Multicast Delivery Using Opportunistic Routing in Wireless Mesh … · 2016. 10. 19. · Multi-hop wireless networks (MWNs) [1,2] have be- ... some attempts to achieve a high performance

Multicast Delivery Using Opportunistic Routing in

Wireless Mesh Networks

Amir Darehshoorzadeh and Llorenc Cerda-Alabern[amirllorenc]acupcedu

Univ Politecnica de Catalunya Computer Architecture Dep Barcelona Spain

June 2013

Abstract

Opportunistic Routing (OR) has been proposed toimprove the efficiency of unicast protocols in wirelessnetworks In OR in contrast to traditional routinginstead of preselecting a single specific node to be thenext-hop forwarder an ordered set of nodes (referredto as candidates) is selected as the next-hop potentialforwarders In this paper we investigate how OR canbe used to improve multicast delivery We proposea new multicast routing protocol based on oppor-tunistic routing for wireless mesh networks namedMulticast Opportunistic Routing Protocol (MORP)MORP opportunistically employs a set of forwardersto send a packet toward all destinations Each for-warder is responsible for sending the packet to a sub-set of destinations Based on the candidates thatsuccessfully receive the packet in each transmissionMORP builds a tree on the fly We compare ourproposal with two well known ODMRP and ADMRmulticast protocols Our results demonstrate thatMORP outperforms ODMRP and ADMR reducingthe number of data transmissions and increasing thedelivery ratio

1 Introduction

Multi-hop wireless networks (MWNs) [1 2] have be-come a very active research field during the lastyears Routing in MWNs is more challenging thanin wired networks because of two fundamental differ-ences The first difference is the heterogeneous char-acteristics of wireless links As a consequence therecan be significant differences in packet delivery prob-abilities across the links of a MWN network Thesecond difference is the broadcast nature of wirelesstransmissions [3] Unlike wired networks where linksare typically point to point when a node transmitsa packet in a wireless network the neighbors of theintended destination node can overhear it

Multicasting in wireless networks has been an ac-tive area of research for quite a long time and a

number of multicast routing protocols have been pro-posed On the other hand Opportunistic Routing(OR) has been investigated in recent years as a wayto increase the performance of wireless networks byexploiting its broadcast nature In this paper we in-vestigate how OR can be used to improve multicastdelivery We do so by proposing MORP a new Multi-cast Opportunistic Routing Protocol and comparingit with two well known multicast protocols proposedin the literature

Generally multicast routing protocols can beclassified into three types based on the multicasttopology tree-based mesh-based and the hybrid-based [4 5] The tree-based multicast protocols(like [6 7 8 9]) establish a single path betweenany two nodes in the multicast group whereas formesh-based multicast protocols (like [10 11]) pack-ets are distributed along mesh structures that are aset of interconnected nodes and multiple paths mayexist between a source-destination pair The meshbased protocols outperform tree-based ones in termsof robustness but in an other hand mesh based pro-tocols suffer from a considerable amount of duplicatepackets Hybrid-based multicast routing protocolscombine the advantages of both tree and mesh-basedapproaches [12 13] However these three types ofmulticast protocols do not fully take advantage ofthe spatial characteristic of wireless communicationsWhen a packet is transmitted it is possible that somenodes in the neighbor nodes receive the packet whilethe designated next-hop does not While there aresome attempts to achieve a high performance in wire-less multicast routing the research field is still open

Opportunistic Routing (OR) also referred to asdiversity forwarding [14] cooperative forwarding [15]or any-path routing [16 17] has recently been pro-posed as a way to increase the performance of unicastin multi-hop wireless networks by taking advantage ofits broadcast nature It improves the throughput andthe transmission reliability in the face of unreliablewireless links [18] In OR in contrast to traditionalrouting instead of preselecting a single specific node

to be the next-hop forwarder an ordered set of nodes(referred to as candidates) is selected as the next-hoppotential forwarders More specifically when the cur-rent node transmits a packet all the candidates thatreceive the packet successfully will coordinate witheach other to determine which one would actuallyforward the packet according to some criteria whilethe other nodes will simply discard the packet

By using OR if a certain wireless forwarder failsor moves out of the radio range during the transmis-sion other possible paths may be used As a resultOR can better cope with lossy unreliable and timevarying link quality It can significantly reduce thenumber of transmissions necessary to deliver a packetto the destination and greatly increases the transmis-sion reliability and the network throughput by tak-ing advantage of the broadcast nature of the wirelessmedium

Previous researches have shown that OR can sig-nificantly reduce the expected number of transmis-sions to deliver a packet to a particular destinationIt is therefore tempting to adopt OR to improve theefficiency of wireless multicast The main challengein adapting of OR with multicast is how to sharethe opportunistic forwarders paths between multipledestinations In OR algorithms for unicast protocolssince a packet is addressed to only one destinationupon transmitting a packet only one of the candi-dates receiving it would actually forward the packetOn the other hand since there are more than one des-tination in the multicast protocols using OR mightcause that more than one candidate has to forwardthe packet to reach all the destinations Anotherchallenge of using OR in multicast in contrast tounicast is that the selected candidates might haveto forward the packets toward more than one desti-nation

This paper presents a new multicast routing pro-tocol that we call Multicast Opportunistic RoutingProtocol MORP Unlike traditional multicast proto-cols there is no designated next-hop forwarder foreach destination in our protocol thus the deliveryratio is maximized by taking advantage of spacialdiversity MORP uses three-way-handshaking ap-proach to transmit the data packet The basic ideaof MORP is as follow when a source node wants totransmit a data packet it creates its candidates setand include it into the packet The candidates whichsuccessfully receive the packet send an acknowledg-ment Then the sender selects some candidates andtowards which destinations they have to forward thepacket This information is sent to the candidateswhich repeat the algorithm until reaching all desti-nations of the multicast group Compared with thetraditional multicast protocols our protocol does notbuild a complete tree or mesh before the transmis-

sions starts Instead MORP builds a tree on the flydepending on the candidates that successfully receivethe packet in each transmission

We compare MORP with two well known proto-cols ODMRP multicast mesh protocol [19 10] andADMR multicast tree protocol [8 20] An additionalreason to choose these protocols for comparison isthat ODMRP is implemented in the simulation toolused to obtain numerical results (GloMoSim [21])The source code of ADMR for GloMoSim was kindlyprovided by the authors in [20]

In summary the main contributions of this paperare

bull We investigate the advantages of using OR tosupport multicast by proposing MORP

bull In contrast to the most of previous workswhich used the two-ray ground or some simpleloss propagation models we use the shadow-ing propagation model for the packet loss of allalgorithms under study

bull Our main conclusion is that OR can be an ef-fective mechanism to achieve reliable multicastdelivery in wireless mesh networks

The rest of this paper is organized as followsWe briefly review the related work on multicast andOR in Section 2 MORP description is presentedin Sections 3 and 4 Section 5 and 6 briefly de-scribe ODMRP and ADMR respectively Section 7explains the evaluation methodology MORPrsquos per-formance is evaluated in Section 8 and concludingremarks are given in Section 9

2 Related work

21 Multicast Routing

Multicast routing protocols come into play when ahost needs to send the same message or data streamto multiple destinations Due to the unique char-acteristics of the wireless networks such as limitedresources and unreliable channels traditional multi-cast protocols in the wired networks do not performwell in wireless and new protocols have been pro-posed One of the most popular methods to classifymulticast routing protocols is based on how distri-bution paths among group members are constructedAccording to this method existing multicast routingapproaches can be classified into tree-based mesh-based and hybrid protocols [4 22 5]

In the tree-based protocols only a single short-est path must be established between source-receiverpair therefore the multicast tree is composed of aunique path from the multicast source to each of themulticast receivers

Tree-based proposals are also divided into twosub-categories source-based tree and shared-basedtree approaches A source-based tree maintains anindividual route towards all the multicast receiversfor each multicast group Some source-based mul-ticast protocols are Differential Destination Multi-cast (DDM) [23] Preferred Link Based Multicast(PLBM) [24] Adaptive Demand-driven MulticastRouting [8] and probabilistically reliable on-demand(PROD) [9]

Since the construction of a separate tree for eachsource is costly some tree-based multicast protocolsuse a shared-based (core-based) tree to distribute themulticast messages In shared-based tree a single treeis constructed to support the whole groups Since theshared-based multicast tree only permits the multi-cast traffic to be sent out from the root to the multi-cast receivers each multicast source must forward itsmulticast traffic to the root initially Multicast traf-fic of each source is then forwarded along the sharedtree Ad-hoc Multicast Routing utilizing Increas-ing ID numbers (AMRIS) [25] Multicast Ad-hocOn-demand Distance Vector routing (MAODV) [26]Multicast Zone Routing (MZRP) [27] and AdaptiveCore based Multicast routing (ACMP) [28] are somepopular shared-based tree multicast routing proto-cols

The main advantage of a tree as the underlyingforwarding structure is that the number of forwardingnodes tends to be reduced However they generallysuffer from fragile tree structure [22] Besides the pre-vious problem source-based tree proposals also sufferfrom large memory space requirements and wastefulusage of limited bandwidth because each source con-structs its own tree But it performs better thanshared-based tree proposals at heavy loads due to ef-ficient distribution of trees Although shared-basedtree proposals are more scalable they have the vul-nerability of the single core problem [29]

In a mesh-based multicast routing protocolmultiple routes may exist between any pair of sourceand destination which is intended to enrich theconnectivity among group members The majordifference between the tree-based and mesh-basedprotocols lies in the manner in which a multicastmessage is relayed In tree-based protocols eachintermediate node on the tree has a well-defined listof the next-hop nodes for a specific multicast sessionIt will send a copy of the received multicast messageto only the neighboring nodes on its next-hop listIn mesh-based protocols each node on the mesh willbroadcast the message upon its first reception of themessage Mesh-based multicast routing protocolsgenerally are robust due to the penalty of multiplepaths between different nodes But many of theseproposals suffer from excessive control overhead

which will affect on scalability and utilizationof limited bandwidth Examples of mesh basedmulticast routing protocols include On-DemandMulticast Routing (ODMRP) [19 10] and itsvariations (PatchODMRP [30] PoolODMRP [31]PDAODMRP [32] EnhancedODMRP [33] ResilientODMRP [34] and limited flooding ODMRP [35])Forwarding Group Multicast Core-AssistedMesh (CAMP) [36] Clustered Group Multi-cast (CGM) [37] Neighbor-Supporting Multicast(NSMP) [38] Dynamic Core based Multicast rout-ing (DCMP) [39] and link stability based multicastrouting in MANETs (LSMRM) [40]

Hybrid multicast routing protocols combine theadvantages of both tree-based and mesh-based mul-ticast approaches ie the robustness of the mesh-based multicast routing protocols and low over-head of tree-based protocols Therefore the hy-brid multicast routing protocols are able to ad-dress both efficiency and robustness issues Multi-cast Core-Extraction Distributed Ad Hoc Routing(MCEDAR) [41] Ad-hoc Multicast Routing (AM-Route) [7] and Efficient Hybrid Multicast Routing(EHMRP) [12] are some well-known hybrid multicastrouting protocols

22 Opportunistic Routing

The majority of previous studies in opportunisticrouting do not use it for multicast routing and mostof them are devoted to the selection of the candidatesthe way of acknowledging packet reception and howto prevent or at least reduce duplicate transmis-sions

Biswas and Morris proposed ExOR [42 18] oneof the firsts and most referenced OR protocols Theselection of candidates in ExOR is based on theExpected Transmission Count (ETX) [43] metricIn [44] Zhong et al proposed a new metric ndashexpectedany-path transmission (EAX)ndash that generalizes ETXto an OR framework They analyzed the efficacy ofOR by using this metric and did a comparison usinglink-level measurements at MIT Roofnet project [45]In [17 16] a distributed algorithm for computing min-imum cost opportunistic routes is presented The au-thors also alert about the risk of using too many relaycandidates In [46] the key problem of how to opti-mally select the forwarder list is addressed and anoptimal algorithm that minimizes the expected totalnumber of transmissions is developed In [47] dif-ferent OR candidate selection algorithms have beencompared

One of the important issues of opportunistic rout-ing is the coordination between candidates in or-der to prevent duplicate transmissions Different co-ordination schemes have been proposed which nor-

mally rely on establishing some priority order andexchanging state information between candidatesIn [14] coordination is achieved by means of a four-way-handshaking the candidates receiving the datapacket send back an acknowledgment to the senderBased on the acknowledgments the sender sends aforwarding order to the best candidate which is alsoacknowledged The coordination used in MORP fol-lows a similar approach In [42] an acknowledgmentbased scheme as the one used in traditional 80211is employed This scheme requires each candidatewhich has received the data packet to broadcast anACK in different time slots according to its prior-ity All the candidates listen to all ACKs before de-ciding whether to forward the data packet Otherapproaches combine OR with network coding pro-viding an elegant method for candidate coordina-tion [48 49 50 51] However using network cod-ing with OR may lead to a high number of potentialforwarders sending coded packets and thus result-ing in redundant transmissions There exists a trade-off between transmitting a sufficient number of codedpackets to guarantee that the destination has enoughcoded packets to reconstruct the native packets andavoiding to inject in the network unnecessary pack-ets [49]

There are some papers which propose analyti-cal models to study the performance of OR Bac-celli et al [52] used simulations to show that ORprotocols significantly improve the performance ofmultihop wireless networks compared to the short-est path routing algorithms and elaborated a math-ematical framework to prove some of the observa-tions obtained by the simulations In [53] an analyt-ical approach for studying OR in wireless multi-hopnetworks have been proposed They used lognormalshadowing and Rayleigh fading models for packet re-ception In their model they assume that the nodesare uniformly distributed over the plane The au-thors did not consider any specific candidate selectionalgorithm but simply compute the expected progressof the packet transmissions based on the probabil-ity of any node in the progressing region successfullyreceives the packet The authors of [54] proposedan utility-based model for opportunistic routing andclaimed that for the optimal solution it is necessaryto search all loop-free routes from the source to thedestination They proposed both optimal and heuris-tic solutions for selecting the candidates according totheir utility function In [55] an algebraic approachis applied to study the interaction of OR algorithmsand routing metrics Zubow et al in [56] claimedthat shadow fading losses for spatially close candi-dates are not independent from each other unlikecommonly assumed They presented measurementsobtained from an indoor testbed and concluded that

correlations can not be neglected if nodes are sepa-rated by less than 2 m In [57 58] a Markov modelto assess the improvement that may be achieved us-ing opportunistic routing was proposed At the sametime Li and Zhang published an analytical frame-work to estimate the transmission costs of packet for-warding in wireless networks [59] Both approachesare similar in their formulation although differ in theway the model is solved our model leads to a dis-crete phase-type distribution while in [59] transmis-sion costs are computed using spectral graph theoryIn [60] the issue of optimal candidates set selectionin the OR has been addressed They provide an an-alytical framework to model the problem of selectingthe optimal candidates set for both the constrained(limited number of candidates) and unconstrained(unlimited number of candidates) candidates set se-lection They proposed two algorithms for optimalcandidates set selection one for the constrained andone for the unconstrained case Finally in [61] someequations that yield the distances of the candidatesin OR such that the per transmission progress to-wards the destination is maximized have been de-rived There we have proposed a lower bound to theexpected number of transmissions needed to send apacket using OR

There are few works that have been made toadapt OR in multicast MORE [50] is a MAC inde-pendent protocol that uses both the idea of OR andnetwork coding It avoids duplicate transmissionsby randomly mixing packets before forwarding Thesender creates a linear combinations of packets andbroadcasts the resulting packet after adding a MOREheader containing the candidates set Each receivingnode discards the packet if it is not linearly indepen-dent from the other packets received before or if itsID does not appear in the candidate list Otherwiseit linearly combines the received coded packets andrebroadcasts the new packet In [62] the source firstcreates the shortest path tree to reach all destinationsbased on the ETX of each link Then the nodes notonly receive packets from their father in the tree butalso can overhear packets from its sibling nodes Ituses random linear network coding to improve mul-ticast efficiency and simplify node coordination Theauthors in [63] used a Steiner tree based on ETX andsent data packets through the links using OR Theirprotocol constrains the nodes involved in routing apacket to be near the default multicast tree The av-erage EAX of each candidate to reach a sub-group ofdestinations is used as the cost of reaching to multipledestinations The authors in [64] proposed a Multi-cast OR (MOR) algorithm It opportunistically em-ploys a set of forwarders to push a packet closer to allreceivers round-by-round They proposed a new met-ric ndashexpect transmission advancement (ETA)ndash which

is the expected number of OR transmissions achievedafter one transmission from a source node towardthe destination using the candidates set of sourceBased on packet receptions at the end of each rounda new forwarder set is constructed to maximize theexpect transmission advancement towards all desti-nations They developed an event-driven simulatorto measure the performance of their proposal Forthe propagation model they used a simple packet losswhich is only related to the geographic distance be-tween two nodes They believe that implementing ofMOR using packet-level simulators is not straightfor-ward The recent work from [65] proposes an overlaymulticast to adapt OR in wireless network Theyconstruct a minimum overlay Steiner tree and mapit into unicast OR relay path connecting the sourcewith all destinations They employed unicast OR oneach link of the tree Their protocol does not exploitopportunistic receptions cross different links in thetree

MORP differentiate from these proposals by thecandidate selection and the coordination mecha-nism between candidates MORP uses a three-way-handshaking where the sending node selects the can-didates and towards which destinations they have toforward the packet By doing this MORP aims toachieve a high delivery ratio with a low number ofdata packet transmissions

3 Multicast Opportunistic Rout-ing Protocol (MORP)

In this section we propose a new multicast routingprotocol that we call Multicast Opportunistic RoutingProtocol MORP In the following we first introducethe network model and notation used in the descrip-tion of MORP then we describe the protocol and itscomponents

31 Network Model

We consider a network of N static wireless nodesincluding 1 source node s and a destinations set Dwith k lt N destinations D = d1 d2 dk

Denote Cidjncand = c1 c2 middot middot middot cncand as the candi-dates set of node i with at most ncand candidates toreach a destination dj using unicast OR (c1 the high-est priority candidate and cncand the least one) Inthis paper we have used ncand = 2 and 10 From this

point forward we shall call Cidj2 and Cidj10 the ldquosmallcandidates setrdquo and ldquolarge candidates setrdquo of node ito reach destination dj respectively Each node inthe network must compute these candidates sets us-ing one of the candidates selection algorithms thathave been proposed in the literature for unicast OR

like ExOR [42] All this information (small and largecandidates sets) is stored in a Candidate-Table

We define the Multicast Candidates Set of asource node s denoted by CsD as a set of candidatesthat allows reaching all destinations in D MORPcomputes this set as the union of the small candi-dates sets of all destinations in D

CsD =⋃djisinD

Csdj2 (1)

Equation (1) uses the small candidates sets insteadof the large candidates set in order to maintain thecardinality of CsD as small as possible The reasonis that the lower is the cardinality of CsD the lessnodes are involved in the packet delivery and thusthe lower is the signaling overhead

MORP also uses a sequence number to distin-guish each data packet created by the multicastsource We shall refer as ID the node identifier usedby MORP

32 Description of MORP

Each time the source s wants to transmit a packetthe following three-way-handshaking is carried outFirst the source inserts its Multicast Candidates Setin the data packet and transmits it The node alsostores the packet in a Message-Cache table to retrans-mit it later if it is necessary

Each node which successfully receives the datapacket checks if its ID is included in the packetrsquosheader If so it stores the data packet in its bufferand sends back an acknowledgment (ACK) other-wise it simply discards the packet Note that a nodemay receive a packet with the same sequence numberfrom different neighbor nodes In this case the nodedoes not consider the packet as duplicated and willprocess it

Upon receiving the ACKs from the candidatesthe source stores candidatesrsquo IDs in an Ack-TableAfter a period of time (TACK) the source checks if itreceived ACKs from enough candidates to reach alldestinations in D If there are not enough ACKs itretransmits the packet which is stored in its Message-Cache This is done up to a maximum number of re-transmissions (MAXReTx) Then according to thecandidates which successfully received the packetthe sender selects the candidates responsible to for-ward the packet and to which destinations Weshall refer these nodes and their destinations as theForwarding-Set and Bind-Destinations respectivelyand denote them as F and Di i isin F If none ofthe destinations are reached the sets Di i isin F aredisjoint and their union is D Otherwise their unionis D di di isin Destinations receiving the packetNote that we can consider the source node s as the

initial Forwarding-Set with Bind-Destinations equalto the multicast destinations set ie Ds = D Thealgorithm to compute the Forwarding-Set and Bind-Destinations is explained in the following section

Then the source s builds a control packet withthe Forwarding-Set and its Bind-Destinations andbroadcast it We shall refer to this packet as theForwarding-Packet Each node i that receives theForwarding-Packet and its ID is included in it mustforward the packet following the same rules as thesource except that its Bind-Destinations Di indi-cated in the Forwarding-Packet will be used insteadof D This process will be continued until the for-warding nodes directly deliver the packet to theirBind-Destinations

33 Forwarding Set

As explained in the previous section upon re-ceiving the candidatesrsquo ACKs the node must se-lect the Forwarding-Set and its Bind-DestinationsIn this section we describe the algorithm used byMORP to select these sets (Forwarding-Set and Bind-Destinations) We classify the candidates which sentback the acknowledgment and the destinations in thefour following sets

Definition 1 Non-Redundant-Destinations-Set(NRDestSet) is the set of destinations reachableby only one candidate Ie for each destinationdj isin NRDestSet there is only one candidate ci in theAck-Table which is able to reach dj Additionallywe shall refer to the set of such candidates as theNon-Redundant-Candidates-Set (NRCandSet)

Definition 2 Redundant-Destinations-Set (RDest-Set) is the set of destinations dk reachable by atleast two candidates eg ci and cj We shall re-fer to the set of such candidates as the Redundant-Candidates-Set (RCandSet) So if a candidate egci is removed from the RCandSet then there is atleast another candidate in RCandSet which is ableto reach any destination dk isin RDestSet

Note that the destination sets NRDestSet andRDestSet are disjoint However this might not betrue for the candidates sets NRCandSet and RCand-Set

To create the non-redundant and redundant setsof candidates and destinations node s uses its large

candidate set Csdj10 dj isin D defined in section 31Here the large candidates set is used instead of thesmall one in order to increase the chance of reachingall destinations with the minimum number of candi-dates For example it may happen that a candidateci does not appear in the small candidates set to reach

Algorithm 1 Computation of the Forwarding-Set and its Bind-Destinations by node s

Require Ds Bind-Destinations of node s

1 Find RCandSet RDestSet NRCandSet andNRDestSet

2 for all dj isin NRDestSet do

3 clarr ci isin NRCandSet and ci isin Csdj

4 Add c to the Forwarding-Set5 Add dj as the Bind-Destinations of c6 end for7 S larr RCandSet8 while TRUE do9 C larr CostFunc(S)

10 R larr arg minT =Sci

CostFunc(T )

11 C prime larr CostFunc(R)12 if (C prime minus C)C gt Threshold then13 break14 else15 S larr R16 end if17 end while18 for all dj isin RDestSet do19 clarr arg min

ciisinS amp ciisinCsdj10

ETX(ci dj)

20 Add c to the Forwarding-Set21 Add dj as the Bind-Destinations of c22 end for

destination dj ci isin Csdj2 but it is in the small can-

didate set of another destination dk ci isin Csdk2 If cireceives the packet and appears in the large candi-

date set of dj (ci isin Csdj10 ) then node s can also use

ci to reach destination dj

Algorithm 1 shows the pseudocode used by anode to compute the Forwarding-Set and its Bind-Destinations The general aim of algorithm 1 isto select few and good candidates to reach all des-tinations such that the expected number of trans-missions is minimized The algorithm works as fol-lows First node s creates the Non-Redundant-Setand Redundant-Set for both candidates and des-tinations (NRCandSet NRDestSet RCandSet andRDestSet) For each destination dj isin NRDestSetthe algorithm assigns the only possible candidateci isin NRCandSet (lines 2-6) Recall that NRDest-Set is the set of destinations dj reachable by only onecandidate Therefore for each destination in the Non-Redundant-Destinations-Set there is only one pos-sible choice from Non-Redundant-Candidates-Set toadd to the Forwarding-Set

Then the algorithm chooses the candidates fromRCandSet to reach the destinations in the RDest-Set For these destinations there are multiple choicesof candidates The optimum choice would mini-mize the expected number of transmissions to reachall destinations However even for a single desti-nation computing the expected number of trans-

missions is an equation with a high computationalcost (see eg [16]) For multiple destinations therehas not been proposed any exact equation to com-pute the expected number of transmissions and inany case the computational cost would be extremelyhigh Additionally in [61] was shown that the per-formance results are not very sensitive to the selec-tion of best candidates Therefore MORP buildsthe Forwarding-Set using the following simple costfunction as an estimation of the expected number oftransmissions to reach all destinations in RDestSetusing the candidates in the set S

CostFunc(S) =sum

djisinRDestSet

minciisinS

ETX(ci dj) (2)

where ETX(ci dj) is the expected transmissionscount [43] from candidate ci to the destination dj Note that equation 2 gives the expected number oftransmissions that would be obtained using unicastdelivery to each destination choosing the candidatein S that is closest to each destination in RDestSetTherefore this will be an upper-bound to the ex-pected number of transmissions obtained using OR

Lines 8-22 of algorithm 1 show the selection ofthe candidates for the destinations in RDestSet Ineach iteration of the while-loop the algorithm runsan exhaustive search over all possible subsets of theset S by removing one candidate The algorithmuses equation (2) to choose the subset having theminimum cost (line 10) If the difference betweenthe cost of new set (C prime) and the previous one (C)to reach the Redundant-Destinations-Set is not verylarge (eg Threshold=1) the algorithm will continuewith the new set to eliminate more candidates

The output of the while-loop of lines 8-17 is a re-duced set of candidates able to reach all destinationsin RDestSet In order to assign the Bind-Destinationsto these candidates it is used the minimum ETX(lines 18-22)

34 Candidate Coordination and Dataforwarding

After running algorithm 1 the source puts theForwarding-Set and its Bind-Destinations in theForwarding-Packet and broadcasts it Each node ireceiving the Forwarding-Packet having its ID in theForwarding-Set will forward the data packet storedin its buffer to its Bind-Destinations The candidateswith IDs not included in the Forwarding-Packet willsimply discard the packet This process will be con-tinued until the forwarding nodes directly deliver thedata packet to their Bind-Destinations

35 Data Structures

This section summarizes the data structures thatnodes running MORP are required to maintain

bull Candidate-Table It is created before the trans-mission starts and stores the candidates setsto reach each destination Each entry in theCandidate-Table is the destination ID the mul-ticast group address and the list of candidatesto reach the destination Recall that we haveused two different maximum number of candi-dates to form the small and large candidatessets Therefore in each node there are twoCandidate-Tables

bull Ack-Table It stores the ID of the candidatesfrom which ACK packets have been receivedEach entry of this table consists of the ID of thecandidate the sequence number of the packetwhich has been received and acknowledged andthe multicast group address of the packet

bull Bind-Destinations-Table When a node for-wards the data packet it stores its Bind-Destinations This information will be usedwhen the ACKs are received and the nodewants to decide to which destination each can-didate should forward the packet IndeedBind-Destinations-Table of node i stores itsBind-Destinations Di for each packet untilthe corresponding Forwarding-Packet is sent

bull Message-Cache The Message-Cache is main-tained by each node to prevent duplicated pack-ets It is also used to retransmit a packetwhich is not acknowledged by enough candi-dates When a node forwards a data packet itstores the source ID the multicast group ad-dress and the sequence number of the packetAn age timer is used to remove old entries

36 An Example of MORP

We finish the description of MORP by means ofa simple example Consider the network topologyshown in Figure 1 Assume that the delivery proba-bility is a function of the distance between the nodesshown in the figure The source node is s and thedestinations set is D = d1 d2 d3 d4 An unicastOR candidates selection algorithm (eg ExOR) isused by all nodes to compute the small and largecandidates sets Table 1 shows these sets for nodes In each row candidates are ordered in descendingpriority from left to right

When s wants to send a packet it puts its mul-ticast candidates set (see equation (1)) which isCsD = a b c d3 d4 f in the data packet and sends

Figure 1 Example of MORP

it The source sets the timer TACK and waits forthe ACKs from the candidates that have received thepacket successfully Assume that only the candidatesa b and d3 receive the data and send back an ACKto the source

When s receives ACK from a b and d3 it storestheir ID in its Ack-Table After TACK expires in nodes it runs the algorithm 1 to find the candidate whichshould forward the packet Since one destination d3has received the packet node s looks for the candi-dates to reach destinations d1 d2 and d4 First itfinds the non-redundant and redundant sets of can-didates and destinations As we mentioned in sec-tion 33 the algorithm 1 uses the large candidatesset to create the non-redundant and redundant setsThe only candidate which has received the packetand can reach the destination d4 is d3 (see large can-didate set in Table 1) Therefore the Non-Redundant-Destinations-Set (NRDestSet) is d4 and the can-didate d3 will be added to the Forwarding-Set withdestination d4 as its Bind-Destination

Table 1 Small and large candidates sets of s(a) Small and large

candidates sets

dest small large

d1 b c a b c e fd2 b a b a c e fd3 d3 f d3 c e fd4 d4 f d4 f d3 e

(b) ETX Table

node d1 d2

a 43 41b 48 38

The benefit of considering the large candidatesset instead of small candidates set becomes appar-ent for destination d4 If the algorithm would havejust considered the small candidates sets since noneof the candidates d4 and f received the packet thedestination d4 would be considered unreachable ands would retransmit the data packet

To reach destinations d1 and d2 there are twocandidates a and b which received the data packetTherefore the Redundant-Destinations-Set (RDest-Set) and Redundant-Candidates-Set (RCandSet) ared1 d2 and a b respectively

In the first iteration of the while-loop of algo-rithm 1 the cost of reaching RDestSet = d1 d2

using S = a b is estimated as C = ETX(a d1) +ETX(b d2) = 81 (see equation 2) Then it reducesthe number of candidates in the RCandSet and usesformula 2 again to find the set with the minimumcost (line 10 in algorithm 1) This is given by theset R = a with cost C prime = 84 Since the rela-tive difference between new cost and the previous one(C = 81) is small the algorithm takes the new setS = a Then the while-loop finishes

Thus the final Forwarding-Set is F = a d3with Bind-Destinations Da = d1 d2 and Dd3 =d4 Node s will put these sets in the Forwarding-Packet and send it Upon receiving the Forwarding-Packet a and d3 will know that they must forwardthe packet to d1 d2 and d4 respectively and willrepeat the forwarding process for these destinations

Note that as the data packets approach the des-tinations the size of the Bind-Destinations sets willbe decreased or remain unchanged Thus it is likeMORP builds a tree on the fly depending on thecandidates that successfully receive the data packetin each transmission

4 Implementation of MORP

As explained in section 31 MORP computes thecandidates sets using one of the candidates selectionalgorithms that have been proposed in the literaturefor unicast OR To do so the nodes need to be awareof the network topology and the delivery probabilityof the wireless links This information can be gath-ered in different ways One possible implementationcould be the method described in ExOR [18] wherenodes collects measurements and send them to a cen-tral server which distributes the required informa-tion to all nodes Distributed algorithms similar tothe topology discovery mechanism used by OLSR [66]would also be possible

MORP could be implemented at link or networklayer A link layer implementation would permitthe design of an efficient signaling protocol For in-stance the three-way-handshaking used by MORP(see section 32) could be implemented using a modi-fied 80211 MAC as shown in Figure 2 In this figurethe Multicast Candidates Set consists of the nodesa b c The candidates send back an ACK whichis immediately followed by the Forwarding-PacketA similar proposal to send the ACKs was proposedin [42]

A network layer implementation would allow us-ing current off-the-shelf 80211 network cards In thiscase ACKs and Forwarding-Packets would be sent us-ing unicast 80211 data frames thus increasing theoverhead and delays of the three-way-handshakingused by MORP Nevertheless for the sake of investi-

data frame

SIFSack a

ForwardingPacket t

Figure 2 Three-way-handshaking of MORP using amodified 80211 MAC

gating the feasibility to implement MORP with cur-rent hardware in the numerical results presented insection 8 we have assumed a network layer implemen-tation using standard 80211 cards

5 Summary of the ODMRP Pro-tocol

The On Demand Multicast Routing Protocol(ODMRP) is a mesh based multicast protocol wheregroup membership and multicast routes are estab-lished and updated by the source on demand [10 2967] It introduces the concept of forwarding groupsA multicast source will transmit packets to the des-tinations via the forwarding group The forwardinggroup is a set of nodes in charge of forwarding mul-ticast packets When a multicast source has datapackets to send but there is no route to the multi-cast group it broadcasts a Join-Query control packetto the entire network This control packet is period-ically sent every REFRESH INTERVAL eg every3 seconds to refresh the membership information andupdate routes When a node receives a non-duplicateJoin-Query it stores the upstream node ID and re-broadcasts the packet

When the Join-Query packet reaches a multicastdestination it creates and broadcasts a Join-Tableto its neighbors This packet is forwarded along theshortest path back to the multicast source that origi-nated the Join-Query When a node receives a Join-Table it checks if its ID matches with the ID of thenext node of one of the entries in the Join-Table If itmatches the node realizes that it is on the path to thesource and thus is part of forwarding group Thenit sets the forwarding flag FG-Flag and broadcastsits own Join-Table The Join-Table is propagated byeach forwarding group member until it reaches themulticast source The FG-Flag of forwarding nodesexpires after a multiple of the interval between suc-cessive Join-Query floods

When a node receives a data packet it forwardsthe packet only when it is non-duplicated and theFG-Flag for the multicast group of this node has notexpired Note that a multicast destination can alsobe a forwarding group node if it is on the path be-tween a multicast source and another destination

These procedures allow for redundant forwarding

to each receiver increasing the packet delivery ratioof the protocol if a packet is dropped on one pathas a result of collision or a link break the receivercan receive it along another path The benefit of thisredundancy comes at the cost of additional overheadand additional load on the network

6 Summary of the ADMR Proto-col

Adaptive Demand-Driven Multicast Routing(ADMR) [8 68] protocol is an on demand protocollike ODMRP It creates a source-based forwardingtree connecting the source with the destinationsof the multicast group Each multicast packetis dynamically forwarded from the source alongthe shortest delay path through the tree to thedestinations of the multicast group In ADMRpacket forwarding is based on two types of floodingtree flood and network flood In the tree flooding thepackets are constrained to the nodes in the multicasttree while network flooding is the flooding amongall nodes in the network Note that the tree floodingin ADMR is similar to the forwarding group conceptin ODMRP

When a source has packet to send but no routingstate yet exists for this sender and group it floods apacket called Source Information to all nodes in thenetwork using network flood Each node in the net-work that receives this packet forwards it unless ithas already forwarded a copy of it In addition thenode records in its Node-Table the ID of the nodefrom which it received the packet When this packetreaches a multicast destination it creates a replypacket called Receiver Join packet back toward thesource The Receiver Join packet is sent automati-cally along the shortest path traversed by the floodback towards the source Each node that forwardsthe Receiver Join creates a forwarding entry in itsMembership-Table indicating that it is a forwarderfor this sender and group

When a destination wants to join a group thenode checks its Membership-Table to determine if itis already connected to the group If it is not itsends a Multicast Solicitation packet as a networkflood Each node in the network forwards the Multi-cast Solicitation In this case if a node receiving theMulticast Solicitation already belongs to the group itwill unicast the Multicast Solicitation only to the pre-vious hop address Therefore the packet follows themulticast tree towards the source speeding up anddecreasing the overhead of the receiver join Whenthe source receives the Multicast Solicitation packetthe source replies to the Multicast Solicitation to ad-vertise to the destination its existence as a sender for

the group

ADMR sends Keep-Alive messages to maintainthe existing forwarding state for the multicast treeThe absence of data packets and Keep-Alive messageswithin a certain period of time is an indication offorwarding tree disconnection Firstly a local repairprocedure is performed to reconnect the tree if itfails a global reconnect procedure is used

7 Evaluation Methodology

To evaluate the performance of MORP we compareit with ODMRP and ADMR which have been shownto perform well in previous studies The simulationcode has been implemented within the Global Mo-bile Simulation (GloMoSim) library [21] The num-ber of multicast groups and sources is set to one in allscenarios Members join the multicast group at thestart of the simulation and remain throughout thesimulation The simulation field consists of a squarewith diagonal equal to 500 m We have run sim-ulations varying the number of nodes in the range20 le N le 100 One node is the source and it islocated in a square corner the others are placed ran-domly inside the square The destinations of the mul-ticast group are chosen randomly among the nodesinside the square Each simulation runs for 300 sec-onds of simulation time Each point in our perfor-mance graphs represents the average of 20 simulationruns For this number of runs we obtained reason-ably small confidence intervals The IEEE 80211Distributed Coordination Function was used as themedium access control protocol

The multicast application-layer source in our sim-ulations generates Constant Bit Rate (CBR) trafficwith 4 packet per second and 64 bytes of payloadThis sending rate was chosen to challenge the routingprotocolsrsquo abilities to successfully deliver data pack-ets in a wireless network It was not chosen to repre-sent any particular or class of applications althoughit could be considered to abstractly model a very sim-ple broadcast audio distribution application [8]

For a more realistic simulation of an 80211 net-work we have considered that packets can be trans-mitted at two different transmission rates a datarate of 11 Mbps and a basic rate of 2 Mbps Mostof previous works used the two-ray ground or somesimple loss propagation models [64 65 63] we usethe shadowing propagation model (below shadowingpropagation model is explained in more detail) forthe packet loss of all algorithms under study Pack-ets transmitted at the data rate are subject to ashadowing propagation model which introduces ran-dom transmission losses Packets transmitted at thebasic rate does not suffer transmission losses We

have assumed that data packets are always transmit-ted at data rate However the protocols can trans-mit signaling packets using the basic rate to preventlosses due to impairments of the radio channel Morespecifically we have assumed that in MORP all sig-naling packets (ie ACKs and Forwarding-Packets)are transmitted at the basic rate In ODMRP Join-Query packets are sent at the data rate This isbecause these packets are used to build the routingtables and thus they need to have the same trans-mission properties over the wireless links as those ofdata packets For the same reason Source Informa-tion and Multicast Solicitation packets are sent at thedata rate in ADMR although Receiver Join packetsare sent at the basic rate

We have assumed that in MORP nodes are awareof the network topology and the delivery probabilityof the wireless links due to the shadowing propaga-tion model of the radio channel MORP uses thisinformation and applies ExOR [42] to compute thecandidates sets

In the shadowing propagation model the wirelesslinks between nodes are not reliable The power re-ceived at a distance d in terms of the transmittedpower is given by

Pr(d)|dB = 10 log10

(PtGtGr λ

L (4π)2 dβ

)+XdB (3)

Where Pr(d) is the power received at a distance dand Pt is the transmitted power The Gt and Gr arethe transmission and reception antenna gains respec-tively L is a system loss λ is the signal wavelength(cf with c = 3 times 108 ms) β is a path loss expo-nent and XdB is a Gaussian random variable withzero mean and standard deviation σdB

Packets are delivered correctly if the receivedpower is greater than or equal to a threshold Rx-Thresh Thus the delivery probability owing to thepropagation model at a distance d is given by

p(d) = Prob(Pr(d)|dB ge 10 log10(RxThresh)) (4)

Table 2 Default GloMoSim values for the shadowingpropagation model

Parameter Value

Pt 003162278 WattRXThresh 7943282times 10minus12 WattGt Gr L 1f 2400 MHz

We have set the model parameters to the defaultvalues used by the GloMoSim given in Table 2 Fig-ure 3 depicts the delivery probability varying thedistance for a path loss exponent with parameters

0 50 100 150 200

β=27 sigma=60

Distance [m]

DeliveryProbability

Figure 3 Delivery probability versus distance for apath loss exponent β = 27 and standard deviationσdB = 6 dBs

β = 27 and σdB = 6 dBs With these parametersthe link delivery probability is approximately 40 atthe distance of 135 m To find the candidates setsusing ExOR we have assumed that a link betweenany two nodes exists only if the delivery probabilitybetween them is greater (or equal) than mindp = 01(see [57] for more details)

For the other parameters it was used Gt = Gr =L = 1 f = 914 MHz and RxThresh = 281 mw Wehave used these values in our simulations With theseparameters the link delivery probability is approxi-mately 40 at the distance of 150 m To find thecandidates sets using ExOR we have assumed that alink between any two nodes exists only if the deliveryprobability between them is greater (or equal) thanmindp = 01 (see [57] for more details)

71 Protocols Parameters

We have evaluated two different variations of theODMRP parameters The rdquoODMRP-3-9rdquo variationrepresents ODMRP using the parameter values cho-sen by ODMRPrsquos designers 3 seconds for the Join-Query flooding interval (REFRESH INTERVAL=3seconds) and a forwarding state lifetime of 3 times ofthis interval (a total of 9 seconds) The rdquoODMRP-3-33rdquo variation reduces the forwarding state lifetime to11 times of the Join-Query flooding interval it showsthe effect of reducing the forwarding redundancy ofODMRP (see section 5) For ADMR parameters wehave used the default values which are used in [8]30 seconds for the periodic data flood interval and2 missing packets to trigger disconnection detectionprocedure

In MORP we have used ExOR [42] as the candi-date selection algorithm Authors in [57] have shownthat using a small number of candidates (like 2) isa sensible choice Therefore we have fixed the max-imum number of candidates for the small and largecandidates sets to ncand = 2 and 10 respectivelyIn our protocol we have assumed that all candidates

that successfully receive the packet send ACK to thesource To do so we have used 12 milliseconds whichis long enough to receive all ACKs from the can-didates set (TACK = 12 ms) The legend MORP-ExOR(n) in the following sections refers to MORPwith MAXReTx = n

72 Performance Metrics

We have evaluated all protocols as a function of num-ber of nodes in the network and number of destina-tions of the multicast group The measures of interestare

bull Packet delivery ratio The ratio of the num-ber of data packets delivered to the destinationsversus the number of data packets supposed tobe received

bull Multicast group reachability Let X be a ran-dom variable equal to the number of destina-tions of the multicast group receiving a givendata packet We have computed the empiricalcomplementary cumulative distribution func-tion (EC-CDF) of X This gives a measureof the number of destinations of the multicastgroup receiving data packets

bull Forwarding cost Total number of data pack-ets transmitted by all nodes in the networkover the total number of data packets sent bythe source This metric represents the deliverycost in terms of transmissions of each multicastpacket Note that to make the comparisonsmore clear in this metric we take as the refer-ence originated instead of delivered packets

bull Normalized packet overhead The total num-ber of all data and control packets transmittedby any node in the network (either originatedor forwarded) divided by the total number ofall data packets received across all multicast re-ceivers

bull End-to-End delay Average end-to-end delay ofall data packets received by the destinations

8 Numerical Results

81 Packet delivery ratio

One important parameter of MORP is the maximumnumber of retransmissions (MAXReTx) Recall thatif the forwarder does not receive enough ACKs fromits candidates it retransmits the data packet up toMAXReTx times before it is forwarded To see theeffect of this parameter Figures 4 and 5 depict thedelivery ratio varying MAXReTx from 1 to 5 The

1 2 3 4 5

MORP-ExOR-Dest(2)MORP-ExOR-Dest(10)

D=500mmindp=01 β=27 σdB=6 N=20

MAXReTx

Packet

DeliveryRatio

Figure 4 Packet delivery ratio of MORP in a sparsenetwork as a function of MAXReTx

1 2 3 4 5

D=500mβ=27mindp=01σdB=6 N=100

MAXReTx

Packet

DeliveryRatio

Figure 5 Packet delivery ratio of MORP in a densenetwork as a function of MAXReTx

two curves correspond to a number of destinationsof the multicast group equal to 2 and 10 The leg-end MORP-ExOR-Dest(n) in these two figures refersto MORP with number of destinations equal to nThese figures have been obtained with a total num-ber of nodes equal to N = 20 (Figure 4) and N = 100(Figure 5) In the rest of this paper we shall refer tothe scenarios having these number of nodes as sparseand dense networks respectively The 95 confi-dence intervals have been added in Figure 4 It canbe observed that the intervals are relatively smalland the same was obtained for the other figures Sofor the sake of clarity confidence intervals are notdepicted in the rest of the figures

As expected Figures 4 and 5 show that the higheris MAXReTx the higher is the delivery ratio Addi-tionally we observe that the maximum delivery ra-tio improvement is obtained when MAXReTx is in-creased from 1 to 2 For instance in the sparsenetwork (Figure 4) we can see that the delivery ra-tio of MORP for 2 destinations with MAXReTx =1 is about 74 while it improves to 94 withMAXReTx = 2 (improvement around 27) Increas-ing from MAXReTx = 2 to 3 yields a delivery ratio of98 (improvement around 4)

Comparing Figures 4 and 5 we can see that packetdelivery ratio is always higher in a dense than in asparse network This comes from the fact that in thedense network MORP uses better candidates thanin the sparse network For instance the packet de-livery ratio of MORP in a sparse network with 2 des-tinations and MAXReTx = 1 is about 74 while itincreases to 90 in a dense network

Figure 6 shows the packet delivery ratio of MORPin comparison with ODMRP and ADMR The curvesare obtained varying the number of nodes from 20 to100 In this figure the number of destinations hasbeen set to 5 (NumDest = 5) The results of MORPare shown for MAXReTx is set to 1 and 2 (MORP-

20 30 40 50 60 70 80 90 100

MORP-ExOR(1)MORP-ExOR(2)ODMRP-3-9ODMRP-3-33ADMR

D=500m NumDest=5 mindp=01 σdB=6

Number of nodes

Packet

DeliveryRatio

Figure 6 Packet delivery ratio for 5 destinations asa function of number of nodes in the network

ExOR(1) and MORP-ExOR(2) respectively)

As we can see in Figure 6 MORP with anyMAXReTx outperforms both ODMRP (ODMRP-3-9and ODMRP-3-33) and ADMR For instance evenwith MAXReTx = 1 MORP has about 92 packetdelivery ratio while ODMRP-3-9 ODMRP3-33 andADMR have about 83 48 and 89 respectivelyThis comes from the fact that the construction of theroutes in ODMRP and ADMR are subject to the ran-dom losses that may have the Join-Query packets inODMRP and the Source Information and MulticastSolicitation packets in ADMR On the other handMORP takes routing decisions ldquoon the flyrdquo (whenthe forwarding nodes are chosen) and thus adaptsfaster to random losses

Figure 6 shows that the packet delivery ratio ofODMRP-3-33 is significantly lower than ODMRP-39 (about 35) As we described in section 5ODMRP creates forwarding groups within nodes inthe network that expires after a fixed timeout InODMRP-3-33 the forwarding state timeout (33 sec-onds) is shorter than in ODMRP-3-9 (9 seconds)

2 3 4 5 6 7 8 9 10

MORP-ExOR(1)MORP-ExOR(2)MORP-ExOR(3)ODMRP-3-9ODMRP-3-33ADMR

D=500m β=27 mindp=01 σdB=6 N=20

Number of Destinations

Packet

DeliveryRatio

Figure 7 Packet delivery ratio in a sparse networkas a function of number of destinations

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Number of destinations

Packet

DeliveryRatio

Figure 8 Packet delivery ratio in a dense network asa function of number of destinations

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Figure 9 Distribution of received packets for 10 des-tinations and 20 nodes

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

Figure 10 Distribution of received packets for 10destinations and 100 nodes

Therefore ODMRP-3-33 has less number of nodesin the forwarding group than in ODMRP-3-9 result-ing in a lower delivery ratio

Figures 7 and 8 show the delivery ratio of theprotocols under study varying the number of desti-nations In these figures we can see that all protocolsachieve a higher delivery ratio in the dense scenariothan in the sparse network In the sparse network(Figure 7) the packet delivery ratio of MORP withany maximum number of retransmissions (MAXReTx)outperforms both variations of ODMRP The samefigure shows that ADMR has a delivery ratio about6 better than MORP-ExOR(1) However in sec-tion 83 we will see that this small improvement is atcost of having much more data transmissions thanMORP Nevertheless Figure 7 shows that MORPoutperforms ADMR when MAXReTx is increased to2 and 3

For a dense network we can see in Figure 8that the packet delivery ratio of MORP with anyMAXReTx is higher than ODMRP and ADMR Thiscomes from the fact that in a dense network MORPcan choose better candidates to forward the pack-

ets For instance the packet delivery ratio of MORP-ExOR(1) and MORP-ExOR(2) in the dense networkis about 94 and 98 respectively Although thedelivery ratio of ADMR in the case of a dense net-work is close to MORP-ExOR(1) we will see in sec-tion 83 that it is at cost of a large amount of datatransmissions

82 Multicast group reachability

In this section we investigate the number of destina-tions of the multicast group receiving data packetsTo do so we have calculated its empirical comple-mentary cumulative distribution function (EC-CDF)This is shown in Figures 9 and 10 in a scenario with10 destinations for the sparse and dense networks(with 20 and 100 nodes) respectively

Figure 9 shows that ODMRP-3-33 performsmuch worst than the others the probability of reach-ing the multicast group decreases sharply with in-creasing the number of destinations and less than20 of the packets reach all destinations

Regarding the other protocols Figure 9 shows

20 30 40 50 60 70 80 90 100

18MORP-ExOR(1)MORP-ExOR(2)ODMRP-3-9ODMRP-3-33ADMR

Number of nodes

ardingcost

Figure 11 Forwarding cost for 5 destinations as afunction of number of nodes in the network

that in the sparse scenario around 5 of packets donot reach any destination in ODMRP-3-9 ADMRand MORP-ExOR(1) Then the curves are approx-imately flat and decrease at the end (specially for10 destinations) This behavior is explained by thesimulation topology Recall that the simulation fieldis a square with the source in one corner and theother nodes distributed randomly This distributionof nodes favors that when the source reach the firstnext hop it reaches also most of the destinations withhigh probability

Figure 9 also shows that the difference betweenMORP-ExOR(2) and MORP-ExOR(3) is small Inboth cases almost 100 of packets reach up to 9 des-tinations and about 90 of packets are deliveredto 10 destinations in MORP-ExOR(2) and 95 inMORP-ExOR(3)

Finally the same conclusions can be derived fromthe dense network scenario shown in Figure 10 Ofcourse the group reachability increases due to thehigher proximity between the nodes

83 Forwarding cost

In this section we compare the forwarding cost ofMORP ODMRP and ADMR Recall that we havedefined forwarding cost as the number of data packetstransmitted by all nodes in the network over the totalnumber of data packets sent by the source

Figure 11 shows the forwarding cost of the proto-cols varying the number of nodes from 20 to 100 inthe case of 5 destinations The results of MORP likein Figure 6 are obtained for MAXReTx = 1 and 2

Figure 11 shows that the forwarding cost ofMORP outperforms ODMRP-3-9 and ADMR Infact the forwarding cost of both variations ofODMRP and ADMR is rather sensitive to the num-ber of nodes while in MORP is not This is becauseusing opportunistic routing as in MORP only some

useful nodes are selected as candidates to forwardthe packets Figure 11 also shows that there is onlya slight increase of the forwarding cost when MORPincreases MAXReTx from 1 to 2

As described in section 5 ODMRP periodicallyfloods a data packet together with a Join-Querypacket Ie it piggybacks the Join-Query informa-tion on the data packet periodically to update theroutes Because of this new nodes may become for-warders while forwarders created during a previousperiodic flood still have a set forwarding flag Conse-quently redundant routes are created and the num-ber of data transmissions increases with increasingthe network density In fact the forwarding costof both variations of ODMRP is dominated by theflooding packets and forwarding state timeout Sincein ODMRP-3-9 the forwarding state timeout (9 sec-onds) is longer than in ODMRP-3-33 there are morenodes with the forwarding flag set in ODMRP-3-9than in ODMRP-3-33 Therefore the forwardingcost of ODMRP-3-9 is much higher than in ODMRP-3-33

The construction of the routes in ADMR is sub-ject to the random losses that may have the SourceInformation and Multicast Solicitation packets Re-call that the absence of some data packets in ADMRis an indication of forwarding tree disconnection andthe local or global repair procedure is triggered torepair the path As we mentioned in section 71 weused 2 missing data packets to trigger disconnectiondetection When a node on the tree does not receive2 consecutive data packets it starts repairing algo-rithm which may add new nodes to the tree Thisis exacerbated with increasing the density of the net-work thus increasing the forwarding cost

Figure 11 shows that for N = 20 ODMRP-3-33is slightly better than MORP However recall fromFigure 6 that in this scenario the delivery ratio ofODMRP-3-33 is much lower than in MORP

In section 81 we have shown that MORP out-performs ODMRP and ADMR in terms of packetdelivery ratio for different number of destinationsFigures 12 and 13 give the forwarding cost for thesame scenarios Figure 12 shows that the forwardingcost of MORP in the sparse network and with anyMAXReTx is much lower than ADMR and ODMRP-3-9 The figure shows that only ODMRP-3-33 isslightly better than MORP However as we showedin Figure 7 the delivery ratio of ODMRP-3-33 ismuch lower than MORP

Regarding ADMR Figure 7 in section 81 showedthat the delivery ratio in a sparse network is slightlybetter than MORP-ExOR(1) However we observein Figure 12 that this is at cost of ADMR having aforwarding cost of about 3 times higher than MORP-ExOR(1)

2 3 4 5 6 7 8 9 10

17MORP-ExOR(1)MORP-ExOR(2)MORP-ExOR(3)ODMRP-3-9ODMRP-3-33ADMR

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

Figure 12 Forwarding cost in a sparse network as afunction of number of destinations

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

Figure 13 Forwarding cost in a dense network as afunction of number of destinations

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

Figure 14 Packet overhead in a sparse network as afunction of number of destinations

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

Figure 15 Packet overhead in a dense network as afunction of number of destinations

Figures 12 and 13 confirm as in Figure 11 thatthe forwarding cost of ODMRP and ADMR is muchhigher in a dense than in a sparse network whileMORP is rather insensitive to network density Nev-ertheless it can be observed that the forwarding costincrement in MORP is slightly higher in a dense thanin a sparse network This may be counterintuitivesince MORP can choose better candidates in a densenetwork However since MORP looks for candidatesto reach all destinations in a dense network MORPchooses more candidates thus increasing the for-warding cost However recall from Figures 7 and 8this slightly higher forwarding cost in a dense net-work is rewarded with a significantly higher packetdelivery ratio

84 Packet overhead

In this section we compare the packet overhead ofthe protocols under study Recall that we com-pute the packet overhead as the ratio of data andcontrol packets transmitted by any node to deliverdata packets We count as the control packets for

ODMRP the Join-Query and Join-Table for ADMRthe Source Information Receiver Join Multicast So-licitation and packets related to the repair processThe ForwardingPacket and ACK packets in MORPare counted as the control packets

Figures 14 and 15 show the packet overhead ofall protocols varying the number of destinations fora sparse and dense network respectively Figure 14shows that in a sparse network MORP has a higherpacket overhead than ADMR and ODMRP This isdue to the ACKs sent by the candidates in MORPNote that in this comparison we are giving the sameweight to data and control packets Recall from Fig-ures 12 and 13 that MORP performs better than theother protocols in terms of forwarding cost whereonly data packets are considered Thus in case ofsending data packets with large payload the over-head considered in this section would be a pessimisticcomparison with respect to MORP Furthermore asexplained in section 4 the overhead of control packetsin MORP could be reduced implementing the three-way handshaking used by MORP at MAC layer

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

Figure 16 End-to-end delay in a sparse network asa function of number of destinations

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

Figure 17 End-to-end delay in a dense network as afunction of number of destinations

85 End-To-End Delay

Figures 16 and 17 show the average end-to-end delayfor different number of destinations for a sparse anddense network respectively

These figures show that the end-to-end delay inMORP is higher than in ODMRP and ADMR Thisis because each time a node transmits a data packetin MORP it waits for the ACKs sent by the can-didates during TACK =12 ms However recall fromsection 4 that implementing the three-way handshak-ing used by MORP at MAC layer could reduce thisdelay significantly

Comparing Figures 16 and 17 we can see that thedecrease of the end-to-end delay in a dense networkfor MORP is much more noticeable than in the otherprotocols This is because using OR in a dense net-work the probability of reaching a candidate closeto the destination increases thus reducing the aver-age number of end-to-end hops It can also be ob-served that the difference between the average delayusing MORP with MAXReTx = 1 and MAXReTx gt 1is higher in a sparse network than in a dense net-work This is because in the dense network the prob-ability that the sending node receives enough ACKsto reach all destinations is higher than in the sparsenetwork Therefore MORP requires less retransmis-sions of data packets in a dense network

9 Conclusion

In this paper we have investigated how OpportunisticRouting (OR) can be exploited to implement a mul-ticast protocol for wireless mesh networks This hasbeen done by proposing a new protocol called Multi-cast Opportunistic Routing Protocol MORP MORPuses a three-way-handshaking approach where candi-dates send ACKs to the sender node upon success-fully receiving data packets Then the sender node

partitions the set of destinations and assigns eachsubset to the most appropriate candidate This in-formation is sent to the candidates which repeat thesame approach for each subset of destinations Com-pared with traditional multicast protocols MORPdoes not build a complete tree or mesh before thetransmissions starts Instead MORP builds a treeon the fly depending on the candidates that success-fully receive the packet in each transmission

We have compared MORP with other relevantmulticast protocols that have been proposed in theliterature On Demand Multicast Routing Protocol(ODMRP) and Adaptive Demand-Driven MulticastRouting (ADMR) The comparison is done takinginto consideration realistic simulations using 80211standard MAC layer A lossy shadowing propagationmodel has been used for the radio channel

Simulation results show that in most casesODMRP and ADMR have a number of data trans-missions much higher than in MORP while theachieved delivery ratio is not as good as in MORPAlthough the signaling overhead and end-to-end de-lay in MORP is a bit higher than in ODMRP andADMR the overhead of control packets could bereduced significantly implementing the three-way-handshaking used by MORP at MAC layer

We conclude that MORP outperforms traditionalODMRP and ADMR multicast protocols reducingthe number of data transmissions and increasing thedata delivery ratio Hence using OR can be a usefultechnique to implement reliable multicast protocolsin wireless mesh networks

References

[1] I Akyildiz X Wang A survey on wireless mesh networks

Communications Magazine IEEE 43 (9) (2005) S23 ndash S30

doi101109MCOM20051509968

[2] I F Akyildiz X Wang Wireless Mesh Networks John

Wiley amp Sons Ltd 2009

[3] M L Kai Zeng Wenjing Lou Multihop Wireless Networks

Opportunistic Routing John Wiley amp Sons 2011 2011

[4] L Junhai Y Danxia X Liu F Mingyu A survey of mul-

ticast routing protocols for mobile ad-hoc networks Com-

munications Surveys Tutorials IEEE 11 (1) (2009) 78 ndash91

[5] R C Biradar S S Manvi Review of multicast routing

mechanisms in mobile ad hoc networks Journal of Network

and Computer Applications 35 (1) (2012) 221 ndash 239 doi

101016jjnca201108003

[6] K Chen K Nahrstedt Effective location-guided tree con-

struction algorithms for small group multicast in manet in

INFOCOM 2002 Twenty-First Annual Joint Conference of

the IEEE Computer and Communications Societies Pro-

ceedings IEEE Vol 3 2002 pp 1180 ndash 1189 vol3

[7] J Xie R R Talpade A Mcauley M Liu Amroute ad

hoc multicast routing protocol Mob Netw Appl 7 (2002)

429ndash439

[8] J G Jetcheva D B Johnson Adaptive demand-driven

multicast routing in multi-hop wireless ad hoc networks

in Proceedings of the 2nd ACM international symposium

on Mobile ad hoc networking amp computing MobiHoc rsquo01

ACM New York NY USA 2001 pp 33ndash44

[9] X Zhao C T Chou J Guo S Jha A Misra Proba-

bilistically reliable on-demand multicast in wireless mesh

networks in World of Wireless Mobile and Multimedia

Networks 2008 WoWMoM 2008 2008 International Sym-

posium on a 2008 pp 1 ndash9

[10] S J Lee W Su M Gerla On-demand multicast routing

protocol in multihop wireless mobile networks Mob Netw

Appl 7 (2002) 441ndash453

[11] B Sun L Li Reliable adaptive multicast protocol in wire-

less ad hoc networks Journal of Systems Engineering and

Electronics 17 (1) (2006) 187 ndash 192

[12] J Biswas M Barai S Nandy Efficient hybrid multi-

cast routing protocol for ad-hoc wireless networks in Lo-

cal Computer Networks 2004 29th Annual IEEE Interna-

tional Conference on 2004 pp 180 ndash 187

[13] A Mnaouer L Chen C H Foh J Tantra Ophmr An

optimized polymorphic hybrid multicast routing protocol

for manet Mobile Computing IEEE Transactions on 6 (5)

(2007) 551 ndash562

[14] P Larsson Selection diversity forwarding in a multihop

packet radio network with fading channel and capture SIG-

MOBILE Mob Comput Commun Rev 5 (4) (2001) 47ndash

[15] D Fuste-Vilella J Garcia-Vidal J Morillo-Pozo Coop-

erative forwarding in ieee 80211-based manets in Wire-

less Days 2008 WD rsquo08 1st IFIP 2008 pp 1 ndash5

[16] H Dubois-Ferriere M Grossglauser M Vetterli Valu-

able detours Least-cost anypath routing Networking

IEEEACM Transactions on 19 (2) (2011) 333 ndash346

[17] H Dubois-Ferriere M Grossglauser M Vetterli Least-

cost opportunistic routing in Proceedings of 2007 Allerton

Conference on Communication Control and Computing

[18] S Biswas R Morris ExOR opportunistic multi-hop

routing for wireless networks ACM SIGCOMM Computer

Communication Review 35 (4) (2005) 133ndash144

[19] S-J Lee M Gerla C-C Chiang On-demand multicast

routing protocol in Wireless Communications and Net-

working Conference 1999 WCNC 1999 IEEE 1999 pp

1298 ndash1302 vol3

[20] M Pandey D Zappala A scenario-based performanceevaluation of multicast routing protocols for ad hoc net-works in Proceedings of the Sixth IEEE InternationalSymposium on World of Wireless Mobile and Multime-dia Networks WOWMOM rsquo05 IEEE Computer SocietyWashington DC USA 2005 pp 31ndash41

[21] X Zeng R Bagrodia M Gerla GloMoSim A Libraryfor Parallel Simulation of Large-Scale Wireless Networksin Proceedings of the 12th Workshop on Parallel and Dis-tributed Simulation (PADSrsquo98) IEEE Computer Society1998 pp 154ndash161

[22] L Junhai X Liu Y Danxia Research on multicast rout-ing protocols for mobile ad-hoc networks Computer Net-works 52 (5) (2008) 988 ndash 997 doi101016jcomnet

200711016

[23] L Ji M Corson Differential destination multicast-amanet multicast routing protocol for small groups in IN-FOCOM 2001 Twentieth Annual Joint Conference of theIEEE Computer and Communications Societies Proceed-ings IEEE Vol 2 2001 pp 1192 ndash1201 vol2

[24] R Sisodia I Karthigeyan B Manoj C Murthy A pre-ferred link based multicast protocol for wireless mobile adhoc networks in Communications 2003 ICC rsquo03 IEEEInternational Conference on Vol 3 2003 pp 2213 ndash 2217vol3

[25] C Wu Y Tay Amris a multicast protocol for ad hocwireless networks in Military Communications Confer-ence Proceedings 1999 MILCOM 1999 IEEE Vol 11999 pp 25 ndash29 vol1

[26] E M Royer C E Perkins Multicast Ad hoc On-DemandDistance Vector (MAODV) Routing - draft-ietf-manet-maodv-00 Tech rep Mobile Ad Hoc Networking WorkingGroup (2000)

[27] X Zhang L Jacobrdquo rdquomzrp An extension of the zonerouting protocol for multicasting in manetsrdquo rdquoJ Inf SciEngrdquo (2004) 535ndash551

[28] B Kaliaperumal A Ebenezer Jeyakumar Adaptive corebased scalable multicasting networks in INDICON 2005Annual IEEE 2005 pp 198 ndash 202

[29] O S Badarneh M Kadoch Multicast routing protocolsin mobile ad hoc networks a comparative survey and tax-onomy EURASIP J Wirel Commun Netw (2009) 261ndash2642doi1011552009764047

[30] M Lee Y K Kim Patchodmrp an ad-hoc multicastrouting protocol in Information Networking 2001 Pro-ceedings 15th International Conference on 2001 pp 537ndash543

[31] S-B Cai X-Z Yang The performance of poolodmrp pro-tocol in MMNSrsquo03 2003 pp 90ndash101

[32] L W S Cai X Yang Pdaodmrp An extendedpoolodmrp based on passive data acknowledgement Jour-nal of Communications and Networks 6 (4) (2004) 362ndash375cited By (since 1996) 3

[33] S Oh J-S Park M Gerla E-odmrp enhanced odmrpwith motion adaptive refresh in Wireless CommunicationSystems 2005 2nd International Symposium on 2005 pp130 ndash 134

[34] D Pathirana M Kwon Rodmrp Resilient on-demandmulticast routing protocol Advanced Information Net-working and Applications Workshops International Con-ference on 2 (2007) 85ndash92

[35] M-A Kharraz H Sarbazi-Azad A Y Zomaya On-demand multicast routing protocol with efficient route dis-covery Journal of Network and Computer Applications35 (3) (2012) 942 ndash 950 iexclcetitleiquestSpecial Issue on TrustedComputing and Communicationsiexclcetitleiquest doi101016jjnca201103012

[36] J Garcia-Luna-Aceves E Madruga The core-assistedmesh protocol Selected Areas in Communications IEEEJournal on 17 (8) (1999) 1380 ndash1394

[37] C Lin S-W Chao A multicast routing protocol for mul-tihop wireless networks in Global TelecommunicationsConference 1999 GLOBECOM rsquo99 Vol 1A 1999 pp 235ndash239 vol1a

[38] S Lee C Kim Neighbor supporting ad hoc multicastrouting protocol in Proceedings of the 1st ACM interna-tional symposium on Mobile ad hoc networking amp comput-ing MobiHoc rsquo00 IEEE Press Piscataway NJ USA 2000pp 37ndash44

[39] S K Das B S B S Manoj C S Ram Murthy Adynamic core based multicast routing protocol for ad hocwireless networks in Proceedings of the 3rd ACM inter-national symposium on Mobile ad hoc networking amp com-puting MobiHoc rsquo02 ACM New York NY USA 2002pp 24ndash35

[40] R Biradar S Manvi M Reddy Link stability based mul-ticast routing scheme in manet Comput Netw 54 (2010)1183ndash1196

[41] P Sinha R Sivakumar V Bharghavan Mcedar multi-cast core-extraction distributed ad hoc routing in Wire-less Communications and Networking Conference 1999WCNC 1999 IEEE 1999 pp 1313 ndash1317 vol3

[42] S Biswas R Morris Opportunistic routing in multi-hopwireless networks ACM SIGCOMM Computer Communi-cation Review 34 (1) (2004) 69ndash74

[43] D S J De Couto D Aguayo J Bicket R Morris Ahigh-throughput path metric for multi-hop wireless rout-ing Wireless Networks 11 (4) (2005) 419ndash434

[44] Z Zhong J Wang S Nelakuditi G-H Lu On selectionof candidates for opportunistic anypath forwarding SIG-MOBILE Mob Comput Commun Rev 10 (4) (2006) 1ndash2

[45] MIT roofnet httppdoscsailmiteduroofnet

[46] Y Li W Chen Z-L Zhang Optimal forwarder list selec-tion in opportunistic routing in Mobile Adhoc and SensorSystems MASS rsquo09 IEEE 6th International Conference on2009 pp 670 ndash675

[47] A Darehshoorzadeh L Cerda-Alabern Candidate selec-tion algorithms in opportunistic routing in PM2HW2Nrsquo10 Proceedings of the 5th ACM workshop on Performancemonitoring and measurement of heterogeneous wireless andwired networks ACM New York NY USA 2010 pp 48ndash54

[48] M A Iqbal B Dai B Huang A Hassan S Yu Sur-vey of network coding-aware routing protocols in wirelessnetworks Journal of Network and Computer Applications34 (6) (2011) 1956 ndash 1970 iexclcetitleiquestControl and Optimiza-tion over Wireless Networksiexclcetitleiquest doi101016j

jnca201107012

[49] R Bruno M Nurchis Survey on diversity-based routingin wireless mesh networks Challenges and solutions Com-puter Communications 33 (3) (2010) 269ndash282

[50] S Chachulski M Jennings S Katti D Katabi Tradingstructure for randomness in wireless opportunistic routingin SIGCOMM ACM New York NY USA 2007 pp169ndash180

[51] Y Yan B Zhang H Mouftah J Ma Practical coding-aware mechanism for opportunistic routing in wireless meshnetworks in Communications 2008 ICC rsquo08 IEEE Inter-national Conference on 2008 pp 2871 ndash2876

[52] F Baccelli B Blaszczyszyn P Muhlethaler On the per-formance of time-space opportunistic routing in multihopmobile ad hoc networks in WiOpt 2008 pp 307ndash316

[53] C-P Luk W-C Lau O-C Yue An analysis of oppor-tunistic routing in wireless mesh network in Communi-cations 2008 ICC rsquo08 IEEE International Conference on2008 pp 2877ndash2883

[54] J Wu M Lu F Li Utility-based opportunistic routingin multi-hop wireless networks in Distributed ComputingSystems ICDCSrsquo08 The 28th International Conference on2008 pp 470ndash477

[55] M Lu J Wu Opportunistic routing algebra and its ap-plications in IEEE INFOCOM 2009 pp 2374ndash2382

[56] M Kurth A Zubow J-P Redlich Cooperative oppor-tunistic routing using transmit diversity in wireless meshnetworks in IEEE INFOCOM 2008 pp 1310ndash1318

[57] L Cerda-Alabern V Pla A Darehshoorzadeh On theperformance modeling of opportunistic routing in Mo-biOpp rsquo10 Proceedings of the Second International Work-shop on Mobile Opportunistic Networking ACM NewYork NY USA 2010 pp 15ndash21

[58] A Darehshoorzadeh L CerdA -Alabern V Pla Mod-eling and comparison of candidate selection algorithms inopportunistic routing Computer Networks 55 (13) (2011)2886 ndash 2898

[59] Y Li Z-L Zhang Random walks on digraphs a theoreti-cal framework for estimating transmission costs in wirelessrouting in 29th conference on Information communica-tions INFOCOMrsquo10 IEEE Press 2010 pp 2775ndash2783

[60] A Cacciapuoti M Caleffi L Paura Optimal con-strained candidate selection for opportunistic routingin Global Telecommunications Conference (GLOBECOM2010) 2010 IEEE 2010 pp 1 ndash5 doi101109GLOCOM

20105683490

[61] L CerdA -Alabern V Pla A Darehshoorzadeh On themaximum performance in opportunistic routing in IEEEWoWMoM 2010 Montreal Canada 2010

[62] D Koutsonikolas Y Hu C-C Wang Pacifier High-throughput reliable multicast without ldquocrying babiesrdquo inwireless mesh networks in INFOCOM 2009 IEEE 2009pp 2473 ndash2481

[63] Y WenZhong Z ZhenYu W Bo W XiaoHong A re-liable multicast for manets based on opportunistic rout-ing in Wireless Communications Networking and MobileComputing (WiCOM) 2010 6th International Conferenceon 2010 pp 1 ndash4

[64] T Le Y Liu Exploring the gain of opportunistic routingin wireless multicast Technical-report Polytechnic Insti-tute of New York University 5 Metrotech Center Brook-lyn NY 11201 US (2009)

[65] T Le Y Liu Opportunistic overlay multicast in wirelessnetworks in GLOBECOM 2010 pp 1ndash5

[66] T Clausen P J (editors) C Adjih A Laouiti P MinetP Muhlethaler A Qayyum L Viennot Optimized linkstate routing protocol (OLSR) RFC 3626 pages 1-75 net-work Working Group (October 2003)

[67] M Naderan-Tahan A Darehshoorzadeh M DehghanOdmrp-lr Odmrp with link failure detection and local re-covery mechanism in Computer and Information Science2009 ICIS 2009 Eighth IEEEACIS International Confer-ence on 2009 pp 818 ndash823

[68] R Sokullu O Karaca Comparative performance study ofadmr and odmrp in the context of wireless lans and wire-less sensor networks in Proceedings of the 7th WSEASInternational Conference on Telecommunications and In-formatics World Scientific and Engineering Academy andSociety (WSEAS) Stevens Point Wisconsin USA 2008pp 183ndash187

1 Introduction

2 Related work



3 Multicast Opportunistic Routing Protocol (MORP)

31 Network Model


33 Forwarding Set

34 Candidate Coordination and Data forwarding

35 Data Structures



5 Summary of the ODMRP Protocol

6 Summary of the ADMR Protocol




8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

to be the next-hop forwarder an ordered set of nodes(referred to as candidates) is selected as the next-hoppotential forwarders More specifically when the cur-rent node transmits a packet all the candidates thatreceive the packet successfully will coordinate witheach other to determine which one would actuallyforward the packet according to some criteria whilethe other nodes will simply discard the packet

By using OR if a certain wireless forwarder failsor moves out of the radio range during the transmis-sion other possible paths may be used As a resultOR can better cope with lossy unreliable and timevarying link quality It can significantly reduce thenumber of transmissions necessary to deliver a packetto the destination and greatly increases the transmis-sion reliability and the network throughput by tak-ing advantage of the broadcast nature of the wirelessmedium

Previous researches have shown that OR can sig-nificantly reduce the expected number of transmis-sions to deliver a packet to a particular destinationIt is therefore tempting to adopt OR to improve theefficiency of wireless multicast The main challengein adapting of OR with multicast is how to sharethe opportunistic forwarders paths between multipledestinations In OR algorithms for unicast protocolssince a packet is addressed to only one destinationupon transmitting a packet only one of the candi-dates receiving it would actually forward the packetOn the other hand since there are more than one des-tination in the multicast protocols using OR mightcause that more than one candidate has to forwardthe packet to reach all the destinations Anotherchallenge of using OR in multicast in contrast tounicast is that the selected candidates might haveto forward the packets toward more than one desti-nation

This paper presents a new multicast routing pro-tocol that we call Multicast Opportunistic RoutingProtocol MORP Unlike traditional multicast proto-cols there is no designated next-hop forwarder foreach destination in our protocol thus the deliveryratio is maximized by taking advantage of spacialdiversity MORP uses three-way-handshaking ap-proach to transmit the data packet The basic ideaof MORP is as follow when a source node wants totransmit a data packet it creates its candidates setand include it into the packet The candidates whichsuccessfully receive the packet send an acknowledg-ment Then the sender selects some candidates andtowards which destinations they have to forward thepacket This information is sent to the candidateswhich repeat the algorithm until reaching all desti-nations of the multicast group Compared with thetraditional multicast protocols our protocol does notbuild a complete tree or mesh before the transmis-

sions starts Instead MORP builds a tree on the flydepending on the candidates that successfully receivethe packet in each transmission

We compare MORP with two well known proto-cols ODMRP multicast mesh protocol [19 10] andADMR multicast tree protocol [8 20] An additionalreason to choose these protocols for comparison isthat ODMRP is implemented in the simulation toolused to obtain numerical results (GloMoSim [21])The source code of ADMR for GloMoSim was kindlyprovided by the authors in [20]

In summary the main contributions of this paperare

bull We investigate the advantages of using OR tosupport multicast by proposing MORP

bull In contrast to the most of previous workswhich used the two-ray ground or some simpleloss propagation models we use the shadow-ing propagation model for the packet loss of allalgorithms under study

bull Our main conclusion is that OR can be an ef-fective mechanism to achieve reliable multicastdelivery in wireless mesh networks

The rest of this paper is organized as followsWe briefly review the related work on multicast andOR in Section 2 MORP description is presentedin Sections 3 and 4 Section 5 and 6 briefly de-scribe ODMRP and ADMR respectively Section 7explains the evaluation methodology MORPrsquos per-formance is evaluated in Section 8 and concludingremarks are given in Section 9

2 Related work

Multicast routing protocols come into play when ahost needs to send the same message or data streamto multiple destinations Due to the unique char-acteristics of the wireless networks such as limitedresources and unreliable channels traditional multi-cast protocols in the wired networks do not performwell in wireless and new protocols have been pro-posed One of the most popular methods to classifymulticast routing protocols is based on how distri-bution paths among group members are constructedAccording to this method existing multicast routingapproaches can be classified into tree-based mesh-based and hybrid protocols [4 22 5]

In the tree-based protocols only a single short-est path must be established between source-receiverpair therefore the multicast tree is composed of aunique path from the multicast source to each of themulticast receivers

31 Network Model

CsD =⋃djisinD

Csdj2 (1)

33 Forwarding Set

CostFunc(T )

ETX(ci dj)

CostFunc(S) =sum

djisinRDestSet

minciisinS

ETX(ci dj) (2)

35 Data Structures

candidates sets

dest small large

(b) ETX Table

node d1 d2

a 43 41b 48 38

data frame

SIFSack a

ForwardingPacket t

the group

Pr(d)|dB = 10 log10

(PtGtGr λ

L (4π)2 dβ

)+XdB (3)

Parameter Value

0 50 100 150 200

β=27 sigma=60

Distance [m]

DeliveryProbability

8 Numerical Results

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

20 30 40 50 60 70 80 90 100

Number of nodes

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

31 Network Model

CsD =⋃djisinD

Csdj2 (1)

33 Forwarding Set

CostFunc(T )

ETX(ci dj)

CostFunc(S) =sum

djisinRDestSet

minciisinS

ETX(ci dj) (2)

35 Data Structures

candidates sets

dest small large

(b) ETX Table

node d1 d2

a 43 41b 48 38

data frame

SIFSack a

ForwardingPacket t

the group

Pr(d)|dB = 10 log10

(PtGtGr λ

L (4π)2 dβ

)+XdB (3)

Parameter Value

0 50 100 150 200

β=27 sigma=60

Distance [m]

DeliveryProbability

8 Numerical Results

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

20 30 40 50 60 70 80 90 100

Number of nodes

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

31 Network Model

CsD =⋃djisinD

Csdj2 (1)

33 Forwarding Set

CostFunc(T )

ETX(ci dj)

CostFunc(S) =sum

djisinRDestSet

minciisinS

ETX(ci dj) (2)

35 Data Structures

candidates sets

dest small large

(b) ETX Table

node d1 d2

a 43 41b 48 38

data frame

SIFSack a

ForwardingPacket t

the group

Pr(d)|dB = 10 log10

(PtGtGr λ

L (4π)2 dβ

)+XdB (3)

Parameter Value

0 50 100 150 200

β=27 sigma=60

Distance [m]

DeliveryProbability

8 Numerical Results

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

20 30 40 50 60 70 80 90 100

Number of nodes

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

31 Network Model

CsD =⋃djisinD

Csdj2 (1)

33 Forwarding Set

CostFunc(T )

ETX(ci dj)

CostFunc(S) =sum

djisinRDestSet

minciisinS

ETX(ci dj) (2)

35 Data Structures

candidates sets

dest small large

(b) ETX Table

node d1 d2

a 43 41b 48 38

data frame

SIFSack a

ForwardingPacket t

the group

Pr(d)|dB = 10 log10

(PtGtGr λ

L (4π)2 dβ

)+XdB (3)

Parameter Value

0 50 100 150 200

β=27 sigma=60

Distance [m]

DeliveryProbability

8 Numerical Results

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

20 30 40 50 60 70 80 90 100

Number of nodes

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

33 Forwarding Set

CostFunc(T )

ETX(ci dj)

CostFunc(S) =sum

djisinRDestSet

minciisinS

ETX(ci dj) (2)

35 Data Structures

candidates sets

dest small large

(b) ETX Table

node d1 d2

a 43 41b 48 38

data frame

SIFSack a

ForwardingPacket t

the group

Pr(d)|dB = 10 log10

(PtGtGr λ

L (4π)2 dβ

)+XdB (3)

Parameter Value

0 50 100 150 200

β=27 sigma=60

Distance [m]

DeliveryProbability

8 Numerical Results

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

20 30 40 50 60 70 80 90 100

Number of nodes

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

CostFunc(S) =sum

djisinRDestSet

minciisinS

ETX(ci dj) (2)

35 Data Structures

candidates sets

dest small large

(b) ETX Table

node d1 d2

a 43 41b 48 38

data frame

SIFSack a

ForwardingPacket t

the group

Pr(d)|dB = 10 log10

(PtGtGr λ

L (4π)2 dβ

)+XdB (3)

Parameter Value

0 50 100 150 200

β=27 sigma=60

Distance [m]

DeliveryProbability

8 Numerical Results

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

20 30 40 50 60 70 80 90 100

Number of nodes

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

candidates sets

dest small large

(b) ETX Table

node d1 d2

a 43 41b 48 38

data frame

SIFSack a

ForwardingPacket t

the group

Pr(d)|dB = 10 log10

(PtGtGr λ

L (4π)2 dβ

)+XdB (3)

Parameter Value

0 50 100 150 200

β=27 sigma=60

Distance [m]

DeliveryProbability

8 Numerical Results

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

20 30 40 50 60 70 80 90 100

Number of nodes

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

data frame

SIFSack a

ForwardingPacket t

the group

Pr(d)|dB = 10 log10

(PtGtGr λ

L (4π)2 dβ

)+XdB (3)

Parameter Value

0 50 100 150 200

β=27 sigma=60

Distance [m]

DeliveryProbability

8 Numerical Results

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

20 30 40 50 60 70 80 90 100

Number of nodes

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

the group

Pr(d)|dB = 10 log10

(PtGtGr λ

L (4π)2 dβ

)+XdB (3)

Parameter Value

0 50 100 150 200

β=27 sigma=60

Distance [m]

DeliveryProbability

8 Numerical Results

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

20 30 40 50 60 70 80 90 100

Number of nodes

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

0 50 100 150 200

β=27 sigma=60

Distance [m]

DeliveryProbability

8 Numerical Results

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

20 30 40 50 60 70 80 90 100

Number of nodes

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

1 2 3 4 5

MAXReTx

Packet

DeliveryRatio

20 30 40 50 60 70 80 90 100

Number of nodes

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

Packet

DeliveryRatio

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

Packet

DeliveryRatio

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

1 2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

20 30 40 50 60 70 80 90 100

Number of nodes

ardingcost

83 Forwarding cost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

ardingcost

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

ardingcost

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=20

alizedpacket

overhead

2 3 4 5 6 7 8 9 10

D=500m mindp=01 β=27 σdB=6 N=100

alizedpacket

overhead

84 Packet overhead

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=20

End-to-E

ndDelay

2 3 4 5 6 7 8 9 10

D=500m β=27 mindp=01 σdB=6 N=100

End-to-E

ndDelay

85 End-To-End Delay

9 Conclusion

References

doi101109MCOM20051509968

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

101016jjnca201108003

429ndash439

Appl 7 (2002) 441ndash453

(2007) 551 ndash562

1298 ndash1302 vol3

200711016

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

jnca201107012

20105683490

1 Introduction

2 Related work




31 Network Model


33 Forwarding Set


35 Data Structures








8 Numerical Results



83 Forwarding cost

84 Packet overhead

85 End-To-End Delay

9 Conclusion

Documents

Multicast Delivery Using Opportunistic Routing in Wireless Mesh … · 2016. 10. 19. · Multi-hop wireless networks (MWNs) [1,2] have be- ... some attempts to achieve a high performance