

M

L

Y

P

R

X

Fig. 12. Error recovery

provided in terms of hints at each node. A hint, hid,computed by node i with respect to the destination noded, is represented by a positive value indicating the chanceof node i residing in the neighborhood of node d. A lowerhint value equates to a higher probability with hid = 0 whennodes i and d are one-hop neighbors.

When a node forwards a packet to the destination d itwill try sending to the neighbor with the best hint whichhas not been tried before for the same packet (effectivelytrying various alternatives in the order of the hint). If nosuch neighbor exist, the packet is discarded. All the nodesbroadcast periodic heartbeat packets which are then usedto create a vector of time information for each neighbor.This time information vector is utilized to calculate thehint values. Hints are disseminated by broadcasting thecontrol messages through the periodic beacon messages.Every node receives hints from nodes at most L hopsaround it. This is the lookahead value for the protocol.The authors show that a small lookahead value is sufficientfor a node to gather correct hints. Smaller lookahead valuesalso have an added advantage of lower control overheads.

Labeled distance routing (LDR) [40]: LDR, by Garcia–Luna–Aceves et al., is based on AODV but uses distance la-bels instead of sequence numbers to ensure loop freedomin the network. It utilizes a loop free invariant for each des-tination with the sequence numbers which can only beincremented by the destinations. The sequence numbersare used for path resets. An advertisement denotes an offerfor a route to the destination while a solicitation denotes arequest for information to a destination. To ensure loopfreedom, RREQs are disseminated within a tree which en-forces a strict ordering of feasible distances along successorpaths. The tree is created by the regular reverse-path flood-ing as in AODV. Ordering of feasible distances from sourceto destination is attained by adhering to the followingconstraints: i) numbered distance condition, ii) feasibledistance condition, and iii) start distance condition. SinceLDR is dynamic, the computed path graph must alsodynamically adapt to changing node positions and linkconditions. A route discovery is initiated by a source nodewith a solicitation for a destination. Nodes relaying thesolicitation participate in the route computation by cach-ing the path data for a fixed period of time. The destina-tions or nodes having a path to the destination then sendthe RREPs along the reverse path.

Dynamic backup routes routing protocol (DBR2P)[125]: Wang and Chao present an on-demand routing

V

D

Connectivity

Original route of the packet

Salvage path

Local recovery path

example [108].


protocol which does not require any routing table. Destina-tion nodes send back entire routes to the source node whilesetting up multiple backup routes dynamically. These back-up routes are used in the event of a link failure. Intermediatenodes also receive and transmit request packets from thesource nodes to gather more information in order to createthe backup routes. During route discovery, the sourcesends an RD-request packet with a unique sequencenumber and a route content field storing the addresses ofall the nodes in the path. The sequence number distin-guishes between route discoveries from different sourceand destination pairs. RD-request packets received withduplicate sequence numbers are discarded. Route mainte-nance uses acknowledgment packets at the data link layerto detect lost or corrupted packets. Passive acknowledg-ments, where a node hears the packet forwarded by itsneighbor to the next node, can also be used by the protocol.On link failure, data packets are buffered at the failed nodetill a backup route comes from an upstream node whichcaches backup routes. These backup routes are saved atspecific backup nodes on the route during path discovery.

Refinement based routing (RBR) [75]: Liu and Lin pro-pose a refinement based route maintenance mechanismwhich adds proactive route selection and maintenance toon-demand routing approaches. RBR consists of two mech-anisms: passive probe route-redirection (P � PR2) andactive probe route-redirection (A � PR2). P � PR2 dynami-cally repairs broken routes through a node called a pas-sive-redirector that redirects the broken source to itselfbefore connecting it to a new node located in the vicinityof the redirector and close to the broken path. A � PR2 usesan active redirector which searches for available shorterpaths progressively by overhearing two nodes on the origi-nal path. These nodes should be two-hops away and withinthe vicinity of the active-redirector. These two mecha-nisms are incorporated into AODV routing protocol.Fig. 13 illustrates the concept of the vicinity of a node.

3.1.1. ComparisonSource routing protocols have been a research topic of

high interest for the ad hoc community. There are severalreasons for this. First, while table driven protocols can beperceived as extensions or adaptations of existing proto-

A

X

D

C

B

Y

HCj(Y) = 2

HCj(X) = 1

vicinity of node Y

vicinity of node X

Fig. 13. Vicinity of node [75].

cols from the wired domain, reactive routing is specific toad hoc networks. Furthermore, reactive routing is bestadapted to the most challenging incarnations of the adhoc networks: situations where the mobility of the nodesis so high that any route found between a source and a des-tination will inevitably be temporary in nature. Not onlythat routes are created on-demand, but the probability ofa route failing is so high that the efficiency of respondingto failures is a major design consideration.

The baseline for this class of protocols is set by AODVand DSR, both of them have several independent imple-mentations for various operating systems. A large numberof other protocols in this class can be seen as attempts toimprove the performance of these protocols under variousoperating scenarios. Some of these scenarios involve addi-tional assumptions about the nodes capabilities, for in-stance the ability to possess a way to accurately measurelocation, such as a GPS, or the ability to measure thestrength of a received signal.

One of the main improvement directions have beenabout the handling of recovery of failed routes. This canbe approached from several directions:

� Improving the speed of rebuilding the routes: protocolsof this class are TORA [89], AODV-ABR[65] (rebuildingwith backup routes), low overhead dynamic routerepair [135], SLR [108] (using bypass routing), DBR2P[125] (backup routes), RBR [75] (using routeredirectors).� Choosing paths which are predicted to be more stable:

ABR [116] (based on associativity), SSBR [37] (basedon signal stability), FORP [113] (by predicting futurechanges), LLR [28] (favoring long lifetime routes), QMRB[52] and LBAQ [136] (based on prediction of future linkavailability).� Based on prediction of failure and preemptive rebuild-

ing of routes: the preemptive routing by Goff et al.[47], SWORP [123] (using a GPS signal to predict whena pair of moving nodes might disconnect).

Another direction of optimization is the lowering of thecost of route discovery. One immediate way to performthis is by taking advantage of location information. Thisway, these protocols represent a bridge towards location-aware routing. From the protocols discussed in this section,polarized gossip [12] and OD-PFS [4] fall in this category.Other protocols work to lower the cost of route discoverywithout relying on location: Adjusted probabilistic routediscovery [1] and recycled path routing (RPR) [38].

Another direction of work is to improve upon the tran-sitory behavior of the baseline protocols – for instance LSR[101] and LDR [40] improve upon AODV’s behavior withregards to temporary loops and the phenomena of count-to-infinity.

Finally, there are some protocols which extend upon thebaseline by considering various additional networkingchallenges, such as Quality of Service (LBAQ [136] andQMRB [52]), interference and channel assignment [48] orintermittent connectivity (SCaTR [17]).

The novel nature of source originated routing encour-aged researchers to branch out from the traditional


comfort zone of networking (which implies reliance ongraph theory and flow optimization as the mathematicalbackground). A number of innovative and interdisciplinaryapproaches have been applied to source originated routing.In the papers surveyed in this section, we have two ap-proaches based on ant colony optimization (ARA [50] andDAR [104]), an approach based on game theory FDG [85]as well as a probabilistic routing approach [13].

Table 3.1.1 summarizes the papers reviewed in this sec-tion and compares some of their key features.

3.2. Table-driven protocols

Table driven protocols always maintain up-to-dateinformation of routes from each node to every other nodein the network. Routing information is stored in the rout-ing table of each node and route updates are propagatedthroughout the network to keep the routing informationas recent as possible. Different protocols keep track of dif-ferent routing state information; however, all of them havethe common goal of reducing route maintenance overheadas much as possible. These types of protocols are not suit-able for highly dynamic networks due to the extra controloverhead generated to keep the routing tables consistentand fresh for each node in the network.

Destination-Sequenced Distance-Vector (DSDV) [93]:Perkins and Bhagwat introduced Destination-SequencedDistance-Vector (DSDV), one of the earliest ad hoc routingprotocols. As many distance-vector routing protocols, it re-lies on the Bellman-Ford algorithm. Every mobile nodemaintains a routing table which contains the possible des-

Table 3.1.1Reactive routing protocols comparison.

Protocol MR Route metric Route reposi

DSR Yes SP or next available RCAODV No Newest route and SP RTTORA Yes SP or next available RTABR No Strongest associativity RTSSBR No Strongestquit signal strength RTGoff et al. Yes SP RTAQOR No Link bandwidth RTARA Yes SP RTROAM Yes SP RTFORP No First created route RT at clusterSCaTR Yes Proxy contact value RTDAR Yes Weighted probabilities Stochastic RTBeraldi No Forwarding probability –QMRB No QoS metrics RTYu et al. No SP or backup route RCLABQ Yes LA, Q, and E RCLSR Yes Relay sequenced label (RRL) RTOD-PFS No VBR and CH RTSWORP No RET, EC, HC RTRPR No Newest route and SP RTGRP Yes SP RCSLR Yes SP or next available RCBeraldi et al. Yes Hint value RCLDR No Newest route and SP RTDBR2P No SP None

MR = Multiple routes.Route metric: SP = Shortest path; LA = Link availability; Q = Quality; E = energ

VBR = Virtual backbone routing; CH = clusterhead.Route repository: RC = Route cache; RT = Routing table.CO = Communication overhead.

tinations in the network together with their distance inhop counts. Each entry also stores a sequence numberwhich is assigned by the destination. Sequence numbersare used in the identification of stale entries and the avoid-ance of loops. In order to maintain routing table consis-tency, routing updates are periodically forwardedthroughout the network. Two types of updates can be em-ployed; full dump and incremental. A full dump sends theentire routing table to the neighbors and can require mul-tiple network protocol data units (NPDUs). Incrementalupdates are smaller (must fit in a single packet) and areused to transmit those entries from the routing tablewhich have changed since the last full dump update. Whena network is stable, incremental updates are forwardedand full dump are usually infrequent. On the other hand,full dumps will be more frequent in a fast moving network.In addition to the routing table information, each route up-date packet contains a distinct sequence number assignedby the transmitter. The route labeled with the most recent(highest number) sequence number is used. The shortestroute is chosen if any two routes have the same sequencenumber.

Analysis of a randomized congestion control schemewith DSDV routing in ad hoc wireless networks [20]:Boukerche et al. describe a randomized version of theDSDV protocol (R-DSDV) where the control messages arepropagated based on a routing probability distribution.Local nodes can tune their parameters to the traffic androute the traffic through other routes with lighter load.This implies implementing a congestion control schemefrom the routing protocol’s perspective.

tory Route rebuilding CO

New route and notify source HighSame as DSR or local repair HighReverse link and repair route HighLocal broadcast MediumNew route and notify source MediumPath discovery before probable route failure MediumInitiated from destination MediumAlternate route or backtrack MediumErase route, start new search Low

head New route and notify source MediumProxy requests HighNew route by sending forward ants MediumBroadcast HighSame as DSR or local repair HighLocal backup MediumSame as DSR or local repair HighSame as DSR or local repair HighLocal repair MediumSame as DSR or local repair HighLocal repair HighBackup route HighNew route and notify source HighLocal broadcast HighSame as DSR or local repair HighLocal repair Low

y used; RET = Route expiration time; EC = Error count; HC = Hop count;


The randomization of the algorithm is with respect tothe routing table advertisement packets and the rate atwhich they are forwarded. In DSDV, whenever there isany change in the routing table, advertisement packetsare propagated to update the state information at eachnode. R-DSDV sends these update messages only at a prob-ability Prn,adv for a node n in the network. This can reducethe control packet overhead; however, there may be a cor-responding delay in updating all the nodes. If there is arouting table update at node n, the node can send a regularmessage with a probability 1 � Prn,adv or an update mes-sage with probability Prn,adv. The sending rate of the tableadvertisement is qn = Fsendx Prn,adv, where Fsend is the fre-quency at which a node is allowed to send a message.The scheme allows the piggybacking of the routing tableupdates on the regular messages.

Optimized link state routing (OLSR) [31]: Clausenet al. designed the OLSR algorithm which improves onthe classical link state protocols through several optimiza-tions targeted towards wireless ad hoc networks. Theseoptimizations are centered on specially selected nodescalled multipoint relays (MPR). First, only MPR’s forwardmessages during the route information flooding process,substantially reducing the total number of messages for-warded. In addition, link state information is generatedonly by the MPRs, further reducing the amount of datawhich needs to be disseminated. Finally, the MPRs mightchoose to report only links between themselves and theirMPR selectors. This last technique of partial link stateinformation is a departure from the customary approachof link state protocols which relay on the disseminationof the full link state.

The multipoint relay aims to reduce retransmissionswithin the same region. Each node selects a set of one-hop neighbors which are called the multipoint relays(MPR) for the node. The neighbors of the node which arenot MPRs process the packets but do not forward themsince only the MPRs forward the packets. The multipointrelay set must be chosen such that its range covers allthe two-hops neighbors. This set must also be the mini-mum set to broadcast the least number of packets. Themultipoint relay set of a node N should be such that everytwo-hops neighbors of N has a bi-directional link with thenodes in the MPR set of N. These bi-directional links can bedetermined by periodic HELLO packets containing infor-mation about all neighbors and their link status. Thus, aroute is a sequence of hops from a source to a destinationthrough multipoint relays within the network. The sourcedoes not know the complete routes only next hop informa-tion to forward the messages.

A hierarchical proactive routing mechanism for mo-bile ad hoc networks (HOLSR) [119]: Villasenor-Gonzalezet al. networks where some nodes have significantly higherresources (transmission range, bandwidth, directional an-tenna and so on). The authors notice that traditional, flatrouting protocols cannot efficiently exploit the capabilitiesof the nodes with high resources. For this scenario, theauthors propose the HOLSR algorithm which builds uponthe OLSR protocol by introducing a hierarchical architec-ture with multiple ad hoc networks at distinct logical lev-els within the network.

The HOLSR network arranges the nodes in three topol-ogy levels, depending on their capabilities. The low capa-bility nodes at topology level 1 have only one wirelessnetwork interface and communicate with nearby nodes.Nodes on topology level 2 are assumed to have two wire-less interfaces, possibly relying on different wireless stan-dards, allowing them to communicate with nodes attopology levels 1 and 2. Finally, the nodes at topology level3 are the most powerful (e.g. airborne nodes) and can haveup to three wireless interfaces, allowing them to commu-nicate with nodes at topology levels 1, 2 and 3. At each le-vel, the mobile nodes can self-organize into clusters, witheach node participating in multiple topology levels auto-matically becoming a clusterhead at the lower level.

Each cluster node maintains a routing table with rout-ing information about nodes in the cluster. Higher cluster-heads contain bigger routing tables since they have tomaintain routes to all nodes at lower clusters. For lower le-vel nodes, this overhead is minimal. A hierarchical topol-ogy control (HTC) message is used to send clustermembership information from a lower to higher level.Topology control is carried out using these control mes-sages using the clusterheads as the gateway nodes.

Clusterhead gateway switch routing (CGSR) [23]: TheCGSR protocol, by Chiang et al., uses a distributed algorithmcalled the Least Cluster Change (LCC). By aggregating nodesinto clusters controlled by the clusterheads, a framework iscreated for developing additional features for channel ac-cess, bandwidth allocation and routing. Nodes communi-cate with the clusterhead which in turn communicateswith other clusterheads within the network (see Fig. 14).

The selection process of a clusterhead is an importanttask since changing clusterheads frequently adversely af-fect the resource allocation algorithms. Thus, cluster stabil-ity is of primary importance in this scheme. The LCCalgorithm is considered stable since the clusterheads willchange only under two conditions: when two clusterheadscome within the range of each other or when a node getsdisconnected from any other cluster.

CGSR is an effective way for channel allocation withindifferent clusters by enhancing spatial reuse. Each clusteris defined with unique CDMA code and hence each clusteris required to utilize spatial reuse of codes. Within eachcluster, TDMA is used with token passing.

Gateway nodes are members of more than one cluster;therefore, they need to communicate using different CDMAcodes based on their respective clusterheads. The main fac-tors affecting routing in these networks are token passing(in clusterheads) and code scheduling (in gateways). Apacket is routed through a collection of these clusterheadsand gateways in this protocol.

Wireless routing protocol (WRP) [83]: Murthy andGarcia–Luna–Aceves propose WRP which builds upon thedistributed Bellman-Ford algorithm. The routing table con-tains an entry for each destination with the next hop and acost metric. The route is chosen by selecting a neighbornode that would minimize the path cost. Link costs are alsodefined and maintained in a separate table and varioustechniques are available to determine these link costs.

To maintain the routing tables, frequent routing updatepackets must be forwarded to all the neighbors of a node

1

2

3

4

5

6

7

8

Node

Clusterhead

Gateway

Fig. 14. Cluster Gateway Switch Routing [23].

a

i

h

g f

e

d

c

b

jm

x

v

s

t

rk

q

w

z

y

hNode selecting MPR One-hop neighbor 2-hop neighbor

Fig. 15. Example for multipoint relay selection [82].


and contain all the routes in which the node is aware of.Since these are just update messages, only the recent pathchanges are included instead of the whole routing table. Tokeep the links updated, empty HELLO packets are for-warded at periodic intervals only if no other update mes-sages need forwarding.

Global state routing (GSR) [27]: Chen and Gerla pro-pose the GSR protocol, where the control packet size is ad-justed to optimize the MAC throughput.

Each node maintains the neighbor list and three routingtables containing the topology, the next hop, and the dis-tance respectively. The neighbor list contains all neighborsof the current node. The topology table contains the linkstate information and a timestamp indicating the time inwhich the link state information is generated. The nexthop table contains a list of next hop neighbors to forwardthe packets while the distance table maintains the shortestdistance to and from the node to various destinations. Aweight function computes the distance of a link whichmay be replaced by other QoS routing parameter.

Initially each node in the network starts with an emptyneighbor list and a topology table. It learns about its neigh-bors by the sender field of the incoming packets. By pro-cessing these packets to obtain link state information, thebest route to the destination is computed. After all therouting messages are processed, the routing table is cre-ated and shared with other nodes by broadcasting it tothe next hop neighbors. This process is carried out period-ically to maintain the most up-to-date information.

Source-tree adaptive routing (STAR) [43]: Garcia–Luna–Aceves and Spohn propose STAR where each nodemaintains a source tree which contains preferred links toall possible destinations. Nearby source trees exchangeinformation to maintain up-to-date tables. A routeselection algorithm is executed based on the propagatedtopology information to the neighbors. The routes aremaintained in a routing table containing entries for thedestination node and the next hop neighbor. The link stateupdate messages are used to update changes of the routesin the source trees. Since these packets do not time out, noperiodic messages are required. The STAR protocol pro-vides two distinct approaches: optimum routing (ORA)

and least overhead routing (LORA). The ORA approach ob-tains the shortest path to the destination while LORA min-imizes the packet overhead. STAR also requires a neighborprotocol to make sure that each node is aware of its activeneighbors. The STAR protocol has been further developedas SOAR [105].

OLSR with quality of service (QOLSR) [82]: Munarettoand Fonseca design the QOLSR protocol by adding the QoSparameters of delay and bandwidth to the standard OLSR.Three new heuristics, QOLSR1, QOLSR2 and QOLSR3, areproposed for multipoint relay selection (see Fig. 15). Theseheuristics select multipoint relays (MPRs) within the net-work based on various QoS parameters. QOLSR1 choosesthe neighbor node which can reach the largest number ofnodes (node with the maximum degree). The priority is gi-ven to the neighbor with smallest delay in the case of mul-tiple neighbors with identical maximum degrees. QOLSR2prioritizes the neighbor with the smallest delay. If multiplenodes have the same minimum delay, then the node withthe highest degree is chosen. The final heuristic selects thenode with the smallest delay among neighbors within atwo hop distance.


3.2.1. ComparisonIn contrast to source initiated routing, table driven

routing has extensive precedents in the research done forrouting in the wired domain. Nevertheless, the require-ments of the ad hoc routing are such that none of the wiredrouting protocols could be successfully transferred into thead hoc wireless domain.

On the other hand, the two main classes of wired rout-ing protocols have inspired their own classes of protocolsin table driven ad hoc routing.

One of these classes is the distance vector protocols,where the nodes maintain only a local topology, and usethe distributed Belman-Ford algorithm (or its variations)to maintain the routing tables. In the wired Internet thisclass contains protocols such as RIP and IGRP. In the adhoc network domain the most popular protocol of thisclass is DSDV [93], another protocol being WRP [83]. Theother class of protocols is the link state routing protocolswhere the routers exchange full topology information,and then use a graph-theoretic shortest path algorithm(such as Dijkstra’s) on the resulting graph. On the wiredInternet this class is represented by algorithms such asOSPF and IS-IS. The most representative example in thead hoc wireless domain is the OLSR [31] protocol. Betweenthese two classes we find protocols which transfer a partialtopology, such as STAR [43].

Most of the recent work on table-driven protocols canbe seen as improvements on these baseline distance vectorand link state approaches.

One direction of research is the adaptation of the rout-ing decisions to the traffic, proposed in the randomizedcongestion control scheme for DSDV [20]. Another classof protocols aim to improve the scalability of table-drivenprotocols. The subnetwork approach, successfully appliedon the wired Internet, cannot be directly applied in adhoc networks due to the much more variable connectionstructure. On the other hand, the connections in ad hocnetworks are strongly correlated with the physical proxim-ity, which allows the development of clustering ap-proaches (e.g. CGSR [23]), possibly integrated with ahierarchical model (such as in HOLSR [119]). Anotherdirection of research is the consideration of quality of ser-vice, as in QOLSR [82].

Table 3.2.1 summarizes the protocols reviewed in thissection and compares some of their key features.

Table 3.2.1Proactive routing protocols comparison.

Protocol Tables Update interval

DSDV 2 PeriodicR-DSDV 2 ProbabilisticOLSR 3 PeriodicHOLSR 3 PeriodicCGSR 2 PeriodicWRP 4 PeriodicGSR 3 Periodic only with neighborsSTAR 1 Only at specific eventsQOLSR 3 Periodic

Routing metric: SP = Shortest path; HC = hop count.CO = Communication overhead [High = H; M = Medium; L = Low].

3.3. Hybrid protocols

The hybrid routing schemes combine elements of on-demand and table-driven routing protocols. The generalidea is that area where the connections change relativelyslowly are more amenable to table driven routing whileareas with high mobility are more appropriate for sourceinitiated approaches. By appropriately combining thesetwo approaches the system can achieve a higher overallperformance.

Zone routing protocol (ZRP) [106]: The ZRP protocol,designed by Samar et al. is designed to be used in largescale networks. The protocol uses a pro-active mechanismof node discovery within a node’s immediate neighbor-hood while inter-zone communication is carried out byusing reactive approaches.

Local neighborhoods, called zones, are defined for nodes(see Fig. 16). The size of a zone is based on q factor definedas the number of hops to the perimeter of the zone. Theremay be various overlapping zones which helps in routeoptimization.

Neighbor discovery is accomplished by either IntrazoneRouting Protocol (IARP) or simple Hello packets. IARP ispro-active approach and always maintains up-to-daterouting tables. Since the scope of IARP is restricted withina zone, it is also referred to as ‘‘limited scope pro-activerouting protocol’’. Route queries outside the zone are prop-agated by the route requests based on the perimeter of thezone (i.e. those with hop counts equal to q), instead offlooding the network. The Interzone Routing Protocol(IERP) uses a reactive approach for communicating withnodes in different zones. Route queries are forwarded toperipheral nodes using the bordercast resolution protocol(BRP). The ZRP architecture can be seen in Fig. 17.

Fisheye state routing (FSR) [90]: Pei et al. propose theFSR protocol which takes inspiration from the ‘‘fisheye’’technique of graphic information compression proposedby Kleinrock and Stevens. When adapted to a routing table,this technique means that a node maintains accuracy dis-tance and path quality information about its immediatevecinity, but the amount of detail retained decreases withthe distance from the node. Each node considers a numberof surrounding fish-eye scopes, areas which can be reachedwith 1, 2, . . . hops. A higher frequency of update packetsare generated for nodes within smaller scope while the

Critical node Routing metric CO

– SP L– SP L– SP H– SP HClusterhead SP L– SP L– SP L– SP L– Degree, delay, & HC H

S

2

1

6

4

5

8

7

9

3

Fig. 16. Example routing zone with q = 2 [106].

IARP IERP

BRP

NETWORK LAYER

MAC LAYER NDP

Packet flow

Interprocesscommunication

Fig. 17. ZRP architecture [106].


updates are fewer in general for farther away nodes. Eachnode maintains a local topology map of the shortest pathswhich is exchanged periodically between the nodes. Withan increase in size of the network, a ‘‘graded’’ frequencyupdate plan can be adopted across scopes to minimizethe overall overhead.

This approach makes FSR an implicit hierarchical proto-col. Its main advantage is the significant reduction of thecontrol overhead.

Landmark ad hoc routing (LANMAR) [91]: Pei et al.propose LANMAR which builds subnets of groups of nodeswhich are likely to move together. A landmark node iselected in each subnet, similar to FSR [90]. The LANMARrouting table consist of only the nodes within the scopeand landmark nodes. During the packet forwarding pro-cess, the destination is checked if it is within the forward-ing node’s neighbor scope. If so, the packet is directlyforwarded to the address in the routing table. If a packeton the other hand is destined to a farther node, it is firstrouted to its nearest landmark node. As the packet gets clo-ser to its destination, it acquires more accurate routinginformation, thus in some cases it may bypass the land-mark node and routed directly to its destination. Duringthe link state update process, the nodes exchange topologyupdates with their one-hop neighbors. A distance vector,which is calculated based on the number of landmarks, isadded to each update packet. As a result of this process,

the routing tables entries with smaller sequence numbersare replaced with larger ones.

Relative distance micro-discovery ad hoc routing(RDMAR) [2]: RDMAR, by Aggelou and Tafazolli, has dis-tinct route discovery and route maintenance phase. How-ever, the route discovery broadcast messages are limitedby a maximum number of hops calculated using the rela-tive distance between the source and the destination. Eachnode also maintains a routing table containing the nexthop neighbor of each known destination, an estimated rel-ative distance between all known source and destinationnodes, a timestamp at which the current entry was made,a timeout field indicating the time at which a particularroute is no longer active and a flag specifying if a route stillexists or not.

The estimated distances are measured by the sourcenodes using the last known distance between the respec-tive nodes and the estimated speed of the destinationnode. Each node also maintains–a data retransmission buf-fer which queues data being transmitted until an explicitacknowledgment is received and a route request tablewhich stores all necessary information pertaining to themost recent route discovery.

Route discovery and route maintenance is carried outby broadcasting route request packet and expecting a routereply packet from the destination. Each node also occasion-ally probes for bi-directional links by sending a packet onthe link where it has just received a packet. Route mainte-nance is performed when a route failure occurs and thenode re-sends the data up to a maximum number of re-tries. This is why the intermediate nodes buffer data pack-ets until they receive link level acknowledgments from thenext-hop node. When a link failure occurs at an intermedi-ate node close to the destination, this node sets the ‘‘emer-gency’’ flag in its route request packets such that itincreases the possibility of a faster recovery time. If theroute has failed, the intermediate node forwards a failurenotification to the source node by unicasting it to all neigh-boring nodes. When a node receives a failure notification, itupdates its routing tables accordingly.

Scalable location update based routing protocol(SLURP) [127]: SLURP, by Woo and Singh, develops anarchitecture scalable to large size networks. A location up-date mechanism maintains location information of thenodes in a decentralized fashion by mapping node IDs tospecific geographic sub-regions of the network where anynode located in this region is responsible for storing thecurrent location information for all the nodes situatedwithin that region. When a sender wishes to send a packetto a destination, it queries nodes in the same geographicsub-region of the destination to get a rough estimate ofits position. It then uses a simple geographic routing proto-col to send the data packets. Since the location update costis dependent on the speed of the nodes, for high speeds,more number of location update messages are generated.By theoretical analysis, it is shown that the routing over-head scales as O(v) where v is the average node speed,and O(N3/2) where N is the number of nodes within thenetwork.

Zone based hierarchical link state routing protocol(ZHLS) [56]: Joa-Ng and Lu propose ZHLS routing protocol


where a hierarchical structure is defined by non-overlappingzones with each node having a node ID and a zone ID.These IDs are calculated using an external location tool suchas GPS. The hierarchy is divided into two levels: the nodelevel topology and the zone level topology. There are noclusterheads in ZHLS.

When a route is required for a destination located in an-other zone, the source node broadcasts a zone-level loca-tion request to all other zones. Once the destinationreceives the location request, it replies with the path. Inthis technique, only the node and zone IDs of a node is re-quired to discover a path. There is no need for updates aslong as the node stays within its own region and the loca-tion update is required only if the node switches regions.The only drawback of ZHLS is that all nodes should havea preprogrammed static zone map to recognize the zonescreated in the network. This may not be possible in scenar-ios where the network boundaries are dynamic in nature.On the other hand, it is suitable for the networks deployedwith fixed boundary lines.

Distributed spanning tree (DST) routing [96]:Radhakrishnan et al. present a routing algorithm whichuses distributed spanning trees. There can be regions ofdifferent stability in the network and a backbone networkmust be created within the stable regions. All the nodesin the network are aggregated into a number of trees rootedat a particular node. These trees are composed of root andinternal nodes. The root node can make various decisionssuch as whether the current tree could join with anothertree, whether other nodes could join at appropriate posi-tions within the tree, and so on. The other nodes in the treeare considered regular nodes with no extra or unique func-tionality. Each node can be in three different states: router,merge, and configure. DST proposes two strategies to deter-mine a route between a source and a destination pair:

1. Hybrid Tree Flooding (HTF): In this scheme, the sourcesends the control packets to all the neighbors andadjoining bridges in the spanning tree. Each packet isremained static at these places for a specific holdingtime. This serves as a buffering strategy for these nodesto send the packets as network connectivity increaseswith time. Hence, the network becomes gradually morestable and lesser number of packets are dropped due tolink failures.

2. Distributed Spanning Tree (DST) shuttling: In thisapproach, the source sends the control packets to thetree edges till each of them reaches a leaf node. Whena packet reaches the leaf node, it is forwarded to a shut-tling level, i.e. a particular height on the tree. When thepacket arrives at the shuttling level, it is sent down theadjoining bridges. This helps to selectively forward thecontrol packets within the network.

The drawback with such an architecture is the existenceof a single point of failure for the entire tree. If the root nodefails, the entire routing structure falls apart. Also the hold-ing time metric may factor in some additional delays andmay not be suitable for transmitting real-time data.

Distributed dynamic routing (DDR) Algorithm [87]:Nikaein et al. propose a tree-based routing protocol with-

out the need of a root node. Periodic beacon messages areexchanged among neighboring nodes to construct a strat-egy tree. These trees within the network form a forest withthe created gateway nodes acting as links between the treesin the forest. These gateway nodes are regular nodesbelonging to separate trees but within transmission rangeof each other. A zone naming algorithm is used to assign aspecific zone ID to each tree within the network. Hence,the overall network now comprises of a number of overlap-ping zones (if each tree is considered to be a zone).

The DDR algorithm comprise of the following sixphases: (i) preferred neighbor election; (ii) intra-tree clus-tering; (iii) inter-tree clustering; (iv) forest construction;(v) zone naming; and (vi) zone partitioning. Initially, eachnode starts with the preferred neighbor election phase inwhich the preferred neighbor is the one with the highestnumber of neighbors. A forest is constructed by connectingeach node to their preferred neighbor after which the in-tra-tree clustering scheme is used to generate the zonestructure. It also helps set up the intra-zone routing tablefor communication within the tree itself. Through the in-ter-tree clustering, connectivity between trees is achieved.The zone naming algorithm is executed to assign zone IDsto each of the zones, resulting of partitioning the networkinto the various zones based on the assigned zone IDs. Thiswork is extended in [86] by introducing a routing frame-work which uses the intra and inter zone routing tablescreated in DDR.

A4LP routing protocol [120,121]: A4LP, by Wang et al.,is specifically designed to work in networks with asym-metric links. The routes to In-, Out-, and In/Out-boundneighbors are maintained by periodic neighbor updateand immediately available upon request, while the routesto other nodes in the network are obtained by a path dis-covery protocol. A4LP proposes an advanced flooding tech-nique – m-limited forwarding. Receivers can re-broadcast apacket only if it qualifies a certain fitness value specified bythe sender. The flooding cost is reduced and shortest highquality path is likely to be selected by using m-limited for-warding. Moreover, the metrics used to choose from multi-ple paths are based on the power consumed per packet andtransmission latency. A4LP, is also both location- and power-aware routing protocol supporting asymmetric links thatmay be suitable for heterogeneous MANET.

Hybrid ant colony optimization (HOPNET) [122]:Wang et al. present a hybrid routing algorithm based onAnt Colony Optimization (ACO) and zone routing. It consid-ers the scenario of ants hopping from one zone to the nextwith local proactive route discovery within a zone and reac-tive communication between zones. The algorithm borrowsfeatures from ZRP and DSR protocols and combines it withACO based schemes. The forward ants are sent only to bor-der nodes. These forward ants are then directed towardsthe destination node by using the nodes’ local routing table.The ants move from one zone to another via border nodesand by using available local routing information. The zoneapproach achieves the scalability. Link failures are handledwithin a zone without flooding the network. Inter and intrazone routing tables are always maintained which can effi-ciently rediscover a new route in case of a link failure. Anexample scenario is shown in Fig. 18.


Link reliability based hybrid routing (LRHR) [129]:Xiaochuan et al. observe that frequent topology changesin MANETs may require the dynamic switching of table-driven and on demand routing strategies. The LRHR proto-col achieves this switching in a smooth and adaptivefashion. Each node operates in a promiscuous receive modeto overhear any packet transmission in the neighborhoodand setup multiple routes from the source to the destina-tion. The LRHR assigns edge weights between links basedon the link reliability. The higher the value of edge weight,the greater the reliability. It finally selects the route whichhas the maximum edge weight sum as the main route. TheLRHR primarily operates in the table-driven mode andswitches to on demand mode when a source either hasno route to a destination or when the time interval be-tween a new route discovery and the previous route dis-covery phase is larger than the minimum route requestinterval. The route discovery process in LRHR is similar toDSR. The route maintenance operations are carried outbased on the edge weights between the nodes.

Fisheye zone routing protocol (FZRP) [133]: Yang andTseng combine the zone routing protocol with the fisheyestate routing mechanism. By using the concept of a fisheye,a multi-level routing zone structure is created where dif-ferent levels are associated with different link state updaterates (see Fig. 19). The source node generates a RREQ pack-et which is bordercast to nodes till the destination isreached. These packets are forwarded along the border ofthe zone. The nodes at the periphery of the zones forwardthe RREQ to a different zone if the destination node is notlocated within the current zone. A time to live (TTL) fieldis used within the forwarding packets to cover the zonein which the update packets are forwarded. The routing ta-ble at each node contains entries for nodes within its ownzone and for those in an extended zone. An extended zoneis a zone located beyond the inner (basic) zone. Extendedzone entries are generally not very accurate due to differ-ent update frequencies in different zones. Other proce-dures such as local route repair in case of a broken linkare similar to ZRP.

Ad hoc networking with swarm intelligence (ANSI)[97]: Rajagopalan and Shen propose a hybrid routing pro-tocol utilizing swarm intelligence (SI) to select good routesin a network. SI allows self-organizing systems and helpsmaintain state information about the network. ANSI em-ploys a highly flexible cost function which uses informa-tion collected from local ant activity. The protocol takesadvantage of the basic principles of ant based routing algo-

AQ

P

O

N

M

L

KJ

I

H

GF

E

DC

BT

U

R

S

Fig. 18. Example scenario [122]. If the radius of the zone is 2, for node S,nodes A, D, F, G are boundary nodes, and nodes B, C, E are interior nodeswhile all other nodes are considered exterior nodes.

rithms which allows the maintenance of multiple routes toa destination. In ANSI, the nodes using proactive routingperform stochastic routing to select the best path, whilethose performing reactive routing use the extra routes inthe event of route failure. The pheromone trail concept al-lows selecting the routes to the destination from everynode. Nodes using reactive routing first broadcast a for-ward reactive ant towards the destination. Once the desti-nation receives the forward ant, it replies with abackward reactive ant which updates routing tables for allnodes in the path. If route failure occurs at an intermediatenode, that node buffers the packets which could not berouted and sends a forward ant initiating route discovery.Nodes which are part of a less dynamic, infrastructurednetwork maintains routes proactively by periodic routingupdates using proactive ants. Proactive ants are not re-turned as the reactive ants, rather they help reinforce thepath taken by the ant. Local route management is achievedby reinforcement due to movement of data packets and anexplicit neighbor discovery mechanism (see Fig. 20). HEL-LO messages are also broadcast periodically through whichnetwork state information is easily exchanged.

Mobility aware protocol synthesis for efficient rout-ing [9]: Bamis et al. propose a new stability metric to deter-mine the mobility level of nodes in a network. Using thismetric, the nodes can be classified into different mobilityclasses in which they in turn determine the most suitablerouting technique for a particular source–destination pair.Stability uses the concept of associativity, which is the to-tal time for which nodes are connected through beacons.With the level of stability defined, a protocol frameworkis designed, which operates above the network layer onthe protocol stack, determines the optimal routing tech-nique. Minimal changes are needed to the original routingprotocols to ensure ease of integration.

Load balancing in MANET shortest path routing[112]: Souihli et al. achieve load balancing to enable effi-cient routing in MANETs. It has been observed that the loadis maximal at the center while it decreases farther from thecenter of the network. Essentially, the load becomes mini-mal at the network edges. The authors state that such aload imbalance takes place due to shortest-path routing

R = 4 R = 2

Basic zone

Extended zone

Fig. 19. Two level routing zone in FZRP [133].

j

i

D

S

j

i

Fig. 20. Local reinforcement in ANSI. (a) Reinforcement by data packets.Node i, upon receiving a data packet from S via node j, reinforces the pathto node j via j and the source S via j. (b) Reinforcement in neighbordiscovery mechanisms. Upon receiving a HELLO beacon from j all nodes ireinforce trails via j [97].


and propose a new routing metric, the node’s centrality,when choosing the best route. Thus, instead of selectingthe shortest routes, nodes with longer distances from thecenter of the network are chosen. A new routing metricis proposed as follows:

Minimize1n

Xn

k¼1

gðkÞ

where n is the number of nodes in the network and g(k)represents the centrality of node k. The g(k) value is calcu-lated based on the size of the node’s routing table. A great-er size denotes closer to the center of the network while alower size indicate otherwise.

3.3.1. ComparisonHybrid approaches are, in general, justified for large

networks – if a network is small, we can usually make aclear decision between source driven or table driven ap-proaches. Similarly, hybrid approaches cover a very widerange of approaches and intellectual ideas. Still we canidentify several guiding ideas.

First, there is a group of approaches which performs adifferential treatment of the network nodes based oneither (a) zones or (b) the nodes participation in abackbone.

Table 3.3.1Hybrid routing protocols comparison.

Protocol MR Route metric Route

ZRP No SP IntraZFSR No Scope range RTsLANMAR No SP RTs aRDMAR No SP RTSLURP Yes MFR: InterZ, DSR: IntraZ RC atZHLS Yes SP IntraZDST Yes Tree neighbor forwarding RTsDDR Yes Stable routing IIntraA4LP Yes Power consumed RTsHOPNET No SP IntraZLRHR Yes Edge weight RC, RFZRP No SP IntraZANSI Yes SP RT

MR = Multiple Routes.Route metric: SP = Shortest path; InterZ = Intrazone; IntraZ = Intrazone.Route repository: RC = Route cache; RT = Routing table.CC = Communication Complexity [High = H; M = Medium; L = Low].

The approaches which classify nodes into zones areusually counting the zones from the perspective of thesource node. The zones are usually defined based on hopcount in protocols such as ZRP [106], FSR [90] and RDMAR[2] although we also have protocols where the zone isbased on physical location such as in SLURP [127] and ZHLS[56]. In some cases, the zones might be mobile zones ofnodes moving together, as in LANMAR [91].

In backbone based approaches a subset of nodes whichhave more stable connections form a backbone which isfrequently organized into a tree structure. The assumptionis that nodes in the backbone will use table driven routing,while nodes outside the backbone will be reached withsource initiated routing. Example protocols are DST [96]and DDR [87]. The A4LP protocol [120,121] form a specialcase as it is designed to work in networks with asymmetricconnections.

A separate class of protocols are those which we couldcall explicit hybridization. In these protocols there are two,clearly separable routing models, which are frequently fullfeatured routing protocols on their own. The main chal-lenge of these protocols, naturally, is the appropriatechoice between the two protocols as well as their integra-tion in the systems – the basis of this decision is frequentlythe defining factor of the protocol. For instance, LRHR [129]takes the decision based on link reliability, while Bamiset al. [9] makes the decision based on mobility classes.There is also a wide variety among the protocols com-bined: Zone routing with ant colony optimization in HOP-NET [122], FSR with ZRP in FZRP [133], reactive ants withproactive ants in ANSI [97].

Table 3.3.1 summarizes the protocols reviewed in thissection and compares some of their features. (See Tables3.4.1 and 3.5.1)

3.4. Location-aware protocols

Location-aware routing schemes in mobile ad hoc net-works assume that the individual nodes are aware of thelocations of all the nodes within the network. The bestand easiest technique is the use of the Global Positioning

repository Route Rebuilding CC

and InterZ RTs Start repair at failure point MNotify source L

t landmark Notify source MNew route and notify source H

location Notify source Hand InterZ RTs Location request sent M

Pause times/Shuttling LZ and InterZ RTs Notify source L

Notify source Mand InterZ RTs Start repair at failure point H

T Route discovery Hand InterZ RTs Start repair at failure point M

Start repair at failure point M

Table 3.4.1Geographical routing protocols comparison.

Protocol Forwarding strategy Route metric Loop-free Scalability Robustness CO

LAR Directional flooding Hop count No No No MediumDREAM Flooding Hop count No No No LowGPSR Greedy SP Yes Yes No HighColagrosso et al. Directional flooding Hop count No No No LowALARM Directional flooding Hops and mobility Yes Yes No MediumREGR Directional flooding SP Yes No No LowLAKER Directional flooding Hop count No No No LowBlazevic et al. Multipath flooding Hop count Yes Yes Yes HighMORA Greedy Weighted hop count No No No HighOGPR Source routing SP Yes Yes Yes LowSong et al. Greedy and directional forwarding Hop count No No No MediumSOLAR Greedy geographic forwarding SP No No No MediumLBLSP Greedy geographic forwarding LSP and WDG Yes Yes Yes LowGLR Source routing SP Yes Yes No LowMER Greedy geographic forwarding Max. expectation No Yes Yes Low

Route metric: SP = Shortest path; LSP = Local shortest path; WDG = Weighted distance gain.CO = Communication overhead.

Table 3.5.1Multipath routing protocols comparison.

Protocol Proactive/Reactive Loops Route metric Route cache

CHAMP Reactive Yes Shortest path YesAOMDV Reactive No Advertised hopcount NoSMR Reactive No Least delay No, table at sourceNTBR Reactive Yes Link active YesDas et al. Reactive No Least delay YesTMRP Reactive No Auction winner NoLiu et al. Reactive No Shortest path YesSMORT Reactive No Shortest path YesRamasubramanian et al. Proactive No Preferred neighbor RTMuSeQoR Reactive No Shortest Path Yes


System (GPS) to determine exact coordinates of thesenodes in any geographical location. This location informa-tion is then utilized by the routing protocol to determinethe routes.

Location-aided routing (LAR) [62,63]: Ko and Vaidyapresent the LAR protocol which utilizes location informa-tion to minimize the search space for route discovery to-wards the destination node. LAR aims to reduce therouting overhead for the route discovery and it uses theGlobal Positioning System (GPS) to obtain the locationinformation of a node.

The intuition behind using location information to routepackets is very simple and effective. Once the source nodeknows the location of the destination node and also hassome information of its mobility characteristics such asthe direction and speed of movement of the destinationnode, the source sends route requests to nodes only inthe ‘‘expected zone’’ of the destination node. Since theseroute requests are flooded throughout the nodes in the ex-pected zone only, the control packet overhead is consider-ably reduced. If the source node has no information aboutthe speed and the direction of the destination node, the en-tire network is considered as the expected zone.

A source node before sending a packet determines thelocation of the destination node and defines its ‘‘requestzone’’, the zone in which it initiates flooding with the route

request packets. In some cases, the nodes outside the re-quest zone may also be included. If the source node isnot inside the destination node’s expected zone, the re-quest zone must be increased to accommodate the sourcenode. Also, a situation may occur where all neighboringnodes of the destination node may be located outside therequest zone. In this case, the request zone must be in-creased to include as many neighboring nodes as possible.

LAR defines two schemes to identify whether a node iswithin the request zone (see Fig. 21).

� Scheme 1: The source node simply includes the smallestrectangle containing the current location of the sourcenode and the expected zone of the destination nodebased on its initial location and current speed. Thespeed factor may be varied to either include the currentspeed or the maximum obtainable speed within thenetwork. This expected zone will be a circle centeredat the initial location of the destination node with aradius dependent on its speed of the movement. Thesource node sends the route request packets with thecoordinates of the entire rectangle. The nodes receivingthese packets check to see whether their own locationsare within the zone. If so, they forward the packet usingthe regular flooding algorithm, otherwise the packetsare simply dropped.

Fig. 21. LAR routing protocol. The diagrams (a) and (b) present LAR1 and LAR2 schemes [62,63].

S

D

Y

Fig. 22. Y is S’s closest neighbor in greedy forwarding [59].


� Scheme 2: The source node calculates the distancebetween itself and the destination node based on theGPS coordinates and includes these values within theroute request packets. An intermediary node receivingthis packet calculates its distance from the destination.If its distance from the destination is greater than thatof the source, the intermediary node is not within therequest zone and hence drops the packet. Otherwise,it forwards the packet to all its neighbors.

LAR essentially describes how location informationsuch as GPS can be used to reduce the routing overheadin an ad hoc network and ensure maximum connectivity.

Distance Routing Effect Algorithm for Mobility(DREAM) [11]: Basagni et al. propose the DREAM protocolwhich also uses the node location information from GPSsystems for communication. DREAM is a part proactiveand part reactive protocol where the source node sendsthe data packet ‘‘in the direction’’ of the destination nodeby selective flooding. Each node maintains table with thelocation information of each node and the periodic locationupdates are distributed among the nodes to keep thisinformation as up-to-date as possible. Collectively updat-ing location table entries indicates the proactive natureof the protocol while the fact that all intermediate nodesin a route perform a lookup and forward the data packetin the general direction of the destination, reflectsDREAM’s reactive properties.

DREAM is based on two classical observations: the dis-tance effect and the mobility effect. The distance effect statesthat the greater the distance between two nodes, theslower they appear to move with respect to each other.Hence, the location information tables can be updateddepending on the distance between the nodes withoutmaking any concessions on the routing accuracy. Twonodes situated farther apart view the other to be movingrelatively slowly, requiring less frequent location updatescompared with nodes closer to each other. The mobility ef-fect determines how often the location information pack-ets can be generated and forwarded. In an ideal scenario,whenever a node moves, it should update entire the net-work but not generate any packets if it remains idle. Thenodes with higher mobility generate more frequent loca-tion update messages. This allows each node to send con-trol packets based on their mobility and helps to reducethe overhead by a great extent.

Greedy Perimeter Stateless Routing (GPSR) [59]:GPSR, by Karp and Kung, also uses the location of the node

to selectively forward the packets based on the distance.The forwarding is carried out on a greedy basis by selectingthe node closest to the destination (see Fig. 22). This pro-cess continues until the destination is reached. However,in some scenarios, the best path may be through a nodewhich is farther in geometric distance from the destina-tion. In this case, a well known right hand rule is appliedto move around the obstacle and resume the greedy for-warding as soon as possible.

Let us note that the location information is shared bybeacons from the MAC layer. A node uses a simplistic beac-oning algorithm to broadcast beacon packets containingthe node ID and its x and y co-ordinates at periodic inter-vals, helping its neighbors to keep their routing tables up-dated. With greater mobility, the beaconing interval mustbe reduced to maintain up-to-date routing tables; how-ever, this results in greater control overhead. To reducethis cost, the sender node’s location information is piggy-backed with the data packets.

Dynamic route maintenance (DRM) for geographicforwarding [30]: Chou et al. propose a dynamic beaconingscheme to be used in geographic forwarding algorithms inMANETs. In beacon based protocols, each mobile nodetransmits periodic beacons to its neighbors to update andmaintain its routing table. The beacons are generally for-warded at fixed intervals of time. During low mobility, alonger interval would be the best as it would reduce con-trol overhead while providing accurate location informa-tion. However, in cases of higher mobility, determiningan appropriate beacon interval is rather difficult. In DRM,beacon interval and route information are carried outdynamically. Based on the node’s mobility information,its beacon interval is computed while the route manage-ment function updates the routing table entries. TheDRM algorithm is applied to GPSR forwarding algorithm.The results show that usage of DRM reduced the cost of


route maintenance in scenarios with low mobility and im-proved packet delivery rates where mobility among thenodes is higher.

Improvements to location-aided routing throughdirectional count restrictions [32]: Colagrosso et al. aimsto reduce the control packet overhead by reducing dupli-cate route formation packets. The enhancements are pro-posed to the LAR Box algorithm which is based on countrestriction [126] of rebroadcasts. A node after receivingthe route request packet waits for an assessment delay(AD) time interval before deciding whether to rebroadcastthe packet or simply drop it. This decision is base on thenumber of duplicate broadcasts received during the ADinterval and if this number is greater than a count thresh-old, then the packet is simply dropped. Using the countthreshold value, the number of control packets can be re-duced by not rebroadcasting them.

Adaptivelocation aided mobile ad hoc network rout-ing (ALARM) [19]: The Adaptive Location Routing (ALARM)algorithm, by Boleng and Camp, uses feedback for adapta-tion and location information for performance improve-ments. While using location information has shown toincrease efficiency, feedback is suggested as a mobilitymetric assisting ad hoc network protocols adapt to the cur-rent network scenario [18]. Link duration is considered as asuitable mobility metric since constant links for long peri-ods denote low mobility while links experiencing shorteraverage durations represent high mobility scenario. Essen-tialy, the link duration feedback agent allows the proposedprotocol to become adaptive. The ALARM protocol aims tooptimize the existing protocols and devise techniques tocombine multiple protocols into a hybrid protocol wheremore suitable techniques based on the present networkconditions can be employed.

ALARM uses the link duration feedback at each of themobile node to determine the appropriate forwardingscheme. A threshold duration is provided to make furtherdecisions. ALARM forwards the packets on the specifiedpath if the link duration of the source node is larger thanthe threshold, otherwise it performs a directed broadcast.In such a scheme, even though a broadcast flood is initi-ated, this flood may be ‘‘damped’’ by nodes closer to thedestination if they have stronger links. Hence, the propaga-tion method of the packets is selected based on the mobil-ity metric, the link duration. The important thing is todeliver the packet through regions of ‘‘high instability’’and it does not matter whether the packet reaches the finaldestination via flooding or through the original route se-lected. The flood horizon parameter limits the number ofhops to flood a packet, keeping the floods within a specifichorizon. No route repairs are required in this scheme sincethe packets can be forwarded over network ‘‘hot-spots’’ bydirected flooding resulting in decreased overall end-to-enddelay.

A region-based routing protocol for wireless mobilead hoc networks (REGR) [76]: The REGR protocol, pro-posed by Liu et al., dynamically creates a pre-routing regionbetween the source and the destination, hence control theflooding of route request packets within this region. Thecorrect selection of the region, which should not be toosmall, is important for the discovery of the optimal routes.

Once the pre-routing region is selected, all nodes withinthis region engage in forwarding route request packets.The two main characteristics of this protocol include a re-gion based route creation and a region based route update.While the route creation is carried out by using existingbroadcast schemes, it is followed by creation of a routingregion in the neighborhood of this preliminary route.When a route update is needed, this routing region is usedto propagate the route update packets. Route creation con-sists of the following phases:

Destination discovery: A destination location packet(DLOC) is first broadcast by the source node. Any existingbroadcast algorithm can be employed in this phase. Allnodes forwarding the DLOC store the path, so when thedestination is reached a preliminary route is ready.

Formation of the pre-routing region: Once the prelimin-ary route is selected, the destination broadcasts the RegionDefinition (RDEF) packets which contain the address of theprevious hop from the preliminary route and a REGION-WIDTH value which defines the maximum number of hopsfrom the nodes on the route. The nodes receiving the for-warded RDEF packets check whether they fall within thisrange or not. If so, they are included within the region.

In-region route discovery: A short period after the RDEF

packet is broadcast, the destination again broadcasts aRREQ packet. Those nodes which are marked within the re-gion by the RDEF packets re-broadcast the RREQ packets.Such a technique helps in limiting the route discovery pro-cedure to a specific region resulting in decreased controloverhead.

Since the route selected is currently in use, the routediscovery process is not needed. Route update packetsare flooded throughout the accepted region to update theroute changes.

Location aided knowledge extraction routing for mo-bile ad hoc networks (LAKER) [70]: Li and Mohapatra TheLAKER protocol, by Li and Mohapatra, minimizes the net-work overhead during the route discovery process bydecreasing the zonal area in which route request packetsare forwarded. During this process, LAKER extracts knowl-edge of the nodal density distribution of the network andremember a series of ‘‘important’’ locations on the pathto the destination. These locations are named ‘‘guidingroutes’’ and with the help of these guiding routes the routediscovery process is narrowed down.

LAKER uses the same forwarding strategy as DSR andcaches the forwarding routes and also creates its own guid-ing routes. While a forwarding route is a series of nodesfrom the source to the destination, the guiding route con-tains a series of locations along this route where theremay be a cluster of nodes. Even though individual nodesmay move around a bit, the basic cluster topology gener-ally remains similar for an extended period of time. Thus,the information found in the first route discovery roundis stored and used during the subsequent route discoveries.Since LAKER uses the guiding route caches in route discov-ery, the mobility model chosen becomes very important.The restricted random waypoint mobility model [15],which is an extension of [21] is used. LAKER uses anon-demand request-reply mechanism for route discovery(see Fig. 23). The control packet format in LAKER contains

S

D

P1

P2

Expected zone

LAKER request zone

LAR Request zoneNetwork Area

Fig. 23. Request zone LAKER vs LAR [70]. It can be seen that the requestzone in LAKER is more specific in comparison with LAR; therefore, there isa greater probability of accuracy in determining the exact location of thedestination node.


its forwarding route and guiding route metrics in order todecrease the forwarding area of these route requestpackets. Knowledge extraction is achieved by keepingtrack of the number of neighbors of each node till a certainthreshold value is reached. The route reply packet forwardsthe forwarding and guiding routes to the source.

A location-based routing method for mobile ad hocnetworks [14]: Blazevic et al. propose Terminode Routing,a combination of a location-based routing protocol calledTerminode Remote Routing (TRR) and a link state routingcalled Terminode Local Routing (TLR). TRR is used fornodes located some distance away from the source node,while TLR is used for local nodes. Terminode routing alsouses a unique flooding scheme called Restricted LocalFlooding (RLF) for flooding control packets during routediscovery. Anchors are geographical points serve as point-ers for source nodes to route the packets.

Naturally, the location independent addressing is usedby TLR; on the other hand, TRR uses direct paths, perimetermodes and anchors. A direct path is an approximation of astraight line from the source to the destination whileperimeter modes help to turn around obstacles if the pack-et is ‘‘stuck’’ at a location with no neighbor node closer tothe destination than itself.

Perimeter modes often result in the discovery of subop-timal paths since there may be multiple routing loops cre-ated. The usage of anchors can avoid this problem. Anchorsare referred as imaginary static geographical locations, notthe nodes within the network. These locations are writtento the packet header and the packet is forwarded accord-ingly to the intermediate node closest to the anchor. A wellchosen set of anchors could decrease the total number ofhops in the route. An addtional technique is proposed to al-low the source node to check whether anchors are neededor not depending on an estimate of the number of hops onthe direct, non-anchored path. Anchors are selected usingthe Friend Assisted Path Discovery (FAPD) or the Geo-graphical Map-Based Path Discovery (GMPD). While FAPDresponders have a stable view of the network, GPMDassumes the availability of the network density maps tothe source nodes. Even though GPMD performs better, it

requires an extra overhead of distributing the maps inthe network.

Movement-based algorithm for ad hoc networks(MORA) [16]: MORA, by Boato and Granelli, takes into ac-count the direction of the movement of the neighboringnodes in addition to forwarding packets based on the loca-tion information. The metric for making the forwardingdecision is a combination of the number of hops whichhave an arbitrary weight assigned and a function indepen-dent of each node.

While calculating this function F, the primary goal re-mains to make full use of the directions of the neighboringnodes’ movement in selecting the optimal path fromsource to destination. The function F should depend uponthe distance of the node from the line joining the sourceand destination (sd) and the direction it is moving towards.The function should reach the maxima when the node ismoving on sd and should decrease with an increase inthe distance from this line. The MORA protocol has twoversions:

1. UMORA: This is the Unabridged-MORA version since itis very similar to source routing on IP networks. Here,a short message (called a probe) is used to localize theposition of the destination. The destination sends aprobe along various different routes. Each node receiv-ing this packet keeps updating its own weight functionaccordingly. After a fixed period of time, the source hasall the paths to the destination and correspondingweight functions and selects the most suitable pathbased on this information.

2. D-MORA: The Distributed MORA is a scalable algorithmand uses a single path from source to destination. Ashort probe message is also forwarded from the desti-nation to the source. In every k hops, the node receivingthe packet polls for information from the neighboringnodes. The packet is then forwarded to the node withthe higher link weight. The path information is attachedto the packet header and forwarded to the next node.

On-demand geographic path routing (OGPR) [46]:Giruka and Singhal propose a geographic path routing pro-tocol which does not depend on a location service to findthe position of the destination. OGPR is stateless and usesgreedy forwarding, reactive route discovery and source-based routing. It is a hybrid protocol incorporating theeffective techniques of other well known routing protocolsfor MANETs. OGPR constructs geographic paths to routepackets between a source and a destination node. A geo-graphic path is termed as a collection of geographic positionsfrom source to destination which decouple node IDs fromthe path, meaning that explicit node IDs are not used toconstruct a path. The addressing scheme is implementedby using a grid-based position encoding strategy. The en-tire network area is divided into square grids with a unitsquare grid assigned as an order-1 grid. Four adjacent or-der-1 grids form an order-2 grid and so on. With the gridstructure set up, a unique binary address is assigned toall order-1 grids. Such a technique helps decouple nodeIDs from the addressing scheme and enables any nodealong the geographic path to forward data packets.

X

S

t

l

Closedvoid X

S

t

Open void

(a) (b)Fig. 24. The blind detouring and triangular routing problems, (a) closedvoid, (b) open void [84].


Path discovery is carried out on-demand by the sourceusing a flooding path request (PREQ) message. This mes-sage accumulates the grid addresses of the intermediatenodes till it reaches the destination node. The path reply(PREP) containing the geographic path is forwarded backto the source node. If path breaks, the nodes use a path-healing technique. Every packet traces the geographic pathfollowed from source to destination. By greedy forwardingalong the geographic path, multiple paths may be discov-ered and when the destination receives a packet from anewer path, it sends a path update (PUPD) message. Con-trol overhead in OGPR is low and paths found are also loopfree.

Secure position-based routing protocol [111]: Songet al. propose a secure geographic forwarding (SGF) algo-rithm which provides source authentication, neighborauthentication, and message integrity. It is combined witha secure grid location service (SGLS) to enable any receiverto verify the correctness of the location messages. SGF usesboth greedy and directional flooding with unicast mes-sages being encrypted with pair-wise shared keys betweensource and destination. The SGLS provides additional secu-rity mechanisms to the original GLS, incoporates securelocation querying and secure HELLO message exchanges.The additional security features prevents message tamper-ing, dropping, falsified injection, and replay attacks.

Sociological orbit aware location approximation androuting (SOLAR) [45]: Ghosh et al. first propose a macro-level mobility framework termed ORBIT. It is a determinis-tic orbital movement pattern of mobile users along specificplaces called hubs. The movement pattern is based on thefact that most mobile nodes are not truly random in theirmovements but actually move around in an orbit fromhub to hub. Each hub may be a rectangle and movementmay take place either inside a hub or in between hubs.Example orbital models discussed are random orbit, uni-form orbit, restricted orbit, and overlaid orbit.

The SOLAR protocol uses the ORBIT framework and thespatial/temporal locality of the nodes around these hubs. Ituses the concept of peer collaboration among acquaintances(nodes whose hub lists are cached by the node). Each nodeonly needs to know the terrain in terms of hub locationsand its own location. Periodic HELLO messages are for-warded to neighbor nodes for both neighbor discoveryand sharing of hub lists. When a source needs to send datato an acquaintance whose hub list is at the source, thepacket is forwarded to the center of the hub. Packet for-warding is achieved in a greedy fashion. If nothing isknown about the destination’s hub list, a new query is for-warded to the hubs in the acquaintance’s hub list. This iscalled a logical hop. The query is forwarded along logicalhops till either the destination’s hub list is obtained orthe number of logical hops exceeds a given threshold. Ifthe query reaches the destination, it responds with itsown hub list and its current hub where the packet isforwarded.

Load balanced local shortest path (LBLSP) routing[25]: Carlsson and Eager propose a distributed routingalgorithm which uses both local shortest path (LSP) andweighted distance gain (WDG) to finalize the forwardingnode. The two non-Euclidian distance metrics provide load

balanced routing around obstacles and hotspots. Staticnodes with lifetimes longer than the time required to routearound an obstacle are considered.

The goal is to find the non-Euclidian metrics which aredistributed, stateless and conform to Euclidian distances.To achieve this, the network traffic is routed from a sourceto its destination further from an obstacle on the path in-stead of routing along the obstacle boundary. In the LSPmetric, only a single such obstacle is considered eventhough there may be multiple obstacles in the network.The source and destination nodes are first determinedwhether they are in line of sight of each other. If so, theLSP is the Euclidian distance, otherwise, it is calculated asa sum of three distances. Each distance is calculated asthe shortest path around an obstacle. Load balancing isachieved by pushing traffic away from the obstacle bound-ary and avoid creating local hotspots. The WDG metric isused to provide various levels of importance of either usinga straight line route or an obstacle avoidance route. Byassigning different magnitudes of weight to either of theseroutes, the WDG metric combines distance gains instead ofabsolute distances between source and destination.

The LBLSP routing algorithm combines both the LSP andWDG. A packet initially starts in greedy mode and maychange its mode to recovery mode till it reaches its desti-nation. In greedy mode, the forwarding node calculatesLSP values for itself and its neighbors. For nodes with lowerLSP values, their respective WDG values are computed andthe one with the largest WDG value is chosen as the nexthop. If there are not any nodes with lower LSP values, thenthe packet mode is changed to recovery mode. In thismode, perimeter routing is carried out till a node with alower LSP value is found.

Geographic landmark routing (GLR) [84]: The GLRalgorithm, by Na and Kim, solves the blind detouring prob-lem and the triangular routing problem in MANETs. Theblind detouring problem occurs when a packet arrives ata dead-end when the next node is blindly selected (seeFig. 24). This could result in a longer detour path evenwhen there are actually shorter paths available. GLR solvesthis by discovering two paths bypassing the void area. Oneof the paths is from source to destination while the otherpath is in the reverse direction. These paths are comparedand the shorter path is selected.


The triangular routing problem occurs due to pathrefraction at dead-end nodes. GLR solves this problem byusing the concept of landmark nodes which are specialintermediate nodes identified during path discovery. Thislandmark node reclaims greedy forwarding after escapingfrom a dead-end. GLR computes straight sub-paths be-tween the landmarks. GLR has two sub-layers: basic andoptimization. GLR uses a regular geographic routing algo-rithm such as GPSR to forward packets to target nodes inits basic sub-layer. In the optimization sub-layer, informa-tion about forward and backward paths such as hop countsand landmark nodes is collected. Then, it selects the short-er detour path and utilizes a loose source routing with thelandmarks.

Maximum expectation within transmission range(MER) [64]: Kwon and Shroff propose a packet forwardingalgorithm for location aware networks. In most cases, loca-tion estimates have significant error rates which may beoverlooked in most location based routing protocols. Theselocation errors could induce either transmission failures orbackward progress in greedy mode. The former occurswhen the selected node is out of transmission range whilethe latter takes place when the next hop node is actuallyfarther than the destination. This leads to looping withinthe network.

The proposed MER algorithm utilizes an error informa-tion field in its messages which is used by the objectivefunction. The greedy routing scheme (GRS) is improvedby the authors where they mitigate the impact of the loca-tion errors. Using the error statistic, the objective functioncalculates a maximum expectation for each node in whichthe nodes with the highest expectation are chosen as thenext node for packet forwarding.

Implementation framework for trajectory basedrouting (TBR) [137]: Yuksel et al. study various implemen-tation issues of TBR in this work. A proposed method en-codes trajectories into packets at the source node beforesending them to the destination. Bezier curves are utilizedas possible path trajectories to efficiently forward thepackets. These curves provide flexibility in the greedy for-warding of TBR with the possibility of multiple types ofcurves. With a given node neighborhood and a trajectoryfor packet forwarding, TBR specifies the followingobjectives:

� Obey the trajectory: The packet stays within the giventrajectory and does not deviate at all.� Reach the destination: If application is more sensitive to

packet loss, then the packets should be forwardeddirectly to the destination node instead of followingthe given trajectory.� Reach quickly: The packets are forwarded such that

they reach the destination with the minimal delay.

Forwarding nodes on the curve may be selected at ran-dom or the nodes closest to the curve (CTC). For floodingapplications, the least advancement on curve (LAC) ensuresthat the maximum nodes receive the packet while a com-bining CTC and LAC may be another possibility for packetforwarding. To reduce delivery delay, the packet may beforwarded to the farthest node on the curve with most

advancement on curve (MAC) algorithm and if the trajec-tory must be strictly obeyed then the lowest deviationfrom curve (LDC) can be implemented. In this case, theroute deviates from the trajectory as little as possible.

3.4.1. ComparisonThe research on location-aided routing has evolved

from the design of the generic routing frameworks to theiroptimization and the solution of specific problems posedby certain ad hoc network configurations.

The earliest proposed protocols were designed to im-prove on the generic ad hoc routing protocols by takingadvantage of the knowledge of the physical location ofthe nodes. Most of these protocols can be classified assource originated, reactive routing: LAR [62,63], DREAM[11], GPSR [59], but they also require the distribution ofthe location information (and in some cases, movementinformation as well).

In this respect, these protocols are proactive in distrib-uting location information. However, location informationinvolves less data than full topological connectivity andhas a more predictable and continuous change pattern.Nevertheless, the distribution of the location informationis a significant overhead and the DRM [30] and improve-ment to LAR [62,63] approaches are directed towardsreducing it.

Another group of protocols are based on the general ap-proach that location based routing is used to get to thegeneral vicinity (‘‘region’’) of the destination, where therouting will be shifted to a different approach. Examplesof this group are REGR [76], LAKER [70], and Terminoderouting [14].

Another group involves algorithms which resemblesource routing from the wired IP networks, but with theexplicit list of IP addresses to be traversed replaced withthe geographical path that needs to be followed or approx-imated by the packet. Examples of this approach are on de-mand geographic path routing (OGRP) [46] and trajectorybased routing (TBR) [137].

Finally, one of the most recent focus on location aidedrouting was an increasing attention paid to the actual dis-tribution of the nodes in the physical space. The particular-ities of the network occasionally include aspects which arehelpful for routing – for instance the SOLAR [45] algorithmwhich exploits the recurrent movement pattern of thenodes. In many cases, however, the physical space of thead hoc network contains obstacles or simply areas whereno forwarding nodes are present. Following a greedy loca-tion-based scheme would lead the packets into a dead endin this situation. Protocols such as LBLSP [25] and GLR [84]have been designed to address exactly these challenges,while MER [64] is designed to reduce the impact of loca-tion estimation errors.

Fig. 4 summarizes the protocols surveyed in thissection.

3.5. Multipath protocols

Multipath routing protocols create multiple routes fromsource to destination instead of the conventional singleroute discovered by other protocols. The main advantage

Fig. 25. Example of a potential routing loop scenario with multiple pathcomputation [79].


of discovering multiple paths is that the bandwidth be-tween links is used more effectively with greater deliveryreliability. It also helps during times of networks conges-tion which may arise due to bursty traffic within the net-work. Multiple paths are generated on-demand or using apro-active approach and is of great significance as routesgenerally get disconnected quickly due to node mobility.

Caching and multipath routing protocol (CHAMP)[118]: Valera et al. propose the CHAMP protocol whichuses data caching and shortest multipath routing. It alsoreduces packet drops in the presence of frequent routebreakages. Every node maintains a small buffer for cachingthe forwarded packets. This technique is helpful in the casewhen a node close to the destination encounters a for-warding error and cannot transmit the packet. In such asituation, instead of the source retransmitting again, anupstream node which has a cached copy of the packetmay retransmit it, thereby reducing end-to-end packet de-lay. In order to achieve this, multiple paths to the destina-tion must be available.

In CHAMP, each node maintains two caches; a routecache containing forwarding information and a route re-quest cache which contains the recently received and pro-cessed route requests. Those entries which have not beenused for a specific route lifetime are deleted from the routecache. A node also maintains a send buffer for waitingpackets and a data cache for storing the recently forwardeddata packets. A route discovery is initiated when there isno available route. The destination replies back with a cor-responding route reply packet. There may be multipleroutes of equal length established, each with a forwardingcount value which starts with a zero from the source and isincreased by one with every retransmission.

Ad hoc on-demand multipath distance vector rout-ing (AOMDV) [79]: Marina and Das present the AOMDVprotocol, which uses the basic AODV route constructionprocess, with extensions to create multiple loop-free andlink-disjoint paths. AOMDV mainly computes the multiplepaths during route discovery process and it consists of twomain components: a rule for route updates to find multiplepaths at each node, and a distributed protocol to calculatethe link-disjoint paths.

In this protocol, each route request and route replypacket arriving at a node is potentially using a differentroute from the source to the destination. All of these routescannot be accepted since they can lead to creation of loops(see Fig. 25). The proposed ‘‘advertised hop count’’ metricis used in such a scenario. The advertised hop count for aparticular node is the maximum acceptable hop count forany path recorded at that node. A path with a greaterhop count value is simply discarded and only those pathswith a hop count less than the advertised value is accepted.Values greater than this threshold means the route mostprobably has a loop.

The following proven property allows to have disjointroutes [79]: Let a node S flood a packet m in the network.The set of copies of m received at any node I (not equal S) eacharriving via a different neighbor of S, defines a set of node-dis-joint paths from I to S. This distributed protocol is used inthe intermediate nodes where multiple copies of the sameroute request packet is not immediately discarded. Each

packet is checked whether it provides a node-disjoint pathto the source.

Split multipath routing (SMR) [68]: The SMR protocol,by Lee et al., establishes and uses multiple routes of max-imally disjoint paths from source to destination. Multipleroutes are discovered on-demand and the route with theshortest delay is selected.

When a node wants to send a packet to a destination forwhich a route is not known, it floods a RREQ packet into thenetwork. Due to flooding, several duplicate RREQ messagesreach the destination along various different paths. Sourcerouting is used since the destination needs to select multi-ple disjoint paths to send the RREP packet.

Unlike the conventional routing protocols such as AODVand DSR, the intermediate nodes in SMR are not allowed tosend back RREPs even if they have the route to the desti-nation. This is because the destination can only make adecision on the validity of maximally disjoint multiplepaths from all of its received RREQ packets. If the interme-diate nodes reply back, it is almost impossible for the des-tination to keep track of the routes forwarded to thesource. Intermediate nodes also use a different packet for-warding approach. Instead of dropping all duplicate RREQs,each node only forwards those RREQ packets arrived usinga different link from the first RREQ packet and having a hopcount lower than the first RREQ packet.

The destination considers the first received RREQ packetas the path with the shortest delay. It immediately sends aRREP back to minimize the route acquisition latency. Tofind the maximal disjoint path to the already replied route,it waits for additional time to determine all possible routeinstances. In some cases, there may be more than one max-imal disjoint route and if so, the shortest hop distanceroute is selected.

Neighbor table based multipath routing (NTBR)[134]: Yao et al. present an initial theoretical analysisshowing that non-disjoint multipath routing has a higherroute reliability than the conventional disjoint multipathrouting. This study has led to the development of a neigh-bor table based multipath protocol which does not requirethe disjoint routes. In NTBR, every node maintains a neigh-bor table which records routing information related to its khop neighbors. While k can be set to any value, the controloverhead also increases accordingly.

The NTBR protocol has a route discovery and routemaintenance mechanisms. It also maintains a route cache


at each node in the network. The route caches are main-tained by the neighbor tables which also serve to makean estimate of the lifetime of the wireless links. This infor-mation is used to keep track of the route lifetime. Everynode transmits periodic beacon packets to its two-hopneighbors. The neighbor table is established based on theinformation from the beacon packet by using any of thefollowing approaches:

� Time driven: In this approach each node essentiallywaits for a predefined timeout interval before decidingwhether a link is active or not. The node waits for a bea-con packet from either its one-hop or two-hop neigh-bors and adds the information to its neighbor routingtable. However, the problem is that there is always atimeout between the actual topology change and thetime in which this information is realized by the node.� Data-driven: This approach alleviates the problem aris-

ing from the time driven mechanism. In this scheme,one field of the beacon packet is used to inform whethera node is unreachable or not. The address of theunreachable node is added into the beacon packet andall nodes receiving the packet update their neighborrouting tables accordingly.

The route cache contains all the routing information fora particular node and it is updated by monitoring anypacket passing through the network. To extract individualroutes, the route extraction reason mechanism is usedwhich simply prioritizes the routes extracted from differ-ent packets. The routes from route replies are assignedthe highest priority while the routes from route requestpackets or neighbor tables become second followed bythe routes from data packets. These priorities are used dur-ing the route selection process. The route discovery androute maintenance are similar to the other routingalgorithms.

Adaptive QoS routing framework through multiplepaths [36]: Das et al. observe that when communicationtakes place between source–destination pairs, an interrup-tion may occur in case of a link breakage due to variousreasons. This interruption leads to degraded QoS as theuser has to wait till another route is discovered. To solvethis problem, an adaptive framework for computing multi-ple paths in both temporal and spatial domains is pro-posed. The data transmission in the spatial domainbalances the traffic load in the network, while the temporaldomain ensures continuity of data transfer.

Two different aspects are considered within this frame-work: (i) preemptive route rediscoveries are performed be-fore route errors occur as data packets are forwarded fromone node to the other. Multiple paths are computed in thetemporal domain and in the event of a route error, one ofthe alternate paths is selected; (ii) multiple paths are alsocomputed in the spatial domain to ensure forwarding ofdata at any instant of time. These packets are transferredsequentially in blocks over the multiple paths computed,helping to reduce further congestion and link delay.

The concepts of link stability and path stability are alsoproposed based on the employed route discovery and routemaintenance schemes. Path stability metric depending on

the route lifetime should be valid as long as the data is for-warded from source to destination. If the path is not stableenough, then a route maintenance scheme is required toeither repair the path or suggest an alternative path. Oncethe alternative set of paths are decided, the total volume ofdata is divided into blocks and forwarded along all the dif-ferent routes to the destination.

Truthful multipath routing protocol (TMRP) [124]:Wang et al. present a multipath routing protocol whichcan be used in networks with non-cooperative nodes, alsotermed as selfish nodes. These nodes are characterized bythe fact that they agree to forward packets only for a re-turn/payment. The cost of forwarding a packet is measuredin terms of resources available at each node. A node usingthe forwarding services of another node would like to min-imize its own cost of forwarding while the selfish nodemay not advertise its true cost. The authors hence designa truthful mechanism to ensure that a node’s payoff is max-imized only when it reveals its true cost.

A generic truthful multipath routing protocol (GTMR) ispresented which can be used to make any table-driven mul-tipath routing protocol a truthful one without additionalcontrol overhead. It only requires the multipath routingprotocol to be loop-free and table-driven. The GTMR buildsa coordination between the neighbor table and the routingtable of the multipath routing protocol and is followed byan auction based approach to select the next hop for packetforwarding. The second-price sealed bid auction, the Vick-rey auction, is used by a forwarding node when it receivesa packet to forward. All one-hop neighbors of the node arequalified bidders who advertise their bids in their HELLOmessages. Based on the lowest cost, the winner is selectedand the packet is forwarded.

The authors implement the GTMR over the AOMDVmultipath routing protocol and call it the TMRP. The TMRPhas two variations depending on whether the nodechanges its packet forwarding cost over time or not. Inthe case of varying the cost, the node varies the cost pro-portionally to the number of packets forwarded until thecost reaches an upper bound. At this moment, the cost isreduced to its lower bound. Nodes use the routing tableand the packet forwarding auction approach suggested inGTMR to decide the next hop where to forward the packet.

On-demand discovery of node-disjoint paths [74]: Liuet al. design an algorithm to find node-disjoint paths in amultipath routing protocol for MANETs. A theoreticalframework proves the equivalence between multipath dis-covery and network flow assignment. The aim is to guaran-tee discovery of node-disjoint paths, a characteristic whichis overlooked by other multipath routing protocols in theliterature. The authors first show that this problem of find-ing node-disjoint paths in a MANET is equivalent to theflow assignment problem in flow networks. A typical flowassigment algorithm (the Ford-Fulkerson method) is con-sidered and with the aid of reverse mapping, the augment-ing path in the flow network is mapped to an auxillary pathin the ad hoc network (see Fig. 26).

Multiple node-disjoint paths (MNDP) is propose whichis an incremental algorithm where an auxillary path iscomputed and merged with the existing path set betweena given source–destination pair. In addition, the auxilary

s

t

s

t

GG

Fig. 26. Equivalent graphic representation of an ad hoc network [74].


path discovery protocol (APDP) is introduced and inte-grated with the MNDP algorithm for discovery and mainte-nance of multiple paths. The APDP is a derivation of DSRand its route discovery is carried out in three phases.

� Phase 1: Initial path from source (s) to destination (t) isdiscovered using DSR and is termed the reference path.� Phase 2: Auxilary path is discovered based on the refer-

ence path.� Phase 3: Reference and auxilary paths are merged to

give two node-disjoint paths.

To compute k node-disjoint paths, the phases 2 and 3need to be executed k times. Route maintenance is alsoachieved by executing phases 2 and 3 with an existing va-lid path as the reference path.

Scalable multipath on-demand routing (SMORT)[102]: Reddy and Raghavan propose a multipath routingprotocol with the aim of minimizing route break recoveryoverhead. A primary path from source to destination ischosen and multiple intermediate nodes are selected onthe primary path to provide multiple paths. The primarypath is most often the shortest path. SMORT has norequirements on discovering node-disjoint paths and aimsto find multiple fail-safe paths. A fail-safe path is auxillaryto the primary path and bypasses at least one intermediatenode on the primary path. Multiple fail safe paths differfrom the node-disjoint and link-disjoint paths by the factthat they may have common nodes and links. These fail-safe paths reduce route recovery time and path mainte-nance overhead more than the node-disjoint routingprotocols. This is because in the latter case, route errorpackets need to be forwarded to the source node every alink breakage occurs. In fail-safe paths, a new path is se-lected right away thereby keeping the control overheadand delay minimal. Route discovery in SMORT is similarto AODV with the only difference being each node is al-lowed to accept multiple copies of route request packets.To avoid end-to-end routing loops from occurring, SMORTdoes not allow extra route reply packets from the interme-diate nodes to be relayed back to the source node. Perfor-mance evaluation of SMORT shows an average of fiftypercent reduction in control overhead in comparison toAODV and forty percent with DMRP.

Secure multipath routing (SecMR) [80]: Mavropodiet al. propose a multipath routing protocol with varioussecurity enhancements to guard against collaboratingmalicious nodes. The SecMR is an on demand routing pro-tocol designed to protect against denial of service (DoS) at-tacks from malicious nodes. Multiple paths may beaffected by several vulnerabilities such as man-in-the-middle attacks, lack of authentication or the racing phe-nomenon. The SecMR guards against these vulnerabilitiesand discovers existing non-cyclic, node-disjoint paths be-tween a source and destination. In the path selection pro-cess, the number of hops is required not to exceed aspecific maximum. The SecMR is divided into two mainphases: neighborhood authentication and route discoveryand maintenance. The former implements the asynchro-nous mutual authentication of nodes in the neighborhoodusing a pair of public secret keys. Each secret key alsoneeds to be certified through a certifying authority (CA).Signed messages are broadcast to immediate neighborsby each node at periodic time intervals containing the cur-rent time and its unique identifier. In the route discovery,the maximum hop count is determined before the randomkey selection. The encrypted key is calculated before themessage construction and then the message is broad-casted. Every route request query comprises of a differentindependent thread received at the destination. The desti-nation node decrypts the message with the secret key tocheck the validity of the received message. The communi-cation is initiated through node-disjoint routing paths.

Disjoint multipath routing using colored trees [99]:Ramasubramanian et al. develop a multipath routing pro-tocol using a pair of colored trees. The proposed red andblue trees are either link disjoint or node disjoint (seeFig. 27). Every node in the network maintains two pre-ferred neighbors for every destination. These neighborscompose the respective red and blue colored trees. Thedistributed algorithm uses only local information to com-pute multiple disjoint paths to a destination node. Thedrain node, an intermediate or final destination is thenode in which the data is forwarded along the trees.Any packet transmitted from the source is marked witheither red or blue colors. The intermediate node receivingthis packet forwards it to the preferred neighbor based onthe color of the packet. The colored trees are constructedusing the depth first search (DFS) technique. A node inthe network could be in one of the following states atany point of time: unvisted, visited, cycle, token and fin-ish. Nodes in the cycle state belong to both trees andthose in the token state initiate path searches. Once anode receives a token message, it initiates a path searchby sending out search messages. These messages are for-warded sequentially to every node in the neighbor listuntil a success message is received from the neighbor. Anode in the cycle state sends this success message. Tokenmessages are then sent to select nodes from which thesuccess message was received and the neighbor list tra-versed in the reverse direction. Every node finishing theoperation sends a return message denoting the end ofthe operation. After all neighbors send the return mes-sages, a final return message is sent to the parent nodeinitiating the path search with a token message. This


procedure results in the creation of a pair of colored treesfrom the originating source to the drain node.

Reliable and efficient forwarding (REEF) [33]: Contiet al. propose an efficient forwarding, based on reputationand reliability, scheme to be used in conjunction with anexisting multipath routing protocol in MANETs. The mech-anism is trustworthy and builds the reputation of eachnode with a set of forwarding policies while avoiding unre-liable routes and balance network utilization simulta-neously. REEF uses only the node’s internal knowledge.The basic principle followed states that cooperation prob-lems should not be solved by using cooperation. This meansthat each node trusts only itself. REEF uses an existing mul-tipath routing protocol to obtain information such as thenext hop and the number of hops towards the destination.Packets are forwarded on a route using either a probabilis-tic scheme or simply the best/most reliable route. The pro-tocol basically forwards the packet on the route with thehighest success probability. The novelty of this protocollies in the fact that the data forwarding is accomplishedin a lightweight manner with no additional overhead.

Multipath security-aware QoS routing (MuSeQoR)[103]: Reddy et al. propose a multipath routing protocolwith QoS guarantees ensuring secure and reliable commu-nication even in the case of multiple path failures. Multiplepaths are selected based on the current state of the net-work and the number of possible paths present betweensource and destination nodes. MuSeQoR considers twotypes of channel models: erasure channels and corruptionchannels. The former deals with channels susceptible topath loss due to link failures while the latter considerschannels which may be corrupted by malicious nodes.The protocol is a modification of DSR with no global topol-ogy information maintenance at the nodes. Sessions are di-vided into time frames during which it is assumed that thenetwork state remains constant. Based on the networkstate during a session, paths are computed. During routediscovery, route request (REQUEST) packets are sent bythe source. These packets contain a sequence number,the path traversed and the reliability of the path traversedtill now, the source and destination IDs, the path band-width and the required bandwidth for this session, thepath failure metric, eavesdropping ratio for this session.Intermediate nodes keep checking the packet route tomake sure no loops are created. REQUEST packets are for-

B

DrainC

A

B

DrainC

A

(a) (b)Fig. 27. Two node disjoint path from nodes A and B to the drain,computed independently by each node [99].

warded a predefined number of times after link and band-width availability on the path is recalculated. Once thedestination receives multiple REQUEST packets, each pathis sorted by their failure probabilities. Using the positionand velocity information of the last node that forwardedthe packet, link availability is estimated. Once a set ofpaths is obtained, each path receives a RESERVE packet.The source begins data transmission along the path inwhich the RESERVE is received. However, the destinationsends a RESERVE packet only if a set of node-disjoint pathsare found.

3.5.1. ComparisonMultipath protocols find multiple paths between the

source and the destination and use these paths to improveon specific performance metrics. The papers discussed inthis section cover a very wide range of approaches. Thereare, however, several discernible trends.

A certain group of papers are striving to add multipathextensions to existing protocols. Thus, AOMDV [79],TMRP[124] and SMORT [102] build on AODV, while thework on on demand discovery of node disjoint paths [74]and MuSeQoR [103] build on DSR. Note that both aresource originated protocols – in general, most of theapproaches in this class had chosen a source originatedapproach. Several approaches are also using source routing(such as split multipath routing (SMR) [68]), giving thesource complete freedom over the path to be followed bythe individual packets.

The paths in the multipath protocols must meet indi-vidually performance requirements, similar to single pathprotocols. In addition, the routes have specific constraintson their relationship: they must be either link-disjoint ornode-disjoint, although other criteria are also possible,such as described in NTBR [134]. This is a much more com-plex optimization problem than, for instance, finding a sin-gle shortest path. Among the papers surveyed, several areusing sophisticated mathematical models for optimizingmultiple paths: e.g. Lagrangian relaxation and subgradientheuristics (Adaptive QoS routing [36]), network flow the-ory (on demand discovery of disjoint paths [74]) and col-ored tree graphs [99].

We get a different cross-section of the field if we con-sider the ultimate goal of building multiple paths. Surpris-ingly, relatively few paper consider load balancing as oneof the primary goals (for instance, SMR [68] and adaptiveQoS routing [36]). For most papers surveyed, the mainobjective of the multipath approach is fault toleranceand/or quick recovery from route failures: CHAMP [118],NTBR [134], adaptive QoS routing [36] and SMORT [102].

A somewhat different approach is taken by paperswhich use multipath routing to improve the overall secu-rity of the transmission and to defend against maliciousnodes (SecMR [80] and MuSeQoR [103]) or to improvethe efficiency of the forwarding in networks with unreli-able nodes (REEF [33]).

3.6. Hierarchical protocols

While traditional internet routing is natively hierarchi-cal, in the first approximation ad hoc networks route over a


flat collection of nodes. As the networks grow in size, theseapproaches lead to increased routing table sizes and con-trol packet overhead. Hierarchical ad hoc routing protocolsbuild a hierarchy of nodes, typically through clusteringtechniques. Nodes at the higher levels of the hierarchy pro-vide special services, improving the scalability and the effi-ciency of routing.

Hierarchical state routing (HSR) [53]: Iwata et al.introduce a class of protocols based on multilevel cluster-ing. The goal is to replace the flooding of the control infor-mation with a local collection of this information in theclusterhead, followed by the propagation of this informa-tion to the other clusterheads.

First, nodes form level 0 clusters based on physicalproximity and elect a clusterhead. Clusterheads connectto each other using virtual links. Multiple clusterheadscan assemble themselves into higher level clusters. Whena node changes its position, link state information is ex-changed between clusterheads using the virtual links.The clusterhead collects link state information about thenodes in its cluster and propagates it to other clusterheadsthrough gateway nodes.

Routing in HSR happens using a hierarchical addressingscheme, with the clusterhead acting as routers. When anode wants to send a packet, it sends it first to the localclusterhead. The clusterhead looks up the destination andsends the packet to its nearest gateway node. The gatewaynode then propagates the packet to the nearest gatewaynode at the next level of the hierarchy. The process contin-ues until the packet reaches the gateway node of the des-tination cluster. The final gateway node routes the packetto the clusterhead of the destination cluster which thenforwards the packet to the destination node.

Core-extraction distributed ad hoc routing (CEDAR)[110]: The CEDAR protocol, proposed by Sivakumar et al.is allows the consideration of QoS requirements in an adhoc setting. The protocol selects a subset of nodes calledthe core of the network (see Fig. 28). Control messages willonly be broadcasted among the nodes of the core, whichcan use any existing ad hoc routing mechanism for com-munication. The core is positioned as a ‘‘self-organizingrouting infrastructure’’ which performs route availability

1

2

3

4

5

67

Fig. 28. Example of a core broadcast. Nodes in black are the core nodes. Solid linthe core graph [110].

computations through waves of messages with dynami-cally limited propagation speed. The availability of in-creased bandwidth is transmitted by slow propagatingincrease waves, while information about decreased band-width is transmitted by fast propagating decrease waves.

Routing in the CEDAR architecture happens as follows.The source node sends a route request packet containingthe source, destination, and the requested bandwidth toits dominator, the local core node. The dominator thencomputes and establishes a QoS route if feasible. The dom-inator nodes in each cluster maintain local state informa-tion and communicate with each other using virtuallinks. Route computation is carried out only on the corepath.

CEDAR aims more at robustness than optimality incomputing the routes. Each core node only knows aboutthe neighboring core node and has no global knowledgeabout the core subgraph. This simplifies the maintenanceof the core network which can be necessary due to topol-ogy changes induced by the mobility or failure of nodes.Core paths are established on-demand for connection re-quests and route computation is carried out only when aspecific request for a route is received.

Eriksson et al.’s dynamic addressing approach [39]:Eriksson et al. propose a dynamic addressing scheme toimprove scalability in ad hoc networks. The approach isbased on adding a location based dynamic address to thenode, in addition to its permanent identifier. A distributedlookup table is used to map the permanent identifier to thedynamic address (see Fig. 29). The main challenge is theassignment and maintenance of dynamic addresses, whichis done through a hierarchical address tree. The approachhad been shown to successfully scale to networks of sev-eral thousands of nodes.

Hierarchical landmark routing (H-LANMAR) [131]:An extension to the LANMAR scheme [91] is proposed byXu et al. The LANMAR protocol is logically hierarchicaland uses landmarks which summarize routing informationto remote nodes. H-LANMAR improves the scalability ofLANMAR through the use of a backbone network. In the firststep, nodes are grouped into dynamic multi-hop clusters,with a clusterhead called the backbone node (BN). Backbone

8

9

10

111

3 6

11 10 8

31 11 6 10

8

es denote links in the ad hoc network. Dotted pipes denote virtual links in

xxx

1xx0xx

00x 01x 10x 11x

110100010000

1.

2.3.

A.D. B. C.

A B

CD

2.3.

SCENARIO

1. B joins via A2. C joins via B3. D joins via A

1.

Fig. 29. Address tree for a small network topology. The numbers 1–3 show the order in which the nodes were added to the network [39].


nodes are then connected using higher-level links and theprocess is continued recursively to create a multilevelhierarchy.

In the original LANMAR protocol the packet is for-warded to the nearest landmark. In contrast, H-LANMARroutes the packet to the nearest BN which then forwardsit along the backbone network to the other BNs until thepacket reaches the BN of the destination node. This BNthen sends the packet either directly to the remote locationor to its nearest landmark. This process helps in reducingthe number of hops to the destination and the overallpacket delay. It is assumed that each node is simulta-neously running the LANMAR scheme locally in case of afailure in the backbone.

3.6.1. ComparisonHierarchical routing protocols, in general are proposed

as a scalability approach. In contrast to wired IP routing,the hierarchy has nothing to do with the hierarchical nat-ure of the address, rather the hierarchy reflects a geograph-ical clustering of the nodes.

Eriksson et al. [39] is an unusual case in the sense that itintroduces a second, dynamic address for the nodes, basedon their geographical location.

Another interesting observation is that all the protocolswe have reviewed create the hierarchy dynamically and ina distributed manner. It is certainly possible to create ahierarchy in a centralized way – yet none of the ap-proaches have chosen to do so.

Another way to think about these protocols is whetherthe protocol have been a hierarchical adaptation of a previ-ously existing protocol (H-LANMAR [131]) or have beendesigned from scratch to be hierarchical (all the other re-viewed protocols).

Another aspect we need to consider is whether QoSconsiderations such as the current load affect the hierar-chical routing or not. From the surveyed protocols only CE-DAR [110] considers QoS considerations.

Table 3.6.1Hierarchical routing protocols comparison.

Protocol Routing tables Update freq

HSR Yes PeriodicCEDAR Yes On demandEriksson et al. Yes PeriodicH-LANMAR Yes Periodic

A number of other comparison criteria has been sum-marized in Table 3.6.1.

3.7. Multicast protocols

Multicasting is the simultaneous transmission of datafrom one sender to multiple receivers. Several widely usedapplications require multicasting at least at the logical le-vel. Examples include audio–video teleconferencing, real-time video streaming and the maintenance of distributeddatabases. In many cases it is advantageous to implementmulticasting at the level of the routing algorithm (otherapproaches would be one-to-all unicast or the implemen-tation of multicasting at the application layer). In the fol-lowing we review several representative examples ofmulticast ad hoc routing protocols.

Dynamic core based multicast routing (DCMP) [35]:Das et al. introduces DCMP, a source-initiated multicastprotocol. DCMP has been designed from the ground up asa multicast protocol, without relying on existing unicastprotocols.

DCMP classifies the sources into active, core active, andpassive as shown in Fig. 30. Active sources use the tradi-tional technique of flooding the network with JoinReq

control packets at regular intervals. Nodes which desireto join the multicast group as a destination, reply with aJoinReply packet along the reverse path to the source.Passive nodes do not participate in the creation of the mul-ticast routes themselves. Instead, a subset of the activenodes, the core active nodes form a shared mesh throughwhich the passive sources transmit their data packets. Asingle core active source can support a maximum of Max-PassSize passive sources and the hop distance betweenthese sources is limited by the MaxHop parameter.

Simulation results show that the control packet over-head of DCMP is up to 30% percent lower compared to mul-ticast routing protocols not employing a core. This leads toan increase in scalability and multicast efficiency of about

uency Hello message Critical node

Yes YesNo YesNo NoNo Yes

A

S3

S1

R

R

R

S4

S

S2

S

R

Node not in multicastgroup

Forwarding node

Normal activesource

Passive source

Core Active source

Receiver

Fig. 30. Mesh topology in DCMP [35].


10–15%, with a slight tradeoff (2%) in the packet deliveryratio for networks with light traffic loads.

Adaptive demand-driven multicast routing (ADMR)[54]: Jetcheva and Johnson propose ADMR, an on-demandmulticast routing algorithm with almost no periodic com-ponents within its system. ADMR dynamically maintainsthe multicast routing state for active groups of nodes.The protocol uses source-based forwarding trees and con-tinuously monitors the traffic pattern of the source. This al-lows ADMR to detect broken links in a tree as well as toidentify the sources which stopped sending data packets.The sender node will transmit ‘‘keep-alive’’ messages tomaintain the forwarding tree. When the source wants toterminate the route, it stops sending the ‘‘keep-alive’’ mes-sages, and the tree expires. Similarly, the receiver nodesneed to keep alive their respective branches by sendingdownstream passive ACK messages. If these messages arestopped, ADMR ‘‘prunes’’ the specific branches of the for-warding tree. Multicast data packets are forwarded fromthe sender to the multicast receivers using MAC layer mul-ticast transmissions along the path of shortest delay.

In a network with highly mobile nodes the cost of main-taining the quickly changing multicast trees is higher thansimply flooding the information to every node and letting

Fig. 31. Tree flood vs. n

the node decide whether it wants to retain it. ADMR de-tects moments when the overall node mobility is very highand automatically starts flooding the network with datapackets. After a short period of time, ADMR switches backto the normal mode since within this time node mobilitymay have reduced. Controlled flooding is highly beneficialand its relation with regular flooding is shown in Fig. 31.

The novel features of the protocol are summarized [54]:(i) no periodic floods of control packets or routing table ex-changes and it also does not require any core node; (ii) pas-sive acknowledgments are used to automatically adjusttrees; (iii) high node mobility can be detected by ADMRand in such a scenario it simply reverts to flooding of datapackets since routes become unstable and change fre-quently; (iv) a limited number of keep-alive packets areforwarded to bursty sources to ensure that their forward-ing trees are maintained as often as possible and distin-guish between lack of data or complete disconnection.

AMRoute: ad hoc multicast routing protocol [130]:Xie et al. propose AMRoute which aims to avoid the highcontrol packet overhead associated with the maintenanceof multicast trees in ad hoc networks with highly mobilenodes. It does not support guarantees for minimal band-width and packet latency; the main design objectives are

Tree flood

Network Flood

etwork flood [54].


robustness and scalability. A conventional unicast routingprotocol is used to keep track of the network dynamics.AMRoute is independent of the underlying unicast proto-col, which can be chosen according to the specific networkrequirements.

Thus, AMRoute is only concerned with the dynamics ofthe multicast groups. The protocol defines user-multicasttrees composed of senders and receivers. Packet forward-ing is carried out by the members of the groups over theunicast tunnels which form the links in the tree. Throughthis approach, in AMroute nodes which are not membersof the multicast trees do not need to support any multicastprotocol and hence are not storing any state information.

AMRoute uses a logical core to discover new groupmembers, create and maintain the multicast tree. The log-ical core is not fixed, it changes with the group dynamics.

Energy efficient multicast routing [69]: Li et al. focuson developing an energy efficient multicast routing proto-col. By assigning the transmission power of each node as aweight, the network graph is transformed to a new graphwith weights between edges. The minimum energy multi-cast (MEM) problem is to find the multicast tree whose totalenergy cost is minimized. The problem now reduces to thedirected Steiner tree (DST) problem. This is a known NP-hard problem, and the authors show that it is unlikely thatthere is an approximation algorithm with a constant perfor-mance ratio to the number of nodes in the network. Theauthors show several heuristic approaches for the problem.

In the node-join-tree (NJT) algorithm a cover set con-taining all non-leaf nodes is built incrementally by select-ing the shortest path from the source node to the nodes inthe uncovered set. The heuristic grows the multicast tree byselecting the nodes with the highest energy efficiency. Theauthors describe a distributed implementation of the algo-rithm where nodes have information only about theirneighbors.

QoS multicast routing protocol for clustering mobilead hoc networks (QMRPCAH) [67]: Layuan and Chunlinpresent a QoS aware multicast protocol for MANETs withclustering. The proposed QMRPCAH protocol allows a nodeto maintain only local multicast information and a sum-mary of other clusters; it does not require knowledge ofthe global network. The protocol supports soft QoS withoutany hard guarantees. There may exist transient periods oftime without the required QoS, for instance during periodsof congestion, link breakage or packet loss.

In QMRPCAH every node periodically measures the de-lay on its outgoing links and broadcasts it to the membersof its cluster. Thus, every node keeps its intra-cluster rout-ing tables updated. Bridge nodes maintain the inter-clusterrouting tables in a similar fashion. When a mobile node en-ters a new domain, it uses a remote subscription method tosubscribe to the new domain and joins the local multicasttree. When a node wants to join a multicast tree a JoinReqmessage is forwarded to the parent bridge node. The bridgeappends its own address to the message and forwards it tothe top level bridge node if it is not aware of the multicasttree itself. Multicast tree (MT) messsages are forwardedfrom the top bridge node to the local node.

QMRPCAH uses a receiver-initiated selection floodingalgorithm where links violating bandwidth constraints

are deleted. Flooded messages also avoid these violatinglinks.

QoS multicast routing using multiple paths/trees[128]: Wu and Jia propose a routing protocol using multipleparallel paths or trees to ensure the bandwidth require-ment of a connection. The protocol is distributed and usesstandard route discovery and route reply techniques. TheQoS requirements include a bandwidth requirement for aroute and a delay bound represented by the number of hopsfrom source to destination. Similar to other on demand pro-tocols, RREQ packets are broadcast from the source andRREP packets are returned by the destination. Since thedestination may receive multiple RREQ packets, it has mul-tiple possible routes to the source. It runs a paths/treeselection algorithm to construct the multicast trees. Threedifferent tree selection algorithms are proposed:

� Shortest path tree based multiple-paths (SPTM): It com-putes the shortest path in terms of hops between sourceand destination. SPTM uses multiple paths from theshortest path tree (SPT), that a branch in the SPT con-sists of several parallel paths.� Least cost tree based multiple-paths (LCTM): It uses QoS

parameters such as bandwidth and delay to compute adelay-bounded least cost tree. The source node is thestart of the tree and the destination node is added if itpresents the least network cost.� Multiple least cost trees (MLCT): The source searches

for least cost trees (LCT) until the aggregate bandwidthfor all LCTs satisfy the net bandwidth requirement. Dueto multiple parallel trees the delay variance could belarge, but the highest delay value is maintained withinthe delay bound.

Genetic algorithms for group multicast [100]: Ran-daccio and Atzori propose a genetic algorithm (GA) basedapproach to the problem of finding multicast trees whichoptimize bandwidth and delay parameters.

The algorithm is initialized by building a population ofmulticast trees in isolation by combining unicast paths be-tween the source and destination pairs. The unicast pathsfollow the shortest path in terms of hops, calculated usingDijkstra’s algorithm. From this initial population, the GAalgorithm generates various (possibly sub-optimal) combi-nations and selects them for fitness. The fitness functionused by the GA is based on the weighted average of trans-mission delay and network resource utilization.

Fireworks [66]: Law et al. propose a 2-tier multicast/broadcast routing protocol, Fireworks, which adapts itselfbased on network topology and group density. At appropri-ate times, it resorts to broadcast instead of multicast. Sen-sor nodes are grouped together with local group leaders orcohort leaders corresponding to areas of high group mem-ber affinity. These cohort leaders establish a sparse multi-cast tree between themselves and the source node whilethey broadcast messages within the local group memberswithin their own cohort. The 2-tier hierarchical structurecomprises of the upper tier formed the source and cohortleaders (see Fig. 32). The lower tier consists of the membersin the cohort. The authors use a new metric termedas cohesiveness which maintains the affinity of group

A

D

E

B

C

S

Group source

Cohort leader

Group member

Ordinary node

Upper tier multicast structure

Cohort region

Fig. 32. Fireworks 2-tier multicast hierarchy structure [66].


members within a node’s k-hop radius. Individual groupmembers are discovered using special ADVERTISE mes-sages which contain the address, multicast address,hopcount and cohesiveness of the node. The joining nodethen determines whether it should join as the cohort leaderfor its k-hop neighborhood. The node becomes a cohortleader if it does not find any cohort leader and broadcastsa LEADER message. The upper tier structure is constructedby the source sending SOURCE-QUERY messages which arereplied to by cohort leaders using SOURCE-REPLY mes-sages. Within group members, Fireworks uses broadcastingto forward data while it uses multicasting between cohorts.

Probabilistic predictive multicast algorithm (PPMA)[95]: The Probabilistic Predictive Multicast Algorithm(PPMA) proposed by Pompili and Vittucci improves therobustness and reliability of multicast trees in the eventof link and/or node failures. The algorithm defines a newway to quantify the suitability of a link, the probabilisticlink cost which is comprised of three terms: energy, dis-tance and lifetime. Using this new metric, the multicasttrees can be computed in the centralized or distributedmanner. In the centralized approach, the algorithm simplysubstitutes the new metric for the other metrics tradition-ally used in the centralized Bellman-Form algorithm (suchas hop count).

In distributed PPMA, the multicast tree is created basedon a private and a public link costs. The private link cost isused by the nodes which are already added to the multi-cast tree, while the reminder of the nodes in the network,used the public link cost to aggregate paths and form mul-ticast trees.

Hierarchical multicast techniques and scalability[49]: Gui and Mohapatra introduce a framework for hierar-chical multicasting in MANET. The proposed approachesinclude a domain-based and an overlay-driven.

In the domain-based scheme, a large multicast group ofnodes is divided into sub-groups. Each sub-group is as-signed as a sub-root, chosen based on topological optimal-ity. The sub-root uses its own lower-level multicastprotocol to create its tree and deliver packets to nodes

within its sub-group. The source nodes of each group andsub-roots form a special sub-group for upper level multi-cast which is used by the source node to deliver packetsto the sub-roots.

In the overlay-driven multicast scheme, a hierarchicaloverlay multicast protocol is used to construct the virtualmulticast tree. In this framework, the upper level multicasttree needs to be logically spanned over all the group mem-bers. Once the tree is constructed, each non-leaf node isresponsible to forward packets to its children in the tree.This mechanism further improves the data deliveryefficiency.

Application layer multicast algorithm (ALMA) [44]:Ge et al. propose an application-layer receiver-driven over-lay multicast protocol. As the ALMA protocol operates atthe application level, it can be used in conjunction withany routing protocol.

ALMA creates a tree of logical links between the groupmembers. If node mobility or congestion makes it neces-sary, the tree can be dynamically reconfigured. Each edgeof the logical multicast tree represents a logical link – a pathat the network layer (see Fig. 33). The members of the thegroup can choose to allow zero, one or more children. Anew member joins the group by sending join messages tomultiple existing members. A member willing to take anew child responds and accepts the new node within thetree. If multiple replies are received by the new node, itchooses the member who replies the first. If a memberwants to leave a group, it sends an explicit leave messageto its parents and children nodes. Node failures are treatedsimilarly to unannounced departures from the tree. Thus,each member sends periodic hello messages to its parentand receives a return response. If no reply is received in apre-defined time interval, the nodes assumes the parenthas failed and it tries to rejoin the tree. Hello messages alsoprovide a mechanism to measure the quality of a path. Chil-dren monitor the quality of the path to their parent andmay switch parents when required. With members joining,moving or leaving, the multicast structure is changingand thus the tree requires constant reconfiguration. New

A

DC

X

B

Y Z

physical linkbetweenX and C

logical link betweenC and D

Fig. 33. Logical links versus physical links [44].


parents are searched by nodes based on the average esti-mated round-trip time (RTT) of a link. New parents shouldhave an RTT less than a pre-specified RTT threshold. Loopscan be detected in ALMA but not entirely avoided when twomembers decide to switch parents at the same time.

QoS aware multicast routing [114]: Sun and Li de-scribe a series of QoS extensions to the MAODV protocol.The approach uses the delay, bandwidth and packet-losscharacteristics of MAODV with no additional signaling. Italso incorporates multicast routing capability with theexisting unicast. A source node sends a QoS route request,RREQ, which is forwarded by intermediate nodes until itreaches the destination. The destination sends back theRREP packet with a delay time corresponding to a prede-fined node traversal time (NTT). Intermediate nodes addtheir own NTTs to the delay value and update their routingtables. Routes with the minimum delay are selected fordata transmissions. A similar technique is applied for thebandwidth requirement where source nodes indicate theirbandwidth requirements and intermediate nodes comparetheir available bandwidth before forwarding the packet.

Ad hoc QoS multicasting (AQM) [22]: Bur and Ersoypropose the AQM protocol which tracks QoS availabilitywithin the neighborhood of every node based on therequirements and announces it during the session initia-tion. In order to join a session, the nodes go through a re-quest-reply-reserve procedure that ensures the QoSinformation is updated and a possible route is selected.

A session is initiated by the initiator (MCN_INIT) nodeby broadcasting a session initiation packet (SES_INIT).This packet consists of the identity number and QoS classof the new session while also setting the bandwidth andhop count rules for the session. Active sessions aremaintained in a table (TBL_SESSION) at each node. A mem-bership table (TBL_MEMBER) maintains the status of pre-decessor nodes. The session information is maintainedwith periodic session updates keeping track of changes inthe QoS conditions and node connectivity. Session updatepackets (SES_UPDATE) refresh the session informationperiodically while session termination (SES_TERMINATE)closes a session. On session termination, nodes clean theirtables and free the reserved resources for that session.Neighborhood maintenance is also carried out by broad-casting periodic hello packets (NBR_HELLO) which informsneighbors of the node’s current bandwidth usage.

Content based multicast (CBM) [138]: Zhou and Singhpresent a multicast model for a scenario where nodes areinterested in obtaining information about specific threatsand resources. These threats and resources are a time t

and distance d away from the current location of the node.Nodes generate information about the movement, inten-sity and location of threats. This information is multicastthrough the network using a sensor-push receiver-pull ap-proach. Here, sensors push the information into the net-work while receivers pull the relevant information. Thenetwork is divided into geographic regions and a sensordetecting a threat broadcasts it into one of these small re-gions. Individual receivers then pull threat warnings fromnodes that lie in the direction of travel. The broadcast mes-sage is forwarded in the projected path of the threat to theleader nodes of each region. Each leader receiving the mes-sage broadcasts it within their groups. Whenever thethreat changes direction, a new message is created and re-broadcast. Nodes pull the messages containing the threatinformation to be warned. Using the PULL_REQUEST mes-sage, the node requests the leader of the block to send thethreat message after a certain time. The leader then sendsback all information it has about the threat. These pull re-quests have a time to live (TTL) field during which the lea-der sends back threat messages to the node issuing the pullrequest. This ensures that pull requests are only used infre-quently thereby not causing flooding in the network.

Differential destination multicast (DDM) [55]: In theDDM algorithm, proposed by Ji and Corson the source nodeof a multicast transmission encodes all the destination ad-dresses within each data packet header in an in-band fash-ion. With this approach, no fixed multicast tree is created,the routing will be soft-state, similar to state routing algo-rithms such as DSR. This allows a lower control overhead,as there is no need for extra packets to maintain multicastforwarding state. Control overhead only occurs when thereis actual data to send. Nodes along the paths also do notneed to maintain alternate backup routes. Once a node re-ceives a data packet, it checks the DDM header to deter-mine which node to forward the packet. This informationwill be remembered to help in the forwarding of futurepackets in the same direction. If the destination nodechanges, the upstream node informs all its immediateneighbors about the difference in the forwarding node.

Robust multicasting in ad hoc networks using trees(ROMANT) [117]: The ROMANT algorithm proposed byVaishampayan et al. uses a receiver-initiated group joiningscheme which does not require any underlying unicastrouting protocol or the pre-assignment of cores to groups.

The cores of the groups are determined as follows.When a receiver joins a group, it checks if it has ever re-ceived a core announcement for that group. If it did, thenode joins the group as a non-core node. Otherwise, thenode considers itself to be the core of the group and startssending core announcement packets with a core ID. If sev-eral receivers join the group simultaneously, the one withthe highest ID becomes the group core.

The periodically broadcasted core announcements areused to establish a connectivity list at each node, storinginformation about the core and routes which lead to thecore. Router nodes use the core announcements receivedto determine where to route data packets. Data packetsare forwarded to nodes from which the core announce-ment with the highest ID has been received and the corethen forwards the data packet across the tree.


Epidemic-based reliable and adaptive multicast formobile ad hoc networks (EraMobile) [88]: Ozkasapet al. propose a reliable and adaptive multicast protocolbased on bio-inspired epidemic methods. Epidemic meth-ods are stateless, thus they are a good match for the rapidlychanging, non-deterministic structure of MANETs. Thealgorithms takes advantage of the broadcast nature of thewireless medium to send gossip messages locally withina multicast group to neighboring nodes.

The traditional approach in gossip based protocols is toselect a random node from a predefined list before unicast-ing the gossip message to the node. In EraMobile, the nodegossips with a random subset of one-hop neighbors, con-stantly changing with node mobility and changes in the lo-cal node density. The frequency of gossip messages are alsoadjusted dynamically.

The periodic gossip messages allow nodes to recovermissing data packets and make the protocol robust againstdelivery failures. The gossip messages also reduce thebandwidth and energy usage and allow a lower controloverhead compared to multicast flooding.

3.7.1. ComparisonOne of the important comparison terms is whether the

multicast happens at the network layer or somewhere else.Most protocols position the implementation of the multi-cast at the network layer. ALMA [44] implements it atthe application layer (more exactly, at what the ISO modelwould call the session layer – but in our current 4-layernetworking hierarchy would be the lowest sublayer ofthe application layer – with multiple applications beingable to be built on top of it).

Most multicast protocols are based on the receiver sub-scribing to the transmissions of a specific sender. An inter-esting exception is CBM [138], which performs multicastbased on the content rather than the source of the messages.

Almost all the protocols are based on building a multi-cast tree, although there are some exceptions. CBM [138]does not build a tree due to its radically different distribu-tion model. Differential destination multicast (DDM) [55]performs an on-demand, soft state based multicastingwithout constructing an explicit tree. Finally, EraMobile[88] replaces the multicast tree with a stateless approachbased on epidemic algorithms.

Another question is whether the algorithm is consider-ing the state of the underlying network in the choice of therouting tree. There is an overall group of protocols whoseapproach is to select a core of the network. These nodeswill serve as forwarding nodes (for instance, as the non-leaf nodes of the multicast tree). Naturally, the nodes inthe core will be nodes with more resources (although othercriteria might also be considered – for instance, the factthat the core must extend in all geographic areas of thenetwork). From the protocols reviewed, core based proto-cols are DCMP [35], AMRoute [130] (‘‘logical core’’), Fire-works [66] (‘‘cohort leaders’’), and ROMANT [117]. Thelatter is an example of those protocols where the choiceof the core is not based on resources (being simply basedon the highest id). Other protocols do not establish a corebut consider the available resources of the nodes on acase-by-case basis: energy efficient multicast routing [69]

and PPMA [95] where energy is part of the probabilisticlink cost.

Most protocols use a distributed implementation, withthe only exception being the genetic algorithm based ap-proach [100].

Finally, there is a question whether the protocol consid-ers QoS features such as minimal bandwidth of the multi-cast. QoS assurance almost always conflicts with resourceconservation, as nodes with move advantageous locationsor higher bandwidth will tend to become overloaded. Fromthe surveyed protocols, the ones considering QoS are:QMRPCAH [67] (soft Qos), QoS multicast routing usingmultiple paths/trees [128], QoS aware multicast routing[114], AQM [22].

3.8. Geographical multicast (geocast) protocols

Geographical multicast (geocast) routing is a variant ofmulticast where the goal is to route the packets comingfrom a source to destinations located within a specific geo-graphical region. Naturally, for geocast to work, the nodesneed to rely on localization techniques (such as GPS). Anearlier survey on geocast protocols can be found in [78].

In the following we survey some of the representativegeocast algorithms. (To unify the terminology, we shalluse the term geocast throughout this survey – different pa-pers use slightly different terminology).

Direction guided routing (DGR) [6]: The DGR algo-rithm, introduced by An and Papavassiliou relies on clustersof nodes, roughly equivalent to the cells in cellular tele-phony. The clusters are created and the clusterheads electedusing dynamic techniques such as Mobile Clustering Algo-rithm (MBC) [5]. DGR then creates a mesh structure ontop of the clusterheads, with the objective to deliver packetswithin the cluster using reduced overhead. The algorithmfor delivering the messages from a source to a specificgeocast zone can be described in the following steps:

1. The source formulates the geocast zone, calculates itsdistance to the center of the zone and sends the geocastmessages to its own clusterhead.

2. When the source clusterhead receives a geocast mes-sage, it creates and broadcasts a GeoJOIN REQUEST toneighboring clusterheads. This is achieved by usingthe Boundarycast and BoundaryCrosscast operations[51]. When a boundary node receives the GeoJOIN

REQUEST from its clusterhead, it executes the Boun-darycast operation, which is used for routing betweenclusters.

3. If a non clusterhead receives the GeoJOIN REQUEST, itdetermines whether it is within the geocast zone ofthe sender by checking the geographical location withinthe message packet header. If it is within the geocastzone, it simply accepts this message. Otherwise thenode makes a decision to forward or not based on dis-tance calculations. These reduce the overhead associ-ated with route discovery by limiting the region inwhich control messages are forwarded.

4. When a clusterhead receives the GeoJOIN REQUEST, itstores the clusterhead ID of the earlier clusterhead andperforms geocast membership management strategies.


If such members are present, it sends GeoJOIN REPLY

messages to its upstream clusterheads. When theseclusterheads receive the messages, they save theclusterhead ID of the node where the message isreceived and send the message further on upstream.Finally, as the source clusterhead receives the GeoJOINREPLY message, the operation is completed.

Geocast adaptive mesh environment for routing (GA-MER) [24]: Camp and Liu propose the GAMER protocolwhich creates a dynamic mesh based on the mobility prop-erties of the nodes: highly mobile nodes create a densemesh, while nodes with a limited mobility create a sparsemesh. The mesh creation approach is based on the ap-proaches by Chiang and Gerla [29] and by Garcia–Luna–Aceves and Madruga [42].

The geocast aspect of GAMER is based on the forward-ing zone strategies used in LAR [61] and DREAM [11]. InDREAM, a circle is created which is centered on the lastknown location of the destination node. The radius of thecircle depends on the velocity of the destination node.After defining the forwarding zone, the source node sendsthe data packet to nodes within one-hop distance of thiszone. The nodes then repeat the process by creating theirown forwarding zones and forwarding the packets. There-fore, instead of flooding the entire network, the nodes onlyflood their respective forwarding zones.

GAMER uses a source routing approach based onDSR[57] to route geocast packets around the network.The main problem with source routing is the fact that largeamounts of overhead is generated by storing the entireroute in the packet itself. Camp and Liu state that GAMERcould be modified to store only the local state informationinstead of source routing information.

Let us now describe the operation of the GAMERprotocol.

When the source node in GAMER wants to send geocastpackets to transmit, it periodically sends JOIN-DEMAND

(JD) packets to the geocast region. As all mesh nodes inthe target region must join the group, the approach mightbe more correctly described as geo-broadcasting. Thesource node chooses one of three possible forwarding ap-proaches (FAs): FLOOD FA, CORRIDOR FA or CONE FA.FLOOD FA floods the network with the JD messages. InCORRIDOR FA, a rectangular forwarding zone is createdand the JD packets are forwarded through this corridor.

succeed

fail

CONE

CORRIDOR

Fig. 34. Flow of the FA used to transmit

Since the area in this case is much smaller than the previ-ous case, a more sparse mesh is created. The CONE FA fur-ther restricts the the forwarding zone to an area enclosedby the source node acting as the vertex and the angle cre-ated from the tangential sides from this vertex (seeFig. 34).

If a mesh node receives a non-duplicate JD packet and itis within the forwarding zone but not within the geocastregion, it adds its own address to the source route inthe header and forwards it to its neighbors. However, ifit is within the geocast region, it responds by sending aJOIN-Table (JT) packet to its neighbors. This packet con-tains the reverse source route from the JD packet which isthen forwarded to the source which can start sending datapackets.

A geocasting protocol for mobile ad hoc networksbased on GRID (GeoGRID) [73]: Liao et al. base GeoGRIDon the unicasting routing protocol GRID [72]. GeoGRIDuses location information to define the forwarding zone.The geographic area of the network is divided into logicalgrids of size d by d. In each grid area a gateway node iselected, whose responsibility is to forward the geocastpackets. GeoGRID uses two geocast forwarding methods:flooding based and ticket based.

� Flooding based GeoGRID: In this method, only gatewaynodes within the forwarding zone are allowed to broad-cast the geocast packets.� Ticket based GeoGRID: In this technique, only selected

gateways are allowed to forward the geocast packets.If the entire region is divided into an m by n region, atotal of m + n tickets are first generated. The source thendistributes these tickets evenly to neighboring gatewaynodes in the forwarding zone which are closer to thegeocast region than the source.

The election of the gateway is an important issue in theGeoGRID protocol. The initial gateway of the grid is thenode closest to the physical center of the grid. One wayin which the gateway of a grid might change is when thegateway node moves out of the grid. Alternatively, a gate-way node can silently turn itself to a non-gateway nodeand relinquish control to the other more suitable nodewho is then elected as the gateway. Another method ofelecting gateways could be by using the concept of nodeweights as in [10].

fail

succeed

FLOOD

the next JD packet in GAMER [24].


Geocasting in mobile ad hoc networks (GeoTORA)[63]: Ko and Vaidya propose the GeoTORA protocol whichbuilds upon the unicast TORA [89] routing protocol. TORAuses the distributed ‘‘link reversal’’ algorithm and providesmultiple routes to the destination. It also uses the conceptof ‘‘heights’’ to determine direction of the links. In GeoTO-RA, the source node anycasts the geocast messages to thegeocast group by using TORA. Once any node in the geocastregion receives the geocast packet, it floods the packet intothe region. This helps in limiting flooding only within itsown region.

3.8.1. Comparison of geographical multicast protocolsGeographical multicast involves multicasting to a collec-

tion of nodes which are defined as destinations by theirphysical location. Many protocols in this class are buildingupon other protocols: often on location aware protocolssuch as LAR, DREAM and GRID, but the starting point canalso be a non-location aware reactive protocol such as TORA.

Despite their various roots, geocast protocols need tosolve a common set of challenges: how to define the geo-cast area and how to efficiently route the messages to allthe nodes in that area. The approaches taken by the proto-cols are very varied. DGR [6] allows the geocast area to bean arbitrary polygon, but for efficiency purposes it approx-imates it with a circle, ellipse or a rectangle. GAMER [24]routes towards a rectangular geographic area using a meshof paths which are either unconstrained, or constrained ina rectangle or a cone. For GeoGRID [73] the destination re-gion is a rectangular subset of the grid. The routing can beeither minimally constrained by a rectangle covering boththe source and the destination rectangle or through a tick-eting approach where the number of tickets issued is con-trolled by the source. Finally, GeoTORA [63] takes anapproach where it first forwards the packet to any nodein the geocast region through unicast, followed by a localflooding phase where the packet is flooded throughoutthe geocast area.

3.9. Power-aware protocols

Power-aware routing makes the routing decisionsdependent on considerations of the available energy ofthe nodes. These considerations can be significantly morecomplicated than simply finding the route with the lowestenergy consumption (in fact, the shortest path is almost al-ways the one with the lowest energy consumption). Theprotocols from this class take into consideration both theheterogeneity of the energy resources of the nodes, as wellas the uneven energy consumption due to the topology ofthe network and the nature of the data flows. For many ofthese protocols, the ultimate objective is to maximize thelifetime of a network with nodes with limited and fixed en-ergy resources.

Power aware routing in mobile ad hoc networks[109]: In an influential early paper from 1998, Singhet al. argue that routing protocols for ad hoc networksshould possibly consider, in addition to the shortness ofthe paths, one or more from a collection of other metricswhich impact power consumption. The paper proposesthe following metrics:

� Minimize energy consumed per packet: Let us consider apacket j traversing a path formed of the nodesn1,n2, . . . , nk. Let us denote the energy needed to trans-fer the packet from node a to b with T(a,b). The goal is tominimize

Pk�1i¼1 Tðni;niþ1Þ for all packets j.

� Maximize time for network partitioning: This require-ment tries to balance the load on the network in sucha way as to delay, as much as possible, the moment inwhich node failures leave the network as two or moredisjoint graphs, without interconnections.� Minimize variance in node power levels: This metric

directly relates to the extension of the network’s lifetime. No node should be penalized with a higher energyconsumption than its peers, assuring that all nodes willremain up and running together for as long as possible.� Minimize cost per packet: The minimization of energy

consumption might lead to situations where somenodes in specific locations will need to forward moretraffic, exhausting their energy supplies. To extend thelife time of the network we need to define a cost metricwhich can ensure that no node with low energy levelwill participate too often in routes. We can define afunction fi(xi) that estimates the ‘‘node cost’’ or ‘‘weight’’of a node i (where xi is the total energy expended by i).This intuitively defines the node’s reluctance to forwardpackets. Possible forms of this function include combi-nations of inverse relations between the current batterylife and the cost at that node (i.e., higher the battery life,lower the cost at that node).� Minimize maximum node cost: If Ci(t) denotes the cost

to route a packet through node i at time t, and C0(t)denotes the maximum C(t) over all the nodes, the goalis to minimize C0(t). In other words, the aim is to mini-mize the maximum node cost after routing N packets totheir destinations or after T seconds. Using this metricwill delay the time point where the first node in thenetwork will fail and will implicitly reduce the variancein the node power levels.

The paper shows through simulations that the pursuitof these optimization criteria is possible and significantcost reductions can be achieved. Naturally, one cannotoptimize simultaneously along all axes, as some of the cri-teria conflict with each other (for instance, for most net-works, the minimum energy consumption routes are notthe one with the lowest variance on the node energyusage). The paper does not propose full fledged routingprotocols, but offer a guidance with respect to what sortof metrics future protocols would need to strive to, andhow these would be evaluated.

More than a decade later, we can, with some confi-dence, say that the authors were very optimistic whenstating that the integration of these metrics in routing pro-tocols will be easy. Many subsequent papers in poweraware ad hoc routing can be seen as attempts to find prac-tical solutions to the challenges outlined here.

Device and energy aware routing (DEAR) [7]:Avudainayagam et al. propose a protocol for a heteroge-neous network where some nodes are on battery powerwhile other nodes are connected to a continuous supplyof power (or can be periodically recharged as needed).


The goal of the protocol is to rely on the latter type ofnodes for most of the routing functionality, thus extendingthe lifetime of the battery powered nodes.

The DEAR procol calls a node device-aware if it can dis-tinguish between whether it is battery powered or exter-nally powered and the cost of using a device of the lattertype is zero. The routing table of device-aware nodes hasan additional field called DeviceType which indicateswhether the destination node is on battery power or isexternally powered. An additional redirect table containspairs of destination addresses and redirect addresses.

After every routing table update, the node calculates theleast cost to any externally powered device from its rout-ing table and updates the redirect table accordingly aswell. During routing, the node compares the cost of reach-ing the destination versus the cost of reaching a powerednode, and if the latter is cheaper, it will redirect the packetto a powered node. The DEAR protocol assumes that anypowered node can boost its transmission power such thatit is at a one-hop distance to any destination, thus thetransmission from the powered node is considered to beof zero cost from an energy consumption point of view.

Routing and channel assignment for low powertransmission in PCS [107]: Scott and Bombos propose atechnique for minimizing the transmission power in PCSnetworks by the simultaneous choice of the route andchannel assignment (this combined problem is called ‘‘callplacement’’). The aim is to increase the lifetimes of theindividual nodes, and hence the network lifetime. The ap-proach chosen is similar to the frequency reuse factor inAMPS cellular service.

Due to the inherent complexity and the overhead in-volved with the continuous optimization of the entire net-work, the authors use an approach which is triggered onlywhen new calls are made. In this least disturbance ap-proach, new calls are placed such that the total power re-quired to sustain the new call is minimal. This willminimize the new call’s interference with the rest of thenetwork.

Energy conserving routing in wireless ad hoc net-works [26]: Chang and Tassiulas start with the observationthat in general, the routes with the lowest energy cost arethe routes with the least hops. It would appear that calcu-lating the shortest path would also minimize the energyuse. However, this technique leads to high energy con-sumption of the nodes which are along the shortest paths,while the battery power of the other nodes in the networkremain largely unutilized. Chang and Tassiulas introduce arouting strategy to maximize the network lifetime basedon sets of source–destination pairs and the traffic genera-tion rates on these flows. A class of flow augmentationand flow redirection algorithms are proposed which bal-ance energy consumption rates in relation to the energy re-sources of the nodes. Simulation results show 60% increaseof the system lifetime compared to the minimum transmit-ted energy routing algorithm.

CLUSTERPOW and MINPOW [60]: Kawadia and Kumardesigned a series of energy aware algorithms specificallytargetting non-homogeneous ad hoc networks. Theauthors notice that the power control and routing is mutu-ally dependent of each other; for instance routing must be

aware of the power control since link quality mainly de-pends on the available battery power at each node.

The CLUSTERPOW protocol uses a dynamic and implicitclustering of nodes in a network. The clustering criteria isthe power levels of each node (in contrast to the traditionalclustering approaches based on addresses or geographicalplacement of nodes). A route is selected by ensuring thateach hop in the route has a maximum transmit powerlevel.

The tunneled CLUSTERPOW scheme augments this pro-tocol with the ability to perform packet encapsulation atlow power levels instead of sending the packet directlyto the next hop.

Finally, the MINPOW protocol relies on the distributedBellman-Ford algorithm with sequence numbers, usingthe total power consumption as a metric instead of hopcount. To calculate the shortest path, any of the existingshortest path algorithm can be used to calculate the small-est transmit power required to traverse the link. MINPOWtakes into consideration the total power consumed in com-munication instead of individual power levels.

Interference aware cooperative routing [77]: Mah-mood and Comaniciu propose an algorithm specifically tar-geted to CDMA-based ad hoc sensor (in contrast to thecollision-based physical layer models assumed by mostother algorithms).

CDMA networks almost always implement a variabletransmission power to prevent the ‘‘nearfar effect’’, wherethe useful signal is drowned by strong interference fromadjacent transmitters.

The two algorithms proposed by Mahmood andComaniciu maximize the throughput and minimize energyconsumption in ad hoc networks. The proposed strategiestry to mitigate the near-far effect at the network layer bymaking appropriate selections of routes within the net-work instead of using more sophisticated power controltechniques.

The first approach is based on minimum energy routingwith additional selection criterion based on the potentialinterference levels. Once a route has been selected, thealgorithm estimates the interference created by each nodeon the route to neighboring nodes. If this level is higherthan a threshold, the node, as a potential near-far effecthot spot, is removed from the route.

The second approach is based on joint minimum energyand near-far minimization routing, using a composite met-ric. Several additive and multiplicative metrics are pro-posed and evaluated.

Simulation results show an increase in net throughputof the network of up to 60 percent in comparison withthe minimum energy routing.

Minimum energy hierarchical dynamic source rout-ing (MEHDSR) [115]: Tarique and Tepe extend the well-known DSR protocol with two protocols which considerenergy consideration.

The first protocol, minimum energy dynamic sourcerouting (MEDSR), extends the well known DSR protocolby modifying the control messages and using a link-by-linkpower adjustment strategy. Source nodes trying to find anew path to the destination first try to find one traversingexlusively links requiring a low power level. If the attempt

(a)

n1

n1

n2

(b) (c)

n2 n3 n4

n3 n4

n1 n2 n3 n4

Fig. 35. (a) Minimum transmit power, (b) high transmit power, (c) low transmit power [115].


is unsuccessful, successively higher power levels can befound. The different power levels are used to identify pathswhich can return low energy routes (see Fig. 35). Multiplepower levels also reduce the route discovery time andoverhead. Once the route discovery process is successful,the transmission power levels of these nodes are adjustedon a link-by-link basis to the minimum required level foreach link.

Although MEDSR is creates highly efficient routes (a25% improvement in energy consumption has been mea-sured), the flooding technique used for route discoveryhas both a large of overhead and is energy inefficient.The second proposed protocol, HMEDSR, adopts a hierar-chical routing approach similar to Hierarchical DynamicSource Routing (HDSR). The reduction in the routing over-head packets provides an additional 12% energyimprovement.

On-line disjoint path routing for capacity maximiza-tion [71]: Liang and Liu aim to maximize the capacity of anad hoc network, defined as the number of successfully rou-ted messages. The assumption is that the algorithms haveno knowledge of future path requests or the traffic onthose paths (as a note, this is the normal operating condi-tion in ad hoc networks). They are considering a networkwhere the transmission power of every node can be ad-justed within the limit of a maximum transmission power.Thus, links can be added or removed to the topology byallowing the nodes to vary their transmission power. Inthis setting, the routing problem requires both the choiceof the nodes on the path as well as the selection of theirtransmission power.

The authors are proposing two centralized on-line algo-rithms for the problem. The first algorithm is based onmaximizing the local network lifetime thereby minimizingthe transmission energy consumption throughout the net-

work. The second algorithm is a heuristic solution based onan exponential function of the energy utilization at thenodes. The minimum energy between two node/edge-disjoint paths are computed and their sum compared witha threshold to determine whether the connection requestis accepted or not.

Power conserving routing with entropy-constrainedalgorithms [58]: Karayiannis and Nadella develop a rout-ing algorithm which utilizes the information-theoreticconcept of entropy aiming to reduce the uncertainty asso-ciated with route discovery.

The paper starts from the idea of entropy constrainedrouting algorithms, which have been introduced by theauthors in several preceeding papers as a mean to imple-ment routing based on multiple performance criteria.The idea is that the optimization criteria is implementedas a series of constraints on the chosen route. In an en-tropy constrained routing, the initial iteration of theroute starts with a level of uncertainty: multiple possiblenodes can compete for each position of the route. Thisuncertainty is gradually reduced by a deterministicannealing process.

In [58] the authors apply this approach to the issueof power conserving routing. As the entropy con-strained routing can possibly optimize for more than oneperformance criteria, two specific implementations areconsidered:

� optimizing for a single performance metric, with themetric chosen to be the link cost,� multiple performance metrics: the metrics chosen are

link cost together with link reliability.

The authors are comparing the performance of the algo-rithms against traditional approaches using a Dijkstra’s

a

d

c

b10 15

15 20

50

75

200100 a

d

c

b

50

75

200100

40

9085

60 55

185

35180

(b)(a)Fig. 36. (a) A graph showing energy levels at nodes and energy requiredto transmit at each edge. (b) Shows the corresponding energy graph [81].


algorithm in a mobile ad hoc network and conclude thatentropy-constrained algorithms can increase the lifetimeof the network (as measured by the time to the first nodeexhausting its energy).

Online energy aware routing [81]: Mohanoor et al.propose polynomial time combinatorial techniques tocompute energy efficient routes in MANETs. The aim is toselect a route which strikes a balance between the residualenergy, the minimum energy level of any node in the pathand the energy consumed along a path. The network is con-sidered as a graph with the edge weights being the energyrequired to transmit (see Fig. 36(a)). The sum of all weightsalong a path corresponds to the total energy consumedalong the path (see Fig. 36(b)).

The authors consider several algorithms, united by theirreliance on a two phase computation. The first technique,called the shortest widest path works as follows. First, a var-iant of Dijstra’s algorithm is applied to the graph con-structed as above, which searches for paths with theminimum residual energy (the ‘‘widest’’ paths). There canbe several paths which satisfy this condition. In the secondphase, the algorithm chooses from these paths the onewhich has the lowest energy consumption.

Starting from this approach, the paper also describesseveral variations on the shortest widest path approach.

Table 3.7.1Multicast and Geo-Multicast routing protocols comparison.

Protocol Core/broadcast Route metric

DCMP Core Newest RouteADMR Neither Link breaksAMRoute Core Unicast operationLi et al. Neither Minimum energyQMRPCAH Broadcast QoSWu and Jia Neither SP and least costRandaccio and Atzori Neither SP, WBW & delayFireworks Both SP, CohesivenessPPMA Core Distance & link costALMA Neither Link breaksAQM Core QoSCBM Core Threat arrivalDDM – SPROMANT Core ConnectivityEraMobile – Randomly selectedDGR Core SPGAMER Core SPGeoGrid Core Hop countGeoTora Broadcast SP

Route metric: SP = Shortest path; WBW = Weighted bandwidth.Route repository: RT = Routing table; RC = Route cache.

The shortest width constrained path algorithm allows theuser to trade off the width of the path for a lower energyconsumption. The used can specify the trade-off factoreta which describes the acceptable trade-off ratio. The sec-ond variation, the shortest fixed width path finds the mini-mum energy path on the path which has a higher widththan a fixed value.

The authors show through simulation results that theuse of this class of power aware routing algorithms can re-sult in significant improvements in the network lifetime.

3.9.1. ComparisonPower aware routing makes the decision of routes

dependent on power consumption characteristics. In gen-eral, the goal is to find routes for a collection of well spec-ified flows. Thus, the routes need to be set up on-demand.Pre-emptive routing algorithms can be used only in con-junction with an estimate of future flows.

This makes routing a multi-criteria optimization prob-lem. For the papers surveyed, the number of criteria canbe usually considered two: one traffic related (shortestpath, bandwidth) and one energy related criteria. A possi-ble exception is the paper by Mahmood and Comaniciu[77] which adds interference to the minimization criteria(although interference, in this case, is strongly tied tobandwidth).

One of the problems is that power-awareness can rarelybe compressed into a neat optimization criteria. The abso-lute amount of power consumed network-wide is almostcompletely irrelevant, in fact, none of the routing algo-rithms discussed here optimizes for it. One reason for thisis that the minimization of energy consumption is quite of-ten can be achieved by shortest path routing.

The challenges instead are not simply to minimize theenergy consumption, but to manage it. The routing algo-rithm can redistribute the routes over the network, suchthat the overall power consumption is redistributed over

Forwarding strategy Route repository

Source routing RTTree based or flooding RTShared trees Depends on unicast algoSource routing RCBordercast RTSource routing RCSource routing RCMulticast & broadcast RCSource routing RCTree based RTSource routing RTLimited broadcast RCSource routing –Source routing RTLocal broadcast –Limited flooding RCSource routing RCFlooding or ticket based NoneLimited flooding RT

Table 3.8.1Power aware routing protocols comparison.

Protocol Type Path strategy Routing metric Scalability Robustness

DEAR G – Based on ‘‘DeviceType’’ No YesScott & Bombos C Single-path Multiple constrained SP Yes NoChang & Tassiulas G Single-path Power cost No NoCLUSTERPOW Cl Shortest path Total consumed power Yes YesMahmood & Comaniciu D Single-path Energy and interference No NoMEHDSR G Single-path SP or next available Yes NoLiang & Liu D Single-path Energy consumption Yes YesKarayiannis & Nadella D Single path Entropy Yes NoMohanoor et al. D Single-path Constrained SP No

Type: C = Centralized; Cl = Clustered; D = Distributed; G = Global.Routing metric: SP = Shortest path.


the network. As the paper of Singh et al. [109] argued in thelate 1990’s, this cost redistribution can be specified in sev-eral alternative ways. In the majority of papers, the ulti-mate objective is to extend the lifetime of the network.There are two dominant definitions for this. Some papersconsider the moment when the first node in the networkfails such as Karayiannis and Nadella [58], while most ofthem consider the moment when the network looses con-nectivity Chang and Tassiulas [26], Liang and Liu [71],Mohanoor et al. [81].

Another aspect is whether the algorithm is imple-mented in a centralized [107,71,58,81] or a distributedway [7,60,115].

Similarly to QoS routing, one of the difficulties of en-ergy efficient routing is that the optimal route for oneparticular flow of data cannot be determined, except byconsidering other currently ongoing paths. Furthermore,theoretically, optimal allocation cannot be achieved byallocating every new flow of data as they are started:we might possibly need to re-route the existing routes,such that a new optimum can be reached. Such a re-routing of existing flows can lead to a better overallroute, but it also creates many problems. Thus some ofthe considered algorithms are explicitly designed to min-imize the disturbances, e.g. Scott and Bombos [107],while many others are designed to operate in an onlinemanner – that is, consider the new data flows as theyappear.

Another cross-cutting consideration concerns node het-erogeneity. Considering this is tricky because differentauthors mean different things by heterogeneity. First ofall, even if we start with perfectly identical nodes, after awhile, they can become heterogeneous because of differentlocations in the network, and implicitly, different batteryconsumption. Some papers consider hard heterogeneity,where the nodes are physically different, either by theirenergy resources, or by their transmission range: for in-stance, DEAR [7].

The paper by Liang and Liu [71] operates with an inter-esting assumption: it assumes that any node can dynami-cally extend its transmission power to cover the entirearea (see Table 3.7.1).

A number of other comparison terms are summarizedin Table 3.8.1.

4. Conclusions

In this paper, we introduced a taxonomy of ad hoc rout-ing protocols. We have divided the ad hoc routing proto-cols into nine categories: (i) source-initiated (reactive oron-demand), (ii) table-driven (pro-active), (iii) hybrid, (iv)location-aware (geographical), (v) multipath, (vi) multi-cast, (vii) geographical multicast, (viii) hierarchical, and(ix) power-aware. For each of these classes, we reviewedand compared several representative protocols. While dif-ferent classes of protocol operate under different scenarios,they usually share the common goal to reduce controlpacket overhead, maximize throughput, and minimizethe end-to-end delay. The main differentiating factor be-tween the protocols is the ways of finding and/or main-taining the routes between source–destination pairs.

The development of the ad hoc routing protocols overthe last 15 years is an example of one of the most system-atic explorations of a design space in the history of com-puter science. Although, clearly, newer protocols havebuilt upon the earlier ones, we cannot identify a single‘‘best’’ protocol. Almost all the protocols we discussed inthis paper have their own sweet spot deployment scenar-ios and performance metric combinations where they out-perform their competitors.

From the point of view of the practitioner, this creates aserious problem. To deploy an ad hoc network with an opti-mal performance, it requires a very careful analysis of thescenario and its requirements, and the appropriate choiceof the routing protocol from the dozens applicable in thecontext. We hope that the taxonomy presented in this pa-per will be a helpful instrument for making this decision.

References

[1] J.-D. Abdulai, M. Ould-Khaoua, L. Mackenzie, Adjusted probabilisticroute discovery in mobile ad hoc networks, Computers andElectrical Engineering 35 (1) (2009) 168–182.

[2] G.N. Aggelou, R. Tafazolli, RDMAR: A bandwidth-efficient routingprotocol for mobile ad hoc networks, in: Proceedings of ACMWOWMOM, August 1999, pp. 26–33.

[3] C. Ahn, Gathering-based routing protocol in mobile ad hocnetworks, Computer Communications 30 (1) (2006) 202–206.

[4] J. Al-Karaki, A. Kamal, Efficient virtual-backbone routing inmobile ad hoc networks, Computer Networks 52 (2) (2008)327–350.


[5] B. An, S. Papavassiliou, A mobility-based clustering approach tosupport mobility management and multicast routing in mobile ad-hoc wireless networks, International Journal of NetworkManagement 11 (6) (2001) 387–395.

[6] B. An, S. Papavassiliou, Geomulticast: architectures and protocolsfor mobile ad hoc wireless networks, Journal of Parallel DistributedComputing 63 (2) (2003) 182–195.

[7] A. Avudainayagam, W. Lou, Y. Fang, Dear: a device and energyaware routing protocol for heterogeneous ad hoc networks, Journalof Parallel Distributed Computing 63 (2) (2003) 228–236.

[8] R. Bagrodia, M. Gerla, L. Kleinrock, J. Short, T. Tsai, A hierarchicalsimulation environment for mobile wireless networks, Technicalreport, Dept. of Computer Science, University of California at LosAngeles, 1996.

[9] A. Bamis, A. Boukerche, I. Chatzigiannakis, S. Nikoletseas, A mobilityaware protocol synthesis for efficient routing in ad hoc mobilenetworks, Computer Networks 52 (1) (2008) 130–154.

[10] S. Basagni, Distributed clustering for ad hoc networks, in:Proceedings of IEEE ISPAN, June 1999, pp. 310–315.

[11] S. Basagni, I. Chlamtac, V.R. Syrotiuk, B.A. Woodward, A distancerouting effect algorithm for mobility (DREAM), in: Proceedings ofthe ACM MOBICOM, October 1998, pp. 76–84.

[12] R. Beraldi, The polarized gossip protocol for path discovery inmanets, Ad Hoc Networks 6 (1) (2008) 79–91.

[13] R. Beraldi, L. Querzoni, R. Baldoni, A hint-based probabilisticprotocol for unicast communications in MANETs, Ad HocNetworks 4 (5) (2006) 547–566.

[14] L. Blazevic, J.-Y.L. Boudec, S. Giordano, A location-based routingmethod for mobile ad hoc networks, IEEE Transactions on MobileComputing 4 (2) (2005) 97–110.

[15] L. Blazevic, S. Giordano, J.-Y.L. Boudec, Self organized routing inwide area mobile ad-hoc networks, in: Proceedings of IEEEGLOBECOM, November 2001, pp. 2814–2818.

[16] G. Boato, F. Granelli, MORA: a movement-based algorithm for adhoc networks, in: Proceedings of The Tenth InternationalConference on Distributed Multimedia Systems (DMS), 2004, pp.171–174.

[17] J. Boice, J. Garcia-Luna-Aceves, K. Obraczka, Combining on-demandand opportunistic routing for intermittently connected networks,Ad Hoc Networks 7 (1) (2009) 201–218.

[18] J. Boleng, Exploiting location information and enabling adaptivemobile ad hoc networking protocols, Ph.D. Thesis, Colorado Schoolof Mines, 2002.

[19] J. Boleng, T. Camp, Adaptive location aided mobile ad hoc networkrouting, in: Proceedings of IEEE International Performance,Computing, and Communications Conference (IPCCC), April 2004,pp. 423–432.

[20] A. Boukerche, S.K. Das, A. Fabbri, Analysis of a randomizedcongestion control scheme with DSDV routing in ad hoc wirelessnetworks, Journal of Parallel and Distributed Computing 61 (7)(2001) 967–995.

[21] J. Broch, D.A. Maltz, D.B. Johnson, Y.-C. Hu, J. Jetcheva, Aperformance comparison of multi-hop wireless ad hoc networkrouting protocols, in: Proceedings of ACM MOBICOM, October 1998,pp. 85–97.

[22] K. Bür, C. Ersoy, Ad hoc quality of service multicast routing,Computer Communications 29 (1) (2005) 136–148.

[23] W.L.C. Chiang, H. Wu and M. Gerla, Routing in clustered multihop,mobile wireless networks, in: Proceedings of IEEE SICON, April1997, pp. 197–211.

[24] T. Camp, Y. Liu, An adaptive mesh-based protocol for geocastrouting, Journal of Parallel Distributed Computing 63 (2) (2003)196–213.

[25] N. Carlsson, D. Eager, Non-euclidian geographic routing in wirelessnetworks, Ad Hoc Networks 5 (7) (2007) 1173–1193.

[26] J.-H. Chang, L. Tassiulas, Energy conserving routing in wireless ad-hoc networks, in: Proceedings of IEEE INFOCOM, March 2000, pp.22–31.

[27] T.-W. Chen, M. Gerla, Global state routing: a new routing schemefor ad-hoc wireless networks, in: Proceedings of IEEE ICC, vol. 1,June 1998, pp. 171–175.

[28] Z. Cheng, W. Heinzelman, Discovering long lifetime routes inmobile ad hoc networks, Ad Hoc Networks 6 (5) (2008) 661–674.

[29] C.-C. Chiang, M. Gerla, On-demand multicast in mobile wirelessnetworks, in: Proceedings of IEEE ICNP, October 1998, pp. 262–270.

[30] C.-H. Chou, K.-F. Ssu, H. Jiau, Dynamic route maintenance forgeographic forwarding in mobile ad hoc networks, ComputerNetworks 52 (2) (2008) 418–431.

[31] T. Clausen, P. Jacquet, A. Laouiti, P. Muhlethaler, A. Qayyum, L.Viennot, Optimized link state routing protocol for ad hoc networks,in: Proceedings of IEEE INMIC, December 2001, pp. 62–68.

[32] M. Colagrosso, N. Enochs, T. Camp, Improvements to location-aidedrouting through directional count restrictions, in: Proceedings ofInternational Conference on Wireless Networks (ICWN), June 2004,pp. 924–929.

[33] M. Conti, E. Gregori, G. Maselli, Reliable and efficient forwarding inad hoc networks, Ad Hoc Networks 4 (3) (2006) 398–415.

[34] M.S. Corson, A. Ephremides, A distributed routing algorithm formobile wireless networks, ACM/Baltzer Wireless Networks Journal1 (1) (1995) 61–81.

[35] S.K. Das, B.S. Manoj, C.S.R. Murthy, A dynamic core based multicastrouting protocol for ad hoc wireless networks, in: Proceedings ofACM MobiHoc, June 2002, pp. 24–35.

[36] S.K. Das, A. Mukherjee, S. Bandyopadhyay, D. Saha, K. Paul, Anadaptive framework for QoS routing through multiple paths in adhoc wireless networks, Journal of Parallel Distributed Computing63 (2) (2003) 141–153.

[37] R. Dube, C.D. Rais, K.Y. Wang, S.K. Tripathi, Signal stability-basedadaptive routing (SSA) for ad hoc mobile networks, IEEE PersonalCommunications Magazine (1997) 36–45.

[38] J. Eisbrener, G. Murphy, D. Eade, C. Pinnow, K. Begum, S. Park, S.-M.Yoo, J.-H. Youn, Recycled path routing in mobile ad hoc networks,Computer Communications 29 (9) (2006) 1552–1560.

[39] J. Eriksson, M. Faloutsos, S.V. Krishnamurthy, Scalable ad hocrouting: The case for dynamic addressing, in: Proceedings of IEEEINFOCOM, vol. 2, March 2004, pp. 1108–1119.

[40] J. Garcia-Luna-Aceves, M. Mosko, C. Perkins, A new approach to on-demand loop-free routing in networks using sequence numbers,Computer Networks 50 (10) (2006) 1599–1615.

[41] J.J. Garcia-Luna-Aceves, Loop-free routing using diffusingcomputations, IEEE/ACM Transactions on Networking 1 (1) (1993)130–141.

[42] J.J. Garcia-Luna-Aceves, E.L. Madruga, A multicast routing protocolfor ad-hoc networks, in: Proceedings of IEEE INFOCOM, March1999, pp. 784–792.

[43] J.J. Garcia-Luna-Aceves, M. Spohn, Source-tree routing in wirelessnetworks, in: Proceedings of IEEE ICNP, October–November 1999,pp. 273–282.

[44] M. Ge, S. Krishnamurthy, M. Faloutsos, Application versus networklayer multicasting in ad hoc networks: the ALMA routing protocol,Ad Hoc Networks 4 (2) (2006) 283–300.

[45] J. Ghosh, S. Philip, C. Qiao, Sociological orbit aware locationapproximation and routing (solar) in manet, Ad Hoc Networks 5(2) (2007) 189–209.

[46] V. Giruka, M. Singhal, A self-healing on-demand geographic pathrouting protocol for mobile ad-hoc networks, Ad Hoc Networks 5(7) (2007) 1113–1128.

[47] T. Goff, N.B. Abu-Ghazaleh, D.S. Phatak, R. Kahvecioglu, Preemptiverouting in ad hoc networks, Journal of Parallel DistributedComputing 63 (2) (2003) 123–140.

[48] M. Gong, S. Midkiff, S. Mao, On-demand routing and channelassignment in multi-channel mobile ad hoc networks, Ad HocNetworks 7 (1) (2009) 63–78.

[49] C. Gui, P. Mohapatra, Hierarchical multicast techniques andscalability in mobile ad hoc networks, Ad Hoc Networks 4 (5)(2006) 586–606.

[50] M. Gunes, U. Sorges, I. Bouazizi, Ara-the ant-colony based routingalgorithms for manets, in: Proceedings of IEEE ICPP Workshop onAd Hoc Networks (IWAHN), August 2002, pp. 79–85.

[51] Z.J. Haas, A new routing protocol for the reconfigurable wirelessnetworks, in: Proceedings of IEEE ICUPC, October 1997, pp. 562–566.

[52] G. Ivascu, S. Pierre, A. Quintero, QoS routing with traffic distributionin mobile ad hoc networks, Computer Communications 32 (2)(2009) 305–316.

[53] A. Iwata, C. Chiang, G. Pei, M. Gerla, T. Chen, Scalable routingstrategies for ad hoc wireless networks, IEEE Journal on SelectedAreas in Communications 17 (8) (1999) 1369–1379.

[54] J. Jetcheva, D. Johnson, Adaptive demand-driven multicast routingin multi-hop wireless ad hoc networks, in: Proceedings of ACMMobiHoc, October 2001, pp. 33–44.

[55] L. Ji, M. Corson, Differential destination multicast-a manetmulticast routing protocol for small groups, in: Proceedings ofIEEE INFOCOM, vol. 2, April 2001, pp. 1192–1201.

[56] M. Joa-Ng, I.-T. Lu, A peer-to-peer zone-based two-level link staterouting for mobile ad hoc network, IEEE Journal on Selected Areasin Communications 17 (8) (1999) 1415–1425.


[57] D.B. Johnson, D.A. Maltz, J. Broch, DSR: the dynamic source routingprotocol for multi-hop wireless ad hoc networks, in: C.E. Perkins(Ed.), Ad Hoc Networking, Addison-Wesley, 2001, pp. 139–172(Chapter 5).

[58] N. Karayiannis, S. Nadella, Power-conserving routing of ad hocmobile wireless networks based on entropy-constrainedalgorithms, Ad Hoc Networks 4 (1) (2006) 24–35.

[59] B. Karp, H. Kung, GPSR: Greedy Perimeter Stateless Routing forWireless Networks, in: Proceedings of ACM MobiCom, August 2000,pp. 243–254.

[60] V. Kawadia, P. Kumar, Power control and clustering in ad hocnetworks, in: Proceedings of IEEE INFOCOM, vol. 1, March–April2003, pp. 459–469.

[61] Y. Ko, N. Vaidya, Geocasting in mobile ad hoc networks: Location-based multicast algorithms, Technical report, TR-98-018, Texas AMUniversity, September 1998.

[62] Y. Ko, N. Vaidya, Location-Aided Routing (LAR) in mobile ad hocnetworks, in: Proceedings of ACM MobiCom, October 1998, pp. 66–75.

[63] Y.-B. Ko, N.H. Vaidya, GeoTORA: A protocol for geocasting in mobilead hoc networks, in: Proceedings of IEEE ICNP, November 2000, pp.240–249.

[64] S. Kwon, N. Shroff, Geographic routing in the presence of locationerrors, Computer Networks 50 (15) (2006) 2902–2917.

[65] W. Lai, S.-Y. Hsiao, Y.-C. Lin, Adaptive backup routing forad-hoc networks, Computer Communications 30 (2) (2007) 453–464.

[66] L. Law, S. Krishnamurthy, M. Faloutsos, A novel adaptive protocolfor lightweight efficient multicasting in ad hoc networks, ComputerNetworks 51 (3) (2007) 823–834.

[67] L. Layuan, L. Chunlin, A QoS multicast routing protocol forclustering mobile ad hoc networks, Computer Communications30 (7) (2007) 1641–1654.

[68] S. Lee, M. Gerla, Split multipath routing with maximally disjointpaths in ad hoc networks, in: Proceedings of the IEEE ICC, June2001, pp. 3201–3205.

[69] D. Li, Q. Liu, X. Hu, X. Jia, Energy efficient multicast routing in ad hocwireless networks, Computer Communications 30 (18) (2007)3746–3756.

[70] J. Li, P. Mohapatra, LAKER: Location aided knowledge extractionrouting for mobile ad hoc networks, in: Proceedings of IEEE WCNC,March 2003, pp. 1180–1184.

[71] W. Liang, Y. Liu, On-line disjoint path routing for network capacitymaximization in energy-constrained ad hoc networks, Ad HocNetworks 5 (2) (2007) 272–285.

[72] W.-H. Liao, J.-P. Sheu, Y.-C. Tseng, GRID: a fully location-awarerouting protocol for mobile ad hoc networks, TelecommunicationSystems 18 (1-3) (2001) 37–60.

[73] W.-H. Liao, Y. Tseng, K. Lo, J. Sheu, GeoGRID: a geocasting protocolfor mobile ad hoc networks based on grid, Journal of InternetTechnology 1 (2) (2000) 23–32.

[74] C. Liu, M. Yarvis, W. Conner, X. Guo, Guaranteed on-demanddiscovery of node-disjoint paths in ad hoc networks, ComputerCommunications 30 (14-15) (2007) 2917–2930.

[75] J.-S. Liu, C.-H.R. Lin, RBR: refinement-based route maintenanceprotocol in wireless ad hoc networks, Computer Communications28 (8) (2005) 908–920.

[76] Y. Liu, X. Hu, M. Lee, T. Saadawi, A region-based routing protocol forwireless mobile ad hoc networks, IEEE Network 18 (4) (2004) 12–17.

[77] H. Mahmood, C. Comaniciu, Interference aware cooperative routingfor wireless ad hoc networks, Ad Hoc Networks 7 (1) (2009) 248–263.

[78] C. Maihöfer, A survey on geocast routing protocols, IEEECommunications Surveys and Tutorials 6 (2) (2004) 32–42.

[79] M.K. Marina, S. Das, On-demand multi path distance vector routingin ad hoc networks, in: Proceedings of IEEE ICNP, November 2001,pp. 14–23.

[80] R. Mavropodi, P. Kotzanikolaou, C. Douligeris, SecMR – a securemultipath routing protocol for ad hoc networks, Ad Hoc Networks 5(1) (2007) 87–99.

[81] A. Mohanoor, S. Radhakrishnan, V. Sarangan, Online energy awarerouting in wireless networks, Ad Hoc Networks 7 (5) (2009) 918–931.

[82] A. Munaretto, M. Fonseca, Routing and quality of service support formobile ad hoc networks, Computer Networks 51 (11) (2007) 3142–3156.

[83] S. Murthy, J.J. Garcia-Luna-Aceves, An efficient routing protocol forwireless networks, MONET 1 (2) (1996) 183–197.

[84] J. Na, C.-K. Kim, Glr: a novel geographic routing scheme for largewireless ad hoc networks, Computer Networks 50 (17) (2006)3434–3448.

[85] M. Naserian, K. Tepe, Game theoretic approach in routing protocolfor wireless ad hoc networks, Ad Hoc Networks 7 (3) (2009) 569–578.

[86] N. Nikaein, C. Bonnet, HARP: hybrid ad hoc routing protocol, in:Proceedings of International Symposium on Telecommunications(IST), 2001, pp. 56–67.

[87] N. Nikaein, H. Labiod, C. Bonnet, DDR: distributed dynamic routingalgorithm for mobile ad hoc networks, in: Proceedings of ACMMobiHoc, August 2000, pp. 19–27.

[88] O. Ozkasap, Z. Genc, E. Atsan, Epidemic-based reliable and adaptivemulticast for mobile ad hoc networks, Computer Networks 53 (9)(2009) 1409–1430.

[89] V.D. Park, M.S. Corson, A highly adaptive distributed routingalgorithm for mobile wireless networks, in: Proceedings ofINFOCOM, April 1997, pp. 1405–1413.

[90] G. Pei, M. Gerla, T.-W. Chen, Fisheye state routing in mobilead hoc networks, in: Proceedings of IEEE ICDCS Workshop onWireless Networks and Mobile Computing, April 2000, pp. D71–D78.

[91] G. Pei, M. Gerla, X. Hong, LANMAR: Landmark routing for large scalewireless ad hoc networks with group mobility, in: Proceedings ofACM MobiHoc, August 2000, pp. 11–18.

[92] C. Perkins (Ed.), Ad hoc Networking, Addison Wesley, 2001.[93] C. Perkins, P. Bhagwat, Highly dynamic destination-sequenced

distance-vector routing (DSDV) for mobile computers, in: ACMSIGCOMM, August–September 1994, pp. 234–244.

[94] C. Perkins, E. Royer, Ad hoc On-demand Distance Vector Routing, in:Proceedings of the Second IEEE Workshop on Mobile ComputingSystems and Applications, 1999, pp. 99–100.

[95] D. Pompili, M. Vittucci, PPMA, a probabilistic predictive multicastalgorithm for ad hoc networks, Ad Hoc Networks 4 (6) (2006) 724–748.

[96] S. Radhakrishnan, N. Rao, G. Racherla, C. Sekharan, S. Batsell, DST –a routing protocol for ad hoc networks using distributedspanning trees, in: Proceedings of IEEE WCNC, September 1999,pp. 100–104.

[97] S. Rajagopalan, C.-C. Shen, ANSI: A swarm intelligence-basedunicast routing protocol for hybrid ad hoc networks, Journal ofSystems Architecture 52 (8-9) (2006) 485–504.

[98] J. Raju, J. Garcia-Luna-Aceves, A new approach to on-demand loop-free multipath routing, in: Proceedings of the Eight IEEEInternational Conference on Computer Communications andNetworks (IC3N), April 1999, pp. 522–527.

[99] S. Ramasubramanian, H. Krishnamoorthy, M. Krunz, Disjointmultipath routing using colored trees, Computer Networks 51 (8)(2007) 2163–2180.

[100] L. Randaccio, L. Atzori, Group multicast routing problem: a geneticalgorithms based approach, Computer Networks 51 (14) (2007)3989–4004.

[101] H. Rangarajan, J. Garcia-Luna-Aceves, Efficient use of route requestsfor loop-free on-demand routing in ad hoc networks, ComputerNetworks 51 (6) (2007) 1515–1529.

[102] L.R. Reddy, S. Raghavan, SMORT: scalable multipath on-demandrouting for mobile ad hoc networks, Ad Hoc Networks 5 (2) (2007)162–188.

[103] T.B. Reddy, S. Sriram, B. Manoj, C.S.R. Murthy, MuSeQoR: multi-path failure-tolerant security-aware QoS routing in ad hoc wirelessnetworks, Computer Networks 50 (9) (2006) 1349–1383.

[104] L. Rosati, M. Berioli, G. Reali, On ant routing algorithms in ad hocnetworks with critical connectivity, Ad Hoc Networks 6 (6) (2008)827–859.

[105] S. Roy, J.J. Garcia-Luna-Aceves, Using minimal source trees for on-demand routing in ad hoc networks, in: Proceedings of IEEEINFOCOM, April 2001, pp. 1172–1181.

[106] P. Samar, M. Pearlman, S. Haas, Independent zone routing: anadaptive hybrid routing framework for ad hoc wireless networks,IEEE/ACM Transactions on Networking 12 (4) (2004) 595–608.

[107] K. Scott, N. Bombos, Routing and channel assignment for low powertransmission in PCS, in: Proceedings of IEEE ICUPC, September–October 1996, pp. 498–502.

[108] C. Sengul, R. Kravets, Bypass routing: an on-demand local recoveryprotocol for ad hoc networks, Ad Hoc Networks 4 (3) (2006) 380–397.

[109] S. Singh, M. Woo, C. Raghavendra, Power-aware routing in mobilead hoc networks, in: Proceedings of ACM MobiCom, October 1998,pp. 181–190.


[110] R. Sivakumar, P. Sinha, V. Bharghavan, CEDAR: a core-extractiondistributed ad hoc routing algorithm, IEEE Journal on SelectedAreas in Communications 17 (8) (1999) 1454–1465.

[111] J.-H. Song, V. Wong, V. Leung, Secure position-based routingprotocol for mobile ad hoc networks, Ad Hoc Networks 5 (1)(2007) 76–86.

[112] O. Souihli, M. Frikha, M. Hamouda, Load-balancing in MANET shortest-path routing protocols, Ad Hoc Networks 7 (2) (2009) 431–442.

[113] W. Sum, M. Gerla, IPv6 flow handoff in ad-hoc wireless networksusing mobility prediction, in: Proceedings of IEEE GLOBECOM,December 1999, pp. 271–275.

[114] B. Sun, L. Li, QoS-aware multicast routing protocol for ad hocnetworks, Journal of Systems Engineering and Electronics 17 (2)(2006) 417–422.

[115] M. Tarique, K. Tepe, Minimum energy hierarchical dynamic sourcerouting for mobile ad hoc networks, Ad Hoc Networks 7 (6) (2009)1125–1135.

[116] C.-K. Toh, Associativity-based routing for ad-hoc mobile networks,Wireless Personal Communications Journal 4 (2) (1997) 103–139.Special Issue on Mobile Networking and Computing Systems.

[117] R. Vaishampayan, J. Garcia-Luna-Aceves, Efficient and robustmulticast routing in mobile ad hoc networks, in: Proceedings ofIEEE International Conference on Mobile Ad-hoc and SensorSystems, October 2004, pp. 304–313.

[118] A. Valera, W.K.G. Seah, S.V. Rao, Cooperative packet caching andshortest multipath routing in mobile ad hoc networks, in:Proceedings of IEEE INFOCOM, vol. 1, April 2003, pp. 260–269.

[119] L. Villasenor-Gonzalez, Y. Ge, L. Lamont, HOLSR: a hierarchicalproactive routing mechanism for mobile ad hoc networks, IEEECommunications Magazine 43 (7) (2005) 118–125.

[120] G. Wang, Y. Ji, D.C. Marinescu, D. Turgut, A routing protocol forpower constrained networks with asymmetric links, in:Proceedings of the ACM Workshop on Performance Evaluation ofWireless Ad Hoc, Sensor, and Ubiquitous Networks (PE-WASUN),October 2004, pp. 69–76.

[121] G. Wang, D. Turgut, L. Bölöni, Y. Ji, D. Marinescu, Improving routingperformance through m-limited forwarding in power-constrainedwireless networks, Journal of Parallel and Distributed Computing(JPDC) 68 (4) (2008) 501–514.

[122] J. Wang, E. Osagie, P. Thulasiraman, R. Thulasiram, Hopnet: a hybridant colony optimization routing algorithm for mobile ad hocnetwork, Ad Hoc Networks 7 (4) (2009) 690–705.

[123] N.-C. Wang, Y.-F. Huang, J.-C. Chen, A stable weight-based on-demand routing protocol for mobile ad hoc networks, InformationSciences 177 (24) (2007) 5522–5537.

[124] Y. Wang, V. Giruka, M. Singhal, Truthful multipath routing for adhoc networks with selfish nodes, Journal of Parallel and DistributedComputing 68 (6) (2008) 778–789.

[125] Y.-H. Wang, C.-F. Chao, Dynamic backup routes routing protocol formobile ad hoc networks, Information Sciences 176 (2) (2006) 161–185.

[126] B. Williams, T. Camp, Comparison of broadcasting techniques formobile ad hoc networks, in: Proceedings of ACM MobiHoc, June2002, pp. 194–205.

[127] S.-C. Woo, S. Singh, Scalable routing protocol for ad hoc networks,Wireless Networks 7 (5) (2001) 513–529.

[128] H. Wu, X. Jia, QoS multicast routing by using multiple paths/trees inwireless ad hoc networks, Ad Hoc Networks 5 (5) (2007) 600–612.

[129] X. Xiaochuan, W. Gang, W. Keping, W. Gang, J. Shilou, Linkreliability based hybrid routing for tactical mobile ad hocnetwork, Journal of Systems Engineering and Electronics 19 (2)(2008) 259–267.

[130] J. Xie, R.R. Talpade, A. McAuley, M. Liu, Amroute: ad hoc multicastrouting protocol, MONET 7 (6) (2002) 429–439.

[131] K. Xu, X. Hong, M. Gerla, Landmark routing in ad hoc networks withmobile backbones, Journal of Parallel and Distributed Computing63 (2) (2003) 110–122.

[132] Q. Xue, A. Ganz, Ad hoc QoS on-demand routing (AQOR) in mobilead hoc networks, Journal of Parallel Distributed Computing 63 (2)(2003) 154–165.

[133] C.-C. Yang, L.-P. Tseng, Fisheye zone routing protocol: a multi-levelzone routing protocol for mobile ad hoc networks, ComputerCommunications 30 (2) (2007) 261–268.

[134] Z. Yao, J. Jiang, P. Fan, Z. Cao, V. Li, A neighbor-table-basedmultipath routing in ad hoc networks, in: Proceedings of IEEEVTC 2003-Spring, vol. 3, April 2003, pp. 1739–1743.

[135] C. Yu, T.-K. Wu, R. Cheng, A low overhead dynamic route repairingmechanism for mobile ad hoc networks, ComputerCommunications 30 (5) (2007) 1152–1163.

[136] M. Yu, A. Malvankar, W. Su, S. Foo, A link availability-based QoS-aware routing protocol for mobile ad hoc sensor networks,Computer Communications 30 (18) (2007) 3823–3831.

[137] M. Yuksel, R. Pradhan, S. Kalyanaraman, An implementationframework for trajectory-based routing in ad hoc networks, AdHoc Networks 4 (1) (2006) 125–137.

[138] H. Zhou, S. Singh, Content based multicast (CBM) in ad hocnetworks, in: Proceedings of ACM MobiHoc, August 2000, pp. 51–60.

[139] Temporally ordered routing algorithm. Available from webpage <http://wiki.uni.lu/secan-lab/temporally-ordered+routing+algorithm.html>.

[140] A survey of defense technology: The software revolution – todissolve, to disappear, June 1995.

Azzedine Boukerche is a full professor andholds a Canada Research Chair position at theUniversity of Ottawa (Ottawa). He is a fellowof the Canadian Academy of Engineering andthe founding director of the PARADISEResearch Laboratory, School of InformationTechnology and Engineering (SITE), Ottawa.Prior to this, he held a faculty position at theUniversity of North Texas, and he was a seniorscientist at the Simulation Sciences Division,Metron Corp., San Diego. He was alsoemployed as a faculty member in the Schoolof Computer Science, McGill University, and

taught at the Polytechnic of Montreal. He spent a year at the JPL/NASA-California Institute of Technology, where he contributed to a project
centered about the specification and verification of the software used tocontrol interplanetary spacecraft operated by JPL/NASA Laboratory. Hiscurrent research interests include wireless ad hoc and sensor networks,wireless networks, mobile and pervasive computing, wireless multi-media, QoS service provisioning, performance evaluation and modeling oflarge-scale distributed systems, distributed computing, large-scale dis-tributed interactive simulation, and parallel discrete-event simulation. Hehas published several research papers in these areas. He served as a guesteditor for the Journal of Parallel and Distributed Computing (special issuefor routing for mobile ad hoc, special issue for wireless communicationand mobile computing, and special issue for mobile ad hoc networkingand computing), ACM/Kluwer Wireless Networks, ACM/Kluwer MobileNetworks Applications, and Journal of Wireless Communication andMobile Computing. He serves/served as an Associate Editor of IEEETransactions on Parallel and Distributed systems, IEEE Transactions onVehicular Technology, Elsevier Ad Hoc Networks, Wiley InternationalJournal of Wireless Communication and Mobile Computing, Wiley’sSecurity and Communication Network Journal, Elsevier Pervasive andMobile Computing Journal, IEEE Wireless Communication Magazine,Elsevier’s Journal of Parallel and Distributed Computing, and SCS Trans-actions on Simulation. He was the recipient of the Best Research PaperAward at IEEE/ACM PADS 1997, ACM MobiWac 2006, ICC 2008, ICC 2009and IWCMC 2009, and the recipient of the Third National Award forTelecommunication Software in 1999 for his work on a distributedsecurity systems on mobile phone operations. He has been nominated forthe Best Paper Award at the IEEE/ACM PADS 1999 and ACM MSWiM 2001.He is a recipient of an Ontario Early Research Excellence Award (previ-ously known as Premier of Ontario Research Excellence Award), OntarioDistinguished Researcher Award, and Glinski Research Excellence Award.He is a cofounder of the QShine International Conference on Quality ofService for Wireless/Wired Heterogeneous Networks (QShine 2004). Heserved as the general chair for the Eighth ACM/IEEE Symposium onModeling, Analysis and Simulation of Wireless and Mobile Systems, andthe Ninth ACM/IEEE Symposium on Distributed Simulation and Real-TimeApplication (DS-RT), the program chair for the ACM Workshop on QoSand Security for Wireless and Mobile Networks, ACM/IFIPS Europar 2002Conference, IEEE/SCS Annual Simulation Symposium (ANNS 2002), ACMWWW 2002, IEEE MWCN 2002, IEEE/ACM MASCOTS 2002, IEEE WirelessLocal Networks WLN 03-04; IEEE WMAN 04-05, and ACM MSWiM 98-99,and a TPC member of numerous IEEE and ACM sponsored conferences. Heserved as the vice general chair for the Third IEEE Distributed Computingfor Sensor Networks (DCOSS) Conference in 2007, as the program cochairfor GLOBECOM 2007-2008 Symposium on Wireless Ad Hoc and Sensor
http://wiki.uni.lu/secan-lab/temporally-ordered+routing+algorithm.html

http://wiki.uni.lu/secan-lab/temporally-ordered+routing+algorithm.html


Networks, ICC 11, and for the 14th IEEE ISCC 2009 Symposium on Com-puter and Commmunication Symposium, and as the finance chair forACM Multimedia 2008. He also serves as a Steering Committee chair forthe ACM Modeling, Analysis and Simulation for Wireless and MobileSystems Conference, the ACM Symposium on Performance Evaluation ofWireless Ad Hoc, Sensor, and Ubiquitous Networks, and IEEE/ACM DS-RT.He serves as the Vice Chair of the IEEE Technical Committee on Ad-Hocand Sensor Networks.

Begumhan Turgut is a Ph.D. candidate at theDepartment of Computer Science at RutgersUniversity. She graduated Cum Laude fromthe Department of Computer Science and atUniversity of Texas at Arlington. Her researchinterests include localization in wireless net-working; clustering, MAC, routing, networktopology and connectivity in ad hoc networks;modeling and enhancing the stealth level insensor networks, self-organization and mobilesinks in sensor networks. She has numerouspublications in these areas including a bestpaper award in IEEE GLOBECOM 2009. She is a

student member of IEEE and the Upsilon Pi Epsilon honorary society.

Nevin Aydin is an Assistant Professor at theDepartment of Industrial Engineering ofIstanbul Arel University. She received her BSin Mathematics at Istanbul University, MS inMathematics at Ankara University, and Ph.D.degrees in Industrial Engineering at IstanbulTechnical University. Her research interestsinclude operations research, agent and holonicmanufacturing, supply chain management,wireless networking, mobile computing, andobject-oriented databases. Dr. Aydin haspublished several technical papers in theseareas.

Mohammad Z. Ahmad is a Ph.D. candidate atthe Department of Electrical Engineering andComputer Science of University of CentralFlorida. He has received his BS degree fromDepartment of Information Science at GoldenValley Institute of Technology and MS degreefrom School of Electrical Engineering andComputer Science at University of CentralFlorida. His current research interests includeInternet measurements studying Internettopology evolution due to exchange pointsand the affects of the infrastructure growth oninter-domain routing. He has previously

worked on ad hoc routing protocols, congestion avoidance and controlalgorithms in sensor networks.

Ladislau Bölöni is an Associate Professor atthe Department of Electrical Engineering andComputer Science of University of CentralFlorida. He received a Ph.D. degree from theComputer Science Department of PurdueUniversity in May 2000. He received an MSdegree from the Computer Sciences depart-ment of Purdue University in 1999 and BSdegree in Computer Engineering with Honorsfrom the Technical University of Cluj-Napoca,Romania in 1993. He received a fellowshipfrom the Computer and Automation ResearchInstitute of the Hungarian Academy of Sci-

ences for the 1994–1995 academic year. He is a senior member of IEEE,member of the ACM, AAAI and the Upsilon Pi Epsilon honorary society.
His research interests include autonomous agents, grid computing andwireless networking.
Damla Turgut is an Associate Professor at theDepartment of Electrical Engineering andComputer Science of University of CentralFlorida. She received her BS, MS, and Ph.D.degrees from the Computer Science andEngineering Department of University ofTexas at Arlington. Her research interestsinclude wireless networking and mobilecomputing, wireless communication andcoordination in embodied agents, and urbansensing. Dr. Turgut has published over fivedozens of refereed technical papers and bookchapters. Dr. Turgut serves as a member of the

editorial board and of the technical program committee of ACM and IEEEjournals and international conferences. She is a member of IEEE, member
of the ACM, and the Upsilon Pi Epsilon honorary society.

Documents

Routing protocols in ad hoc networks: A surveycs752/papers/routing-001.pdfSurvey Paper Routing protocols in ad hoc networks: A survey Azzedine Boukerchea,1, Begumhan Turgutb,2, Nevin