A Cooperative Quality-aware Service Access System for ... - IEEE Xplore

0 downloads 0 Views 2MB Size Report
3Department of Electrical and Computer Engineering, New Jersey Institute of Technology, NJ, USA ... opened up a new paradigm in vehicular networks by ena-.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

A Cooperative Quality-aware Service Access System for Social Internet of Vehicles Zhaolong Ning1, Xiping Hu2,*, Zhikui Chen1,*, Mengchu Zhou3, Fellow of IEEE, Bin Hu4,*, Jun Cheng2, and Mohammad S. Obaidat5, Fellow of IEEE 1

Key Laboratory for Ubiquitous Network and Service Software of Liaoning Province, School of Software, Dalian University of Technology, Dalian, China. 2 3

Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, China

Department of Electrical and Computer Engineering, New Jersey Institute of Technology, NJ, USA 4 5

School of Information Science and Engineering, Lanzhou University, China

Department of Computer and Information Science, Fordham University, NY, USA

Abstract—Due to the enormous potential to guarantee road safety and improve driving experience, Social Internet of Vehicle (SIoV) is becoming a hot research topic in both academic and industrial circles. As the ever-increasing variety, quantity and intelligence of on-board equipment, along with the ever-growing demand for service quality of automobiles, the way to provide users with a range of security-related and user-oriented vehicular applications has become significant. This paper concentrates on the design of a service access system in SIoVs, which focuses on a reliability assurance strategy and quality optimization method. First, due to the instability of vehicular devices, a dynamic access service evaluation scheme is investigated, which explores the potential relevance of vehicles by constructing their social relationships. Next, this work studies a trajectory-based interaction time prediction algorithm to cope with an unstable network topology and high rate of disconnection in SIoVs. At last, a cooperative quality-aware system model is proposed for service access in SIoVs. Simulation results demonstrate the effectiveness of the proposed scheme. Key Words—Internet of Vehicles, access systems, quality-aware, social relationship, routing protocol

1

INTRODUCTION

Smart city is a term leveraged for the integration of innovative technologies and solutions to manage a city’s assets, including information systems, power plants, transportation, and other critical infrastructures and key resources [1]. The ultimate goal of smart city is to offer the high quality of citizens’ life. In this regard, the transportation sector has drawn intense interests due to the rapidly increasing number of vehicles in big cities. Moreover, people spend a large amount of time in vehicles on roads after homes and offices. Therefore, efficient management of heavy traffic to avoid congestion and

accidents, as well as providing interactive services to the drivers and passengers on roads is an urgent demand nowadays [2, 3]. Vehicular Ad Hoc Network (VANET) is one part of the earliest technologies designed for vehicular networks. It is essentially a subset of mobile Ad Hoc networks, where vehicles carry the wireless communication devices with high mobility. Due to the high mobility and rapid change of a network topology, VANETs offer a challenging task of efficiently routing information to a specified destination with high link connectivity. Therefore, efficient routing protocol design has become a hot research topic for several years. To overcome the connectivity issue in VANETs, researchers have been trying to explore alternative solutions. One promising solution is Delay Tolerant Networks (DTNs), whose vital feature is that an end-to-end connection is not necessary for link establishment. Instead, they are employed in a store-carry-and-forward manner. To achieve this connectionless communication, DTN protocol stack offers a special layer called a bundle layer. Its function is to store the initial packet and pass it on as a bundle to other nodes upon contact, until the packet reaches its destination. Smaldone et al. [4] proposed a social framework for vehicular communication, along with an application named Roadspeak. Through this application, drivers on the road can join interest-based voice chat groups to engage in dialogue with each other for entertainment. They called this framework as Social Internet of Vehicle (SIoV) and defined it as: A social network of vehicles, enabled by spatial and temporal localities on the road. This has opened up a new paradigm in vehicular networks by enabling drivers and passengers to socialize on roads. SIoVs make use of the human social behaviors (relationship, similarity, community, mobility, and social ties etc.) and incorporate them into physical vehicular networks. Due to the huge potential for the realization of smart cities, the utilization of SIoVs has been extensively stud-

2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

ied in road safety guarantee, transportation efficiency optimization, conveniences and comfort improvement for driving. As the continual integration of Table PC and on-board controller, the communication requirement between human to human, Vehicle to Vehicle (V2V), and Vehicle to Infrastructure (V2I) can be satisfied in SIoVs. Furthermore, as on-board equipment becomes more and more intelligent. Users’ requirements on the diversification and optimization of vehicle services increase largely, which bring challenges to the service access of SIoVs, such as network complexity and personalization [5, 6]. As the foundation of vehicular applications, service discovery and network access directly affect the safety and availability of SIoVs, which determine the whole quality of users’ experience. Furthermore, the related methods have to provide vital information for the upper layer vehicular applications. Generally speaking, current studies mainly focus on three aspects: network frameworks, discovery techniques and routing protocols [7, 8]. Although the intelligence degree of vehicular devices improves continually, the expectation of users’ requirements for network services also increases. Because a single device cannot meet all the requirements due to its limitation of information acquiring and processing, the establishment of information sharing, service access and mutual cooperation mechanism among vehicular devices becomes inevitable. However, service dimensionality, and network complexity together with security risk would increase if multiple nodes are engaged for information interaction, such that vehicular applications are inclined to reliable and safe information access and services. This is the motivation of our study. Our work designs a Cooperative Quality-aware Service access (CQS) system for SIoVs. Its main contribution can be summarized as follows: 1. We put forward a dynamic access service evaluation scheme. It comprehensively considers the direct and indirect service quality evaluations, and can cope with the interference and effect brought by the dynamic change of network topology and node instability by introducing a time attribute, attenuation mechanism, and feedback regulation mechanism of the historical record. 2. We propose a social relationship construction method for intelligent vehicles. Based on the foundation of occupation similarity and social relationship in SIoVs, we establish a realistic and scalable social relationship model by exploring the implicit internal similarity inside vehicles to promote the accuracy and success rate of a service access method. 3. We propose a prediction method according to vehicle movement trajectory for interaction time estimation. By analyzing the movement rate and attributes of vehicles, their movement trajectory can be predicted, and the variation tendency can be calculated for the reference of moving vehicles. 4. We construct a CQS system for SIoVs. At first, a

node-centric generation tree is leveraged to evaluate the access quality. Then, different access service routing selection strategies are employed to select the access path according to the current network state. After that, a bi-direction buffering algorithm is given to promote the response efficiency and accuracy of routing. Performance evaluations demonstrate that, comparing with the existing schemes, our CQS scheme can increase the average service quality, number of packets sent per access and network success rate by around 25%, 20% and 5%, respectively. The rest of this paper is organized as follows. Section 2 illustrates the related works. An access service selection scheme is introduced in Section 3. Section 4 presents the presented CQS system, containing topology construction, routing strategy selection and a buffer mechanism. Section 5 demonstrates the simulation results, and Section 6 concludes this paper.

2

RELATED WORKS

The corresponding related works can be generally categorized into three aspects, namely service discovery and network access, quality and security of an access service, and routing in SIoVs. 2.1 Service discovery and network access The service discovery can be classified into three categories: push-based, response-based, and hybrid-based methods. In [9], the authors have studied an adaptive user request-based packet request mechanism, which permits different applications generating request packets according to their observations to guarantee transmission stability. In order to obtain high-efficiency bandwidth access, an infrastructure-supported service discovery protocol is presented in [10]. It has considered not only search service and routing, but also bandwidth resource limitation, such that a trade-off is made for the balance between network load and service discovery efficiency. In order to guarantee Video-on-Demand (VoD) transmission in mobile wireless networks, virtual community-based methods are proposed to balance the efficiency of content sharing and proposed network cost. The authors in [11] have studied a performance-aware mobile community based VoD streaming solution in VANETs, which focuses on mobile community detection and community member management. The former is fulfilled by leveraging a fuzzy ant-inspired clustering scheme and a mobility similarity estimation model. The latter is realized by defining the role and task, join and leave, and collaborative store of community members. Since the network service would be seriously degraded by intermittent link connectivity, a cooperative caching solution has been studied to promote the quality of experience of multimedia streaming services by encouraging social cooperation among vehicles in highways [12]. With the objective of promoting quality of life in cities while not influencing the re-

2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

quirement of human mobility, the authors in [13] attempt to reveal the mobility patterns and discover the influential factors to human beings. 2.2 Quality and security of access service Current studies on vehicular networks mainly focus on the success rate of service discovery, response time, bandwidth utilization, and transmission delay. However, the studies on the quality and security of access services are far from enough. As the ever-increasing requirements of personalization in vehicle equipment, the reliability, security and privacy of an access service are becoming increasingly important, which would impact the success rate of connection and bandwidth utilization [14]. Because of the challenges in SIoVs, such as topology change, frequent linkage interruption, and uncertain network size, how to conduct node detection, construct trust relationship and manage network flexibility and expandability have become crucial. Recently, SIoVs have provided a possible solution for the construction of social relationship among nodes, whose core idea is to build a socialized network connected by intelligent devices. In [15], the authors have presented an event-driven trust management model, which focuses on node subjective experience and the recommendation of common associated devices. However, the dynamic and adaptive set of suitable values for transmission threshold and time window are difficult to obtain. The rapid development of SIoVs accelerates the evolution from vehicular networks toward vehicular cyber-physical systems. A multi-layer context-aware architecture has been presented in [16], where two service components, i.e., SIoVs and context-aware vehicular security, have been detailed. Furthermore, a context-aware dynamic parking service-based application is studied. For the sake of coping with the dynamic topology, low bandwidth, and fleeting connections in SIoVs, a framework, i.e., VeDi, has been investigated for vehicular video social network, where suitable video can be selected with the help of shared metadata scores [17]. 2.3 Routing in SIoVs Although a routing study in SIoVs has received wide attention in recent years, the specific constraints and communication environment make it more complex than that in VANETs. A survey of the existing routing protocols in Internet of vehicles has been made in [18], which focuses on routing in Internet of vehicles in smart cities. Advantages and limitations of route planning and traffic prediction services have also been studied for accurate and efficient transmission. In order to stimulate selfish nodes forwarding data cooperatively, a copy adjustable incentive scheme has been studied in [19], which leverages social credit and non-social credit toward the nodes relaying data for other nodes inside their community or outsiders, respectively. For the sake of enabling social communica-

tions and interactions among users on highway travels, a framework of SIoV has been presented in [20], which motivates road users to provide updated information through inter-vehicle communications. By leveraging social properties of the involved entities in SIoVs, a data forwarding method inspired by the artificial bee colony has been investigated in [21], which takes individual’s learning capability and awareness into consideration. Recently, based on crowded GPS traces, an approach for night-bus route planning has been studied in [22], and the clustering-based method has been utilized for hot area identification by mining GPS traces. Thus, the path with a high number of passengers has a priority. Various methods for data dissemination in SIoVs have been summarized in [23-27], and the authors illustrate social behavior and mobility pattern of nodes are essential parts when designing a content dissemination protocol.

3

A QUALITY-AWARE ACCESS SERVICE SELECTION SCHEME

In this section, we first illustrate our concerned network system, and define the parameters. Finally, we describe a novel quality-aware access service selection scheme. 3.1 System description As shown in Fig. 1, we consider a heterogeneous and decentralized network scenario, where no fixed trust authority exists to provide trust evaluation. Node equipment moves with users, and joins or leaves the network dynamically. Each device contains its own information, such as the interest and provided service. User interest represents the characters and focuses of the provided service, which can be viewed as a map between user social connection in the real world and the social relationship among devices in the network system.

Fig. 1 System model

Since devices with various feature information would provide different evaluations to the same service, nodes form different groups and exhibit distinct social relationship. Since similar feature preference and connection with semblable evaluation criterion exist in the same group, similar ability of access services can be provided. Each node in the network may perform normally or abnormally. The former can provide qualified service and suitable recommendation to other devices positively. However, inferior service, incorrect feedback evaluation or recommendation may be provided in the latter circum-

2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

Tab. 2 Social relationship forms Relationship

Description

Weighting factor

Co-Work Co-Location Social Parental

Among objects that collaborate to provide common services Among objects that are in the same or similar working environment Among objects that occasionally or continuously interact with each other Among objects that belong to the same manufacturer or production batch.

0.4 0.3 0.2 0.1

stance, such that network service quality and stability would be affected. In our study, nodes may transform the behaviors between normal and abnormal as in real network situations. The main notations in this paper are summarized in Tab. 1. Tab. 1 Definition of Relative Parameters Definition

Si Ru , v (t ) TTL(u, si) I(u,v,t) C(u,v,t) S u ,v (t )

Pu,v(t) dir u,v

R (t ) Ruind, v (t )

j Gu

send(si) NodeList(u) Hop(u,v) MAXHOP

Description Type of access service Evaluation of an access service for nodes u and v during time t Time to live of node u for service Si Access count of all the services between nodes u and v before time t Collection of connected devices of nodes u and v during time t Node behavior similarity between nodes u and v during time t Connection time between u and v during t Direct observation between nodes u and v during time t Indirect observation between nodes u and v during time t Movement direction of node j Structure of a spanning tree for node u Requested transmission time of Si Neighbor node list of node u Routing distance (hop) between nodes u and v Maximum number of hops for transmission

During the process of network access, each node can either be a requester or provider of network services. Equipment maintenance records the access service and configuration information of evaluation and recommendation for other devices. The direct evaluation of an access service, indirect recommendation evaluation, socialized connection and interaction time prediction can affect access service selection directly. After each access service finishes, node updates the corresponding local data and configuration information depending on the feedback of service quality. 3.2 Parameter definition The studied access service selection scheme consists of the following three parts: A. Access service evaluation

Direct access service evaluation and indirect recommendation are employed for a node’s historical performance. The former’s parameters include total access count (I), and feedback of service quality (Q). By considering the features of different services (such as importance, and accuracy), the weighting factor of an access service (W) has also been embraced into the evaluation algorithm to lower the risk of shilling attacks. From the aspect of indirect recommendation, the common associated devices (C) are considered for recommendation information. To satisfy the requirement of dynamic evaluation, time attribute (t) is considered during node service access, evaluation, recommendation to record the timing differences of the current evaluation. B. Social relationship evaluation Since different groups are constructed by nodes with diverse features, similar with [28], we define four basic social relationships, i.e., the relationships of co-work, co-location, social and parental ones. Their corresponding description and weighting factors are illustrated in Tab. 2. With the objective of increasing service quality, we assign a high weighting factor to the service with high quality and correlation due to its high reliability. We use relationship dimension to denote the value of social relationship. C. Interaction time prediction The prediction model of interaction time is based on node movement information, including: movement velocity, coordinate position, time and direction. A time window of size N is leveraged to monitor the movement of other nodes. We introduce trajectory radius (R) and gradient (a) in the prediction process, while considering distance and connectivity evaluation in the prediction process to reflect a connection state. 3.3 Access service selection When a node needs network access, it first evaluates other surrounding nodes according to service quality, service relevance, and interaction time. The corresponding information can be estimated by the historical access record and configuration information. By considering a dynamic network scenario, each record and configuration information can be illustrated by a specific attribute at time t. This is because the timeliness of data should be considered for evaluation. Service quality evaluation is based on a node's historical access record and recommendation information, and future service quality can be predicted based on the pre-

2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

vious one. At time t, the evaluation of an access service ( Ru,v (t ) ) can be obtained by the integration of node own direct observation ( Rudir, v (t ) ) and common associated device ( Ruind, v (t ) ), as follows: ind R u , v ( t )  a  R udir , v ( t )  (1   )  R u , v ( t )

(1)

where weighting factor ɑ[0, 1]. The direct observation is calculated by: I ( u ,v , t )

Rudir, v (t ) 



I ( u ,v , t )

 ( t , i )  Q ( v , i )  W ( v, i ) /

i 1



W (v, i ) (2)

i 1

where I(u,v,t) represents the access count of all the services between nodes u and v before time t. Q(v,i) and W(v,i) illustrate the quality assessment and importance factor of node v during the ith access service, respectively. By considering the dynamic character of vehicular networks, a decay coefficient is introduced, and can be computed as:  ( t, i )  1 / ( t  t (i ) ) , where t(i) is the time of the ith network access. Indirect evaluation can be acquired by the computation of recommendation information from the common related devices. We consider the common related device of nodes u and v at time t, which can be viewed as a collection of node devices that have accessed these two nodes before time t. In our scheme, the evaluated result can be regarded as the confidence level between access node u and the co-related device v, which can be calculated as follows: C ( u , v ,t )

Ruind,v (t ) 

 j 1

C ( u , v ,t )

Rudir, j (t )  R dir j ,v ( t ) /



Rudir, j (t )

(3)

j 1

Herein, C(u,v,t) = {w| I(u,w,t)>0 and I(w,v,t)>0}, and it is the collection of co-associated connected devices of devices u and v during time t. It is noted that a high-evaluated value implies with highly credible relationship among devices, which would decrease the effect brought by false recommendation and malicious evaluation to an access quality, such that the accuracy of recommendation can be promoted. Socialized relationship among nodes is utilized as the mapping between the relationship in the real world and the characteristics of service demand. This is another index for evaluation to explore the potential social relationship. Its objective is to effectively and rapidly perceive and recognize trust nodes such that the efficiency and correctness of an access service can be evaluated. In order to deeply and comprehensively construct the social relationship among devices, behavior similarity is leveraged. We consider the similarities of both internal and external behaviors to evaluate it as follows: S u ,v ( t )    Suint,v ( t )  (1   )  Suext,v ( t )

(4)

where  is a weighting factor to indicate the importance of internal and external behaviors in overall similarity between nodes u and v. For the sake of quantitatively assessing internal behavior similarity, Pearson correlation coefficient is used for similarity computation. It means if the evaluations of two nodes tend to be the same from other devices, they are more likely to possess similar access requirement, and construct high-quality cooperation and information sharing. Otherwise, node devices with large evaluation difference would result in providing an inferior service or false evaluation. According to the four kinds of basic socialized relationship in Tab. 2, the similarities of internal and external behaviors can be calculated as follows: C (u,v ,t )

Suint,v (t ) 



dir dir dir Rudir ,k (t )  Ru (t )  Rv, k (t )  Rv (t )

(5)

k 1 C (u,v,t)

C (u,v ,t )

 (R

)

dir dir 2 u, k (t )  Ru (t )

k 1



 (R

)

dir dir 2 v ,k (t )  Rv (t )

k 1



Suext,v (t )  V (i )  Fu ,v (i, t )

(6)

i 1

where Rndir ( t ) illustrates the average evaluation value of node n.  is the external socialized relationship dimension. V(i) is the standard weighting value of node i. Fu,v(i,t) is the relationship strength of node i in each dimension, such as the service type of occurrence density of co-workers. Mobility of on-board equipment in SIoVs is one important issue for access service selection, since the predictions of node mobility and effective connection time affect network access and Quality of Service (QoS) directly. The analysis and computation of connection time are important for service access and routing selection. The related studies can be generally classified as linear and probabilistic methods. The former constructs a mobility prediction model according to a node’s current movement speed and direction, which requires low data size, and the computation is simple and fast. However, the prediction accuracy is low, especially under the situations of curve and accelerated movements. The latter predicts the possibility of effective connection in the near future based on the current movement speed, direction, accelerated speed and initial distance. Although relatively accurate result can be obtained, the requirements for parameters and computational cost are high. In this study, we present an interactive time prediction method based on movement trajectory. Our method analyzes node trajectory according to the time window, and considers the gradient velocity in both lateral and vertical directions, by which node connectivity can be quantified for trajectory prediction. Consider a time window with length n (tcur-n+1, …, tcur-1, tcur) and a prediction period of length m (tcur+1, …, tcur+m), where tcur is the current time point. Corresponding to each time slot in the time window, node movement direction can be expressed as cur-n+1, …, cur-1, cur. By simulating node steering motion in the real world to a circular

2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

motion with radius equals to r, the movement direction varies with path length can be illustrated as:  t  v t  v  , i.e., r  (7)  2 2 r where t is the time variation,  is the variation of movement angles corresponding to t . The movement speed and radius of a node can be represented by v and t, respectively. By (7), the corresponding radius ri from time window ti to the current time tcur can be obtained, such that a series of center points can be obtained, by which the anticipated trajectory radius rp can be predicted by: 1 cur 1 rp    v  (tcur  ti ) / ( cur  i ) (8) n cur n 1 The current location is set as the initial point. According to the predicted radius rp and (7), the corresponding movement direction j (j = 1,2,…,m) can be computed by: 1  j   cur   (t j  tcur )v (9) rp The velocities in horizontal and vertical directions are v xj  v  cos( j ) and v yj  v  sin( j ) , respectively. The accelerated speeds in x axis and y axis are a xj   sin( j ) and a yj  cos( j ) ,

respectively. By integrating the movement direction j in (9), node velocities and accelerated speeds, the corresponding displacement distances x j and y j at each time point can be worked out by: 1  2 x j  x j 1  v  cos( j 1 )  (t j  t j 1 )  2  sin( j 1 )  (t j  t j 1 )  1 y j  y j 1  v  sin( j 1 )  (t j  t j 1 )   cos( j 1 )  (t j  t j 1 ) 2  2

(10) where j  1,2, m . x j and y j satisfy x0  0, y0  0 . By (10), nodes can obtain the coordinates of the nodes to be estimated and themselves, thus the distance varying with time (Distu,v(t)) can be obtained. By combining the obtained distance and effective communication range (Cr) among nodes, the change curves among nodes can be derived as time goes on. After uniformization, the evaluation of connection time between nodes u and v (Pu,v(t)) is expressed as: T

Pu ,v (t ) 

Distu ,v (t )  1   1   dt T 0  Cr 

(11)

By considering the evaluation indexes from service quality, socialized relationship and anticipated connection, the access evaluation Tu,v(t) between nodes u and v during time t can be evaluated by: Tu ,v ( t )  1  Ru ,v ( t )  2  S u ,v (t )  3  Pu ,v (t )

(12)

where 1  3 are the corresponding weighting values of service quality, socialized relationship and anticipated connection, respectively, satisfying 1  2  3  1 .

After that, a selective list for access service evaluation can be generated, and the service evaluation is started from the nodes with high evaluation value. Then, the obtained QoS is fed back from the request node u to provider v, i.e.,  u,v(t). To guarantee high-efficiency transmission, the corresponding stimulation or punishment mechanism is also made, node u evaluates the collection of associated node j in collection C(u,v,t) by:

 u , j ( t )  1   u ,v ( t )  R dir j ,v ( t )

(13)

If the recommended information from node j is close to the obtained access service quality from node u, i.e., node j has provided a proper suggestion, then it receives a positive evaluation. The feedbacks no matter from access service or recommendation are recorded into the local interactive data for a stimulation or punishment decision. Positive feedback distinguishes services with different features and access quality, such that the QoS provided by nodes can be promoted.

4

COOPERATIVE QUALITY-AWARE SERVICE ACCESS

Since link connections are unstable in SIoVs, a network structure varies over time and interference conditions are complex. Therefore, how to select a suitable access object and transmission link is an important issue. Unfortunately, it is difficult to optimize the selection of both access node and transmission link. In this section, by fully considering network access from the viewpoint of both network nodes and links, we present the CQS method.

4.1 Network model According to the studied scheme in Section 3, nodes can perceive and evaluate other devices from the aspects of access QoS, node social relationship and connection time prediction, based on which access request can be launched by suitable devices. In SIoVs, since nodes communicate with each other by wireless technologies (such as Bluetooth and WiMAX), communication among devices is restricted by the propagation path. In order to effectively detect network service with a larger range, we further extend the QoS-aware access service selection scheme to multi-hop dynamic network environment. With the objective of interacting with other devices outside their communication range, a local dataset, including access service record and configuration information, is leveraged to evaluate service quality and explore node social relationship. Furthermore, the information of routing, buffer, and neighbor should also be maintained by node devices. Details are illustrated below. According to the played role during the process of network access, node devices can be generally classified into: 1) Service requester, i.e., the node that launches a service access request; 2) Service provider, including devices that can provide services to others; 3) Service

2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

collaborator, i.e., the node that provides the service of network access or route. It is noted that the role of any node device may change or play multiple roles. A node receives packets from its surrounding nodes periodically, and performs routing according to their types. During each cycle, a node also broadcasts its configuration and neighbor information, and sends packets for a service request if necessary. Based on graph theory, the studied network is defined as a self-centered width-first spanning tree G = (V, E), where V is the number of nodes, and E represents the relationship among nodes. If nodes u and v have an interaction, e = (u, v)  E exists. We assume the maximum height of the spanning tree is 3, i.e. the maximum routing distance is three hops from the root node to the end node. The main considerations are: 1) From the aspect of QoS, long routing distance accompanies with high interruption risk due to the low stability in vehicular networks. However, the communication distance limits network QoS of the spanning tree. 2) From the consideration of network cost and computational cost, a larger spanning tree requires more beacons for maintenance. Besides, it is also resource-consuming for nodes to compute and maintain a neighbor network spanning tree. A node first checks its received queue, and verifies whether a received packet is repetitive. If so, the node judges whether a more reliable routing path exists according to the perception result of the access QoS, and determines whether to update its routing table or not. For the provided neighbor spanning tree, nodes maintain the structure of a spanning tree by pruning and merging, as follows: Gu  Gu  Gv3 hop ( u ,v )

where G vx  G v , and satisfies for any

(14) w  G V

,

x.w  Gvx

hop ( v, w )   V . The maximum hop number between nodes u and v is 3 (i.e., 3  hop ( u, v ) ). Then, a node decides whether to forward its packet according to the routing hops of the float packet or not. Thus, the number of packet transmissions can be limited to a certain range, such that network efficiency can be guaranteed.

4.2 Cooperative routing selection By integrating the proposed access service selection scheme in Section 3 and the construction algorithm of an adjacent network in Section 4.1, nodes can evaluate the access QoS of nodes in the spanning tree, and record the evaluation result in the local neighbor node table. Due to the characters of a dynamic network structure and low stability of link connection, selecting an access node and link with high quality is significant. In this section, we present a cooperative routing selection protocol by considering the strategies of QoS-priority, link quali-

ty-priority, and hybrid selection. The definitions of the relevant parameters can be seen in Tab. 3. 4.2.1 QoS-priority routing strategy Under the QoS-priority routing strategy, a request node starts the access requirement to the node with the highest evaluation value of an access service. If the service failure packet (SNTF) is received, the corresponding node sends the request to the node with the next highest evaluation value of the access services and so on. As shown in Algorithm 1, a node first ranks its local neighbors according to the evaluation result of the access services. When the requirement for an access service is generated, a node first sends the request to the node with the best evaluation result. Then, it sets Time-To-Live (TTL) as the product of single-hop Round Trip Time (RTT) and the number of routing hops between these two nodes. After that, node starts to sense packet transmission in the network. If the received packet exceeds its TTL, the node would discard the current request and restart the request toward the suboptimal node in the neighbor node table and so on until the node has successfully acquired the access service or all the requests of nodes in the neighbor node table have been traversed. Algorithm 1. QoS-priority routing strategy Requester Process: 01 when u genetates an access requirement request do 02 si⟵request.servceid 03 TTL(u, si) ⟵ hop(u, NodeList(u).Top) RTT 04 tsend(si)⟵tcur 05 Send SREQ(u, NodeList(u).Top, si) to NodeList(u).Top 06 end 07 when u receives SRSP(v, u, si) from v do 08 Update receivelist(u) and save Si to the local 09 end 10 when (u receives SNTF(v, u, si) from v )||( tcur >tsend(si) + TTL(u, si)) do 11 if NodeList(u).Next != NULL then 12 TTL(u, si) ⟵ hop(u, NodeList(u).Next) RTT 13 tsend(si) ⟵tcur 14 Send SREQ(u, NodeList(u).Next, si) to NodeList(u).Next 15 else 16 Rebuilt NodeList(u) and Restart from Top 17 endif 18 end Collaborator Process: 19 when w receives SREQ(u, v, si) form u do 20 Transfer the packet to Route(w, v) 21 end 22 when w received SRSP(v, u, si)||SNTF(v, u, si)|| STRN(v, u, si) from v do 23 Transfer the packet to Route(w, u) 24 end





4.2.2 Link quality-priority routing strategy Compared with the method focusing on the evaluation of access QoS, the proposed link quality-priority strategy is more stable to fulfill the service requirement and network access. When a service request is generated, a node selects the nodes according to the optimal evaluation value in the node collection within one hop. The evaluation value is computed by the ranking consequence of node

2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

table in the neighbor network. Then, the node sends a service request packet with a cooperative request identifier. If the service forwarding packet from its cooperative node is received by the request node, it modifies the TTL of this request and keeps monitoring the new feedback information until the service is successfully accessed. If the received request from the request node is failed or exceeds the corresponding TTL, this request is given up. If all the attempts have failed, its neighbor network would be reconstructed and node table would be recomputed to restart request from the node with the optimal evaluation value in the node collection. When the request packet is received by the cooperative node, it first checks whether this packet was received before. Then, this node checks whether the service type of the request can be supplied. If so, the response packet is returned to the request node directly. Otherwise, this node judges whether the packet contains a cooperative request identifier and the current routing hop satisfies the predefined requirement. If so, its own network identification is filled into the segment of the forwarding node, the routing hop segment is increased by one, and the optimal node from its own neighbor network node table is selected to fill into the objective segment as a forwarding node. At last, the cooperative node forwards the packet information of service to the next hop node and starts to monitor the feedback information of the next hop node. For the received service forwarding packet or response packet, the cooperative node forwards the corresponding packet to the request node according to its routing table. 4.2.3 Hybrid routing selection It is obvious that both QoS priority and link quality based routing selection strategies have their advantages. First, the former is more likely to obtain the globe optimal access service according to the evaluation of the selection method, but the risk of link instability increases. Second, the latter employs an iterative one-hop connection method to extend the selection range. However, during each selection process, it only obtains the local optimal objective. The stability of the current neighbor network is the fundamental factor for access method selection. When node density is high in a network, more end-to-end connection paths can be selected to decrease the risk of link interruption, such that the QoS-priority selection method can help a node access a high-quality service effectively. When the network is unstable and few devices are available for connection, the iterative link quality-based selection method can provide a stable routing path by link connection selection with multiple single-hop transmission to get a relatively high-quality access service. Based on the above observation, we propose a hybrid selection strategy by taking the advantages of both strategies mentioned before. Node density is considered for routing selection, i.e., if the density of connection nodes in a neighbor network is higher than a threshold, QoS-priority

routing is selected to launch access services towards the node with the highest evaluation. Otherwise, link quality-priority routing is employed to guide a request node to obtain a local optimal service by constructing a stable routing path. 4.3 Bidirectional caching algorithm By limiting the number of packets and node response time, a cache management mechanism can effectively improve network performance. Besides, the correctness of routing selection and access service can also be promoted by integrating a cache mechanism into a routing strategy, and the access QoS in a dynamic network can thus be promoted. We investigate a bidirectional cache algorithm for the studied CQS scheme by maintaining a node’s cache table. The table records the complete information of an access process, no matter of the network access service from a requester or collaborator. For the requester, once a service request is accessed by a node successfully, the information of its corresponding identification, access type and quality is added into its cache table. For the provider of the access service, when a node receives a request packet, the service type information is added into the cache table. For the access service of the collaborator, when receiving packet from a cooperative node, the service type information of both request node and cooperative node is recorded into the cache table of the request node. Details of the algorithm steps can be found in Algorithm 2. Algorithm 2. Bidirectional caching algorithm Requester Process: 01 when u receives SRSP(v, u, si) from v do 02 if (v, si) already in cache then 03 Update cache: t(v, si)⟵tcur 04 else 05 Insert cache: new log(v, si, tcur) 06 endif 07 end Provider Process: 08 when v receives SREQ(u, v, si) from u do 09 if (u, si) already in cache then 10 Update cache: t(u, si) ⟵tcur 11 else 12 Insert cache: new log(u, si , tcur) 13 endif 14 end Collaborator Process: 15 when w receives SREQ(u, v, si) from u do 16 if (u, si) already in cache then 17 Update cache: t(u, si) ⟵tcur 18 else 19 Insert cache: new log(u, si , tcur) 20 endif 21 end 22 when w receives SRSP(v, u, si) from v do 23 if (v, si) or (u, si) already in cache then 24 Update cache: t(v, si) ⟵tcur or t(u, si) ⟵tcur 25 else 26 Insert cache: new log(v, si, tcur), (u, si, tcur) 27 endif 28 end

2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

It should be noted that on one hand, overlong record challenges the real-time network performance, affects the selection of access service and lowers the success rate of a routing strategy. On the other hand, frequent data exchanges would cause additional system storage and computational resource cost. Therefore, the cache time has to be properly set.

5

SIMULATION RESULTS

The performance evaluation is conducted by Matlab 2012. As shown in Fig. 2, we consider a 3Km  3Km street scenario, and the distance between two streets is 500 meters. There are three kinds of intersections, i.e., Cross-intersection, T-intersection, and L-intersection. 1000 vehicles (nodes) travel randomly in the streets. When a vehicle reaches a cross-intersection, we assume that there is 40% probability to go ahead following the original direction, 25% probability to turn left or right, and 10% probability to turn back. Similarly, when a vehicle reaches a T-intersection, there is 40% probability to go left, 40% probability to go right, and 20% probability to turn back. When a vehicle reaches an L-intersection, the probabilities for vehicles to turn and go back are 70% and 30%, respectively. The speed of vehicles changes between 60 Km/h and 120 Km/h, and the wireless communication range between vehicles varies between 100 meters and 500 meters. According to a series of experiment for parameter selection, we define 1 , 2 and 3 as 0.4, 0.3 and 0.3, respectively.  and  are set to 0.7 and 0.6, respectively. Since network security and malicious attack is not our main consideration, we set the probability that network performs abnormally to 20%. For simplicity, we consider all the requests and access services have the same importance. The time interval for iteration is 100ms, and the one-round simulation lasts for 60 minutes.

Fig. 2

Diagram of the simulation environment

Fig. 3 Illustration of simulation scenarios

For the sake of evaluating the performance of access services, we consider 10 types of services, denoted as S1  S10 , and each vehicle (node) has the ability to provide different kinds of services. The concerned service type of each vehicle is set randomly according to the service type, which affects the access probability of each service. Due to the limited computational ability of vehicles, an iterative method is employed by setting time interval at 100ms. During each iteration, vehicles handle the data in its receiving buffer and write the data into its sending buffer to determine the following movement situation. After each iteration, the system updates the information of the buffer area and coordinate position of each vehicle. The simulation process lasts for 60 minutes. The considered system model has been illustrated in Fig. 3, which consists of the most common moving behaviors of vehicles, i.e., parallel driving, separated driving, and encounter driving. In the parallel driving, nodes A and B keep the same movement in the horizontal direction. In the separated driving scenario, node A keeps moving in the horizontal direction, while node B travels from the horizontal direction to the vertical one. In the encounter driving situation, node A keeps moving in the horizontal direction, while node B travels from the vertical direction to the horizontal one. Tab. 3 is the parameter setting of different simulation situations. The linear link terminated time prediction method (LET) in [29] and the Wiener process-based probability density algorithm (WP) in [30] are utilized for comparison. Network connectivity of three situations in Fig. 3 can be observed in Fig. 4. We notice that although the LET method seems to have larger network connectivity values under different circumstances. Actually, it has the worst network performance, since it derives the value of theoretical value too much. We can observe that our CQS method has good consistency with the theoretical value. For the considered three network scenarios, the average inaccuracies of our method are 1.3%, 2.9% and 10.1% respectively, while they are 4.5%, 4.3% and 11.7% respectively for WP method. The reason behind is that our method is dynamic and is based on vehicle trajectory. To analyze the studied CQS method further, we evaluate the success rate of access services, average access QoS, and average transmitted number of packets. The Receive on Most Stable Group-Path (ROMSGP) method [29], the Hybrid Location based Ad hoc Routing (HLAR)

2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

Fig. 4 The predicted accuracy of connection time Tab. 3 Parameter settings of simulation scenarios Scenario Parallel driving Separated driving Encounter driving

Fig. 5 Success rate of network

Fig. 6 Average service quality

Fig. 7 Average number of packets sent per access

Parameter description Velocity: Va=5m/s; Vb=10m/s; Acceleration: aa=1m/s2; ab=2m/s2; Initial distance: d=0m. Velocity: Va =5m/s; Vb =10m/s; Acceleration: aa =0m/ s2; ab =0m/ s2; Initial distance: d=0m. Velocity: Va =5m/s; Vb =10m/s; Acceleration: aa =0m/ s2; ab =0m/ s2; Initial distance: d=100m.

method [31], and the Hybrid Cooperative Service Discovery (HCSD) [32] are chosen for performance comparison. The success rate of access services is defined as the percentage between the number of the successfully received request services and the number of the total request services. This index illustrates the effectiveness of access service and routing selection among different methods. As shown in Fig. 5, the success rates of different algorithms increase obviously as the extension of a node’s communication range. The reason is that as the more equipment can be accessed, nodes can sense more access services, such that the selection of a routing path becomes plentiful and high success rate of access services can be achieved. When the node communication range is between 100m and 200m, the sensed number of the neighbor node is few, and network is not so stable. Therefore, HLAR performs better since it mainly focuses on link stability. As the communication range becomes large, especially between 300m and 500m, the CQS method and HCSD algorithm perform better due to the increase of the available equipment for link connection. Since our method jointly considers the QoS-priority and link-priority strategies, it can effectively accommodate different network situations. We then evaluate network performance of average access QoS, and set the communication range of node as 300m. The obtained QoS value under successful requests is the value that can be provided to a network service for a node. This value would be 0 if the request is failed due to link failure. As depicted in Fig. 6, compared with the random selection method for an access node, the values of the service quality of the other four algorithms increase obviously. For HLAR and ROMSGP, the corresponding increase values of service quality are relatively

2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

slow. This is because these two algorithms concentrate on link state while lacking the awareness of service quality. By employing a cooperative node and cache mechanism for node access selection, the service quality in HCSD increases obviously. However, because of the limitation of the number and ability of cooperative nodes, the achieved network performance is inferior to CQS scheme. At last, we evaluate the performance of the average number of packets sent by each network access, which is the ratio of the number of sent packets by nodes to the number of services accessed by nodes. This index affects the network cost by node access for high quality service. As depicted in Fig. 7, the number of packets increases as a node’s communication range extends. The reason behind it is the rapid growth of transmitted packets for topology maintenance of a neighbor network. Since nodes always keep connecting in HLAR even with few packets, the average number of packets sent by HLAR is the lowest. When the node communication range is small, both HCSD and ROMSGP attempt to access the network by broadcasting a request, such that the network cost is high. This is because link connection is unstable and it is difficult to acquire the information of a neighbor network. As the communication range increases, node sensing ability and evaluation accuracy increase, and thus both HCSD and CQS approach the performance gained by HLAR. Thus, we can draw the conclusion that our method can significantly increase the success rate and quality of access services with the cost of adding network resource consumption slightly.

[2] J. Dias, J. Rodrigues, and L Zhou, Cooperation Advances on Vehicular Communications: A Survey, Vehicular Communications, vol. 1, no. 1, pp. 22-32, 2014.

6

[14] Z. Ning, F. Xia, X. Hu, Z. Chen, M. Obaidat. Social-oriented Adaptive Transmission in Opportunistic Internet of Smartphones, IEEE Transactions on Industrial Informatics, vol. 13, no. 2, pp. 810-820, 2017.

CONCLUSION

In this paper, we construct a CQS system, focusing on reliability guarantee and service quality promotion in SIoVs. We first study a dynamic access service evaluation scheme to cope with the effect brought by the dynamic network change. Then, we present a social relationship evaluation method to explore the internal and external similarities among vehicles. Furthermore, we investigate a prediction method according to vehicle movement trajectory for interaction time estimation. At last, we introduce a CQS method, which first constructs a node-centric generation tree structure to compute the access quality, then selects an access path according to the current network state. A bidirection buffering algorithm is also investigated to improve the response efficiency and accuracy of our method. Simulation results demonstrate the superiority of CQS method in terms of request success rate, quality of service, and average request cost over its peers in [29-32].

REFERENCE

[3] J. Dias, J. Rodrigues, N. Kumar, K. Saleem, Cooperation Strategies for Vehicular Delay-Tolerant Networks, IEEE Communications Magazine, vol. 53, no. 12, pp. 88-94, 2015. [4] S. Smaldone, et al, RoadSpeak: Enabling Voice Chat on Roadways using Vehicular Social Networks, Social Network Systems, pp. 43–48, 2008. [5] Z. Ning, F. Xia, N. Ullah, X. Kong, and X. Hu, Vehicular Social Networks: Enabling Smart Mobility, vol. 55, no. 5, pp. 49-55, IEEE Communications Magazine, 2017. [6] J. Dias, J. Rodrigues, C. Mavromoustakis, F Xia, A Cooperative Watchdog System to Detect Misbehavior Nodes in Vehicular Delay-Tolerant Networks, IEEE Transactions on Industrial Electronics, vol. 62, no. 12, pp. 7929 - 7937, 2015. [7] F. Wang, N. Zheng, D .Cao, Parallel Driving in CPSS: A Unified Approach for Transport Automation and Vehicle Intelligence, vol. 4, no. 4, pp. 577-587, IEEE/CAA Journal of Automatica Sinica, 2017. [8] Z. Chen, Z. Ning, Q. Xiong, M. Obaidat, A Collaborative Filtering Recommendation based Differentiated Access Service Selection Scheme in Large-Scale WLANs, IEEE Systems Journal, DOI: 10.1109/JSYST.2016.2542179, pp. 1-11, 2016. [9] K. Shafiee, V. Leung, R. Sengupta, Request-Adaptive Packet Dissemination for Context-Aware Services in Vehicular Networks, IEEE Vehicular Technology Conference, vol. 13, no. 1, pp. 1-5, 2012. [10] K. Abrougui, A. Boukerche, R. Pazzi, et al, A Scalable Bandwidth-Efficient Hybrid Adaptive Service Discovery Protocol for Vehicular Networks with Infrastructure Support, IEEE Transactions on Mobile Computing, vol. 13, no. 7, pp. 1424-1442, 2014. [11] C. Xu, S. Jia, M. Wang, H. Zhamg, G. Muntean, Performance-Aware Mobile Community-Based VoD Streaming Over Vehicular Ad Hoc Networks, IEEE Transactions on Vehicular Technology, vol. 64, no. 3, pp. 1201-1217, 2015. [12] W. Quan, C. Xu, J. Guan, H. Zhang, L. Grieco, Social Cooperation for Information-Centric Multimedia Streaming in Highway VANETs, IEEE WoWMoM, pp. 1-6, 2014. [13] P. Mastroianni, et al, Local Optimization Strategies in Urban Vehicular Mobility, PLOS ONE, vol. 10, no. 12, pp. 1-13, 2015.

[15] F. Bao and R. Chen. Trust Management for The Internet of Things and Its Application to Service Composition, IEEE WoWMoM, pp. 1-6, 2012. [16] J. Wan, D. Zhang, S. Zhao, L. Yang, J. Lioret, Context-Aware Vehicular Cyber-Physical Systems with Cloud Support: Architecture, Challenges, and Solutions, IEEE Communications Magazine, vol. 52, no. 8, pp. 106-113, 2014. [17] K. Alam, M. Saini, D. Ahmed, A. Saddik, VeDi: A Vehicular Crowd-Sourced Video Social Network for VANETs, IEEE LCN, 738-745, 2014. [18] J. Cheng, J. Cheng, M. Zhou, Q. Zhang, C. Yan, Y. Yang, C. Liu, Routing in Internet of Vehicles: A Review, IEEE Transactions on Intelligent Transportation Systems, vol. 16, no. 5, pp. 2339-2352, 2015. [19] Z. Ning, L. Liu, F. Xia, B. Jedari, I. Lee, W. Zhang, CAIS: A Copy Adjustable Incentive Scheme in Community-based Socially-Aware Networking, IEEE Transactions on Vehicular Technology, 66(4): 3406-3419, 2017. [20] T. Luan, R. Lu, X. Shen, F. Bai, Social on the Road: Enabling Secure and Efficient Social Networking on Highways, IEEE Wireless Communications, vol. 22, no. 1, pp. 44-51, 2015. [21] F. Xia, L. Liu, J. Li, A. Ahmed, L. Yang, J. Ma, BEEINFO: An Interest-based Forwarding Scheme Using Artificial Bee Colony for Socially-aware Networking, IEEE Transactions on Vehicular Technology, vol. 64, no.3, pp. 1188-1200, 2015 [22] X. Zuo, C. Chen, W. Tan, M. Zhou, Vehicle scheduling of an urban bus line via an improved multiobjective genetic algorithm, IEEE Transactions on Intelligent Transportation Systems, vol. 16, no. 2, pp. 1030-1041, 2015.

[1] M. Zhou, et al, Guest editorial special section on advances and applications of Internet of Things for smart automated systems, IEEE Transactions on Automa[23] M. Wang, H. Liang, R. Zhang, R. Deng, X. Shen, Mobility-Aware Coordinated Charging for Electric Vehicles in VANET-Enhanced Smart Grid. IEEE tion Science and Engineering, vol. 13, no. 3, pp. 1225-1229, 2016. 2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/JIOT.2017.2764259, IEEE Internet of Things Journal

Journal on Selected Areas in Communications, vol. 32, no. 7, pp. 1344-1360, 2014. [24] N. Kumar, S. Misra, J. Rodrigues, M. Obaidat, Coalition Games for Spatio-Temporal Big Data in Internet of Vehicles Environment: A Comparative Analysis, IEEE Internet of Things Journal, vol. 2, no. 4, pp. 310-320, 2015. [25] R. Deng, R. Lu, C. Lai, T. Luan, H. Liang, Optimal Workload Allocation in Fog-Cloud Computing Towards Balanced Delay and Power Consumption, IEEE Internet of Things Journal, vol. 3, no. 6, pp. 1171-1181, 2016. [26] B. Silva, J. Rodrigues, N. Kumar, M. Proença, G. Han, MobiCoop: An Incentive-Based Cooperation Solution for Mobile Applications, ACM Transactions on Multimedia Computing Communications and Applications, vol. 12, no 4, 49, 2016. [27] Z. Ning, F. Xia, X. Kong, Z. Chen, Social-Oriented Resource Management in Cloud-Based Mobile Networks, IEEE Cloud Computing, vol. 3, no. 4, pp. 34-41, 2016. [28] L. Atzori, A. Iera, G. Morabito, SIoT: Giving a social structure to the internet of things, IEEE communications letters, vol. 15, no. 11, pp. 1193-1195, 2011. [29] T. Taleb, E. Sakhaee, A. Jamalipour, et al, A Stable Routing Protocol to Support ITS Services in VANET Networks, IEEE Transactions on Vehicular Technology, vol. 56, no. 6, pp. 3337-3347, 2007. [30] T. Luan, R. Lu, X. Shen, F. Bai, Social on the Road: Enabling Secure and Efficient Social Networking on Highways, IEEE Wireless Communications, vol. 22, no. 1, pp. 44-51, 2015. [31] M. Al-Rabayah and R. Malaney, A New Scalable Hybrid Routing Protocol for VANETs, IEEE Transactions on Vehicular Technology, vol. 61, no. 6, pp. 2625-2635, 2012. [32] A. Lakas, M. Serhani, M. Boulmalf, A Hybrid Cooperative Service Discovery Scheme for Mobile Services in VANET, IEEE WiMob, pp. 25-31, 2011.

2327-4662 (c) 2017 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

Suggest Documents