A geographical segment architecture for vehicular ad hoc networks

However, the data explosion puts further strain on the broadcasting and routing protocols which are already challenged by the high mobility vehicular environment.. Leveraging on its two-

TOWARDS THE INTERNET OF VEHICLES: A

GEOGRAPHICAL SEGMENT ARCHITECTURE FOR VANET

2015

I

Declaration

I hereby declare that this thesis is my original work and it has been written by me in its entirety I have duly acknowledged all the sources of information which have

been used in the thesis

This thesis has also not been submitted for any degree in

any university previously

_

AIDI HUANG

19 Jan 2015

II

Acknowledgements

I would like to dedicate this section to all who helped me in this project

I must thank my supervisor, A/P Mehul Motani, for his guidance, invaluable time, and encouragement throughout the project Every meeting with Prof Mehul, his suggestion helps push my work forward

I am also grateful for the company of my fellow graduate students Cheng Huang and Anshoo Tandon Cheng Huang provided valuable comments on the project and Anshoo Tandan recommended helpful references when I started this project I always gain new findings discussing with them

III

Table of Contents

Acknowledgements II Table of Contents III Summary VII List of Tables VIII List of Figures IX

CHAPTER 1 Introduction 1

1.1 Objective 1

1.2 Challenges of Ad-hoc networking in the vehicular environment 1

1.2.1 The High Mobility Environment 2

1.2.2 The Channel Collision Problem in the High Density Network 2

1.2.3 The Network Fragmentation Problem in the Low Density Network 3

1.3 Literature Survey 3

1.3.1 Broadcasting Protocols 4

1.3.2 Routing Protocols 6

1.3.3 Vehicular Cloud (VC) 9

1.3.4 Clustering Technologies in VANET 10

1.3.5 VANET Standards 11

1.4 Thesis Contribution 12

1.5 Organization of this Thesis 14

IV

CHAPTER 2 The Geographical Segment Architecture 15

2.1 The Structure of the Geographical Segment Architecture 15

2.2 Representation of Segments and Vehicle Localization 17

2.3 The Head Generation Algorithm for the Segment 19

2.4 Discussion on Abnormal Situations 27

2.4.1 Failure of Receiving an HCM 27

2.4.2 Failure of Receiving an HRM 28

2.5 Information Carried by the Header 29

CHAPTER 3 Application of the Geo-Segment Architecture 30

3.1 Broadcasting Strategies 30

3.1.1 Directional Broadcasting 31

3.1.2 Intersection Broadcasting 31

3.1.3 Designated Area Broadcasting 33

3.1.4 The Hidden Node Problem of Broadcasting 33

3.2 Routing Protocols 35

3.2.1 The GSA based Source Routing (GSA-SR) 36

3.2.2 The GSA based Geographical Routing (GSA-GR) 41

3.2.3 The Infrastructure and GSA based Routing (IGSAR) 43

3.3 Optimization of Traffic Flows of the Vehicular Network 48

3.4 Vehicular Cloud 49

V

CHAPTER 4 Simulations 52

4.1 Simulation Setup 52

4.1.1 Simulation Tools 52

4.1.2 Simulation Parameters 53

4.2 Simulation of the City Scenario 55

4.2.1 Simulation Topology 55

4.2.2 Evaluation of Throughput 56

4.2.3 Evaluation of Packet Delivery Rate (PDR) 58

4.2.4 Evaluation of Packet End-to-End Delay 59

4.2.5 Evaluation of Protocol Overhead 60

4.3 Simulation of the Rural Scenario 63

4.3.1 Simulation Topology 63

4.3.2 Evaluation of Throughput 64

4.3.3 Evaluation of Packet Delivery Rate (PDR) 65

4.3.4 Evaluation of Packet End-to-End Delay 66

4.3.5 Evaluation of Protocol Overhead 67

4.4 Simulation of the Highway Scenario 68

4.4.1 Simulation Topology 68

4.4.2 Evaluation of Throughput 68

4.4.3 Evaluation of Packet Delivery Rate (PDR) 69

VI

4.4.4 Evaluation of Packet End-to-End Delay 70

4.4.5 Evaluation of Protocol Overhead 71

4.5 Discussion on the Application of GSA based Protocols 72

CHAPTER 5 Conclusions and Future Research 73

Bibliography 75

VII

Summary

Equipped with wireless transceivers, vehicles can be connected into a vehicular ad hoc network (VANET) to collect, share and utilize data for a variety of purposes, such as safety, navigation and entertainment As diverse sensors are integrated into vehicles and large numbers of access points (AP) are deployed within cities, VANETs are being transformed into the Internet of Vehicles (IoV), where far more data can be collected and utilized to support smarter applications However, the data explosion puts further strain on the broadcasting and routing protocols which are already challenged by the high mobility vehicular environment

In this thesis, a geographical segment architecture (GSA) which clusters vehicles based on their geographic locations is proposed Leveraging on its two-tier

architecture, the GSA can exploit cluster identifiers to develop efficient

broadcasting strategies and routing protocols Extensive simulations of vehicular networks in three road topology scenarios (city, rural, and highway) show that GSA based protocols can achieve high throughput, high packet delivery rate, low delay and low overhead

On the foundation of GSA based broadcasting and routing, we discuss that how the data collection and utilization can be further enhanced by (i) providing a solution to the hidden node problem in a broadcast scenario, (ii) facilitating the optimization of multiple network traffic flows and (iii) supporting the

implementation of the vehicular cloud

VIII

List of Tables

Table 4-1: Parameter of VANETMOBISIM 53 Table 4-2: Parameter of NS2 54

IX

List of Figures

Fig 2-1: Demonstration of the geographic segment architecture 16

Fig 2-2: The two-tier structure of the geographical segment architecture 17

Fig 2-3: Demonstration of Vehicle Localization for Simulation 18

Fig 2-4: Localization mistake in roads with curve shape 18

Fig 2-5: An example of localization mistake of inaccurate GPS 19

Fig 2-6: The flow of the head generation algorithm 20

Fig 2-7: A vehicle enters an empty segment 21

Fig 2-8: A vehicle enters a non-empty segment 22

Fig 2-9: A head vehicle leaves a segment 25

Fig 3-1: Demonstration of the directional broadcasting 31

Fig 3-2: Demonstration of the intersection broadcasting 32

Fig 3-3: Demonstration of addressing the hidden node problem based on GSA 35

Fig 3-4: The destination discovery process to learn the E2E routing path 36

Fig 3-5: A segment map at time point 1 and 2 38

Fig 3-6: Demonstration of the packet routing based on GSA 40

Fig 3-7: Demonstration of neighbor table updating based on GSA 42

Fig 3-8: Demonstration of a vehicle associating to an AP in multi-hop 45

Fig 3-9: An example of a segment map and a routing path based on it 46

Fig 3-10: An example of vehicle density based algorithm 47

Fig 3-11: Demonstration of managing vehicles as computation resource 50

Fig 4-1: The Road Topology for City Scenario 55

X

Fig 4-2: Throughput Comparison for the City Scenario 56

Fig 4-3: PDR Comparison for the City Scenario 58

Fig 4-4: Comparison of Packet End-to-End Delay for the City Scenario 59

Fig 4-5: Comparison of Protocol Overhead for the City Scenario 60

Fig 4-6: The road topology for city scenario 63

Fig 4-7: Throughput Comparison for the Rural Scenario 64

Fig 4-8: PDR Comparison for the Rural Scenario 65

Fig 4-9: Comparison of Packet End-to-End Delay for the Rural Scenario 66

Fig 4-10: Comparison of Protocol Overhead for the Rural Scenario 67

Fig 4-11: The Road Topology for the Highway Scenario 68

Fig 4-12: Throughput Comparison for the Highway Scenario 69

Fig 4-13: PDR Comparison for the Highway Scenario 70

Fig 4-14: Comparison of Packet End-to-End Delay for the Rural Scenario 70

Fig 4-15: Comparison of Protocol Overhead for the Highway Scenario 71

1.2 Challenges of Ad-hoc networking in the vehicular environment

The high mobility environment is a main obstacle for efficient data collection and utilization It makes design and implementation of broadcasting and routing protocols very challenging, since the network topology changes rapidly and neighbor relations among vehicles are extremely unstable The node density can

be remarkably high during the rush hour period, leading to an increased

possibility of channel collisions The node density may also be relatively low, resulting in a failure to support communications among vehicles

2

1.2.1 The High Mobility Environment

In the vehicular environment, vehicles are moving in a high speed, allowing very limited time for them to communicate with one another The communication session is highly unstable When the distance between two vehicles is

comparatively large that multi-hop relay is required, the validity of the routing path would be more vulnerable to the high mobility To obtain and maintain valid routing paths for high dynamic VANETs is a main task for the routing protocol

1.2.2 The Channel Collision Problem in the High Density Network

The density of vehicles would be particularly high during the rush hour period Many vehicles would be within the communication range of one another They may communicate simultaneously and result in frequent channel collisions Besides, due to the lack of effective solutions to address the hidden node problem for broadcasting, channel collisions can easily occur even in medium density conditions Flooding is regularly performed in vehicular networks for various purposes, e.g., disseminate safety messages, which would cause excessive

broadcastings

experience bad quality, e.g., large delay and low throughput

sensors are integrated into vehicles and large numbers of access points (APs) are deployed in cities, the idea of internet of vehicles (IoV) is prompted, of which the connotation is more generalized than that of the vehicular cloud Vehicles are not merely deemed as mobile hardware resources, but also information sources, e.g.,

1.3.1 Broadcasting Protocols

Extensive research has been carried out for the broadcasting protocol in the vehicular environment These protocols are classified into the single-hop scheme and the multi-hop scheme in [4] The difference between the two schemes lies in whether receivers would re-broadcast a packet immediately to flood it through the network

Single-hop protocols [5], [6] store received messages and re-broadcast them periodically In this fashion, a message could be disseminated to a wider area even in the low vehicle density condition However, a large delay would be introduced for spreading the message Besides, single-hop protocols might

become less popular when the infrastructure is readily accessible Vehicles can employ the infrastructure to disseminate information and save the channel

5

resource from periodical broadcasting In the GSA, we regard the infrastructure as

an active component and the single-hop protocol is beyond our consideration

Multi-hop protocols can flood packets through the network in short time A key issue for the multi-hop protocols is to choose proper relay nodes to flood packets There are basically three strategies, i.e., delay-based, probability-based, and network coding, to choose relay nodes For delay-based protocols [7]–[10],

vehicles will initiate a timer to contend for a re-broadcasting chance The node whose timer expires earliest would win the contention The expiration time is commonly related with the distance between the receiver and the sender For example, in [9], [10], the receiver in the farthest distance (within the

communication radius of the sender) will be assigned the shortest expiration time and become the re-broadcasting node For probability-based protocols [11], [12], the receiver calculates a probability to indicate whether to re-broadcast the packet,

so the number of vehicles who re-broadcast the packet is reduced For network coding protocols [13]–[15], received messages are combined to reduce duplicate transmissions Redundant broadcasting can be reduced and collisions can be alleviated subsequently

Multi-hop protocols achieve good results in handling channel collisions according

to their simulation results, but they are not flexible or smart enough to support the growing demand of intelligent applications For example, if we wish to control the

broadcasting protocols can be applied to further improve the broadcasting

efficiency Moreover, the cluster identifiers can be utilized to realize smart

broadcasting strategies which will be presented in chapter 3

1.3.2 Routing Protocols

Routing in networks has been an active research field and received a lot of interest Existing routing protocols can generally fall into three categories: proactive

routing, reactive routing, and position-based routing

In proactive routing protocols [16], [17], every node calculates the routing path based on a global map of the network connections The correctness and accuracy

of the map should be guaranteed which is very challenging Topology changes (e.g., link up and down) need to be quickly synchronized to the entire network It would generate a lot of overhead and easily exhaust channel resources Thus proactive routing is less likely to fit VANETs well

7

Reactive routing protocols [18], [19] discover the routing path on demand It helps avoid the global synchronization problem faced by proactive protocols However, they still face difficulties from the highly dynamic and high density environment The discovered routing path is vulnerable to the high mobility of vehicles that requires constant repair Both the discovery and the repair process of

a path would easily cause channel collisions if conducted by uncontrolled

flooding such as the ad hoc on-demand distance vector (AODV) routing does It may take multiple attempts to successfully discover or repair a routing path The protocol throughput will be low and the average packet delay will be large

Targeting at the high dynamic vehicular environment, position based or

geographical routing protocols [20]-[23] have attracted a lot of attention

Geographical routing protocols route packets without discovering an end-to-end (E2E) path beforehand Routing decisions are made at the intermediate nodes locally through applying a greedy algorithm For example, the distances of

neighbors to the destination node are calculated and the one with the smallest value would be chosen as the next hop However, the performance of

geographical routing is limited by the dead end problem, in which the algorithm may be trapped in a local optimum that no next hop can be found Various

strategies are proposed to resolve the dead end problem but each has its

disadvantages

8

In greedy perimeter stateless routing (GSPR) [22], the right-hand rule is used to recover the route from a dead end The work is improved by considering concrete city scenarios in [23] Vehicles at road intersections are introduced as

coordinators to handle the dead end problem, which also applies the right-hand rule However, the efficiency of applying right-hand rule commonly yields longer path which decreases the throughput dramatically In [21], a digital map is

employed to calculate the routing path which is in the form of road junctions Packet routing among junctions follows the greedy algorithm A problem for this protocol is that the connectivity among junctions is not considered during the path calculation The network might be fragmented among junctions and the validity of the path is not guaranteed

In addition to the dead end problem, the accuracy of the neighbor table is also a problem Vehicles use the neighbor table to choose the next hop Inaccurate table entries may misroute packets Update of the neighbor table in shorter period can help achieve a comparatively higher accuracy but would cause excessive

overhead, especially when the vehicle density is high

Based on GSA, efficient reactive routing and geographical routing protocols can

be implemented For reactive routing, a path is learnt or described in the form of GSA clusters, which will help improve its resistance against vehicles’ high

mobility The route discovery process will be carried through controlled flooding The discovery success ratio is much higher than AODV and the throughput is

9

much higher as well For geographical routing, GSA’s protocol packets are reused

to maintain the neighbor table The neighbor table can be highly accurate based

on our clustering mechanism Road intersections, in the form of intersection clusters, are employed to handle the dead end problem

on studying the applications (e.g., distributed data retrieve) of distributed and ever-changing feature of the VC The necessity of filtering and aggregating data was also discussed, prompted by the fact that data collected from sensors in adjacent vehicles is locally relevant

A key problem for VC, as in above scenarios, is how to track and manage

computation and storage resources, which highlights the demand for smart

broadcasting strategies and high throughput routing protocols In chapter 3.3, we discuss the organization of the VC supported by GSA based protocols

10

1.3.4 Clustering Technologies in VANET

Clustering is a powerful technology to address the network scalability issue Existing clustering methods basically adopt election-based (on various criteria, e.g., node ID, connectivity, speed, etc.) algorithms to form clusters, both for mobile ad hoc network (MANET) [25]–[31] and VANET [32]–[36] These

algorithms mainly focus on maintaining a stable cluster organization of which the effort is to prolong the lifetime of cluster heads To achieve this, they may need to constantly broadcast various types of protocol messages (e.g., hello message), producing a lot of overhead Even worse is that, when node density grows high, the overhead would increase significantly and cause channel collisions Moreover,

it would be challenging for these clustering schemes to support communications across clusters They may need to maintain several gateways, which is a costly operation as well

In the GSA, we use geographical road segments to cluster vehicles and apply a time-estimation based contention algorithm to generate the cluster head In

comparison to election-based method, our approach does not perform status monitoring, association, and election, which reduces overhead and increases scalability

11

1.3.5 VANET Standards

Dedicated Short Range Communications (DSRC) [37] are a group of protocols and standards for communications in the vehicular environment Previously, it is mainly used for electronic toll collection and works on 915 MHz Wireless

Access in Vehicular Environments (WAVE) [38] extends the 802.11 standard to the vehicular network as the lower layer standard (i.e., MAC and PHY) for DSRC Several physical features are standardized in WAVE based DSRC to

accommodate extreme conditions of the vehicular environment Narrower channel bandwidth (5Mbps and 10Mbps) is adopted to handle the adjacent channel

interference and Orthogonal Frequency Division Multiplexing (OFDM) is

employed upon such channels Besides, the maximum communication range could reach up to 1000m

The GSA is basically a clustered architecture, which is made necessary and

feasible by the large communication radius With a large radius, a small group of vehicles (i.e., cluster heads) can be chained up to provide the connectivity for a wide area The scalability issue such as the channel collision problem can be handled Note that GSA does not fit the case when the communication radius is small It would fail to obviously improve the network scalability since the cluster size which is limited by the communication range would be typically small

12

1.4 Thesis Contribution

This thesis contributes to the area of vehicular ad hoc network Concretely, we propose the geographical segment architecture (GSA) to support efficient data sharing, collection and utilization To do this, the GSA should be able to handle the high mobility and scalability issue of the vehicular network

Recapitulating, this thesis makes the following contributions:

 A novel clustering method for vehicular networks is presented to address the high mobility problem (chapter 2) A time-estimation based contention algorithm is proposed which could generate cluster head with low overhead The algorithm scales well when vehicle density grows, i.e., no obvious increase of overhead would be caused

 The GSA can facilitate the design and implementation of the communication system or protocols

• The GSA can provide a solution for the hidden node problem in a broadcast scenario by applying the request to send (RTS) / clear to send (CTS) [39] mechanism among cluster heads The performance of multi-hop broadcasting protocol can be improved

• The GSA can easily support implementing smart broadcasting protocols in a controlled manner, e.g., directional broadcasting, intersection broadcasting and restricted area broadcasting

13

• We propose and simulate a source routing protocol based on GSA The GSA can also help maintain a digital map with cluster information The map is called segment map, based on which the routing protocol can achieve good performance

• We propose and simulate a geographical routing protocol based on GSA We reuse the protocol packet for GSA clustering to maintain the neighbor table

We also utilize the segment map to tackle the dead-end problem

• We propose and simulate an infrastructure dependent source routing protocol based on GSA We update vehicles’ position information to a server

connected by the infrastructure and use the server to calculate end-to-end routing paths The protocol can achieve better performance once access points are widely deployed

• We discuss that the network traffic can approach global optimal with the segment map (chapter 3.3) Calculation of routing paths can be deemed as a multi-commodity flow [45] problem

 The GSA-based broadcasting and routing protocols can further help

implement the vehicular cloud (chapter 3.4)

14

1.5 Organization of this Thesis

The remainder of this thesis is structured as follows We will explain the

geographical segment architecture in chapter 2, including geographical segment representation, vehicle localization (into segment) and the head generation algorithm In chapter 3, we demonstrate that how GSA can be exploited to achieve efficient broadcasting and routing protocols We also discuss how GSA can facilitate handling of other significant issues: hidden node problem of broadcasting, network traffic optimization and vehicular cloud In chapter 4, we perform simulations to evaluate the overhead of GSA clustering and analyze the performance of GSA based protocols In chapter 5, we make conclusions and discuss future works

15

CHAPTER 2

The Geographical Segment

Architecture

2.1 The Structure of the Geographical Segment Architecture

In VANETs, the connectivity among vehicles is very unstable due to their high mobility To mitigate this, we propose a geographical segment architecture (GSA) which divides roads into segments to cluster vehicles Each segment represents a cluster, thus we use cluster and segment interchangeably throughout the thesis Provided at least one vehicle is driving in a segment (cluster), the segment can continuously behave as an intermediate cluster to sustain communications This means the connectivity among vehicle clusters should be more stable than that among individual vehicles This addresses the high mobility issues Note that the density issue can be handled naturally by a clustered structure

Fig 2-1 provides an illustration of the GSA and we can see that the road is

divided into segments To guarantee the communication among adjacent segments, the length of a segment needs to be limited to a certain range In our evaluations, the segment length is set from 50m to 200m for different scenarios, which is well within the vehicular communication range specified in industry standards such as WAVE (wireless access in vehicular environment) [38] Each cluster generates a

16

head to provide the connectivity among clusters A two-tier structure can be established on this basis

Fig 2-1: Demonstration of the geographic segment architecture

As shown in Fig 2-2, the first tier corresponds to the network of vehicles within the same segment and the second tier refers to the network formed by cluster heads Vehicles in the first tier rely on head vehicles forming the second tier to communicate with peers in two or more hops A vehicle can choose any cluster head in its communication radius as the first hop, for the purpose of reducing the hop number For instance, if a vehicle in segment 2 intends to communicate with another vehicle in segment 4, the first hop could directly go to the head of segment 3; its own segment head can be skipped

17

Fig 2-2: The two-tier structure of the geographical segment architecture

In Fig 2-1 and 2-2, all black vehicles are heads of corresponding segments and all white vehicles are member nodes of these segments We follow the same pattern throughout the thesis

2.2 Representation of Segments and Vehicle Localization

We assume every vehicle is equipped with a positioning device, such as the global positioning system (GPS) We also assume that a digital map (we call it segment map) containing the segment information is pre-stored in every vehicle A

segment was represented by its central point A vehicle can localize itself to a segment through comparing its distances to central points of adjacent segments

As in Fig 2-3, the vehicle with a star can locate itself being in segment 2, since d1

is smaller than d2 and d3 However, there are exceptions, such as for the

intersection segment, we need to specify its range in the segment map

18

Fig 2-3: Demonstration of Vehicle Localization for Simulation

We also consider the road direction and the vehicle’s moving trajectory for

localization It could correct localization mistakes when the shape of a segment is irregular An example is shown in Fig 2-4 for a curved road The vehicle may regard itself as being in segment 1 by simply comparing d1 and d2 (d1 is smaller) Such mistake will affect our head generation algorithm and needs to be corrected

Fig 2-4: Localization mistake in roads with curve shape

Fig 2-5: An example of localization mistake of inaccurate GPS

2.3 The Head Generation Algorithm for the Segment

A key issue for GSA clustering is how to generate and maintain valid segment heads with low overhead We propose a time-estimation based competition

strategy for this purpose The idea is to measure vehicles’ lifetime (driving time)

in a segment and set their timers based on the measured time to compete for the head position The lifetime is measured by assuming vehicles are travelling at the maximum speed allowed on the road To reduce the overhead, a new vehicle that

20

enters a segment should have a higher chance to win the competition than

previous vehicles, which means its timer should expire earlier The competition should be finished before the expiration of the current head’s lifetime, or an HRM (Head Renounce Message) will be broadcasted The specific flow of our strategy

is shown in Fig 2-6 and we explain it in detail via scenarios below

Fig 2-6: The flow of the head generation algorithm

• Scenario 1: A vehicle enters an empty segment

21

Fig 2-7: A vehicle enters an empty segment

In Fig 2-7, the vehicle with a star (called V1) drives to segment 2 (called S2)

It searches the local memory to check if ever received a head claim message (HCM) for S2 An HCM is broadcasted when a node claims itself as the head

of its segment Since S2 contains no vehicle now, there should be no valid HCM for this segment and thus no record can find by V1 V1 then broadcasts

an HCM to claim itself as the head of S2 This HCM can be received and cached by vehicles in segment 1 and 3 (called S1 and S3) When vehicles in S1 enter S2 a while later, they will know V1 is the head of S2

22

• Scenario 2: A vehicle enters a non-empty segment

Fig 2-8: A vehicle enters a non-empty segment

In Fig 2-8, the white vehicle with star (called V2) enters S2 V2 should have received an HCM broadcasted by the black vehicle (head vehicle, called V1)

in S2 Assume the HCM is found in the local memory, V2 then checks if it is valid by applying (2.1):

𝑉𝑉 = 𝑇𝑇ℎ𝑐𝑐+ 𝑇𝑇𝑒𝑒 − 𝑇𝑇𝑐𝑐 (2.1) where 𝑉𝑉 indicates the remaining lifetime of the head, 𝑇𝑇ℎ𝑐𝑐 refers to the time

instant when HCM is broadcasted or received (time difference caused by transmission can be ignored), and 𝑇𝑇𝑐𝑐 is the current time The quantity 𝑇𝑇𝑒𝑒

estimates the total lifetime of a head and is calculated as follows:

𝑇𝑇𝑒𝑒 = |𝐶𝐶ℎ𝑐𝑐 − 𝐶𝐶𝑒𝑒|

𝑉𝑉𝑚𝑚 (2.2) where 𝐶𝐶ℎ𝑐𝑐 refers to the coordinate of the node when it becomes the head, 𝐶𝐶𝑒𝑒 is the coordinate of the egress edge of the segment (𝐶𝐶𝑒𝑒 can be read from or

23

calculated based on the segment map), 𝑉𝑉𝑚𝑚 indicates the maximum allowed

speed on the road

If the HCM is valid, V2 marks V1 as the head and set a timer to compete for the head position The length of the timer is calculated following (2.3) For case that HCM is not received or the received HCM is found invalid, V2 would claim itself as the new head All vehicles in the vicinity can receive this HCM V1 and the other white vehicle (called V3) in S2 will mark V2 as the segment head and reset the timer to compete for the head position of the next round (2.4) is applied to calculate the length of the timer in this case

𝑇𝑇 = 𝑇𝑇𝑒𝑒 − (𝑇𝑇𝑐𝑐 − 𝑇𝑇ℎ𝑐𝑐) − 𝑊𝑊𝐶𝐶+ 𝑊𝑊𝐶𝐶 × (𝑇𝑇2 × 𝑇𝑇𝐶𝐶𝑒𝑒 + 𝑇𝑇ℎ𝑐𝑐 − 𝑇𝑇𝑒𝑒𝑒𝑒)

𝑠𝑠 (2.3)

𝑇𝑇 = 𝑇𝑇𝑒𝑒 − (𝑇𝑇𝑐𝑐 − 𝑇𝑇ℎ𝑐𝑐) − 𝑊𝑊𝐶𝐶 × 𝑇𝑇2 𝑒𝑒′ (2.4) where 𝑇𝑇 represents the length of the timer, 𝑇𝑇𝑒𝑒 , 𝑇𝑇ℎ𝑐𝑐 and 𝑇𝑇𝑐𝑐 have been defined before, 𝑇𝑇𝑒𝑒𝑒𝑒 is the time instant when a vehicle enters a segment, 𝑇𝑇𝐶𝐶𝑠𝑠 is the estimated time cost for a vehicle to pass a segment with maximum allowed speed, which is computed by

𝑇𝑇𝐶𝐶𝑠𝑠 = 𝑉𝑉𝐿𝐿𝑠𝑠

𝑚𝑚 (2.5) where Ls is the segment length, 𝑊𝑊𝐶𝐶 represents the predefined contention

window, 𝑇𝑇𝑒𝑒′ is calculated as:

𝑇𝑇𝑒𝑒′ = |𝐶𝐶𝑐𝑐 − 𝐶𝐶𝑒𝑒|

𝑉𝑉𝑚𝑚 (2.6) where 𝐶𝐶𝑐𝑐 is the coordinates of the node at the current time

24

A point to make is that why we introduce 𝑊𝑊𝐶𝐶 into formulas A typical value of

𝑇𝑇𝐶𝐶𝑠𝑠 (i.e., 200m segment, maximum allowed speed as 90km/h) would be 8s

Without 𝑊𝑊𝐶𝐶, the timer length of vehicles can vary from 0s to 8s When the vehicle density is high, some vehicles would set their timers with very short length The head switching will be frequently triggered, producing a lot of overhead However, once with 𝑊𝑊𝐶𝐶, we can restrict the timer length to a

reasonable range (e.g., 6s to 8s) and avoid the problem

Both (2.3) and (2.4) follow the idea that a vehicle which is newer entering a segment should gain a higher chance to be the new head In this way, the lifetime of a head node can be prolonged to reduce the protocol overhead The difference between (2.3) and (2.4) is that the former is for the vehicle which just enters a new segment to set the timer and the latter is for the vehicle which receives an HCM of its current segment to reset the timer

25

• Scenario 3: A head vehicle leaves a segment

Fig 2-9: A head vehicle leaves a segment

In Fig 2-9, the head vehicle (called V1) of S2 is leaving the segment In common situation, before V1 leaves, the timer of other vehicles should have expired and a new head would have been generated However, if V1 is driving too fast that exceeds the maximum allowed speed or the segment is an

intersecting segment where V1 drives within for a short distance, V1 may still

be the segment head while leaving this segment In this situation, V1

broadcasts a head renounce message (HRM) to indicate the absence of the head Upon receiving the HRM, vehicles in segments other than S2 would delete the record of the corresponding HCM; vehicles in S2 would set a timer

to compete for the head position by computing:

𝑇𝑇 = 𝐶𝐶𝑤𝑤′ × 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑚𝑚(0,1) (2.6)

26

where 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑚𝑚(𝑎𝑎, 𝑏𝑏)is a uniformly distributed random variable, and 𝐶𝐶𝑤𝑤′ is

the contention window which is much smaller than𝐶𝐶𝑤𝑤, as we wish to generate

a new head node within this short window

• Scenario 4: Set timer at intersection segments

For segments at road intersections, vehicles are driving in multiple directions and it is difficult to estimate their lifetimes Road intersections can also have irregular shapes and configurations As a result, the timer is set according to (2.7)

𝑇𝑇 = 𝐷𝐷𝑐𝑐 + 𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑈𝑚𝑚(0,1) (2.7) where 𝐷𝐷𝑐𝑐 is a constant delay which is set according to an estimated driving time in the corresponding intersection segment For example, in our

simulation for the city scenario, we set 𝐷𝐷𝑐𝑐 to 3s, since the length of an

intersection segment is 50m and the maximum allowed speed is 60 km/h

27

2.4 Discussion on Abnormal Situations

We use HCM and HRM to maintain the clustered structure, thus the clustering process may be affected when some nodes fail to receive the HCM or HRM We discuss these abnormal situations case by case

2.4.1 Failure of Receiving an HCM

When a vehicle fails to receive an HCM of its own segment, the timer of this vehicle would not be reset In this case, the timer might expire earlier than other nodes’ timers and this node would become the new head The drawback is that more overhead would be generated since the lifetime of current head node is shortened

When a vehicle fails to receive an HCM of an adjacent segment, it may directly claim itself as the new head when enters that segment This would also increase the overhead as the old head's lifetime is shortened In addition, the routing efficiency of the GSA based graphical routing protocol might be affected (see chapter 3.2.2)

Increasing of the overhead is affordable for GSA, if the failure rate of receiving

an HCM is not too high According to our simulation results (see chapter 4), the overhead of GSA clustering is pretty low which leaves enough margin for extra overhead Note that the simulation results have already included extra overhead generated due to abnormal situations

28

For the neighbor table problem, possible solutions are to use shorter segments or broadcast an HCM multiple times The former solution could provide more neighbors as potential next hop and the latter solution would be able to promote the success rate of receiving an HCM packet Both solutions are to trade the success rate with the overhead Actually, based on our simulation results, the GSA based geographical routing protocol performs very well

2.4.2 Failure of Receiving an HRM

When a node fails to receive an HRM of its own segment, the node would be unaware of the missing of the head node If the node is the only one left in that segment, the segment would have no head node till the timer of this node expires Packet forwarding might be affected during this period Actually, this situation would rarely happen Because the timer length of a node is calculated by

assuming vehicles are travelling with the maximum allowed speed on the road Before a head leaves its current segment, other nodes’ timers should have expired and a new head will have been generated

When a node fails to receive an HRM of an adjacent segment, it would fail to delete the corresponding head node record from the memory promptly It may affect the routing or broadcasting efficiency as well Due to similar reason as above stated, such situation occurs rarely Even it happens, the impact should be trivial The node would delete invalid records when performing periodical

checking

29

2.5 Information Carried by the Header

To implement the clustering process described above and to support broadcasting and routing strategies that will be introduced in next chapter, the header of the GSA protocol packet should contain the following information:

struct geocluster {

GEOCLUSTER_PACKET_TYPE packetType; // Sub-type, e.g., HCM, HRM or Data packet

u_int32_t SID; // Segment ID of the curresent segment

union info{

PT_INFO pt; // protocol information

RT_INFO rt; // route information

BC_INFO bc; // broadcasting information

“GSA_HRM_T”, the union “info” will be parsed as “pt” which contains the

information for clustering The data packet contains several sub-types

corresponding to different broadcasting or routing strategies applied For example,

if the value of “packetType” is set to “GSA_IGSAR_T”, the union “info” actually carries the information of “rt” which is an E2E routing path in the form of

segment ID.