Mingsi Zhang, Changle Li, Tao Guo, Yuchuan Fu

State Key Laboratory of Integrated Services Networks, Xidian University, Shaanxi 710071 China.

I. INTRODUCTION

Vehicular Ad-hoc Networks (VANETs) enable communications among vehicles and between vehicles and infrastructure, aiming to support a wide range of applications to make travel safe,informed, and pleasant [1-2]. The applications envisioned vary from transportation safety and efficiency to infotainment. Apparently, the effective dissemination of content is the basis of any type of applications in VANETs. Particularly, multimedia content is able to provide data visualization and to improve users’ driving experience, so its dissemination becomes increasingly vital [3]. Based on DSRC (Dedicated Short Range Communication), content dissemination can be realized in V2V (Vehicle-to-Vehicle), V2I (Vehicle-to-Infrastructure)or hybrid communication mode.

Due to vehicles’ high mobility, severe channel fading and intensive mutual interferences[4-5], the V2V communication is transient and unstable, and the service provided is con fi ned to the download of small-size content. Vehicles shall have opportunities to temporarily access to the Internet when driving into infrastructure’s coverage, such as roadside units (RSUs).However, V2I communication still cannot meet the download requirements of large-size content. Firstly, there exists many no-signal areas on the road, also referred to dark areas due to sparsely deployment of RSUs. As reported in [6], generally, the distance between two adjacent RSUs is 8-16km. This means that vehicles have to drive in the dark areas and have no downloads from RSU for a long time. Secondly, vehicles have so high speed that it can only sojourn in the RSU’s coverage for a short period of time. This makes networks cannot offer resource-intensive services to vehicles, e.g., video sharing. Thirdly, it may occur that a multitude of vehicles in the RSU’s coverage share or contend the channel at the same time, which extremely restricts the data download volume of individual vehicle.Therefore, it is necessary to convert vehicles from competitors to cooperators to enlarge the data download volume.

Based on content download in highway VANETs, a preliminary version [7] proposed prefetching-based content download scheme,and exploited reverse vehicles to cooperatively facilitate data download of individual vehicle in dark areas. Different from our previous work, we incorporate clustering with vehicles in the same lane and carry-and-forward scheme with reverse vehicles to further increase data download volume of the target vehicle in this paper.

The major contributions of this paper are threefold:

·Combinative mobility model: The macroscopic traffic flow model and microscopic accelerated overtaking model are integrated for making a classified discussion of vehicle’s behavior.

·Cooperative cluster: we consider a cooperative scheme to boost content download volume through forming cluster. Meanwhile, we set vehicles in the cluster margin as relay points, and thus the tagged vehicle could receive direct data and two-period forwarding data from reverse cooperators.

·Analysis and evaluation: Detailed analysis and simulation results show that the proposed scheme significantly increases the achievable data download volume.

The remainder of this paper is organized as follows. Section II reviews related work in the literatures and Section III presents the system model. Section IV discusses cooperative download scheme followed by detailed analysis. Section V evaluates the performance of proposed scheme and presents the results.Finally, Section VI closes the paper with conclusions.

In this paper, we have proposed a cluster-based cooperative scheme to efficiently increase data download volume in the dark area.

II. RELATED WORK

Efficient content dissemination in VANETs is a challenging and vital problem, and many efforts have been devoted to devising content dissemination solutions. Meanwhile many related research results provide fundamental reference on design of vehicular mobility models[8], data forwarding protocols [9], transmission schemes [10], relay algorithms [11] and on the performance analysis [12-13]. Cooperative communication has also drawn wide research attention, and extensive cooperative schemes have been proposed accordingly.

For V2V cooperative communications, the authors in [14] investigated a V2V collaboration scenario to distribute different types of service messages, and applied a cooperative relay selection algorithm to improve the packet delivery ratio and reduce delay. Yanet al.[15] analyzed the achievable throughput of cooperative mobile content distribution from RSUs, where packet-level network coding and symbol-level network coding were both exploited. Liet al.[16] proposed the connectivity-aware data dissemination scheme based on the node forwarding capability metric to improve the data transmission capacity in partially connected VANETs.

In terms of cooperative V2I communications, the authors in [17] introduced a representative-based prefetching mechanism to distribute contents, in which a set of representative infrastructures are carefully selected.The authors in [18] proposed opportunistic RSU-aided content download schemes to address heterogeneous networks in terms of data items and users. The authors in [19] devised a vehicle route-based data prefetching scheme to provide reliable data delivery services from a cloud data center to vehicles through roadside wireless access points (APs) with local data storage.

Recent researches pay more attention to co-operative content distribution based both V2V and V2I communication. The authors in [6]designed a cooperative downloading scheme,namely DSRelay, which utilized the passing vehicles to help clients’ downloading for scheduling the dark areas out of RSU coverage areas. The authors in [20] devised a cooperative carry-and-forward scheme to download data. They focused on reducing transmission outage time caused by dark area between two neighboring RSUs. In order to maximize the throughput of content download in dark areas,the authors in [21] researched cooperative vehicles selection strategy and stimulated vehicles with a virtual check to participate in cooperative download.

III. SYSTEM MODEL

As shown in figure 1, in our envisaged mobile content cooperative download system, RSUs are regularly deployed on the highway with distanced. In order to provide content services for vehicles, we assume that the RSUs can be connected to the backbone network through wireless or wired links. In what follows, we introduce the models that are adopted to study the content download problem in VANETs,including vehicle mobility model, V2V communication model and V2I communication model.

3.1 Vehicular mobility model

Fig. 1. System model.

Vehicular mobility models can be roughly divided into macroscopic mobility model and microscopic mobility model. The combination of them is promising to address issues for vehicular mobility model [22]. We apply the macroscopic traffic flow model and microscopic accelerated overtaking model to make a classi fi ed discussion of vehicle’s behavior.

3.1.1 Driving at constant speed

We consider bidirectional traffic flows with mean densityρsveh/m andρrveh/m, and approximate the vehicle arrivals by a Poisson process with parametersλsandλr, respectively. Additionally, we assume that vehicles in each direction move at speeds ofvsm/s andvrm/s. According to the macroscopic traffic flow model [23], the above mentioned variables are related by the following relationship

Based on the typical speed-density relationship presented in [23], we have

wherevfis the maximum allowed speed, andρjamis the vehicle jam density at which traffic flow comes to a halt.

3.1.2 Overtaking model

A lane changing model are widely applied to reproduce the overtaking behavior [24]. However, it is not suitable for our scenario due to the prerequisite that the following vehicle’s speed is greater than the preceding vehicle’s.The introduced microscopic vehicular mobility can be treated as the accelerated overtaking model with corresponding speed and distance constraints. We assume that the tagged vehicle and cooperators drive at a constant speed at the initial moment of overtaking when they have left the coverage of RSU. The overtaking process can be illustrated as showed in figure 2:

1) The accelerated process

We apply the kinematic equation to model the microscopic mobility behavior for the free driving of vehicles on highways.The velocity of cooperative vehicleCciand tagged vehicleCtare the same at the begin-ning of the accelerated process. At timet,Ccichanges into the overtaking lane to catch upCtfrom A to B. And the accelerated distance at thetactime interval can be calculated bywhereais the acceleration ofCci.vci(t) is a velocity of cooperatorCciat timet,vtis a velocity of the target vehicleCtandWe denote the velocity of cooperative vehicleCcidriving after the accelerated process by

2) Constant-speed driving process

At thet+tactime, cooperative vehicleCcistarts to keep the constant speed and drives into the original lane intctime, when the overtaking has been completed. Here the new distance between the consecutive vehiclesCciandCtcan be represented by

In addition, vehicle mobility confines to two rules, namely, speed and distance constraints as follows:

Speed constraint:In the highway scenario,the velocity of each vehicle is restricted to the intervalwherevmin=60km/handare set in the manner. The overtaking will not happen when the velocity of cooperative vehicle is

Distance constraint:The distance constraint between any consecutive vehicles includes minimum of the inter-vehicle driving distance and maximum of the distance between cooperators and the tagged vehicle. The former is an appropriate car-following distancedsafeby considering calculating value of the safety distance cited by [25], which depends on the vehicle’s real-time velocity. Thereforedsafecan be set between 60 and 120 m. And the latter is denoted bydcluster, which can be set to be less than 300 m.

3.2 Communication model

Fig. 2. Overtaking process.

We consider the same communication model as our previous work [7]. The communication range of RSU is divided into different zones corresponding to different transmission rates[26]. In addition, the communication range of RSU is denoted byR. We assume that a vehicle cannot communicate with other vehicles when it is downloading data from RSU. For the ease of illustration, we consider an ideal MAC protocol similar to [27], in which the RSU airtime is shared among all vehicles in a fair manner.

We adopt the unit disk protocol model to study the V2V communication performance[23]. Arbitrary two vehicles can directly communicate with each other if their distance is less than the communication range of each vehicler. We omit physical layer details of V2V communication.

IV. PROTOCOL DESCRIPTION AND ANALYSIS

Due to the existence of dark area and the high speed of vehicles, data download on the highway is largely capped, especially the large-size content download. In this part, we introduce our proposed cooperative scheme. For the easy of illustration, we denote the nearest RSU from the tagged vehicle byRSUi.

The main idea of the proposed scheme is shown as following:

?Cluster forming phase: Before entering the RSU’s coverage, the tagged vehicle hunts for following vehicles which have no downloads as cooperators to form a cluster for downloading from RSU. In some sense, the tagged vehicle and cooperators can be called the cluster head and the cluster members, respectively.

?Data download: Within the communi-cation range ofRSUi, the tagged vehicle and the cooperators in cluster directly download data fromRSUithrough V2I communication.Meanwhile, the tagged vehicle will request the next RSU to prefetch an appropriate portion of the remaining content that will be downloaded by theRSUi+1selected vehicles.

?Data forwarding: Once leavingRSUi’s coverage and driving in the dark area, the tagged vehicle receives data from multiple vehicles within its cluster and encountered reverse vehicles through carry-and-forward method.

Without loss of generality, we define the period from the tagged vehicle arriving at the communication range of a RSU to the next RSU’s coverage as one download cycle. Denote the expected data download volume of the tagged vehicle in one cycle receives from helpers in the same lane and reverse helpers asE[Ds] andE[Dr] respectively. We aim to use our proposed scheme to find throughput of which the tagged vehicle, where the throughput Γ can be calculated as

In what follows, we will introduce details of the proposed scheme and present corresponding analysis.

4.1 Cluster forming phase

Here, we implicitly assume that all vehicles are equipped with GPS devices for timely updating location information. Before entering into the RSU’s coverage, the tagged vehicle will hunt for vehicles who have no download requirements to form a cluster for downloading from RSU. In detail, the tagged vehicle first broadcasts a cooperative request to the following vehicles within its communication range. Vehicles willing to coordinate will make a confirmation to it. In particular, cooperative vehicle who needs to overtake the tagged vehicle in the data forwarding process will notify the target vehicle that it has urgency events in advance, where we denominate this mechanism as Proactive Urgency Notifi cation (PUN). In this section, some relative analysis will be described in detail.

4.1.1 Cluster maintenance

We consider the cluster maintenance to address this issue that vehicle leave beyond the communication range of the tagged vehicle when the overtaking happens.

The overtaking model has been described in the vehicular mobility model. We will fill in other details as follows: ifthe communication between the tagged vehicle and the overtaking vehicle cannot be affected;the tagged vehicle and cooperative vehicles will be informed to adjust speed to follow the overtaking vehicle for downloading segmental large-size content.

4.1.2 Cluster size

1) Probability of being the last vehicle in a cluster

Instead of communicating with the tagged vehicle singly, we suppose that reverse cooperators can also interact with the last vehicle in cluster for boosting the data download volume of the tagged vehicle. Thus, we are interested in introducing the probability that cooperative vehicle will be the last vehicle in a cluster, denoted byPd[28].Pdis simply given by

whereis the Cumulative Distribution Function (CDF) of the inter-vehicle spacing. The parameteris relative to the vehicle arrival time and traffic volume.

2) Average cluster size

In order to gather the data download volume where vehicles within a cluster can contribute, the expected number of vehiclesE[CN][28] in a cluster needs to be calculated as:

3) Average cluster length

Average cluster length can be described by the length between the first vehicle and the last vehicle in a cluster [28]. It can be held by

4.2 Data download

As mentioned above, the data download phase includes the tagged vehicle downloading data with its cluster members and reverse cooperators downloading data through V2I communication.

During the period, we consider a communication scheme that once moving into RSU’s coverage, the tagged vehicle will subscribe to a file immediately. According to the evaluated download throughput of each vehicle, the file will be divided intoNpchunks in order to enable multiparty downloads. The total download content chunks are marked bycorresponds to the download file for tagged vehicle’s cluster members, andcorresponds to the prefetching part, i.e., theRSUi+1selects reverse vehicles to cooperatively download the rest part of the file.

4.2.1 Downloading data with cooperators in the same lane from RSU

Considering that VANETs on highway is prone to network fragmentation, we capture this characteristic for a better design of data download process. Based on V2I communication model, we divide the coverage of RSU intoMzones corresponding to the transmission rate set

We denote the distance of each zone asandthe sojourn time of the tagged vehicle in each zone is denoted asand

Therefore, according to previous work [7],the average data download volume can be calculated as

wherep0andpCare the probability that there exists at least one vehicle within RSU’s coverage, and that maximize vehicles in the RSU’s coverage.

Then the expected amount of data the tagged vehicle’s cluster can download from RSU can be obtained by

4.2.2 Reverse helpers download data from RSU

The tagged vehicle cannot receive data completely from reverse helpers that download the integrated data from RSU. Therefore, we consider that the volume of data which reverse cooperators download from RSU depends upon the connected time with tagged vehicle’s cluster in the data forwarding phase.

4.3 Data forwarding

When driving in the dark area, the tagged vehicle receives data from cluster members and encountered vehicles through carry-and-forward method.

4.3.1 Receiving data from vehicles in its cluster

Cluster members will send data downloading from RSU to its cluster head. We consider that there areE[CN]-1 helpersVi(i=1,2,…,E[CN]-1) in its communication range attempting to access to the tagged vehicle simultaneously to forward carried data.

We apply the IEEE 802.11b DCF scheme for MAC scheduling with the RTS/CTS scheme to eliminate the hidden terminals problem. Denote the constant contention window size of vehicleViasCW. Then we can obtain the average transmission probability ?of each vehicle as [29]

Therefore, the probability thatVisuccessfully transmits packets in any slot to the target vehicle is

whereis the probability that there is at least one transmission in a considered slot time,is the proba-bility thatVisuccessfully transmits packets in a slot time.

Then the tagged vehicle will receive packets from its cluster members until all the forwarding content are done completely. Appling the results in the [29], the MAC throughputSfromVito the target vehicle can be expressed mathematically as

wherePSis the average packet payload size,σis the duration of an empty slot time,Tsucis the duration of a busy slot time (i.e., the average time when the channel is sensed busy because of a successful transmission), andTcois the average time the channel is sensed busy by each vehicle during a collision, which is formulated in [29].

The tagged vehicle will receive packets from its cluster members one by one. Consequently, the data forwarding time of concurrent cooperation can be

4.3.2 Receiving data from reverse vehicles

Upon finishing receiving data from cluster members, the tagged vehicle will achieve data from encountered vehicles in the reverse lane through carry-and-forward method. In order to avoid the case that reverse cooperators cannot forward data to the target vehicle in succession, we set vehicles in the cluster margin as relay points, and thus the tagged vehicle could receive direct data and two-period forwarding data from reverse cooperators. With the consideration of the former constraint, the cluster length will be explained occupied in two aspects.

Fig. 3. Data forwarding when driving at the constant speed.

1) The encountered time when driving at the constant speed

When none of following vehicles will overtake the tagged vehicle, the entire cluster will drive at the same speed.Step1: On the grounds of cluster forming property, reverse vehicles will deliver packets to the tagged vehicle until they are unable to communicate with it.Step2:Then they will transmit the continued packets to the last vehicle within the tagged vehicle’s cluster. We can see the last vehicle as the relay point.Step3: Upon the last vehicle moving packets to the tagged vehicle (i.e., two-period forwarding data from reverse cooperators),the data forwarding process is completed. We consider the encountered time interval of one reverse helper and cluster, denoted by

2) The encountered time when overtaking

When there is a vehicle attempting to overtake, the tagged vehicle’s cluster will make a difference. Due to PUN mechanism proposed above, the tagged vehicle’s speed will be consistent with the overtaking vehicle’s speed and its cluster length will grow large.Step1: Reverse vehicles will first deliver packets to the overtaking vehicle.Step2: Until they enter into communication range of the tagged vehicle,its packets transmission will do communication switchingStep3: They will transmit the continued packets to the last vehicle.Step4:The overtaking vehicle moves packets to the tagged vehicle. What is noteworthy is thatStep3andStep4are simultaneous.Step5: All the two-period data forwarding process will be completed. The last vehicle within the tagged vehicle’s cluster and the overtaking vehicle are seen here as relay points. We assume that the overtaking vehicle is not the last vehicle in the cluster. The encountered time interval of one helper in the reverse lane with cluster at this point, calculated by

Considering the communication rateωbetween the tagged vehicle and reverse helpers, the achievable data volumeE[Dr] can be obtained by

The value oftwill betecorteo, as appropriate. Moreover,E[Dr] effects definitively the volume of data which cooperators in the reverse lane download from RSU.

V. PERFORMANCE EVALUATION

In this section, we verify the performance of our proposed cooperative download scheme and the accuracy of the theoretical analysis through extensive simulation conducted in MATLAB. Meanwhile we compare our proposed scheme with no-cooperator scheme,DSRelay [6] scheme and Chain Cluster [27]scheme. In the bidirectional highway scenario,infrastructures are deployed at constant intervals 8km. The density of cooperators varies from 0.001veh/m to 0.01veh/m. The communication range of RSUs and vehicles are 400 m and 300 m, respectively. the adaptive transmission rates and nodes’ contention windows in MAC layer of V2I communication model are specified in table 1. MAC parameters of V2V communications between the target vehicle and helpers in the same direction are shown in table 2 and V2V transmission rateωbetween the target vehicle and reverse helpers is set to 1.6 Mbps.

Figure 5. shows the relationship between data download volume and driving speed. It is obvious that slow driving speed is beneficial to data download volume. It is due to that each vehicle’s sojourn time within infrastructure’s coverage becomes shorter with the increase of speed. When the driving speed varies from 60km/h to 120km/h, the data download volume varies from 540Mb to 175Mb, and the simulation results match with the analytical values well.

Fig. 4. Data forwarding when overtaking.

Table I. Parameters of different zones.

Table II. Default setting of DCF.

Fig. 5. Data download volume vs. speed.

Figure 6. illustrates comparison of the total amount of download data by different schemes when the tagged vehicle enters from one RSU’s coverage to the next RSU’s. Without cooperative download, the tagged vehicle can only download data during the sojourn time within RSU’s coverage. Note that the cooperative download arises at about 210s when DSRelay is adopted. It is due to that the requested data is only prefetched to the next RSU, and reverse cooperators take a period of time to encounter the tagged vehicle in the dark area. Chain Cluster obtains its cooperative downloads at about 180s, for its strategy that downloads from cooperators are forwarded from the rear to the head one by one. With our proposed scheme, we can see the tagged vehicle can achieve downloads incessantly from its cluster members and reverse cooperators when it just leaves the RSU’s coverage. Obviously,our proposed scheme increases the total download volume greatly in each download cycle.

Fig. 6. The total download volume vs. download time.

Fig. 7. The forwarding volume vs. download time.

Figure 7. demonstrates the impacts of various mobility models in the forwarding process. We simulate a case that three vehicles form a cluster with mean density 0.004 veh/m and cooperatively receive data from reverse vehicles. Upon forwarding twice, the regrouping cluster for overtaking boost the download volume of the tagged vehicle. In addition, that the tagged vehicle drives further through speed adjustment should be paid attention, and the simulation results match with analytical values well when vehicles drive at the constant speed.

Figure 8. shows the relationship between throughput and density of cooperators in the same lane. It is obvious that ChainCluster’s throughput increases with the increase in density. The main reason is that higher density of cooperators makes the expected number of cluster member increase. However, it is observed that the throughput of our proposed scheme decreases with the increase in density of cooperators, which is due to a decrease in download forwarding throughput. DSRelay and No-cooperator scheme depend on the meeting time and download data volume from RSU, respectively, so there is no change in their throughput. The overall picture for our proposed scheme is positive compared to other schemes.

VI. CONCLUSION

In this paper, we have proposed a cluster-based cooperative scheme to efficiently increase data download volume in the dark area. The proposed scheme takes advantage of multiple infrastructures, cluster and reverse vehicles to extend the access time. Through combining microscopic and macroscopic mobility models, we have applied extensive simulations to study the performance of the proposed scheme in terms of data download volume and throughput. The results have shown that our proposed scheme outperforms other content download schemes for highway VANETs,and that our scheme can increase total data download volume by 43.75% compared with DSRelay scheme who has considerable performance. Moreover, our results shed insight into the design of cooperative communication scheme to maximize the utilization of benefits in dark area to extend the limited RSU’s coverage and short contact time between RSU and the tagged vehicle.

ACKNOWLEDGEMENTS

This work was supported by the National Natural Science Foundation of China under Grant No. 61571350, Key Research and Development Program of Shaanxi (Contract No.2017KW-004, 2017ZDXM-GY-022), and the 111 Project (B08038).

Fig. 8. Throughput vs. density of cooperators.

References

[1] X.X. He, H. Zhang, W.S. Shi, et al., “Transmission Capacity Analysis for Linear VANET under physical model,”China Communications, vol. 14, no.3, 2017, pp. 97-107.

[2] F. Cunha, L. Villas, A. Boukerche, et al., “Data Communication in VANETs: Protocols, Applications and Challenges,”Ad Hoc Networks, vol. 44,2016, pp. 90-103.

[3] H.G. Gong, L.F. Yu, N.B. Liu, et al., “Mobile Content Distribution with Vehicular Cloud in Urban VANETs,”China Communications, vol. 13, no. 8,2016, pp. 84-96.

[4] L.Y. Fu, X.Z. Fu, Z.Y. Xu, et al., “Determining Source-Destination Connectivity in Uncertain Networks: Modeling and Solutions,”IEEE/ACM Transactions on Networking, vol. 25, no. 6, 2017,pp. 3237-3252.

[5] L.Y. Fu, X.B. Wang, P.R. Kumar, “Are We Connected? Optimal Determination of Source-Destination Connectivity in Random Networks,”IEEE/ACM Transactions on Networking, vol. 25, no. 2,2017, pp. 751-764.

[6] J.H. Liu, J.P. Bi, Y.C. Bian, et al., “DSRelay: A Scheme of Cooperative Downloading Based on Dynamic Slot,”P(pán)roc. IEEE ICC, 2012, pp. 381-386.

[7] T. Guo, C.L. Li, Z.F. Zhi, et al., “Prefetching-Based Content Download for Highway Vehicular Ad Hoc Networks,”P(pán)roc. IEEE ICCC, 2017, pp. 1-6.

[8] H.Z. Zhu, L.Y. Fu, G.T. Xue, et al., “Recognizing Exponential Inter-Contact Time in VANETs,”P(pán)roc. IEEE INFOCOM, 2010, pp. 1-5.

[9] M. Wang, H.G. Shan, T.H. Luan, et al., “Asymptotic Throughout Capacity Analysis of VANETs Exploiting Mobility Diversity,”IEEE transactions on vehicular technology, vol. 64, no. 9, 2015, pp.4187-4202.

[10] X.B. Wang, L.Y. Fu, C.H. Hu, “Multicast Performance with Hierarchical Cooperation,”IEEE/ACM Transactions on Networking, vol.20, no. 3,2012, pp. 917-930.

[11] X.B. Wang, W.T. Huang, S.X. Wang, et al., “Delay and Capacity Tradeoff Analysis for Motioncast,”IEEE/ACM Transactions on Networking, vol.19,no. 5, 2011, pp. 1354-1367.

[12] N. Lu, X.M. Shen, “Capacity Analysis of Vehicular Communication Networks,”Springer New York,2014.

[13] N. Lu, T.H. Luan, M. Wang, et al., “Bounds of Asymptotic Performance Limits of Social-Proximity Vehicular Networks,”IEEE/ACM Transactions on Networking, vol. 22, no. 3, 2014, pp. 812-825.

[14] B. Das, S. Misra, U. Roy, “Coalition Formation for Cooperative Service-Based Message Sharing in Vehicular Ad Hoc Networks,”IEEE Transactions on Parallel & Distributed Systems, vol. 27, no. 1,2015, pp. 144-156.

[15] Q.B. Yan, M. Li, Z.Y. Yang, et al., “Throughput Analysis of Cooperative Mobile Content Distribution in Vehicular Network Using Symbol Level Network Coding,”IEEE Journal on Selected Areas in Communications, vol. 30, no. 2, 2012,pp. 484-492.

[16] Z.Y. Li, Y. Song, J.L. Bi, “CADD: Connectivity-Aware Data Dissemination Using Node For-warding Capability Estimation in Partially Connected VANETs,”Wireless Networks, vol. 2017,no. 6, 2017, pp. 1-20.

[17] D. Zhang, C.K. Yeo, “Enabling Efficient Wi-Fi-Based Vehicular Content Distribution,”IEEE Transactions on Parallel and Distributed Systems,vol. 24, no. 3, 2013, pp. 479–492.

[18] Y. Li, X.M. Zhu. D.P. Jin, et al., “Multiple Content Dissemination in Roadside-Unit-Aided Vehicular Opportunistic Networks,”IEEE Transactions on Vehicular Technology, vol. 63, no. 8, 2014, pp.3947-3956.

[19] R. Kim, H. Lim, B. Krishnamachari, “Prefetching-Based Data Dissemination in Vehicular Cloud Systems,”IEEE Transactions on Vehicular Technology, vol. 65, no. 1, 2016, pp. 292-306.

[20] Y.J. Wang, Y.S. Liu, J.Y. Zhang, et al., “Cooperative Store-Carry-Forward Scheme for Intermittently Connected Vehicular Networks,”IEEE Transactions on Vehicular Technology, vol. 66, no. 1,2017, pp. 777-784.

[21] C.Z. Lai, K. Zhang, N. Cheng, et al., “SIRC: A Secure Incentive Scheme for Reliable Cooperative Downloading in Highway VANETs,”IEEE Transactions on Intelligent Transportation Systems,vol. 18, no. 6, 2017, pp. 1559-1574.

[22] W. Xiong, Q.Q. Li, “Performance Evaluation of Data Disseminations for Vehicular Ad Hoc Networks in Highway Scenarios,”The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, vol. 37,2015, pp. 1015-1020

[23] W.L. Tan, W.C. Lau, O. Yue, et al., “Analytical Models and Performance Evaluation of Drive-thru Internet Systems,”IEEE Journal on selected areas in Communications, vol. 29, no. 1,2012, pp. 207-222.

[24] T.Q. Tang, H.J. Huang, S. Wang, et al., “A New Overtaking Model and Numerical Tests,”P(pán)hysica A: Statistical Mechanics and its Applications, vol.376, 2007, pp. 649-657.

[25] B. Karel, D.D. Waard, B. Mulder, “Measuring Driving Performance by Car-Following in Traffic,”Ergonomics, vol. 37, no. 3, 1994, pp. 427-434.

[26] K. Ota, M.X. Dong, S. Chang, et al., “MMCD:Max-Throughput and Min-Delay Cooperative Downloading for Drive-thru Internet Systems,”P(pán)roc. IEEE ICC, 2014, pp. 83-87.

[27] H.B. Zhou, B. Liu, T.H. Luan, et al., “ChainCluster:Engineering a Cooperative Content Distribution Framework for Highway Vehicular Communications,”IEEE Transactions on Intelligent Transportation Systems, vol. 15, no. 6, 2014, pp. 2644-2657.

[28] N. Wisitpongphan, F. Bai, P. Mudalige, et al.,“Routing in Sparse Vehicular Ad Hoc Wireless Networks,”IEEE Journal on Selected Areas in Communications, vol. 25, no. 8, 2007, pp. 1538-1556.

[29] T.H. Luan, X.M. Shen, F. Bai, “Integrity-Oriented Content Transmission in Highway Vehicular Ad Hoc Networks,”P(pán)roc. IEEE INFOCOM, 2013, pp.2562-2570.