An Adaptive Clustering Scheme for Improving the Scalability in Intelligent Transportation Systems

Intelligent Transportation Systems (ITSs) aim at making the vehicle users be safer, comfortable, and better informed. Internet-enabled ITS applications get increasingly attractive recently. In this context, gateway is the portal of accessing the Internet. The efficiency of gateway management in ITSs thus profoundly affects the applicability of Internet-enabled applications. When the number of vehicles on the road greatly increased, the effectiveness of gateway requisition challenges the usage of the ITS services. To remedy the scalability problem in accessing the Internet in ITSs, an adaptive clustering scheme for gateway management is proposed in this paper. In the proposed scheme, the gateways are organized as a two-level cluster architecture. The gateways are dynamically employed according to the QoS requirements of the applications and current networking conditions. As shown in the simulation results, the proposed scheme is able to achieve the QoS requirements with fair gateway deployment cost.


Introduction
Intelligent Transportation Systems (ITSs) [1] focus on supplying transport services and traffic management. Advances of vehicular ad hoc network (VANET) technologies [2] make ITS services effectively accessible. Recently, Internetenabled ITS applications become more and more appealing. A gateway is the gate to access the Internet in the context of ITSs. Internet access in VANETs is usually supplied by way of fixed gateways [3,4] along the road. Oppositely, for super high mobility of the VANET users [2], mobile gateways [3][4][5][6][7][8] are introduced for stably connecting to the Internet for vehicle users. Therefore, the performance of gateway dispatching in ITSs significantly impacts the applicability of Internet-enabled applications. Especially, when the number of vehicles on the road remarkably increased, the efficiency of gateway requisition challenges the willingness of using the services.
To remedy the scalability problem of accessing the Internet in ITSs, a cloud-assisted adaptive gateway dispatching scheme is proposed in this paper. In this proposed scheme, the gateways are organized as a two-level cluster architecture. The gateways are employed dynamically according to the QoS requirements of the applications and current networking conditions. Briefly, in our design, four fixed gateways are uniformly deployed in the system in advance and sixteen mobile gateways are dynamically called up to increase network capacity for improving the QoS. The simulation results will show that the proposed scheme is able to achieve requested QoS requirements with less gateway deployment cost.
The rest of this paper is organized as follows. Background and related work is described in Section 2. Section 3 overviews the proposed scheme. The simulation results are discussed in Section 4. A brief conclusion is presented in Section 5.

Preliminary
ITSs [1] aim at making the vehicle users be better informed, safer, and smarter. ITSs applications [1] include emergency vehicle notification, automatic road enforcement, variable speed limits, collision avoidance, and dynamic traffic light sequence. VANET [2] is one of the most important enabling technologies for ITS. The special features of VANETs [2] include sufficient transmission power, powerful computation capability, huge scale, high and predictable mobility, partitioned network, and changing network topology and connectivity. Among others, due to typically large scale of the VANETs, the scalability [9] of proposals remarkably constrains the applicability of VANET applications.
Quite a few studies [9][10][11] investigate the scalability problem in MANETs/VANETs. Where, clustering schemes are suggested in the studies [12][13][14] for solving the scalability problems in VANETs. The authors [13,14] also mention that the effectiveness of the selection of cluster head seriously affects the efficiency of VANET applications. Various cluster head selection methods are introduced in the studies [13][14][15][16]. In the studies [14][15][16], the geographical information (i.e., the location, moving direction, and moving speed, etc.) of the nodes is taken into consideration for choosing the cluster head. Oppositely, the study [13] chooses the node supporting broadest coverage as the cluster head. For vehicle to infrastructure (V2I) communications, the authors of [17][18][19] take the gateway as the cluster head for accessing the Internet.
A hybrid architecture integrating V2V (Vehicle to Vehicle) and V2I (Vehicle to Infrastructure) is proposed in [11] to solve the scalability problem in VANETs. The considerations of AP discovery, handoff management, and the settings of TTL are discussed in the study. In [20], a solution is proposed to access the Internet in MANET. Three methods for finding gateways, namely, reactive, proactive, and hybrid gateway discovery, are introduced and compared in this study. In [21], the packet delivery rate can be greatly improved by combining static node mechanism and mobile cluster approach. A static node is set up at the road junction to meet the high mobility and scalability requirements while mobile cluster is used for helping routing in VANETs. However, Internet access is not remarkably addressed. In [17], a position-based routing protocol, namely, MGRP (Mobile Gateway Routing Protocol), is proposed for routing in VANETs. However, the scalability problem is not considered in MGRP routing protocol. In [22], the vehicles are dynamically clustered according to various metrics. A mobile gateway management method is proposed in [22]. Also, the ways of discovery, selection, and handoff handling of the mobile gateways are described. The study [9] offers a survey of cluster schemes in MANETs. The mechanisms, evaluations, and pros-and-cons of related algorithms are addressed.
However, it is quite laborious to discover fitting gateways to obtain steady connection to the Internet resulting from high mobility and changing topology of the VANETs [2]. Cloud computing represents the new model for supplement, consumption, and delivery of information technology based on the Internet protocols. The powerful processing and storage capability can thus be used for complicated gateway management.

Problem Statement.
Quite a few issues have to be solved to make ITS applications satisfactorily accessible. Among them, the efficiency of Internet access challenges the quality of Internet-enabled services for vehicular users. Here, Gateway is recognized as the portal to the Internet. Thus, gateway management significantly affects the system performance and quality of services (QoS). Apparently, more gateways can supply more networking capacity and better QoS. Traditionally, fixed gateway is deployed with the roadside infrastructure. More gateways thus imply expensive infrastructure cost; that prohibitively hinder ubiquitous deployment of the fixed International Journal of Distributed Sensor Networks  gateways. To remedy this problem, an adaptive cluster scheme is proposed in this paper to solve the scalability problem in ITSs. In the proposed scheme, mobile gateways are dynamically employed according to the QoS requirements of the applications and current networking conditions. That is, when the traffic load gets heavier and/or the vehicle's velocity becomes faster than the predetermined thresholds, in addition to fixed gateways, the proposed adaptive clustering scheme hires additional mobile gateways to supply sufficient networking capacity to satisfy requested QoS requirements. Oppositely, it dismisses superfluous mobile gateways to reduce the corresponding cost when the traffic load becomes light. The considerations of the QoS requirements and the settings of the thresholds will be addressed next.  Clustering server (1) The gateway willing to supply gateway service periodically broadcasts its advertisement message in its service coverage (2) When a client vehicle moves into the coverage of the gateway, it will receive the advertisement message. Then, the client vehicle sends a join request message to the associated gateway cluster server (3) The gateway forwards the join request message to the (4) The cluster server runs the adaptive clustering scheme. When current capacity cannot satisfy needed QoS requirement, the system will upgrade to Level-2 cluster architecture. Also, the cluster server updates related tables accordingly.
(5) The cluster server sends a clustering message to the gateway for informing the decision. Properly, the associated gateway will update related tables and adjust its deployment.  cluster server.
(2) The gateway forwards the leave request message to the (3) The cluster server runs the adaptive clustering scheme. When the current capacity is more than necessary, the system will downgrade to Level-1 cluster architecture. The clustering server updates related tables agreeably.
(4) The cluster server sends a clustering message to the gateway for informing the decision. Correspondingly, the associated gateway will update related tables and adjust its deployment.
(5) The gateway then sends a leave response message to the requesting client vehicle. The session between the client vehicle and the associated gateway is then closed. SGs are uniformly furnished in the system. While the Mobile Gateways (MGs) are the vehicles on the road which are able to directly connect to the Internet. The serving coverage of a MG is called a Microcell (U Cell). In this paper, the MGs are engaged dynamically according to the networking conditions and the applications' QoS requirements. A Client Vehicle (CV) is the vehicle desiring to access the Internet. A CV can access the Internet either via direct connection or via a few Relay Vehicles (RV) to a Gateway. That is, in case that a CV is not currently located within the coverage of any Gateway, the CV then connects to the Internet via one or more vehicles relaying packets from/to a Gateway. A few cloud servers act as the Cluster Servers that are in charge of the clusters management and handling the mobility of the vehicles.
As will be displayed in the simulation results of Section 4.2, more gateways offer more networking capacity but unfortunately increased deployment cost. To obtain satisfactory QoS with affordable deployment cost, an adaptive cluster architecture is used in the proposed system. Initially, the system adopts Level-1 (L1) cluster architecture. That is, four fixed roadside APs/BSs are setup and act as the Stationary Gateways (SGs) to support Internet-enabled services. In this situation, a Client Vehicle (CV) connects to one of these four SGs to access the Internet, as depicted in Figure 2(a). When the networking capacity cannot meet the applications' QoS requirement, the Cluster Server then will upgrade the system to Level-2 (L2) cluster architecture by dynamically employing sixteen Mobile Gateways (MGs).   In this case, a CV connects to one of these sixteen MGs for accessing the Internet, as depicted in Figure 2(b). After that, the Cluster Server is responsible of adaptively deciding the system architecture (i.e., L1 or L2) according to the QoS requirements and current networking condition.

The Adaptive Clustering Scheme.
According to the QoS requirements and current networking status, the adaptive clustering scheme proposed in this paper is able to dynamically adjust the cluster architecture. That is, to obtain the needed QoS requirements, when the traffic load gets heavy and/or vehicle's velocity gets fast, Level-2 cluster architecture will be adopted. Otherwise, Level-1 cluster architecture will be used. Specially, the communication range of the gateways (either SGs or MGs) will be adaptively shrunk to reduce the probability of traffic collisions and get better performance when the traffic load becomes super high. Thus, the settings of    the thresholds triggering the cluster architecture adjustment significantly affect the system performance. Next, the basis of selecting these thresholds will be described. Related parameters used in the following discussion are collected in Table 1. Currently, in this paper, three QoS requirements (denoted as QoS i , = 85, 90, and 95) mean that 85%, 90%, and 95% packet delivery rate can be guaranteed, respectively. To properly set these thresholds, a series of simulations are carried out in this paper. Details of these experiments will be described in Section 4.3. Also, you can find the

Simulations
To systematically evaluate the performance of the proposed scheme, a series of simulations are presented hereinafter. and 250 m. IEEE 802.11b is used as the MAC layer protocol; whose bandwidth is 2 Mbps. The transmission type is CBR (Constant Bit Rate), whose transmission rate is 512 Kbps. The interval of sending packet is 0.5 s. In this paper, the traffic load depends on the number of vehicles participated in the packet transmission. For now, the performance is investigated with various traffic loads (i.e., 20%, 40%, 60%, and 80%). The routing protocols including MGPR [17] and AODV+ [20] are studied and compared with the proposed scheme. In the following simulations, Manhattan mobility model [23] is used. The settings of the parameters are collected in Table 2.
Next, the metrics considered in the simulations are introduced. Firstly, the Packet delivery rate, as expressed in (1), is the rate of successfully delivered packets sent from the source node to the destination node. The source node here is the client vehicle while the destination node means the gateway (either roadside AP or mobile Gateway on the road.) Secondly, the End-to-end delay, as expressed in (2), is the average end-to-end delay of all successfully delivered data packets sent from the source node to the destination node. End-to-end delay = ∑ All End-to-end delay The number of data packets successfully delivered .
Thirdly, the Signaling load, as expressed in (3), is the average number of control messages needed to deliver a data packet.
Fourthly, to quantify the cost-effectiveness of deploying APs, a new metric, namely, the rewarding index, is coined in increased vehicle velocity and/or increased traffic load. Generally, a route is prone to break when the vehicle's velocity increases. Consequently, the signaling overhead yielded by rerouting thence significantly increase. Also, the number of affordable connections in wireless communications is constrained for limited network capacity. Therefore, when the traffic load increases, the success rate of getting related networking resources is correspondingly reduced. As a result, the signaling overhead caused by re-routing remarkably increases as the traffic load increases. Additionally, Figures 6(a), 6(b), and 6(c) display that, given the traffic load (i.e., 20, 40, 60 sessions), the signaling overhead increases with increased vehicle velocity. Also, apparently, the signaling overhead of 4 APs is larger than that of the 16 APs. Naturally, 16 APs offer larger coverage than 4 APs. Therefore, with 16 APs, the nodes have higher probability of staying in the communication range of a certain AP and connect to the AP directly. Oppositely, with 4 APs, the nodes need to connect to the AP by multi-hops; that yields   Discussions. More APs are able to supply more networking capacity but more deployment cost. As suggested in the simulation results, 4 APs presents higher rewarding index than that of 16 APs. Thus, in the proposed scheme, 16 mobile gateways will be dynamically hired in case that 4 APs deployment cannot sufficiently meet the QoS requirements.

Selecting the Thresholds.
To properly set the thresholds used in the proposed Adaptive Clustering Scheme (in Algorithm 1), related simulations are carried out. The effects of a few factors including the traffic load, the velocity of vehicles, and the communication range of the nodes are studied. Four APs and one hundred vehicles (including 16 mobile Gateways) are set up. Four traffic loads, namely, 20%, 40%, 60%, and 80%, are considered in the following simulations.

The Effects of Traffic Load and Vehicle Velocity.
In this experiment, the transmission range is 250 m. For both Level-1 (in Figure 8(a)) and Level-2 (in Figure 8(b)) cluster architecture, the packet delivery rate decreases as either the traffic load increases or the vehicle's velocity increases. Because the success rate of obtaining the needed networking resource reduces when the competing traffic load increases; that results in decreased packet delivery rate. Also, the increased vehicle's velocity makes the topologies highly changing, which results in frequent broken connections and low packet delivery rate.
When the traffic load is not seriously heavy (i.e., depicted in Figures 9(a), 9(b), and 9(c) for 20%, 40%, and 60%, resp.), Level-2 cluster architecture generates higher packet delivery rate than that of Level-1 cluster architecture. In particular, as shown in Figure 9(d), when the velocity is faster than 40 km/hr and the traffic load becomes super high (i.e., 80%), the packet delivery rate of Level-2 drops dramatically. The possible reason of this result is that the broadly overlapped communication range of the nodes possibly cause lots of collisions and worsen the packet delivery rate. The effects of communication range of the nodes will be investigated later.

The Effects of Communication Range.
When the traffic load is not seriously heavy (shown in Figures 10(a), 10(b), and 10(c) for 20%, 40%, and 60%, resp.), longer communication range presents higher packet delivery rate. Because longer communication range needs less hops before the data packets arrives at the destination node. Generally, more hops imply higher probability of broken connections in MANETs/VANETs; that results in lower packet delivery rate. Particularly, when the traffic load becomes super high (i.e., 80% in Figure 10

4.3.3.
Setting the Thresholds. These investigations presented above are digested in the following tables. As suggested in Table 3, to achieve QoS 85 , is set as 250 m, when traffic load (denoted as ) is higher than 60% and vehicle's velocity (denoted as ) is faster than 40 km/hr, the cluster architecture should be upgraded to Level-2. Thus, and V are set as 60% and 40 km/hr, respectively. Also, in Table 4, to achieve QoS 90 , is set as 250 m, when is higher than 40% and is faster than 40 km/hr, the cluster architecture should be upgraded to Level-2. Thus, and V are set as 40% and 40 km/hr, respectively. Similarly, in Table 5, to achieve QoS 95 , is set as 250 m, when is higher than 20% and is faster than 30 km/hr, the cluster architecture should be upgraded to Level-2. Thus, and V are set as 20% and 30 km/hr respectively. The settings of these thresholds are summarized in Table 6.
Specially, when the traffic load gets super high (i.e., 80%), the cluster architecture will be upgraded to Level-2. Also, in Table 7, according to the vehicle's velocity, the communication range will be limited for reducing traffic collisions and achieving the needed QoS requirements. In this context, with communication range , two velocity thresholds V and ℎV are set as 30 km/hr and 60 km/hr, respectively.   Table 6, when the traffic load is less than 60% and the vehicle velocity is below 40 km/hr, the proposed scheme adopts Level-1 cluster architecture (i.e., 4 fixed APs). Or else, it uses Level-2 cluster architecture (i.e., additional 16 mobile gateways). As depicted in Figures 11(a), 11(b), and 11(c), the packet delivery rate of Level-2 is higher than that of Level-1 for various vehicles' velocity and different traffic load. Also, as depicted in Figures 12(a), 12(b), and 12(c), the end-to-end delay of Level-2 is shorter than that of Level-1 for various vehicles' velocity and different traffic load. Besides, as depicted in Figures 13(a), 13(b), and 13(c), the signaling overhead of Level-2 is less than that of Level-1 for various vehicles' velocity and different traffic load. The proposed scheme is able to dynamically upgrade/downgrade to Level-2/Level-1 to achieve the QoS requirement with less deployment cost.

QoS Requirement 90%
. For QoS 90 , as mentioned in Table 6, when the traffic load is less than 40% and the vehicle velocity is below 40 km/hr, the proposed scheme adopts Level-1 cluster architecture (i.e., 4 fixed APs). Or else, it uses Level-2 cluster architecture (i.e., additional 16 mobile gateways). As depicted in Figures 14(a), 14(b), and 14(c), the packet delivery rate of Level-2 is higher than that of Level-1 for various vehicles' velocity and different traffic load. Also, as depicted in Figures 15(a), 15(b), and 15(c), the endto-end delay of Level-2 is shorter than that of Level-1 for various vehicles' velocity and different traffic load. Besides, as depicted in Figures 16(a), 16(b), and 16(c), the signaling overhead of Level-2 is less than that of Level-1 for various vehicles' velocity and different traffic load. The proposed scheme is able to dynamically upgrade/downgrade to Level-2/Level-1 to achieve the QoS requirement with reduced deployment cost.

QoS Requirement 95%
. For QoS 95 , as mentioned in Table 6, when the traffic load is less than 20% and the vehicle velocity is below 30 km/hr, the proposed scheme adopts Level-1 cluster architecture (i.e., 4 fixed APs). Or else, it uses Level-2 cluster architecture (i.e., additional 16 mobile gateways). As depicted in Figures 17(a), 17(b), and 17(c), the packet delivery rate of Level-2 is higher than that of Level-1 for various vehicles' velocity and different traffic load. Also, as depicted in Figures 18(a), 18(b), and 18(c), the endto-end delay of Level-2 is shorter than that of Level-1 for various vehicles' velocity and different traffic load. Besides, as depicted in Figures 19(a), 19(b), and 19(c), the signaling overhead of Level-2 is less than that of Level-1 for various vehicles' velocity and different traffic load. The proposed scheme is able to dynamically upgrade/downgrade to Level-2/Level-1 to achieve the QoS requirement with decreased deployment cost.

Discussion.
The proposed scheme adopts Level-1 cluster architecture when both the traffic load and the vehicle velocity are below the predetermined thresholds. If not, it employs Level-2 cluster architecture. The proposed scheme is able to dynamically upgrade/downgrade to Level-2/Level-1 cluster architecture to achieve the QoS requirements with less deployment cost.

Comparing Different Methods.
The performance of two well-known routing protocols, including MGRP [17] and AODV+ [20] are compared with the proposed scheme. In particular, without losing the fairness, 16 fixed APs are deployed for AODV+ while 16 mobile gateways are deployed for MGRP. It means that the AODV+ needs more deployment cost. And, the mobility of gateways in MGRP results in unstable connections that yields low packet delivery rate, long end-to-end delay, and heavy signaling overhead. As summarized in Table 6, when both the traffic load and the vehicle's speed is below the predecided thresholds, the proposed scheme will adopt Level-1 cluster architecture. Otherwise, it will use Level-2 cluster architecture. The proposed scheme will dynamically upgrade/downgrade to Level-2/Level-1 to achieve the QoS requirements with minimum deployment cost.
Level-2. The performance of the proposed scheme, including packet delivery rate (in Figures 20(b), 20(c), and 20(d)), the end-to-end delay (in Figures 21(b), 21(c), and 21(d)), and the signaling overhead (in Figures 22(b), 22(c), and 22(d)), fall between those of the ADOV+ and MGRP. In particular, given super high traffic load (i.e., 80 sessions), the performance of the proposed scheme, including packet delivery rate (in Figure 20(d)), the end-to-end delay (in Figure 21(d)), and the signaling overhead (in Figure 22(d)), approximates that of the AODV+.
Summary. These results imply that the proposed scheme can achieve the QoS requirements with lower deployment cost.

Conclusions
In this paper, a cloud-assisted adaptive gateway dispatching scheme is proposed to solve the scalability problem of accessing the Internet in ITSs. With the proposed scheme, the gateways are dynamically used according to the applications' QoS requirements and current networking conditions. Four stationary gateways are initially deployed in the system in advance, while sixteen mobile gateways are dynamically employed to increase network capacity to improve QoS. As displayed in the simulation results, the adaptive clustering scheme proposed in this paper is capable of meeting the QoS requirements of the applications with fair gateway deployment cost.