E cient packet replication control for a geographical routing...

Scientia Iranica B (2015) 22(4), 1517{1533

Sharif University of TechnologyScientia Iranica

Transactions B: Mechanical Engineeringwww.scientiairanica.com

E�cient packet replication control for a geographicalrouting protocol in sparse vehicular delay tolerantnetworks

I.-C. Changa;�, C.-H. Lia and C.-F. Choub

a. Department of Computer Science and Information Engineering, National Changhua University of Education, Changhua, Taiwan.b. Department of Computer Science and Information Engineering, National Taiwan University, Taipei, Taiwan.

Received 22 May 2014; received in revised form 20 October 2014; accepted 2 December 2014

KEYWORDSVehicular ad hocnetwork;Vehicular delaytolerant network;Controlled replication;Binary spraying;IG-Ferry;Delay evaluationfunction.

Abstract. To date, many vehicular ad hoc network unicast routing protocols have beenproposed to support e�cient packet transmission between vehicles in urban environments.However, when there is insu�cient vehicle density during non-rush hour times, the vehicularad hoc network is often intermittently connected. These unicast routing protocols,therefore, perform poorly when forwarding packets over this vehicular disruption tolerantnetwork. This paper adopts the controlled replication approach, in a proposed IG-Ferryrouting protocol, to spray a limited number of packet copies, denoted by packet tokenvalues, to relay vehicles in a vehicular disruption tolerant network. We then identify threekinds of relay vehicle, i.e. direct buses, non-direct buses and private cars according to theirtravel itineraries. Based on the proposed delay evaluation function for the three types ofintermediate vehicle, the IG-Ferry packet spraying mechanism, instead of that of traditionalbinary spraying, can e�ciently spray appropriate packet tokens to vehicles. Finally,intensive NS2 simulations are conducted using the realistic Shanghai city vehicle tra�ctrace, IEEE 802.11p protocol, with EDCA and the Nakagami radio propagation model,to show that IG-Ferry outperforms three well-known VDTN routing protocols, in termsof average packet delivery ratios, end-to-end transmission delays and packet replicationoverheads, with respect to various combinations of �ve communication parameters.© 2015 Sharif University of Technology. All rights reserved.

1. Introduction

In order to achieve e�cient unicast routing in vehicularad hoc networks (VANETs), tra�c information, suchas positions, direction of movement, speed and distri-bution of all vehicles and real-time tra�c events, areessential to derive optimal routes for multi-hop wirelesspacket transmission. Well-known position-based rout-ing protocols, like Greedy Perimeter Stateless Routing

*. Corresponding author. Tel.: +886-4-7232105;Fax: +886-4-7211284E-mail addresses: [email protected] (I.-C. Chang);[email protected] (C.-H. Li); [email protected](C.-F. Chou)

(GPSR) [1] and Greedy Perimeter Coordinator Rout-ing (GPCR) [2], use the greedy forwarding approachfor an intermediate node to forward a packet to itsdirect neighbor that is known in real-time and whichis closest to the geographic position of the destina-tion. However, both of them must execute the repairstrategy, as with the perimeter mode of GPSR, toescape the local maximum [1] on the path where greedyforwarding fails in real-time. Vehicle-Assisted Data De-livery (VADD) [3], Road-Based using Vehicular Tra�c(RBVT) [4], GeoCross [5], Intersection Graph (IG) [6]etc. further improve the unicast routing performancein VANET. However, when there is low tra�c densityduring non-rush hour periods, the network is oftennot fully connected, with intermittent and opportunis-

1518 I.-C. Chang et al./Scientia Iranica, Transactions B: Mechanical Engineering 22 (2015) 1517{1533

tic connectivity. These unicast routing protocols,therefore, exhibit poor packet forwarding performanceover these kinds of intermittently connected mobilead hoc networks, which are called Disruption TolerantNetworks or Delay Tolerant Networks (DTN) [7,8].

DTNs feature sparse and intermittent connectiv-ity, long and variable delay, high latency, high errorrates, and no stable end-to-end path [9]. To achievepacket transmission between source and destinationnodes over a DTN, two types of routing approachhave been proposed to overcome the characteristicsof DTNs. One is the single-copy protocol thatnever replicates a packet [10], and the other is themulti-copy protocol that does replicate packets [11].The single-copy protocols, like Message Ferrying [12]and GeOpps [13], usually adopt the Store-Carry-and-Forward (SCF) technique [14] to keep only a single copyof a packet in the DTN at any given time. They, there-fore, introduce a very low packet delivery ratio, but ahigh end-to-end packet delay. Conversely, nodes usingEpidemic routing [15], one of the multi-copy protocols,continuously replicate and transmit packets to newlydiscovered nodes that have not already received a copyof the packet. Epidemic routing can, thus, achieve thehighest packet delivery ratio in DTNs by adopting thiskind of uncontrolled replication approach. However,replicating packets without any control is extremelywasteful in terms of wireless bandwidth and bu�erspace.

Consequently, several protocols, like Spray andWait [16], Spray and Focus [17], Selectively MAk-ingpRogress Toward delivery (SMART) [18] andGeoSpray [19], have adopted the controlled replicationapproach to spray a small, �xed number of packetcopies to di�erent relay nodes. The source or therelay node uses the binary spraying scheme [16] toopportunistically forward one-half of the carried packetcopies to a new contact until it meets the destination.This approach has advantages in terms of reducingthe enormous resource overhead of Epidemic routing.However, protocols adopting the controlled replicationapproach may su�er from low delivery ratios, longtransmission delays and/or require extra space forstoring needed information, compared to Epidemicrouting.

When further considering packet routing amongvehicular nodes in Vehicular Delay Tolerant Networks(VDTN) [20], di�erent types of vehicle may own hetero-geneous information about their movements and cur-rent geographical locations. For example, public busesand trains know their current movement directions,schedules, stops, and their maximal allowed speeds,etc. on their strictly prede�ned itineraries. However,taxis will not necessarily move along a �xed route, evenwhen driving to a prede�ned destination. Additionally,privately owned vehicles, with or without navigation

systems, will not necessarily follow a planned route, ormay change their destinations en route [21]. Becausetraditional DTN spraying protocols do not considerthe heterogeneous characteristics of these vehicles,they cannot achieve optimal routing performances forVDTN.

In this paper, intermediate vehicles, which meetthe vehicle carrying the packet copy, are classi�ed intothree categories in VDTN, according to their movementitineraries. The �rst type is the direct bus, which canmove from the contacted position to the destination.The second type is the non-direct bus, which does notleave for the destination from the contacted position.The third type is the private car, which may changeits destination while travelling. Major contributions ofthis paper are listed as follows:

1. By extending the controlled multi-copy replicationapproach over the intermittently connected VDTN,we will propose the IG-Ferry protocol, instead ofthat of traditional binary spraying, to e�cientlyspray appropriate packet tokens to the maximumnumber of relay vehicles.

2. The IG-Ferry packet spraying mechanism dependson the proposed Delay Evaluation Function (DEF)for three types of intermediate vehicle.

3. Due to short contact durations and limited wirelessbandwidth between vehicles within wireless trans-mission range, we also propose the shortest remain-ing Time-To-Live (TTL) �rst packet schedulingmechanism to transfer the restricted amount ofpackets in a contact opportunity.

4. Based on the geographical location and mobilityinformation of heterogeneous vehicles, IG-Ferry cansigni�cantly improve its average packet deliveryratio, reduce its average end-to-end delay anddecrease its average replication overhead, comparedto traditional replication-based or non-replication-based routing protocols.

The remainder of this paper is organized as follows.Related work is compared in Section 2. Details ofIG-Ferry are described in Section 3. In Section 4,NS2 simulations are conducted to show that IG-Ferryoutperforms four well-known VDTN protocols, i.e.Epidemic, Spray and Focus, SMART and GeoSprayin terms of average packet delivery ratios, end-to-enddelays and replication overheads. Finally, conclusionsand suggestions for future work are given in Section 5.

2. Related work

As mentioned above, the controlled replication ap-proach is able to reduce the enormous resource over-head introduced by the uncontrolled packet replicationof Epidemic routing. Speci�cally, the Spray and Wait

I.-C. Chang et al./Scientia Iranica, Transactions B: Mechanical Engineering 22 (2015) 1517{1533 1519

scheme initially generates L packet copies for everypacket originating at a source node. The source orthe relay node uses the binary spraying scheme [16] toopportunistically forward one-half of the carried packetcopies to a new contact in the spray phase. In thewait phase of Spray and Wait, at most, L relay nodescarrying a packet copy forward the copy only to itsdestination. However, the destination su�ers from alow packet delivery ratio because the relay itself maynot move into the wireless transmission range of thedestination; even so, the packet transmission delay willbe signi�cant. Therefore, the relay in the focus phaseof the Spray and Focus scheme can forward its copy toa further relay, depending on the value calculated by autility function, rather than waiting for the destinationto be encountered. In addition, Spray and Focusadopts the value of the forwarding token to representthe number of packet copies carried by a relay, whichreduces a relay bu�er space for storing multiple packetcopies.

Selectively MAking pRogress Toward delivery(SMART) adopts repeated mobility patterns of mobilenodes, i.e. encounter histories with other nodes, to de-termine the destination's travel companions, i.e. nodesthat frequently encounter the destination. The sourcenode �rst injects a �xed number of packet copies intothe network to opportunistically forward the packet tothe destination's companions. When the packet reachesa companion, this companion only forwards receivedpackets to other companions, instead of all contactednodes. SMART therefore achieves a higher delivery ra-tio and lower delivery latency than schemes like Sprayand Focus, which only use controlled opportunistically-forwarding mechanisms. However, SMART needs a

lot of space to record encounter histories with othernodes.

GeoSpray makes routing decisions based on ge-ographical location data provided by a positioningdevice like GPS. It assumes that each vehicle has tomove along a route suggested by its navigation systemto determine the Nearest Point (NP) on its route to thedestination. Therefore, the Minimum Estimated Timeof Delivery (METD) of the vehicle to the destinationis equal to the sum of the time from the vehicle to theNP, and from the NP to the destination. GeoSpraystarts with a multiple-copy scheme to spread a limitednumber of packet copies, and then switches to a single-copy forwarding scheme to seek additional contactopportunities. Instead of performing opportunisticforwarding, as proposed in Spray and Wait and Sprayand Focus, GeoSpray also adopts the binary sprayingscheme to guarantee that one half of the packet copiesis only spread to intermediate vehicles that are closerto the packet's destination vehicle, i.e. intermediatevehicles that have smaller METDs than those of thecurrent vehicle. However, not all vehicles have on-board navigation systems to plan their routes in ad-vance and then calculate corresponding METDs; theGeoSpray routing mechanism may fail in these cases.

Table 1 lists important characteristics of multi-copy routing protocols for VDTN. The �rst featureis whether the token is used to replace the packetcopy. The protocols use of tokens can reduce consumedwireless bandwidth and bu�er spaces to spray andstore packet copies, especially with a large number ofmaximum allowed replicable packet copies. Vehicletra�c information, including geographical positions,real-time movement information like movement direc-

Table 1. Characteristics of multi-copy routing protocols for VDTNs.

CharacteristicsToken

(instead ofthe packet

copy)

Drivingitinerary

Vehicletype

Geographicalinformation

Real-timemovement

information

Speci�c(utility)function

Sprayingmechanism

Pro

toco

ls

IG-Ferry X X X X X X(DEF)

Four cases(depending ondriving itinerary,vehicle type,DEF value)

GeoSpray X X X X(METD)

Binary spraying

SMART X X Binary sprayingSpray and focus X X Binary sprayingSpray and wait Binary spraying

Epidemic Uncontrolledreplication


tions, instantaneous speeds and driving itineraries,is essential in deriving optimal routes for VANETs.As the input to a speci�c decision function, thisinformation also helps the vehicle to select the bestcandidate neighbor to e�ciently spray the packet. Ad-ditionally, di�erent types of vehicle have various drivingbehaviors and itineraries. The binary spraying schemeused in traditional controlled replication protocols doesnot consider these vehicle heterogeneities. Therefore,protocols with binary packet spraying and replicationamong vehicles in VDTNs are de�cient. Comparedto these well-known multi-copy routing protocols, theproposed IG-Ferry features important characteristicsto further improve routing performance in VDTNs.Details of the IG-Ferry will be described below.

3. IG-Ferry routing protocol

3.1. IG-Ferry conceptsIG-Ferry consists of three important design concepts,described below:

1. IG-Ferry classi�es contacted vehicles into threetypes, i.e. direct buses, non-direct buses and pri-vate cars according to their heterogeneous mobilityinformation. It further adopts di�erent token repli-cation mechanisms for them to e�ciently forwardpacket copies. For example, when the current relayvehicle contacts a direct bus, it will forward itscarried packets with the appropriate token value,at least one, to that direct bus. In the worst casescenario, even when all other relay vehicles fail toreach the destination, this replication mechanismstill guarantees that the direct bus can carry thepacket with the token value by itself to reach thedestination. Details of replication mechanisms aredescribed below.

2. IG-Ferry records the number of packet copies car-ried by a relay vehicle, as the value of the forwardingtoken in the packet header, as Spray and Focus,in order to reduce the relay's required bu�er spacefor storing multiple packet copies and consumedwireless bandwidth for their exchange.

3. IG-Ferry proposes the DEF function to e�cientlyspray packet token values to the maximum numberof appropriate relay vehicles, depending on theirtype, geographical location and mobility informa-tion. Thus, IG-Ferry yields better performancethan traditional multiple-copy routing protocols.

3.2. IG-Ferry delay evaluation functionIn order to spray more token values to vehicle that canreach the destination soonest, we propose the DelayEvaluation Function (DEF), as shown in Figure 1, toestimate how much delay time the vehicle will requireto carry the packet token from the current location

to the destination in the worst case scenario. Thedelay time consists of two parts. Assume that thecurrent relay vehicle, which could be a public bus Bkor a private car Ck, enters the wireless communica-tion range of the contacted vehicle Bl=Cl. Distknextand Distlnext are de�ned as the curve-metric distancesfrom the current contacted location, where Bk=Ckand Bl=Cl meet, to the �rst intersections, Iknext andI lnext, that Bk=Ck and Bl=Cl will reach, respectively.These curve-metric distances are measured accordingto geometric shape [22]. Thus, Bk=Ck and Bl=Cl mustspend the �rst part of the delay time, i.e. T knext andT lnext, carrying the packet token by themselves to Iknextand I lnext. Eq. (1) formulates T knext and T lnext, whereV k and V l denote the average speeds of Bk=Ck andBl=Cl, respectively.

T knext =DistknextV k

; T lnext =DistlnextV l

: (1)

If Bl=Cl follows a path, i.e. I lnext ! � � � ! Im ! In !� � � ! Id, from the �rst intersection, I lnext, to the closestintersection, Id, which directly connects to the roadsegment on which the destination is currently situated,it must spend the time, Distm;n=Vm;n, carrying thepacket token by itself to pass the curve-metric distance,Distm;n, of each road segment, Rm;n, where Vm;ndenotes the average speed of Bl=Cl on road segmentRm;n. Therefore, Bl=Cl must spend the second part oftime, T lcur, formulated by Eq. (2), carrying the packettoken by itself along each road segment of this path tothe destination, where the set, RSl, contains each roadsegment, Rm;n.

T kcur =X

8Rm;n2RSkDistm;nVm;n

;

T lcur =X

8Rm;n2RSlDistm;nVm;n

: (2)

Thus, Bk=Ck also uses Eq. (2) to calculate the secondpart of time, T kcur. Finally, the estimated DEF values,i.e., T k and T l, of Bk=Ck and Bl=Cl, are equal to thesums of the two parts of their delay times, formulatedas T knext + T kcur and T lnext + T lcur, respectively, and asshown in Eq. (3):

T k = T knext + Tkcur; T

l = T lnext + Tlcur: (3)

In addition, we propose di�erent approaches to calcu-late curve-metric distances for three types of vehiclein VDTNs. This curve-metric distance represents themaximum estimated distance for the current vehicleto carry the packet tokens by itself to reach thedestination.


Figure 1. The delay evaluation function for vehicle k at its current location loc.

1. If the vehicle is a direct bus, its curve-metricdistance, i.e. Distcur, is equal to the total lengthof the road segments along its itinerary from the�rst intersection, Inext, after the contact, to theclosest intersection, Id, of the destination. Asshown in the upper right part of Figure 2, thoughtwo direct buses, Bk and Bl, drive to the same�rst intersection, I15, their itineraries after I15contain di�erent intersections, i.e. I12, I7 and IEfor Bk, and I16, I13, I8 and ID for Bl, to reach theclosest intersection to the destination. Thus, theircurve-metric distances, i.e. Distkcur and Dist

lcur, are

calculated along their individual itineraries.

2. If the vehicle is a non-direct bus, it cannot carry thepacket tokens by itself to the destination. Thus,its curve-metric distance, Distcur, is estimated asthe sum of two parts. The �rst part is the curve-metric distance from the �rst intersection of thevehicle to the Nearest Point (NP) on its pre-de�neditinerary to the destination. The second part is thedistance the packet token is carried by the relay

vehicle r from the NP to the destination. Here,the NP is considered the new starting locationin order to estimate the curve-metric distance ofthe second part. Data-mining schemes, like Se-mantic Trajectory Mining [23], etc., are usuallyadopted to calculate contact probabilities betweentwo vehicles. Hence, they could be used hereto predict the closest vehicle, which will contactthe current packet-carried vehicle with the highestcontact probability to the NP, as the relay vehicler. As shown in Figure 2, the non-direct bus, Bl,between I10 and I11 can carry the packet tokenby itself as far as the NP, i.e. I2, along its pre-de�ned itinerary, which is shown as the orangedashed line beside Bl. Then, Bl replicates thepacket tokens at I2 to the closest contact vehicle,which is moving toward I3. In this way, thepacket tokens can reach the closest intersectionIE, and, �nally, the destination. The curve-metricdistance, Distlcur, of Bl is, therefore, estimated asthe total length of the road segments along thegreen path.


Figure 2. DEF examples for three types of vehicles.

3. According to results observed in real life, mostdriver trips were duplicated. Hence, sometrajectory-based scheme, like the Shared-Trajectory-based Data Forwarding Scheme (STDFS) [24], His-tory Based Predictive Routing (HBPR) [25], etc.can be used in this paper to predict the routes ofprivate cars. Hence, if the vehicle is a private car, itscurve-metric distance, Distcur, is estimated as thesum of the following three parts. The �rst part isthe curve-metric distance from the �rst intersectionto one of its adjacent intersections; the secondis the curve-metric distance of a path startingfrom this adjacent intersection to the Nearest Point(NP) on the vehicle's predicted itinerary to thedestination; the third part is the distance the packettoken is carried by the closest contact vehicle rfrom the NP to the destination. As shown inFigure 1, the �rst intersection along the route ofCk is I11, which has three adjacent intersections,i.e., I12, I14 and I6. The curve-metric path fromI11 to the closest intersection, i.e. ID or IE, ofthe destination must pass by one of these threecandidate intersections, which is I6 in Figure 2.Then, Ck carries the packet token by itself, as faras the NP, i.e. I3, along its itinerary, predictedby the trajectory-based scheme. After that, Ckreplicates the packet tokens at I3 to the closest con-tact vehicle moving toward the closest intersection,IE, such that the packet tokens can �nally reachthe destination. Consequently, the curve-metricdistance, Distkcur, of Ck is, therefore, estimated asthe total length of all the road segments alongthe purple path, consisting of I11, I6, I3 andIE.

Ekl, in Eq. (4), is used to express the DEF ratioby dividing the DEF value of Bl=Cl by that of Bk=Ck

upon an encounter.

Ekl =T l

T k: (4)

If Ekl is smaller than 1, i.e., T l is smaller thanT k, Bl=Cl will incur a lower estimated delay thanBk=Ck to carry the packet token by itself to thedestination, which means that Bl=Cl is likely to reachthe destination sooner, and the IG-Ferry will, therefore,spray more packet tokens to Bl=Cl. Details of the IG-Ferry spraying mechanism are described below.

3.3. IG-Ferry protocol owThe IG-Ferry protocol starts when any vehicle thatdoes not carry the packet token makes contact with thesource vehicle, i.e. Csrc. In Figure 3, Csrc is locatedbetween intersections I11 and I12. According to thebinary spraying scheme, Csrc forwards one-half of itscarried packet tokens to this vehicle to be relayed tothe destination vehicle, i.e. Cdst. The initial TimeTo Live (TTL) values are assigned to these packets torepresent the maximal remaining lifetimes of carriedpackets. After the packet tokens have been carried bythe �rst relay vehicle, there are four cases for sprayingpacket tokens between the current relay vehicle and thecontacted vehicle, depending on vehicle type. Whenthe carried packet is replicated or forwarded to acontacted vehicle, the TTL value of this packet isdecreased by the time elapsed since the last packetreplication or forwarding. As soon as the TTL valueof a carried packet is reduced to zero, the packetshould be dropped from the bu�er. During the limitedcontact period, two vehicles adopt the largest TTL �rstpacket scheduling mechanism to exchange their carriedpackets. An example of packet token spraying betweentwo vehicles is shown in Figure 3. Figure 4 illustratesthe complete IG-Ferry packet token spraying ow.


Figure 3. An example of packet token spraying between two vehicles.

Figure 4. The IG-Ferry packet token spraying ow.

Case 1: When a direct bus carrying packet tokensmeets a direct bus.

As shown in Case 1 of Figure 3, when directbus, Bk, which is carrying the packet with the tokenvalue, TOk, encounters direct bus Bl, which can driveto one of the closest intersections, i.e., I4 and I5, todestination Cdst along its itinerary, Bk and Bl �rstexchange a list of carried packets and corresponding

token values through the HELLO message. There aretwo di�erent situations for spraying the packet tokensbetween these two vehicles. First, if Bk and Bl carrythe same packets with token values TOk and TOl,both will re-spray all packet tokens, i.e., TOk + TOl,according to Eq. (5), to Bk and Bl, respectively. Thevehicle with the smallest estimated DEF value is likelyto reach the destination soonest.


8>:TOk =

l(TOk + TOl)� � T lTk+T l�m

TOl =j(TOk + TOl)� � TkTk+T l�k (5)

Thus, Eq. (5) implies that the direct bus with thesmaller estimated DEF value will carry larger packettoken values in order to spray them to more relayvehicles. In this way, end-to-end transmission delaysfor these packets to arrive at the destination can bereduced. Second, if Bl does not carry the same packetsas Bk, Bk will replicate these packets with appropriatetoken values to Bl. The IG-Ferry uses the followingrule, which is formulated as Eq. (6), to re-spray tokenvalues between them.8>>>>>>>>>>>>>>>>>>>>>>>>>:

(TOk = 1TOl = TOk � 1 ; if E

kl < 1

(TOk = TOk=2TOl = TOk=2

; if Ekl = 1

(TOk = TOk � 1TOl = 1

; if Ekl > 1

(6)

If Ekl < 1, which means that the estimated DEF, i.e.T l of Bl, is smaller than that of Bk, i.e. T k, the packettoken values of Bk and Bl are assigned to 1 and TOk�1, respectively. Conversely, if Ekl > 1, the packet tokenvalues of Bk and Bl are assigned as TOk � 1 and 1,respectively. However, if Ekl = 1, the binary sprayingapproach is used to equally spray token values betweenBk and Bl. In this way, the IG-Ferry packet deliveryratio can be improved by allowing each direct bus tocarry at least one packet token to the destination.

Case 2: When a direct bus carrying packet tokensmeets a non-direct bus or a private car.

As shown in Case 2 of Figure 3, when direct bus,Bk, which is carrying the packet with the token value,TOk, encounters a non-direct bus, Bl, or a private car,Cl, denoted as Bl=Cl, Bk and Bl=Cl, �rst, exchange alist of carried packets and corresponding token valuesthrough the HELLO message, as in Case 1. There arealso two di�erent situations for spraying the packettokens between these two vehicles. First, if Bk andBl=Cl carry the same packets with token values TOkand TOl, both will re-spray all packet tokens, i.e.TOk + TOl, to Bk and Bl=Cl with the same rule,i.e. Eq. (5), as Case 1. Second, as mentioned above,if Ekl < 1, the estimated DEF, i.e. T l of Bl=Cl, issmaller than T k ofBk, which means thatBl=Cl is likelyto reach the destination sooner. Therefore, if Bl=Cldoes not carry the same packets as Bk and Ekl < 1,the IG-Ferry re-sprays 1 and (TOk � 1) packet tokens

to Bk and Bl=Cl, respectively. Conversely, if Ekl > 1,which means that the current direct bus, Bk, is likely toreach the destination sooner, Bk keeps all packet tokensand does not need to replicate any packets to Bl=Cl.As in Case 1, the binary spraying approach is used toequally spray token values between Bk and Bl=Cl, ifEkl = 1. With the rule formulated in Eq. (7), the IG-Ferry reduces the total number of replicated packetsamong vehicles in VDTNs by only re-spraying packettokens to the direct bus, which reaches the destinationby itself, or to the non-direct bus or private car, whichwill reach the destination sooner than the current relayvehicle.

If Bl=Cl does not carry the same packet of Bk:8>>>>>>>>>>>>>>>>>>>>>>>>>:

(TOk = 1TOl = TOk � 1 ; if E

kl < 1


; if Ekl = 1

(TOk = TOk

TOl = 0; if Ekl > 1

(7)

Case 3: When a non-direct bus or a private carcarrying packet tokens meets a direct bus.

As shown in Case 3 of Figure 3, when the non-direct bus or private car, Bk=Ck, carrying the packetwith the token value, TOk, encounters direct bus, Bl,Bk=Ck and direct bus Bl will exchange their lists ofcarried packets and corresponding token values throughthe HELLO message. After this, they will re-spraytheir packet tokens with the following two rules. First,if Bk=Ck and Bl carry the same packets with tokenvalues, TOk and TOl, both will re-assign all packettokens, i.e. TOk+TOl, to Bk=Ck and Bl, with Eq. (5)as Case 1. Second, if Bl does not carry the samepackets as Bk=Ck and Ekl < 1,Bk=Ck, will forwardthe carried packets with all token values, i.e. TOk, toBl, because Bl is a direct bus and is likely to reachthe destination sooner than Bk=Ck. Then, Bk=Ckwill clear these packets from its bu�er. Conversely,if Ekl > 1, which means that current Bk=Ck is likelyto reach the destination sooner than Bl, Bk=Ck mustreplicate each packet to direct bus Bl with the tokenvalue of 1. It then modi�es the token values of thesereplicated packets to TOk � 1. Consequently, theIG-Ferry packet delivery ratio can be increased byallowing Bl to carry a packet copy to the destinationwith the aforementioned rule, which is formulated asEq. (8).

If Bl does not carry the same packet of Bk=Ck:


8>>>>>>>>>>>>>>>>>>>>>>>>>:

(TOk = 0TOl = TOk

; if Ekl < 1


; if Ekl = 1

(TOk = TOk � 1TOl = 1

; if Ekl > 1

(8)

Case 4: When a non-direct bus or private car car-rying packet tokens meets a non-direct bus or privatecar.

As in the three above cases, after Bk=Ck andBl=Cl exchange their lists of carried packets and cor-responding token values through the HELLO message,as shown in Case 4 of Figure 3, they will re-spray theirpacket tokens with the following two rules. First, ifBk=Ck and Bl=Cl carry the same packets with tokenvalues, TOk and TOl, both, will re-assign all packettokens, i.e. TOk + TOl, to Bk=Ck and Bl=Cl withEq. (5) as Case 1. Second, the token spraying rule ofthis case is formulated as Eq. (9) if Bl=Cl does notcarry the same packets of Bk=Ck.8>>>>>>>>>>>>>>>>>>>>>>>>>:

(TOk = 0TOl = TOk

; if Ekl < 1


; if Ekl = 1

(TOk = TOk

TOl = 0; if Ekl > 1

(9)

If Ekl < 1, Bk=Ck will forward carried packets with alltoken values, i.e. TOk, to Bl=Cl; it is likely to reachthe destination sooner than Bk=Ck. Then, Bk=Ck willclear these packets from its bu�er. Conversely, if Ekl >1, which means that current Bk=Ck is likely to reachthe destination sooner than Bl=Cl, Bk=Ck does notneed to replicate packets to Bl=Cl because Bl=Cl isnot guaranteed to reach the destination by itself. Inthis way, the total number of replicated packets in theIG-Ferry is reduced.

4. Simulations

4.1. Simulation environmentIn [26], the authors derived �ve core aspects de�ningInter-Vehicle Communication (IVC) simulations, i.e.network simulator, radio propagation model, mediumaccess protocol, road tra�c mobility and scenariodescription. First, we adopt the well-known network

simulator, NS-2.34 [27], to perform simulations in thispaper. Second, Nakagami-m [28] is a mathematicalgeneral modeling of a radio channel with fading. Byvarying the shape factor m value, a high fadingscenario, like a city environment or a freeway orhighway, can be formed [29]. As the radio propaga-tion model used in [30-32] for VANETs, Nakagami-3, which has been supported by NS-2.34, is adoptedin our simulations. Some recent work [33,34] hasfound that obstacles have a signi�cant e�ect when twovehicles are driving on roads separated by buildingsor vehicles. These studies tried to develop realisticpath loss models to improve the quality of wide rangeVANET simulations. Whenever NS-2 supports theseproposed models, we will re-verify the validity of thissimulation in the near future. Third, each vehicle isassumed to communicate with other vehicles by theIEEE 802.11p protocol [35] with Enhanced DistributedChannel Access (EDCA). According to the steps andparameters described in [36], we apply 802.11p to theseNS-2 simulations. The transfer rate is chosen as thelowest rate supported by 802.11p, namely 3 Mbps,and the Communication Range (CR) is set at 250m [30,31,34].

The sparse mobility settings simulate the Ve-hicular Delay-Tolerant Network (VDTN) scenario asfollows. As in [37], we also consider the real motiontraces from 1051 operational taxis for about one monthin Shanghai city, collected by GPS [38]. The locationinformation of the taxis is recorded within a 10�10 kM2area, shown in Figure 5. The itinerary, the NP andthe contact vehicle nearest the NP of every taxi, i.e.the private car in this paper, are extracted from thistrace. Further, the instantaneous speed of each taxi iscon�ned within 8-15 m/s. Based on Shanghai city businformation, there are 458 public buses that move along

Figure 5. Simulation topology of Shanghai city.


their real itineraries with velocities of 10-12 m/s. Themaximum acceleration and deceleration for the publicbuses are +0:5 m/s2 and �0:5 m/s2, respectively. Theperiod of each tra�c light and the number of lanesof each road segment are recorded in this real trace.Finally, in the 1400-second period, i.e. 8:00 pm-8:23pm on 1 March 2007 in the trace of Shanghai city [38],we conduct a realistic scenario that contains CBR owstransmitted with the UDP transport protocol betweendi�erent source-destination vehicle pairs, which areuniformly chosen from all vehicles in this area. AllCBR ows begin their transmissions at 35 secondsuntil the end of the simulation, with a default packetsize of 512 bytes, and 10 packets per second, whichintroduces transmission rates of 40 kbps. The initialTTL value of each CBR packet is set to 50 seconds indefault to represent the maximum remaining lifetimesof carried packets. As soon as the TTL value ofa carried packet reaches zero, that packet should bedropped from the default 0.6M byte bu�er of eachvehicle. Conversely, if the destination vehicle receivesa packet copy with a positive TTL value, this packet iscalled the successfully received one by the destination.Parameters used for these simulations and their defaultvalues are listed in Table 2.

In order to evaluate the transmission performanceof the aforementioned CBR/UDP tra�c with ourproposed IG-Ferry routing protocol in a VDTN, weadopt four well-known DTN routing protocols, i.e.Epidemic, Spray and Focus, SMART and GeoSpray forperformance comparison. Note that each vehicle withthe IG-Ferry �rst exchanges its carried packet with

the largest TTL value with encountered vehicles. Inthe following, we compare the average values of threeperformance metrics, i.e. Average Packet DeliveryRatio (APDR), Average End-to-End Delay (AEED)and Average Replication Overhead (ARO), for all �verouting protocols with respect to �ve parameters, i.e.the number of CBR source-destination pairs, the valueof initial tokens, the initial TTL of each packet issuedby the source, packet size and bu�er size. We conducttwenty independent runs for each evaluation in whichthe simulation time of each run is 1400 seconds, i.e.the non-rush period from 8:00 pm to 8:23 pm on 1March 2007 in the trace. The error bars in the following�gures represent the 95% con�dence intervals. APDRis de�ned as the quotient of dividing the number ofpackets successfully received by the destination overthe number of packets sent from the source vehicle, andAEED is that of dividing the total end-to-end delaysof all successfully received packets by the number ofsuccessfully received packets. In this paper, ARO isde�ned as the average value of total packet copiesreplicated over vehicles in a VANET when the TTLvalue of the packet reduces to zero for a single sourcevehicle. Therefore, the DTN routing protocol withthe smallest average replication overhead value is theone which replicates the least copies to all contactedvehicles, which means this protocol consumes the leastwireless bandwidth for spraying copies.

4.2. Simulation resultsFigures 6 and 7 show the APDRs and AEEDs achievedby the �ve VDTN routing protocols for successfully

Table 2. NS2 Simulation parameters.

Parameter Value

Simulation time 1400 seconds (8:00 pm - 8:23 pm on 1 March 2007 in the trace ofShanghai city [38])

Simulation area 10 km � 10 kmNumber of private cars 1051 (according to the trace of Shanghai city)Number of public buses 458 (according to bus information of Shanghai city)Number of CBR source-destination pairs 10, 20, 30 (default), 40, 50Number of initial tokens 100Maximal Communication Range (CR) 250 mVelocity range of private cars 8-15 m/sec (according to the trace)Velocity range of public buses 10-12 m/sec (according to the trace)CBR rate 10 packets per secondMAC protocol IEEE 802.11pPhysical propagation model Nakagami, m = 3 [30-32]IEEE 802.11p wireless bandwidth 3 Mbps [30,31,34]CBR packet size 512 bytesInitial TTL value of each CBR packet 50 secondsBu�er size of each vehicle 0.6 M bytes


Figure 6. Average packet delivery ratio vs. the numberof source-destination tra�c pairs.

Figure 7. Average end-to-end delay vs. the number ofsource-destination tra�c pairs.

received packets in Cdst. As the number of CBRtra�c pairs increases, the APDRs of the �ve VDTNrouting protocols decrease slowly, but their AEEDsincrease accordingly, due to increased collisions ofCBR packets. Because Epidemic is a ooding-basedprotocol, it achieves the highest APDRs and the lowestAEEDs. Conversely, it is extremely wasteful of wirelessbandwidth and bu�er space in its replication of thehuge number of copies over vehicles, as shown by theAROs in Figure 8. Three traditional controlled repli-cation approaches, i.e. Spray and Focus, SMART andGeoSpray, achieve signi�cant improvements on theirAROs over Epidemic by controlled opportunistically-forwarding and binary spraying mechanisms. However,they do not consider heterogeneous vehicle characteris-tics in VDTNs. These three protocols, therefore, su�erfrom much lower APDRs and higher AEEDs than Epi-demic, as shown in Figures 6 and 7, respectively. Basedon the proposed token spraying ow and estimatedDEF values for three kinds of heterogeneous vehicle,

Figure 8. Average replication overhead vs. the number ofsource-destination tra�c pairs.

Table 3. Three normalized metrics vs. source-destinationtra�c pairs.

APDR AEED ARO

Epidemic 100% 100% 100%IG-Ferry 75% 134% 7%GeoSpray 64% 194% 9%SMART 59% 210% 9%Spray and focus 54% 231% 9%

the IG-Ferry re-sprays at least one packet token to di-rect buses, and more packet tokens to the vehicle mostlikely to reach the destination sooner upon a contact,which can raise its APDRs and reduce its AEEDs, ascompared to those of Spray and Focus, SMART andGeoSpray. It also achieves lower AROs than these threecontrolled replication approaches. Note that AROs ofall protocols in Figure 8 remain stable because theyare irrelevant to the number of CBR source-destinationpairs. By dividing the average metric value (vs. source-destination tra�c pairs) of each protocol over thatof Epidemic, Table 3 lists the normalized values ofthree metrics of all protocols. It is clear that IG-Ferry outperforms GeoSpray, i.e. the best controlledreplication protocol among Spray and Focus, SMARTand GeoSpray, with 11% (75%-64%) on APDR, 60%(194%-1334%) on AEED, and 2% (9%-7%) on AROin this sparse VDTN scenario. These results showthe signi�cant performance improvements of IG-Ferryover traditional controlled replication protocols at thecost of only a 25% decrease and 34% increase on theAPDR and AEED of Epidemic, respectively. However,IG-Ferry resolves the ooding defect of Epidemic byobtaining 7% of the AROs of Epidemic.

As mentioned above, the ooding-based protocol,i.e. Epidemic, continuously replicates and transmitspackets to newly discovered nodes that have not al-


Figure 9. Average packet delivery ratio vs. the numberof initial tokens.

Figure 10. Average end-to-end delay vs. the number ofinitial tokens.

ready received a copy of the packet. Therefore, threemetrics of Epidemic are independent of the numberof initial tokens given in the source vehicle, as shownin Figures 9-11. Conversely, as the number of initialtokens increases, the APDRs and AROs of these fourcontrolled replication VDTN routing protocols, i.e.Spray and Focus, SMART, GeoSpray and IG-Ferry,increase, while their AEEDs decline accordingly, asshown in Figures 9 to 11, respectively. This is becausethey can spray more packet copies to more vehicles and,�nally, to the destination, with higher possibilities andshorter delays, when the initial token value is larger.Moreover, these four protocols have much smallerAROs than Epidemic because the maximum number ofreplicated copies over all vehicles in VDTNs is limitedto the number of initial tokens. Of these four controlledreplication VDTN routing protocols, IG-Ferry has thelargest APDRs and the lowest AEEDs/AROs due toits proposed token spraying ow. Moreover, as shown

Figure 11. Average replication overhead vs. the numberof initial tokens.

Table 4. Three normalized metrics vs. the number ofinitial tokens.

APDR AEED ARO


in Figures 9-11, respectively, its APDRs and AEEDsapproach those of Epidemic, but its AROs grow slowerthan Spray and Focus, SMART and GeoSpray whenthe token number reaches 300. This means that theIG-Ferry achieves a much better performance over thethree traditional controlled replication protocols, espe-cially when the number of tokens increases. Finally,Table 4 lists the normalized values of three metricsof all protocols. It is clear that IG-Ferry outperformsGeoSpray with 11% (81%-70%) on APDR, 57% (190%-133%) on AEED and 3% (16%-13%) on ARO, at acost of only a 19% decrease on the APDR and a 33%increase on the AEED of Epidemic, respectively. Theseresults also prove that IG-Ferry resolves the oodingdefect of Epidemic with only 13% of the AROs ofEpidemic with respect to di�erent numbers of initialtokens.

With a larger TTL value, carried packets cantravel further, such that they have a higher probabilityof reaching the destination, but a lower probability ofbeing dropped by intermediate vehicles. Thus, as theTTL value of each packet increases, the APDRs andAEEDs of the �ve VDTN routing protocols all increase,as shown in Figures 12 and 13, respectively. Note thatregardless of how large the TTL is, the maximal AROof each controlled replication VDTN routing protocolis con�ned by the number of tokens, which is 100, by


Figure 12. Average packet delivery ratio vs. TTL.

Figure 13. Average end-to-end delay vs. TTL.

default. Therefore, their AROs are relatively stableas the TTL increases, as shown in Figure 14. Onthe other hand, in the worst case, those of Epidemiccan only increase to the total number of all vehicles,consisting of public buses and private cars, when eachvehicle receives a copy of the CBR packet. In thesethree �gures, it is clear that IG-Ferry outperformsthe three traditional controlled replication protocols onthese three metrics. Table 5 lists the normalized valuesof three metrics of all protocols over those of Epidemic.It can be seen that IG-Ferry has a 12% (79%-67%)

Table 5. Three normalized metrics vs. TTL.

APDR AEED AROEpidemic 100% 100% 100%IG-Ferry 79% 146% 6%GeoSpray 67% 211% 8%SMART 63% 228% 8%Spray and focus 57% 248% 8%

Figure 14. Average replication overhead vs. TTL.

higher APDR, a 65% (211%-146%) lower AEED, anda 2% (8% - 6%) lower ARO than GeoSpray at a cost ofonly a 21% decrease on the APDR, and a 46% increaseon the AEED of Epidemic, respectively. Moreover,IG-Ferry obtains only 6% of the AROs of Epidemic,even when the TTL increases to 110 seconds, which issigni�cant for VDTN protocols.

In the following, the results of three metrics of�ve VDTN routing protocols are presented with variedpacket sizes. Because the contact time of two vehiclesmay not be long enough to replicate a large packetbetween them, the carried packets with larger packetsizes would be dropped by the intermediate vehicle andcollide with other packets during transmission withhigher possibilities before reaching the destination.Therefore, as the size of each packet increases, theAPDRs of the �ve VDTN routing protocols decrease,as shown in Figure 15. Conversely, because each

Figure 15. Average packet delivery ratio vs. the packetsize.


Figure 16. Average end-to-end delay vs. the packet size.

intermediate vehicle incurs greater delays in order toprocess and transmit larger packet copies, the AEEDsof the �ve protocols increase with the larger packetsize, as shown in Figure 16. Aside from the e�ects ofpacket dropping and collision as mentioned above, theaverage values of total packet copies of all protocolsare independent of the packet size when the TTL valueof the packet reaches zero. However, Epidemic su�ersfrom higher packet dropping and collision probabilitythan the other four controlled replication protocolsdue to its unlimited replication on contact, which getsworse with larger packet sizes. Thus, the AROs of thefour controlled replication protocols are relatively morestable than those of Epidemic, as shown in Figure 17.According to the normalized values of three metricsof all protocols listed in Table 6, the IG-Ferry hasan 11% (77%-66%) higher APDR, and a 74% (209%-135%) lower AEED than GeoSpray at a cost of only

Figure 17. Average replication overhead vs. the packetsize.

Table 6. Three normalized metrics vs. the packet size.

APDR AEED AROEpidemic 100% 100% 100%IG-Ferry 77% 135% 7%GeoSpray 66% 209% 7%SMART 61% 218% 8%Spray and focus 55% 241% 8%

Figure 18. Average packet delivery ratio vs. the bu�ersize.

a 23% decrease on the APDR and a 35% increaseon the AEED of Epidemic, respectively. Though IG-Ferry has a 1% ARO improvement, at most, over theother three traditional controlled replication protocols,all four protocols result in up to 8% of the AROs ofEpidemic with respect to the varied packet size.

Furthermore, with larger bu�er sizes, each in-termediate vehicle can carry more packet copies suchthat fewer packets will be dropped due to bu�eroverow by the intermediate vehicle before reachingthe destination. Thus, as the bu�er size of eachvehicle increases, the APDRs of the �ve VDTN routingprotocols increase, but their AEEDs decrease, as shownin Figures 18 and 19. As mentioned above, theAROs of Epidemic and the four controlled replicationprotocols, shown in Figure 20, are limited to themaximum number of all vehicles in a VDTN and thenumber of initial tokens, respectively. Consequently,the four controlled replication protocols have stableARO values under the number of initial tokens, i.e.100, but Epidemic su�ers from higher AROs as thebu�er size increases. The normalized values of threemetrics of all protocols, with respect to bu�er size,are listed in Table 7. IG-Ferry has a 12% (77%-65%)higher APDR, a 58% (182%-124%) lower AEED, anda 1% (8%-7%) lower ARO than GeoSpray, at a cost ofonly a 23% decrease and a 24% increase in the APDRand AEED of Epidemic, respectively. Moreover, IG-


Figure 19. Average end-to-end delay vs. the bu�er size.

Figure 20. Average replication overhead vs. the bu�ersize.

Table 7. Three normalized metrics vs. the bu�er size.

APDR AEED ARO


Ferry only results in 7% of the AROs of Epidemic. Wesummarize aforementioned simulation results for thethree metrics of the �ve routing protocols as follows:

1. Though the four controlled replication routing pro-tocols su�er from lower APDRs and higher AEEDsthan those of Epidemic, they signi�cantly reducetheir AROs to less than 10% of Epidemic's, whichpresents the e�ect of controlled replication for

spraying packets. Of the four controlled replicationrouting protocols, IG-Ferry outperforms the otherthree traditional ones on these three metrics, withrespect to �ve simulation parameters. It signi�-cantly achieves, on average, about 12% higher AP-DRs, 60% lower AEEDs, and 2% lower AROs thanthose of GeoSpray, which performs best among thetraditional binary spraying controlled replicationrouting protocols surveyed.

2. As any of the two parameters, i.e. the number ofinitial tokens and bu�er size, increases, all con-trolled replication routing protocols achieve higherAPDRs and lower AEEDs simultaneously in oursimulations. Conversely, they su�er from lowerAPDRs and higher AEEDs with larger tra�c pairs,TTL and packet size. However, if the numberof initial tokens and bu�er size grow too large,much heavier network tra�c and higher packetcollision/dropping/error probabilities are incurredby too many replicated packet copies generated inthe VDTN with too much signal attenuation. Inthe future, we will study optimal values of initialtokens and bu�er size for di�erent realistic VDTNtra�c scenarios, mobility models and propagatione�ects due to obstacles and obstructing vehicles.

3. No matter how large the tra�c pair, TTL, packetsize and bu�er size, the maximal ARO of eachcontrolled replication VDTN routing protocol iscon�ned by the number of initial tokens. However,the AROs of Epidemic may grow to the totalnumber of all vehicles in the worst case scenario.We, therefore, conclude that the number of initialtokens is the most important simulation parameterof the �ve examined in this simulation.

5. Conclusion and future work

In this paper, we proposed the IG-Ferry routing pro-tocol for e�ciently replicating packet copies amongvehicles in a VDTN. IG-Ferry classi�es contactedvehicles into three types, and re-sprays packet tokensto di�erent types of contacted vehicles according totheir estimated end-to-end delays calculated by theproposed delay evaluation function. We conductedNS2 simulations of a sparse urban environment, basedon the realistic vehicle tra�c trace of Shanghai city,IEEE 802.11p protocol, with EDCA and the Nakagamiradio propagation model. Simulation results showedsigni�cant performance improvements over three well-known binary spraying DTN routing protocols onaverage packet delivery ratios, average end-to-end de-lays and average replication overheads with respectto �ve parameters. Consequently, IG-Ferry facilitateshigher usability on the controlled packet replicationmechanism than traditional approaches in VDTN.


To extend the results of this paper, there aretwo major pieces of research work for the near future.First, we will study how to predict the exact routeof a private car, and then propose an accurate routeestimation module in DEF to further improve theperformance of the IG-Ferry, according to several realtrace data of private cars. Second, we will study opti-mal values of communication parameters for di�erentrealistic VDTN tra�c scenarios, mobility models andpropagation e�ects, due to obstacles and obstructingvehicles in urban environments.

Acknowledgment

This work was supported by the National ScienceCouncil (NCS), Taiwan, under Grant Number NSC100-2221-E-018-018.

References

1. Karp, B. and Kung, H. \GPSR: Greedy perimeterstateless routing for wireless networks", 6th ACMMobile Computing and Networking (2000).

2. Lochert, C., Mauve, M., Fussler, H. and Hartenstein,H. \Geographic routing in city scenarios", ACM Mo-bile Computing and Communications Review, 9(1), pp.69-72 (2005).

3. Zhao, J. and Cao, G. \VADD: Vehicle-assisted datadelivery in vehicular ad hoc networks", IEEE Trans-actions on Vehicular Technology, 57(3), pp. 1910-1922(2008).

4. Nzouonta, J., Rajgure, N., Wang, G. and Borcea,C. \VANET routing on city roads using real-timevehicular tra�c information", IEEE Transactions onVehicular Technology, 58(7), pp. 3609-3626 (2009).

5. Lee, K-C., Cheng, P-C. and Gerla, M. \GeoCross: Ageographic routing protocol in the presence of loops inurban scenarios", Ad Hoc Networks, 8(5), pp. 474-488(2010).

6. Chang, I-C., Wang, Y-F. and Chou, C-F. \E�cientVANET unicast routing using historical and real-timetra�c information", 17th IEEE International Confer-ence on Parallel and Distributed Systems (ICPADS2011), Tainan, Taiwan (2011).

7. Cerf, V., Burleigh, S., Hooke, A., et al. \Delay-tolerantnetworking architecture", IETF RFC 4838 (2007).

8. Fall, K. \A delay-tolerant network architecture forchallenged Internets", The Conference on Applica-tions, Technologies, Architectures, and Protocols forComputer Communications (SIGCOMM '03) (2003).

9. Gong, H. and Yu, L. \Study on routing protocols fordelay tolerant mobile networks", International Journalof Distributed Sensor Networks, Article ID 145727, 16pages (2013). DOI:10.1155/2013/145727

10. Spyropoulos, T., Psounis, K. and Raghavendra, C.S.\E�cient routing in intermittently connected mobile

networks: The single-copy case", IEEE/ACM Trans-actions on Networking, 16(1), pp. 63-76 (2008).

11. Spyropoulos, T., Psounis, K. and Raghavendra, C.S.\E�cient routing in intermittently connected mo-bile networks: The multiple-copy case", IEEE/ACMTransactions on Networking, 16(1), pp. 77-90 (2008).

12. Zhao, W., Ammar, M. and Zegura, E. \A messageferrying approach for data delivery in sparse mobilead hoc networks", 5th ACM International Symposiumon Mobile Ad Hoc Networking and Computing (2004).

13. Leontiadis, I. and Mascolo, C. \GeOpps: geographicalopportunistic routing for vehicular networks", IEEEInternational Symposium on World of Wireless, Mo-bile and Multimedia Networks (2007).

14. Zhang, Z. \Routing in intermittently connected mo-bile ad hoc networks and delay tolerant networks:Overview and challenges", IEEE CommunicationsSurveys & Tutorials, 8(1), pp. 24-37 (2006).

15. Vahdat, A. and Becker, D. \Epidemic routing forpartially-connected ad hoc networks", Tech. Rep.,Duke University (2000).

16. Spyropoulos, T., Psounis, K. and Raghavendra, C.S.\Spray and wait: An e�cient routing scheme forintermittently connected mobile networks", ACMSIGCOMM Workshop on Delay-Tolerant Networking(2005).

17. Spyropoulos, T., Psounis, K. and Raghavendra, C.S.\Spray and focus: E�cient mobility-assisted rout-ing for heterogeneous and correlated mobility", FifthAnnual IEEE Conference on International PervasiveComputing and Communications Workshops (2007).

18. Tang, L., Zheng, Q., Liu, J. and Hong, X. \Selec-tive message forwarding in delay tolerant networks",Mobile Networks and Applications, 14(4), pp. 387-400(2009).

19. Soares, V.N.G., Rodrigues, J.P.C. and Farid, F.\GeoSpray: Ageographic routing protocol for vehic-ular delay-tolerant networks", Information Fushion(2011).

20. Pereira, P., Casaca, A., Rodrigues, J., et al. \Fromdelay-tolerant networks to vehicular delay-tolerantnetworks", IEEE Communications Surveys & Tutori-als, 14(4), pp. 1166-1182 (2012).

21. Cheng, P.-C., Lee, K.-C., Gerla, M. and Harri, J.\GeoDTN+Nav: Geographic DTN routing with nav-igator prediction for urban vehicular environments",Mobile Networks and Applications, 15(1), pp. 61-82(2010).

22. Li, H., Guo, A. and Li, G. \Geographic and tra�cload based routing strategy for VANET in urban tra�cenvironment", IET 3rd International Conference onWireless, Mobile and Multimedia Networks (ICWMNN2010) (2010).

23. Ying, J., Lee, W., Weng, T. et al. \Semantic trajec-tory mining for location prediction", The 19th ACMSIGSPATIAL International Conference on Advancesin Geographic Information Systems (2011).


24. Xu, F., Guo, S., Jeong, J. et al. \Utilizing sharedvehicle trajectories for data forwarding in vehicularnetworks", IEEE INFOCOM (2011).

25. Huang, Z., Wang, S., Ding, B. et al. \Historybased predictive routing in multi-lane delay tolerableVANETs", IEEE International Conference on Com-munications (ICC) (2012).

26. Joerer, S., Sommer, C. and Dressler, F. \Towardreproducibility and comparability of IVC simulationstudies: A literature survey", IEEE CommunicationsMagazine, pp. 82-88 (2012).

27. Browse nsnam�les on SourceForge. net, http:// source-forge.net/projects/nsnam/ �les/.

28. Beaulieu, N.C. and Cheng, C. \E�cient nakagami-m fading channel simulation", IEEE Transactions onVehicular Technology, 54(2), pp. 413-424 (2005).

29. Bi, Y., Cai, L.X., Shen, X. and Zhao, H. \E�cientand reliable broadcast in intervehicle communicationnetworks: Across-layer approach", IEEE Transactionson Vehicular Technology, 59(5), pp. 2404-2417 (2010).

30. Li, M., Yang, Z. and Lou, W. \CodeOn: Cooperativepopular content distribution for vehicular networksusing symbol level network coding", IEEE Journal onSelected Areas in Communications, 29(1), pp. 223-235(2011).

31. Yang, Z., Li, M. and Lou, W. \CodePlay: Livemultimedia streaming in VANETs using symbol-levelnetwork coding", IEEE Transactions on WirelessCommunications, 11(8), pp. 3006-3013 (2012).

32. Yu, T.-X., Yi, C.-W. and Tsao, S.-L. \Rank-basednetwork coding for content distribution in vehicularnetworks", IEEE Wireless Communications Letters,1(4), pp. 368-371 (2012).

33. Sommer, C., Eckho�, D., German, R. and Dressler,F. \A computationally inexpensive empirical modelof IEEE 802.11p radio shadowing in urban environ-ments", IEEE Eighth International Conference onWireless On-Demand Network Systems and Services(2011).

34. Boban, M., Vinhoza, T.T.V., Ferreira, M., Barros, J.and Tonguz, O.K. \Impact of vehicles as obstacles invehicular ad hoc networks", IEEE Journal on SelectedAreas in Communications, 29(1), pp. 15-28 (2011).

35. IEEE Std. 802.11p-2010 \Wireless LAN medium accesscontrol (MAC) and physical layer (PHY) speci�cationsamendment 6", wireless access in vehicular environ-ments (2010).

36. NS2 simulation with 802.11p.https://sites.google.com/site/nahoons2/english-con/ns2-simulation-with-802-11p.

37. Li, Y., Jin, D., Su, L., Zeng, L. and Wu, D.O. \Optimalmobility control with energy constraint in delay toler-ant networks", Wireless Communications and MobileComputing, DOI:10.1002/wcm.2247 (2012).

38. S.J.U. Tra�c Information Grid Team, Grid ComputingCenter, Shanghai Taxi Trace Data. http:// wireless-lab.sjtu.edu.cn/.

Biographies

Ing-Chau Chang received his BS degree from theDepartment of Computer and Information Science atthe National Chiao Tung University, Hsinchu, Taiwan,the Republic of China, in 1990, and MS and PhDdegrees from the Institute of Computer Science andInformation Engineering at the National Taiwan Uni-versity, Taipei, Taiwan, the Republic of China, in 1992and 1997, respectively. He is currently Professor inthe Department of Computer Science and InformationEngineering at the National Changhua University ofEducation, Changhua, Taiwan, the Republic of China.His current research topics include wireless networks,multimedia network protocols, and multimedia sys-tems. He is the member of the Institute of Electricaland Electronics Engineers (IEEE).

Chien-Hsun Li received his MS degree in the In-stitute of Computer Science and Information Engi-neering at the National Changhua University of Ed-ucation, Changhua, Taiwan, the Republic of China,in 2012. His research interest includes the wirelessnetwork.

Cheng-Fu Chou received MS and PhD degrees fromthe University of Maryland, College Park, MD, USA,in 1999 and 2002, respectively, and is currently in theComputer Science and Information Engineering De-partment at the National Taiwan University, Taiwan,the Republic of China. He has also been a visitingscholar in the Computer Science Department at theUniversity of Southern California. His current researchinterests are in distributed multimedia systems, peer-to-peer computing, wireless networks, and their perfor-mance evaluation.

E cient packet replication control for a geographical routing...

Documents

Transcript of E cient packet replication control for a geographical routing...