2018 21st International Conference on Intelligent Transportation Systems (ITSC)

Maui, Hawaii, USA, November 4-7, 2018

Design and Analysis of a Communication Protocol for Dynamic

Vehicular Clouds in Smart Cities
Aida Ghazizadeh1 and Puya Ghazizadeh2

Abstract— The rise of Internet of Things (IoT) and adoption II. VEHICULAR CLOUD COMPUTING
of smart cities create opportunities for creative and efficient
Vehicular Cloud was motivated by the realization that
management and utilization of the available resources. One of
the characteristics of smart cities is the interconnectivity of present-day vehicles are endowed with powerful on-board
the city’s infrastructure, which allows data to be transferred computers, powerful transceivers and an impressive array of
between vehicles, infrastructures and road side units. Future sensing devices. The computing, storage and communication
vehicles with powerful on-board computers, communication resources available in vehicles are chronically under-utilized.
capabilities, and vast sensor arrays are perfect candidates in this
Putting these resources to work in a meaningful way will
hyper-connected environment to utilize into a fluid Vehicular
Cloud (VC) capable of performing large-scale computations. have significant impacts in many areas such as environment,
In this paper, we propose a vehicle-to-infrastructure (V2I) energy and safety. Vehicular Cloud can provide many types
communication protocol for enabling a dynamic Vehicular of cloud services [10] such as Network as a Service (NaaS),
Cloud Computing system in smart cities. Collaboration as a Service (CaaS) and Storage as a Service
(StaaS).
Individuals can benefit from VCs by renting out their
I. INTRODUCTION
excessed on-board capabilities such as computational or
storage units to customers that need it, in a nearly similar
Many big cities around the world are accepting and
fashion to CC paradigm. Vehicles parked in the parking lot of
adopting to the new concept of smart cities. Recently, the
a major airport or a large company with the help of proposed
US Department of Transportation (USDOT) has invested
communication model by [11], are perfect candidates for
$350 million to support projects related to the advanced
establishment of a VC. The long residency time of these
technology transportation and smart city technologies around
vehicles, along with the unused resources create opportuni-
the US. USDOT has launched the ”Smart City Challenge” in
ties for building datacenter in such parking lots. Although
December 2015 to encourage cities to propose ideas for new
vehicles that are parked in large parking lots seem to be
smart transportation systems that improve the currently avail-
excellent possible resources to be used for Vehicular Cloud,
able transportation solutions [1]. Intelligent Transportation
the idea of using moving vehicles on roads as resources for
System (ITS) is among the most important and key solutions
VCs have been overlooked. There are many beltways or ring
for smart cities. With the aid of ITS, connected vehicles [2]
roads around major cities that are built for the purpose of
that are equipped with sensors [3] can collect, create, analyze
traffic control and improving the speed and safety of the
and share data which can be used for finding solutions to
travel. Travelers use these beltways to travel faster from one
many problems such as improving public safety, emergency
point to another distant point in the city. These roads consist
management and better predictions of travel demands.
of several lanes, capable of serving millions of vehicles daily.
In 2010, inspired by the success and promise of ITS Vehicles that are traveling for long periods of time, create a
as well as Conventional Cloud Computing (CC) [4], a large pool of underutilized computing resources that can be
number of papers have introduced the concept of a Vehicular used to solve many of the problems that the large cities face.
Cloud (VC), a nontrivial extension of the conventional Cloud In the next section we describe the system requirements and
Computing paradigm [5][6]. This concept comes with several architecture of a dynamic VC.
challenges such as resource allocation and communication
model. Ghazizadeh et al. have proposed series of strategies III. SYSTEM REQUIREMENTS AND
and models to address some of these challenges [7][5][8][9]. ARCHITECTURE
In this paper we first briefly describe the concept of Vehicular A. Background and Motivations
Clouds, then we propose an architecture and communication In this section we propose and explain a possible archi-
protocol to be used in dynamic VCs, we then show our tecture and describe the system requirements needed for
predictions, numerical results as well as validations via enabling dynamic Vehicular Cloud in future smart cities.
simulations. We believe that in the near future, with the advances in
technology [12], it is possible to use the underutilized
1 Aida Ghazizadeh is with Department of Computer Science, Old Domin-
resources of vehicles on road for solving complicated traffic
ion University, Norfolk, VA 23529, USA [email protected] and non-traffic related problems.
2 Puya Ghazizadeh is with Department of Mathematics and Com-
puter Science, Seton Hall University, South Orange, NJ 07079, USA There are many huge beltways or ring roads around major
[email protected] cities that are built for the purpose of traffic control and

978-1-7281-0323-5/18/$31.00 ©2018 IEEE 1

improving the speed and safety of the travel. Examples
of such beltways are the Brussels Ring, the Amsterdam
Ring and the Boulevard Priphrique around the city of Paris.
Travelers use these beltways to travel faster from one point
to another distant point in the city. These roads consist
of several lanes, capable of serving millions of vehicles
daily. As an instance, 1.1 million cars use the Boulevard
Priphrique of Paris, each day. Large number of vehicles
that are traveling for long periods create a huge pool of
underutilized computers that can be used to solve the many
problems that the cities face.

To use the underutilized resources available on highways

we need a form of communication mean that can be used to
transfer data and applications from vehicles to roadside units Fig. 1: Data transfer between vehicles and integrated on-road
and vice versa. Wireless Access for Vehicular Environments APs around a major beltway.
(WAVE), also known as IEEE 802.11p, is an amendment
to the IEEE 802.11 standard, which was created to support
wireless communication in VANETs. B. Hypervisor and Container based virtualization
Virtualization is a foundational element in Cloud Com-
In 1999, the FCC has allocated 75 MHz of spectrum in the puting that aids in creation of virtual resources such as
5.9 GHz band, known as Dedicated Short-Range Communi- Operating Systems, storage or network resources to be shared
cations (DSRC), which is a wireless technology, to facilitate between multiple users. This is achievable by creating inter-
vehicle-to-infrastructure (V2I) and vehicle-to-vehicle (V2V) mediate software layers, called Virtual Machines (VMs) that
communications. This service provides high data transfer are emulations of isolated computer systems. There are many
rates with minimized latency, which is convenient for the virtualization techniques available. Hypervisor based virtual-
highly mobile nature of vehicles and transportation systems. ization, which is the most popular virtualization technique,
The motivation behind creating DSRC was to improve safety requires a hypervisor or Virtual Machine Monitor (VMM) to
and reduce the number of collisions by providing the vehicles create and run VMs. The computer that runs the hypervisor
with means to transfer traffic data easily and frequently. is called the host and the VMs are called guest machines.
There are six 10 MHz channels for DSRC that are avail- VMs are provided with their own OS and can run separately
able in the frequency band of 5.85-5.925 GHz, defined as from each other and therefore different types of OS can run
service channels (SCHs) and one 10 MHz channel defined on a host machine.
as the control channel (CCH). The CCH is reserved for Recently, Container based virtualization or system-level
high priority control and safety-related messages [13]. These virtualization was introduced for faster and more lightweight
technologies create opportunities for building more sophisti- virtualization. The hypervisor-based virtualization provides
cated vehicular applications. Palazzi et al have suggested that abstraction for guest OSs while in Container based virtual-
by using smart Access Points (APs) on roads, travelers can ization, guests share the same OS and kernel with the host
experience faster online gaming experience [14]. In a similar OS and abstraction is provided directly for guest processes
manner, we suggest using APs on the road-sides for utilizing (Fig 2) [16].
the computational power of vehicles. We assume that APs Container based virtualization provides several advantages
can be installed around major beltways in cities in a way for Vehicular Cloud Computing, such as faster startup time
that covers the drivers’ traveling time on that particular ring compared to Hypervisor based virtualization. It usually takes
road (Fig 1). seconds or minutes to deploy a VM and its OS, however
Containers are lighter and can be booted up in millisecond.
Using the network provided by several APs, jobs can be Containers also take lower amount of storage as the OS is
uploaded on moving vehicles and the results will be sent back shared, compared to the VMs that take much more storage
to the nearest AP. Olariu et al. [15] have suggested that each for running separate guest OSs and their required associated
vehicle is preloaded with a Virtual Machine Monitor (VMM) programs. [17]
and therefore jobs can be transferred on Virtual Machines
(VM) to vehicles that are participating in the VC. Users can IV. COMMUNICATION PROTOCOL
pick their desired operating system (OS) and receive a VM In this section we discuss the proposed communication
with an install instance of their desired OS. In this paper, we protocol and frame structure for V2I communications in
suggest using Docker containers, assuming that each vehicle dynamic VCs. As shown in Fig 3, each communication frame
is preloaded with an OS. Container based virtualization is consists of multiple fields with fixed durations. The frame
a recent technology that is a lightweight alternative to the begins with a synchronization byte (SYNC) which is used for
hypervisor virtualization. informing the vehicles that a new packet is arriving as well

2
 Fig. 4: Structure of T0.

that are available to vehicles and one of these policies is

selected based on the congestion level of the network and
this will lead to a higher probability of success. We will
use s(k) to represent the probability of success that a vehicle
obtains a communication slot, given that k other vehicles are
competing for a slot.
A : vehicle obtains a slot f or communication
Fig. 2: a) Hypervisor and b) Container based virtualization
model for VCs. B : k other vehicles are competing f or a slot
s(k) : P r(A|B) (1)
as synchronizing the receiver’s clock with the transmitter’s There are two main bidding policies that the vehicle can
clock. Synchronization sequences are also used in between use to compete for obtaining a communication slot which
fields due to the dynamic characteristics of the vehicular we will describe in detail.
environments. The start and the end of a frame are indicated 1) Bidding Policy 1: In a less congested area, a vehicle
by a start of frame delimiter (SFD) and an end of frame that wants to communicate selects a random slot number
delimiter (EFD), respectively. The remaining of the fields from 1 to M to transmit information in that slot in T1. If
are explained in detail in the next subsections. no other vehicle picks the same number then the vehicle can
transfer information in that slot, otherwise, if another vehicle
picks the same number then a collision occurs and neither of
the vehicles that picked the same slot number can transmit
in that slot. The vehicles get another chance to compete for
a slot in T2. In this case, the probability of success is the
Fig. 3: Proposed frame structure for V2I communication in probability that at least one of the two attempts to obtain a
dynamic VCs. slot in one of the M slots in either recognition periods T1
or T2 is successful, which can be computed as:

A. T0: Open Communication Period 1 k 1 2k

s(k) : 2(1 − ) − (1 − ) (2)
This field is used for broadcasting the access point (AP) M M
identification number and the layout of the frame to all 2) Bidding Policy 2: In a congested area, a vehicle that
vehicles that are in the coverage area of the AP. Fig 4 shows wants to communicate chooses if it wants to compete in T1
the structure of T0. Vehicles that request to participate in the or T2 and gets only one chance per frame to select a random
VC should be aware of the structure of the communication slot number from 1 to M and transmits in that slot in T1 or
frame, therefore it is necessary that the AP broadcasts the T2. Similar to the previous bidding policy, if another vehicle
basic information about the layout of the frame, such as the picks the same number then a collision occurs and neither of
current frame number, number of the fields in the frame, the vehicles that picked the same slot number can transmit
length of each field in bits, beginning of the next field, in that slot. In this case, the probability of success is the
number of slots in the establishment period, and a bidding probability that the attempt to obtain a slot in one of the M
sequence. The bidding sequence is an indicator of the bidding slots in recognition periods T1 or T2 is successful, which
policy in the establishment period which should be revealed can be computed as:
to the vehicles participating in the VC. The bidding policy
1 k
changes based on the congestion level of the network and s(k) : (1 − ) (3)
the number of vehicles in the coverage area which will be 2M
discussed in detail in the next subsection. The transmitted information in T1 and T2 differs based
on the status of the vehicle and can be classified as follows:
B. T1 and T2: Establishment Period a) Initial request: A vehicle that contacts the AP for the
After receiving the initial signal and the frame layout first time to receive a job does not have an ID and therefore in
from the AP, each vehicle that wants to communicate should T1 the vehicle transmits an estimated number of miles that it
compete for a communication slot based on the bidding will be on the highway and the slot number that it will select
sequence received in T0. There are two main bidding policies in T2. In T2, the vehicle sends the estimated number of miles

3
that it will be on the highway, and the slot number selected d) Job submission and migration acknowledgment: To
in T1. This will help the system to identify the vehicles that enhance the performance of the system and reduce number
successfully obtained a slot in both transmission periods. of failures, vehicles with higher priority should receive a
b) Job download request: Vehicles that are requesting slot in the next consecutive frames. Vehicles that complete
to download an assigned job should compete for a slot and a job and request job submission have the highest priority.
therefore send the vehicle ID and a sequence indicating a The next priority class belongs to the vehicles that request
request to download the job. job migration. In this acknowledgment period vehicles will
c) Input data request: Vehicles that successfully down- received the assigned slot number in the transmission period,
loaded the assigned job and need to download the inter- along with a reservation sequence that contains details of
mediate input data should compete for a slot and therefore the reservation. The last slot contains information regarding
send the vehicle ID and a sequence indicating a request to the length of each slot and the number of vehicles that are
download the input data. guaranteed a slot in the transmission period.
d) Job submission and migration request: Vehicles that
complete the assigned job and are ready to transmit the final
produced data or want to migrate an incomplete job should
send their vehicle ID, along with a sequence indicating the
request and also the size of the data that should be uploaded
in megabytes. Sending the size of the data helps the system Fig. 6: Structure of T3.
to accordingly reserve slots for the vehicles with higher
priorities in the next consecutive frames.
D. T4: Transmission Period
In this period the acknowledged vehicles can transmit or
receive information in the assigned slot. The number of slots
in this period N and the size of each slot is adjusted based on
the number of vehicles that were guaranteed a transmission
slot.
Fig. 5: Structure of T1 and T2.

C. T3: ID and Availability Acknowledgment Fig. 7: Structure of T4.

This period has 2M+1 slots and the purpose of it is to send

acknowledgments to the vehicles that successfully obtained E. T5: Transmission Acknowledgment
a slot in the establishment period. As a general rule, each In this period the vehicles that received the requested
vehicle that obtained the slot number k in the establishment information from the AP send an upward ACK and similarly,
period should listen to the slot number k in this period the APs can send downward ACKs to the vehicles, indicating
to receive the acknowledgment. Each acknowledgment is the receipt of the message. The number of slots in this period
based on the messages that vehicles transmitted in the N and the size of each slot is adjusted based on the number
establishment period and can be classified as follows: of vehicles that were guaranteed a transmission slot.
a) Initial acknowledgment: Vehicles that contacted the
AP for the first time to receive a job will receive the assigned
slot number in the transmission period, along with the size of
the assigned job. These vehicles then compute their unique Fig. 8: Structure of T5.
vehicle ID, composed of the AP identification number, the
frame number and the assigned slot number. Each vehicle
then uses its unique ID in the next communication attempts V. PROBLEM FORMULATION AND SIMULATION
to download the assigned job. RESULTS
b) Job download acknowledgment: Vehicles that con- In this section we examine the probability of successful
tacted the AP, requesting to download an assigned job, will slot allocations for each of the proposed bidding policies and
receive the assigned slot number in the transmission period, present our simulation results. Table I shows the parameters
along with the size of the remaining files that should be and notations that are used in our statistical modeling and
downloaded. simulations.
c) Input data acknowledgment: Vehicles that contacted Vehicles arrive to the coverage area according to a Poisson
the AP requesting to download the input data, will receive the process with a mean arrival rate λ. As explained in the
assigned slot number in the transmission period, along with previous section, s(k) is defined as the conditional probability
the size of the remaining files that should be downloaded. that at least one of the two attempts to acquire a slot in either

4
 TABLE I: Simualtion parameters and notation
∞ m m
X (λt) X 1 k
Symbol Description Value R= e−λt (1 − )
M Number of slots in T1 and 20
m=0
m! 2M
T2 k=0
 
r Access point range 40 m m k m−k
v Average traveling speed 80 (km/h) α(n) [1 − α(n)]
k
λ Arrival rate of vehicles [0-30] (vehicle/s)
t Time a vehicle in under (sec) We then obtain:
the coverage area of one
AP ∞
X [λt(1 − α(n)
2M )]
m
R = e−λt
m=0
m!
Which simplifies to:
establishment period T1 or T2 is successful given there are
k other vehicles that are competing for recognition. We then λtα(n)
R = e− 2M (8)
define the following probabilities:
Fig 9 shows the success probability of the two bidding
C : k other vehicles attempt recognition when there policies as functions of λ and M. The success probability of
are m vehicles present in the coverage area of the AP the first bidding policy (when vehicles compete in T1 and
T2) is higher than the second bidding policy (when vehicles
q(m, k) : P r(C) (4) compete in either T1 or T2) for the values of λ less than
or equal to 10 and is lower for values greater than 10. This
Within the coverage area of each AP, each of the ve-
shows that the second bidding policy is more suitable than
hicles attempt to start communication with the probability
the first bidding policy for high-traffic areas or rush hours
α(n), n = 1, ..., N . Since each of the vehicles attempt
and highest-traffic periods.
communication with probability α(n), q(m, k) follows the
We then examine the robustness of our predictions by
binomial distribution:
simulating a highway with properties similar to that shown
 
m in Table I. Vehicles arrive to the coverage area of an AP
k m−k
q(m, k) : α(n) [1 − α(n)] (5) according to a Poisson process with a mean arrival rate
k
λ. Each vehicle that receives the AP identification number
The success probability R for the first bidding policy, when and the layout of the frame in T0, can then compete in
vehicles compete in T1 and T2 can be expressed as: the Establishment period. For the first bidding policy each
vehicle chooses a random slot number from 1 to M in T1
∞ m m
X (λt) X and again in T2 and for the second bidding policy each
R= e−λt s(k)q(m, k) (6)
m! vehicle randomly chooses T1 or T2 and selects a slot in T1
m=0 k=0
or T2. We average the number of successful slot allocations
Then, using the results in equations (2, 5, 6), we obtain: in each scenario and compare it with our prediction results.
The success probability obtained from our simulations and
∞
X m m
(λt) X 1 k 1 2k the predicted values are shown in Fig 10 and Fig 11 .
R= e−λt 2(1 − ) − (1 − )
m! M M VI. CONCLUSIONS
m=0 k=0
The decreased cost of sensing devices and storing infor-
 
m k m−k
α(n) [1 − α(n)] mation along with the recent advances in technology, notably
k
Cloud Computing, Big Data and IoT have motivated the
After algebraic manipulations, we then obtain cities and the Intelligent Transportation Systems to move
towards the adoption of smart cities. Vehicular Cloud Com-
∞
[λt(1 − α(n) m puting, a newborn concept, was inspired by the success of
M )]
X
−λt
R = 2e − the conventional CC. Although the modern vehicles have
m!
m=0 an amazing array of sensing devices and great computing
∞ α(n) 1 m power, the rise of autonomous and self-driving cars have
X [λt(1 − M (2 − M ))]
e−λt encouraged companies to build even smarter vehicles with
m=0
m!
better computing power to assist in making smarter and faster
Which could be simplified to: decisions. For instance, recently Toyota has announced that
it will use supercomputers for its autonomous vehicles. In
α(n) α(n) 1
R = 2e−λt M − e−λt M (2− M )
(7) the near future, there will be millions of supercomputers on
wheels moving on roads or parked in the parking lots. These
Similarly, for the second bidding policy, when vehicles resources are often underutilized and can be used in VCs
compete in either T1 or T2, using the results in equations (3, to solve many problems that travelers and pedestrians face
5, 6), the success probability R can be obtained as follows: every day.

5
 Fig. 11: Success probability obtained from simulation vs the
predicted values for the case when vehicles compete in T1
or T2.
Fig. 9: Success probability as a function of the traffic flow
λ.
[4] R. Fathi, M. A. Salehi, and E. L. Leiss, “User-friendly and secure
architecture (ufsa) for authentication of cloud services,” in Cloud
Computing (CLOUD), 2015 IEEE 8th International Conference on.
IEEE, 2015, pp. 516–523.
[5] S. Olariu, M. Eltoweissy, and M. Younis, “Towards autonomous
vehicular clouds,” ICST Transactions on Mobile Communications and
Computing, vol. 11, no. 7-9, pp. 1–11, July-September 2011.
[6] S. Olariu, I. Khalil, and M. Abuelela, “Taking VANET to the clouds,”
International Journal of Pervasive Computing and Communications,
vol. 7, no. 1, pp. 7–21, February 2011.
[7] P. Ghazizadeh, S. Olariu, A. G. Zadeh, and S. El-Tawab, “Towards
fault-tolerant job assignment in vehicular cloud,” in Services Comput-
ing (SCC), 2015 IEEE International Conference on, June 2015, pp.
17–24.
[8] P. Ghazizadeh, R. Mukkamala, and R. Fathi, “Modeling and predicting
fault tolerance in vehicular cloud computing,” in Computing and
Network Communications (CoCoNet), 2015 International Conference
on. IEEE, 2015, pp. 395–400.
[9] P. Ghazizadeh, “Resource allocation in vehicular cloud computing,”
PhD Thesis, Old Dominion University, July 2014.
Fig. 10: Success probability obtained from simulation vs the [10] R. Meneguette, R. E. De Grande, and A. A. F. Loureiro, “Vehicular
predicted values for the case when vehicles compete in T1 cloud: Aspects and challenges of vehicular networks and cloud,” pp.
and T2. 113–145, 08 2018.
[11] P. Ghazizadeh and R. Fathi, “A propagation based model for commu-
nication in vehicular cloud,” in 2018 IEEE International Conference
on Pervasive Computing and Communications Workshops (PerCom
In this work, we proposed a new system architecture and Workshops), March 2018.
[12] B. Islam, M. Uddin, S. Mukherjee, and S. Nirjon, “Rethinking
communication protocol for V2I communications in dynamic ranging of unmodified ble peripherals in smart city infrastructure,” in
VCs in smart cities. We then investigated the probability Proceedings of the 9th ACM Multimedia Systems Conference, MMSys
of successful slot allocations for each proposed bidding 2018. Association for Computing Machinery, Inc, 6 2018, pp. 339–
350.
policy and then evaluated the robustness of our predictions [13] C. Campolo, A. Molinaro, and A. Vinel, “Understanding the per-
with simulations. This paper is a groundwork for our future formance of short-lived control broadcast packets in 802.11p/wave
works and in the future, we will investigate job assignment, vehicular networks,” in 2011 IEEE Vehicular Networking Conference
(VNC), Nov 2011, pp. 102–108.
migration and replication strategies, as well as network [14] C. E. Palazzi, S. Ferretti, and M. Roccetti, “Smart access points on
capabilities, and a detailed study of the types of tasks that the road for online gaming in vehicular networks,” Entertainment
can be supported by dynamic VCs. Computing, vol. 1, pp. 17–26, 2009.
[15] S. Olariu, T. Hristov, and G. Yan, The Next Paradigm Shift: From
Vehicular Networks to Vehicular Clouds. Wiley-Blackwell, 2013,
R EFERENCES ch. 19, pp. 645–700. [Online]. Available: https://onlinelibrary.wiley.
com/doi/abs/10.1002/9781118511305.ch19
[1] http://www.transportation.gov/, January 2017. [16] A. Celesti, D. Mulfari, M. Fazio, M. Villari, and A. Puliafito, “Explor-
[2] Y. Li and H. Zhao, “A car-following model for connected vehicles ing container virtualization in iot clouds,” in 2016 IEEE International
under the bidirectional-leader following topology,” in 2017 IEEE Conference on Smart Computing (SMARTCOMP), May 2016, pp. 1–6.
20th International Conference on Intelligent Transportation Systems [17] R. Dua, A. R. Raja, and D. Kakadia, “Virtualization vs containerization
(ITSC), Oct 2017, pp. 1–5. to support paas,” in 2014 IEEE International Conference on Cloud
[3] M. Tellez, S. El-Tawab, and H. M. Heydari, “Improving the secu- Engineering, March 2014, pp. 610–614.
rity of wireless sensor networks in an iot environmental monitoring
system,” in 2016 IEEE Systems and Information Engineering Design
Symposium (SIEDS), April 2016, pp. 72–77.