Received 26 December 2022, accepted 4 January 2023, date of publication 9 January 2023, date of current version 12 January 2023.

Digital Object Identifier 10.1109/ACCESS.2023.3234998

Vehicular Cloud Forming and Task Scheduling for

Energy-Efficient Cooperative Computing
MINYEONG GONG 1 , YOUNGHWAN YOO 2, (Member, IEEE),
AND SANGHYUN AHN 3 , (Member, IEEE)
1 Department of Electrical and Computer Engineering, University of Seoul, Seoul 02504, South Korea
2 School of Computer Engineering and Science, Pusan National University, Busan 46241, South Korea
3 Department of Computer Science and Engineering, University of Seoul, Seoul 02504, South Korea
Corresponding author: Sanghyun Ahn ([email protected])
The work of Minyeong Gong and Sanghyun Ahn was supported by the National Research Foundation of Korea (NRF) Grant funded by the
Korea Government [Ministry of Science and ICT (MSIT)] under Grant 2021R1A2C1011184. The work of Younghwan Yoo was supported
by the NRF Grant funded by the Korea Government (MSIT) under Grant 2021R1A2C1095635.

ABSTRACT A vehicular cloud (VC) is a network of vehicles that perform cooperative computing through
vehicle-to-vehicle (V2V) communication. Existing research on vehicular cloud computing (VCC) is mostly
based on cloud servers or edge servers, not VCs. However, vehicles, by constructing a Vehicular Ad-Hoc
Network (VANET), can perform applications requiring the large amount of computation cooperatively on
their own without the help of edges or cloud servers. One of important issues for the VANET cooperative
computing is how to handle the frequent topology change due to vehicle mobility. The unstable network
topology limits the advantage of cooperative computing and even makes its operation stop sometimes.
This paper proposes a cooperative computing method based on vehicle-to-vehicle (V2V) communication.
For stable and energy-efficient cooperative computing, the proposed method considers the distance when
selecting vehicles that it will cooperate with and delays task offloading back as far as possible. The proposed
method outperforms previous static scheduling methods in terms of energy efficiency and network stability.

INDEX TERMS Computation offloading, cooperative computing, VANET, vehicle-to-vehicle communica-

tion, vehicular cloud.

I. INTRODUCTION are located far away. Thus, the Cloud-tier is good for applica-
Recently, vehicles have not only been connected to Internet tions that require large amounts of memory and computation
via wireless communication, but have also become network but do not require short latency, such as infotainment services.
nodes that can perform various applications in real-time Information services are information and entertainment ser-
[1]. Particularly, applications such as big data analysis and vices that are not related to driving safety, requiring a latency
image processing using machine learning require high com- of 1 second and a bandwidth of 80 Mbps or more [5].
puting power. Vehicular Cloud Computing (VCC) might On the contrary, the Vehicle-tier consists of vehicles.
be a promising solution to run these kinds of application Because a vehicle can receive data from other vehicles
smoothly [2], [3], [4]. directly, this tier requires the shortest latency. Therefore, this
The VCC has a 3-tier architecture of Cloud-tier, Edge-tier, tier is appropriate for applications requiring very short latency
and Vehicle-tier. Each tier has different features, so each is like collision warning of which the acceptable end-to-end
suitable for different VCC applications. First, the Cloud-tier delay is just 20-50 ms [5]. However, this tier is not good
has the highest computing power and largest memory, but it for performing computation-intensive applications due to the
may take the longest latency to reach because cloud servers lack of computing power.
Finally, the Edge-tier can provide shorter latency than the
The associate editor coordinating the review of this manuscript and Cloud-tier; and more computing power and memory than the
approving it for publication was Barbara Masini . Vehicle-tier. Thus, it is suitable for applications such as traffic

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
3858 VOLUME 11, 2023
M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

information sharing and analysis services, which need a delay focused only on minimization of this execution time. On the
of 100-500 ms and a throughput of 10-45 Mbps [5]. other hand, we consider not only execution time but also
Due to the moderate latency and computing power, some energy costs when designing our task scheduling method.
papers [6], [7] suggest that most of the computation work on The energy cost is particularly important for electric vehicles
vehicles is offloaded to edge servers. However, edge servers since it can affect their driving range.
always have certain constraints on both computing power and This paper proposes Stepwise Computation Offloading for
memory unlike cloud servers, thus if too many tasks are given Cost-efficient Cooperation (SCOCC) to form a vehicular
to edge servers, the service quality might be degraded and cloud and find an energy-efficient task schedule. Main idea
even the security might be endangered [6], [8]. of SCOCC is to minimize the time interval between task
In this situation, vehicles can relieve the burden of edge offloading and task execution, and to consider the distance
servers. Vehicles in the previous VCC model [9] were con- between a CV and WVs when choosing WVs. An appli-
sidered as just end-terminals sending requests to cloud or cation is decomposed into tasks, which are logical mini-
edge servers, but this is not all that a vehicle can do. If they mum units of work, being represented in a Directed Acyclic
construct a Vehicular Ad-hoc Network (VANET) and cooper- Graph (DAG) form by a CV. The tasks are then offloaded
atively run many applications by themselves through vehicle- only to a fixed number of WVs in the order closest to
to-vehicle (V2V) communication, both the burden on edge the CV. Note here that not all tasks are offloaded at once,
servers and the network traffic load can be reduced. In addi- but rather some parts of them are offloaded sequentially to
tion, it is better for privacy since private information does not minimize the time difference between scheduling and task
pass through external networks [10]. execution.
We suggest a platform that enables vehicles to efficiently It is known as an NP-complete problem to find an optimal
and reliably perform cooperative computing in VANETs built schedule for the tasks represented in a DAG [14], [15]. Thus,
through V2V communication. In the cooperative computing, SCOCC is a heuristic method for performing reliable and
tasks of a vehicle are offloaded to other vehicles. The vehicle energy-efficient task offloading. To evaluate our heuristic
that requests task offloading is called client vehicle (CV), and in practical networks, we adopted the Veins simulator [16].
the vehicle that helps a CV is called worker vehicle (WV). The simulation results showed that SCOCC in a straight-road
The task offloading takes two steps: A CV finds some can- environment can reduce the probability of WVs going out of
didate WVs around itself and then distributes tasks to them CV’s range to just 1.7% from 15.8% of conventional static
according to the task execution schedule. The task offloading algorithms. Furthermore, the energy consumption for data
improves application running speed, but on the other hand, transfer is reduced by 69.5%.
it requires additional cost to transfer data. Minimizing both The contributions of this paper are as follows:
time and energy for data transfer is a key performance metric • SCOCC builds vehicular clouds that maintain stable
for a vehicular cloud. V2V connectivity despite vehicle mobility.
The latest vehicle network modules support both direct • Additional energy consumption for cooperative comput-
V2V communication using Dedicated Short Range Com- ing is minimized by reducing the average data transmis-
munication (DSRC) [11] or C-V2X mode4 [12]; and cellu- sion distance.
lar connectivity using LTE C-V2X [13]. The cellular-based • Performances are evaluated with DAGs based on real
connection can cover a wider range, but is expensive, so it applications in practical networks.
is more economical to transmit over V2V connections as
much as possible. However, V2V connections may not be Before moving on to the next section, we would like to
reliable because vehicles move fast. The distance between discuss why WV serves CV in V2V computation offloading.
two vehicles in a VANET continuously changes due to their First, we could expect voluntary involvement from all vehi-
different speed. Whereas the cellular communication can cles to achieve common goals such as smooth autonomous
reach several kilometers, the transmission range of the DSRC driving service. In fact, the Internet service also depends on
V2V communication is just from several hundred meters to the voluntary participation of all network equipment. Second,
1 km, so vehicles can easily be out of the range from each a platform can be devised in which WVs receive a reward
other. In this case, an attempt can be made to resume the for their services. Actually, we are developing a blockchain-
communication over the cellular network, but it increases cost based, tamper-proof credit system. The nonfungible token
and delay. Therefore, in order to improve VANET stability (NFT) might also be used as the credit. However, this paper
and minimize transmission cost, a CV needs to select WVs just assumes that all vehicles are willing to cooperate with
that are least likely to go out of communication range of others, since blockchain and NFT-based systems are beyond
the CV. the scope of this paper.
Another important factor in improving the efficiency of The rest of the paper is organized as follows. Section II
task offloading is task scheduling to find the best order to run introduces related work. In Section III, we define the sys-
tasks. The task scheduling determines the amount of data to tem model and analyze it with mathematical equations.
be transferred, the associated transfer cost, and eventually the Section IV describes our SCOCC heuristic algorithm, and
final completion time for all tasks. Most previous studies have Section V presents the simulation results, and Section VI
VOLUME 11, 2023 3859
 M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

discusses our analysis on the results and limitations of the

study. Finally, Section VII concludes this paper.

II. RELATED WORK

Task offloading using cloud computing was initially proposed
in a mobile environment [17] and then extended to VCC.
A typical example of using VCC is the intelligent transporta-
tion system (ITS) [2], [18], [19], which provides services to
reduce traffic accidents and facilitate traffic flow.
Mao et al. [7] presented computation offloading based on
the Lyapunov function to minimize cost on edge servers using
dynamic voltage and frequency scaling (DVFS) and power
control. Assuming that each task was independent of each
other, they decided whether to offload each task to edge server
or not.
Sun et al. [20] proposed a scheduling scheme that runs on
vehicular cloud or edge cloud. Their scheme firstly computes
the cell dwell time of each vehicle based on the distance
to a base station. This cell dwell time is used to form a FIGURE 1. Vehicular cloud overview.
vehicular cloud, and then tasks are scheduled to minimize the
completion time. It was a genetic algorithm based scheduling
method with low complexity considering task dependency for tasks in advance, it has long been considered a leader in DAG
the optimization. scheduling due to outstanding performance improvement.
Unlike the above methods, Dai et al. [6] studied the case The duplication based algorithms [28], [29], [30], [31]
where multiple edge servers exist. For load balancing, tasks create copies of tasks on participating devices to minimize
in a vehicle were split and offloaded to multiple edge servers. data transfer overhead. Significant performance gains can be
Xu et al. [21] performed a multi-objective optimiza- expected if the resources of the devices are abundant, but we
tion, which reduces both execution time and energy cannot say that it is efficient given the amount of resources it
consumption and prevents privacy conflict, by adopting uses.
non-dominated sorting genetic algorithm II (NSGA-II). Hu et al. [31] proposed a platform in which dynamically
However, they did not consider the dependency between formed vehicle clusters act as edge servers that perform
tasks when finding routes from vehicles to suitable edge computations for surrounding mobile devices. It is notewor-
servers. thy that vehicles on this platform do not request offloading,
While most studies enable a client to select workers, but act as edge servers. An application to be offloaded is
Zhou et al. [22] suggested a method in which candidate work- scheduled with the Greedy-based Task Scheduling Algorithm
ers choose an actual worker by voting. This voting system (GBTSA), which uses a greedy-based task copy technique,
works since it is assumed that all workers in this paper are considering the inter-task dependency. This scheduling tech-
wired edge servers and all other servers’ computing resource nique is to apply the task scheduling proposed by [32] to the
information is known to all. However, this system is not vehicle cloud.
suitable for wireless vehicular cloud networks because too Existing studies mostly assumed offloading to edge servers
many packets must be exchanged between vehicles to share and proposed scheduling techniques to minimize execution
resource information in real time. time. This paper attempts to minimize energy consumption
One of the most important issues in cooperative computing and the probability of WVs out of CV’s coverage by consid-
in VCCs is how to schedule the execution of tasks. List ering the distance between vehicles when selecting WVs and
scheduling [23], [24], [25], [26], a typical DAG scheduling assigning tasks.
method, consists of two steps. First, priorities are calculated
and assigned to tasks in order. For the priority calculation, III. SYSTEM MODEL
the cumulative sum of node cost and edge cost from the All vehicles mentioned in this paper are electric vehicles
starting point to each task is used in general [25], [27]. And equipped with GPS and transmission devices capable of both
in the second step, a node or device that will run each task is direct V2V communication and cellular communication, and
determined using its own algorithm. they can perform tasks of other vehicles, which are requested
Heterogeneous Earliest Finish Times (HEFT) [26], a rep- by virtual machine (VM) technology.
resentative list scheduling method, assigns tasks to devices Fig. 1 shows the overall system flow. For computa-
that can complete them in the shortest amount of time, using tion offloading, a CV first requests information from other
the insertion-based scheduling. Although HEFT is a static vehicles within the transmission radius via direct V2V com-
algorithm that requires information about all devices and munication, and then selects WVs based on the returned

3860 VOLUME 11, 2023

M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

information. Detailed criteria to select WVs among the B. PROBLEM DEFINITION

searched vehicles will be discussed later. An application appc A DAG G = (T , E) is adopted to model tasks of an appli-
of CV vc is decomposed into tasks, which are the smallest cation appc . T is a set of nodes meaning tasks ti ’s of appc
logical units that can no longer be decomposed, then those and E is a set of directed edges representing the dependencies
tasks are presented as nodes in a DAG. The problem of decid- between tasks. A task without any incoming edge is called a
ing which vehicle should perform each task is an important source task. This paper deals with DAGs that have only one
factor affecting the overall execution time, which is known source task. But, this does not limit the practicality of the
as NP-complete [14], [15]. The transmission of data and the proposed method because DAGs with multiple source tasks
result for task offloading is attempted by default through can be easily converted to DAGs with one source task by
direct V2V communication. However, sometimes a CV and adding one void task that has outgoing edges to all original
a WV may be farther apart than the V2V communication source tasks. If there exists an edge from ti to tl , ti is called an
range when offloaded works are completed, and in this case immediate predecessor (i-predecessor) of tl . Conversely, tl is
the result is transmitted through cellular networks. In this an immediate successor (i-successor) of ti . All predecessors
paper, appc is an application having a constraint that must must be completed before their successors.
be completed within the deadline. And CV vc requests the After CV vc makes a set V w of WVs with some 1-hop
appropriate WV vj to execute task ti of appc so that it can neighbor vehicles that respond to its broadcast message, each
be completed within the deadline, considering the computing task in T is assigned to a WV in V w . The binary variable xij
power and data transfer time of the neighbor vehicles in indicates that task ti is assigned to WV vj .
one hop. The task scheduling (or task mapping) that deter- (
mines which WV to be assigned ti is an important factor 1 if ti is allocated to vj by vc
xij = (1)
to determine the application completion time and energy 0 otherwise
consumption.
X is a set of xij , which represents a task map assigning tasks
A. SYSTEM DESCRIPTION
in T to WVs in V w .
X
A CV vc tries to offload tasks of an application appc to other xij = 1 (2)
vehicles. vc first searches for nearby vehicles, chooses WVs, j
and then requests them to process tasks of appc expressed
vj receives the assigned task via V2V communication and
in a DAG. To generate every 1 byte of output data for
checks the DAG to determine when to start executing the
task ti , as many instructions as ipbi must be executed on
task. vj can execute ti after receiving the results of the ti ’s
average. So, if the output data of task ti is Do (ti ) bytes, a total
i-predecessors. When the execution of ti is completed, the
of Do (ti ) × ipbi instructions are needed. Each vehicle has
resulting data of size Do (ti ) should be transmitted to vehicles
different computing power, which is expressed as the number
performing ti ’s i-successors. The time spent executing ti in vj ,
of instructions per second, ips, and data transmission between
denoted by wij , is the number of instructions to be processed
vehicles is simply modeled as the amount of data transmitted
for ti , Do (ti ) × ipbi , divided by the number of instructions that
per second, bytes/sec.
can be processed by vj every second, ipsj .
In order to collect information from surrounding vehi-
cles, vc first broadcasts a message requesting information Do (ti ) × ipbi
wij = (3)
on them. Among the vehicles receiving this message, the ipsj
vehicle willing to provide its computing resources returns
information about its current location, current movement sta- And the expected finish time of ti can be calculated as the
tus, and computing resources that can be provided to vc . These sum of the starting time and processing time in vj .
resources are the remainder besides those reserved for their FTi = STi + wij (4)
own work, as all vehicle prioritize their own tasks. Then vc
has to decide which tasks to be assigned to which vehicles In order for vj to start ti , all i-predecessors of ti are com-
based on the received information. Since the completion time pleted and their results must be transmitted to vj . Also, among
of the application and the amount of energy consumed vary the tasks assigned to vj , all tasks with higher priority than ti
depending on the scheduling that allocates tasks, it is neces- should be completed before. The latest finish time of all ti ’s
sary to find a scheduling that minimizes energy consumption i-predecessors, LFTi , is defined as the latest time when their
on the condition that the application is completed before its operation and the transfer of the result to vj are completed.
deadline T . Data transmission overhead is very important in 
Do (tl )

a vehicular network since the channel is wireless. Wireless LFTi = max FTl + , ∀i ∈ {1, 2, . . . , nw },
tl ∈pred(ti ) b
channels require more time and energy to send data to other (5)
vehicles than wired lines. If the computation offloading cost,
including the data transmission cost, exceeds its gain, there is where pred (ti ) is a set of ti ’s i-predecessors, and b is the data
no need to offload tasks. transmission speed between vehicles.

VOLUME 11, 2023 3861

 M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

TABLE 1. Notation definitions.

The execution order of tasks that are assigned to a WV clouds, changes in task mapping cause more change in data
is already represented in a DAG. A predecessor must be transfer energy than task execution energy, assuming pro-
completed before a successor. The time when vj is available cessors in all vehicles consume similar power to execute an
for ti , since all vj ’s tasks with higher priorities than ti are instruction. This assumption makes sense, as electric vehicles
completed, is denoted by ATij . are likely to feature the best processors of their time, and
as is the case with processors for PCs and laptops these
ATij = max {xlj FTl } (6)
l∈{1,2,..,nt } days, these best processors are commonly produced by a few
manufacturers. As a result, the total Eexe is almost the same
The earliest start time when ti can run on vj , ESTij , is the
no matter where tasks are performed, thus this paper focuses
latter of LFTi and ATij .
on data transmission energy.
ESTij = max{LFTi , ATij } (7) To calculate the data transmission energy Etr , we adopt the
energy model in Liu’s study [33]. In the paper, the energy for
The earliest finish time of ti in vj , EFTij , is as follows: transferring M bytes between vj and vk is given as:
EFTij = ESTij + wij (8)
Eamp_jk = M × εfs × djk2 (13)
tri is the time it takes to transfer the result of ti from vj
to WVs having ti ’s i-successors, which can be obtained as
Here, εfs is the energy to transmit one byte in free space, and
follows.
djk is the distance between vj and vk . Thus, for all tasks in
Do (ti ) an application, Etr , the total energy needed to transmit result
tri = (9)
b data from vehicles performing predecessor tasks to vehicles
For ti to be executed in vj , all higher priority tasks must with successor tasks, can be computed as follows.
be completed, so ti ’s starting time, STi , should satisfy the X
following inequality. Etr = Eamp
X X XX
ATij ≤ STi (10) = εfs {Do (ti ) xij xlk djk2 } (14)
i tl ∈succ(ti ) j k
In order for appc to be completed before T , all tasks must
be completed before T .
where succ (ti ) is a set of ti ’s i-successors.
max {FTi } ≤ T (11) By default, data and result of cooperative computing are
i∈{1,2,..,nt }
transmitted through direct V2V communication. However,
The total energy Etotal consumed by vj for ti is the sum of at the time a WV finishes its task operation, it is often
the energy Eexe to perform the task ti and the energy Etr to out of the direct communication range of the vehicles that
transmit its results to vehicles with ti ’ i-successors. should receive the result. In this case, data is transmitted
through cellular networks. If pout is the probability that WV
Etotal = Eexe + Etr (12)
vj with ti is outside the direct communication range at the
In general, data transmission energy dominates total energy time ti is completed and ns (ti ) is the number of vehicles with
consumption in wireless networks. Likewise, in our vehicular i-successors of ti , the expected data transmission cost can be

3862 VOLUME 11, 2023

M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

obtained as follows. Algorithm 1 Stepwise Computation Offloading for

X Cost-Efficient Cooperation (SCOCC)
E[Ctr ] = {pout cce + (1 − pout )cdi } Do (ti )ns (ti ) (15) Require: A set of tasks T , Task DAG G
i
1: Texe ← ∅
where cce is the cost of transmitting data over cellular net- 2: Tready ← tsrc
works and cdi is the cost of transmitting data directly over 3: while T ̸ = ∅ do
V2V communication. The more transmissions over cellular 4: V ← Searching for nearby vehicles
networks, the higher the overall cost, because the cellular 5: for ti ∈ Texe do
network costs more. When it comes to the cost, we consider 6: Tready ← tm where tm ∈ succ(ti )
the fee that should be paid to network operators in case we use 7: end for
the cellular network. However, aside from money, DSRC is 8: Texe ← ∅
more efficient than cellular communication in terms of energy 9: V w ← CV and η vehicles nearest to the CV among
and transmission time. The current DSRC and C-V2X are vj ∈ V
based on WiFi and LTE respectively, then the energy effi- 10: Calculating b-levels for all ti ∈ Tready
ciency for WiFi is about 60% higher than that for LTE [34]. 11: Sorting tasks in Tready in the descending order of
Also, the transmission times of DSRC and C-V2X are 0.4 ms b-level
and 1 ms, respectively. Therefore, we argue that it is better 12: while Tready ̸= ∅ do
to use DSRC direct communication as much as possible than 13: Taking the first task ti in Tready
cellular communication for computation offloading. 14: Calculating EFTij for all vj ∈ V w
As a result, we need to find a task mapping X that min- 15: xij ← 1 where j = arg minvj ∈V w EFTij
imizes E[Ctr ] under the conditions of (10) and (11). It can 16: Tready ← Tready − ti
be defined as an optimization problem that has the following 17: end while
objective functions. 18: for xij ∈ X do
19: Assigning ti to vj if xij = 1
minimize Etr 20: T ← T − ti
minimize E[Ctr ] 21: end for
s.t (10), (11) (16) 22: Waiting for a message saying any task ti has started
23: Texe ← ti
24: end while
IV. PROPOSED METHOD
This section describes our Stepwise Computation Offload-
ing for Cost-efficient Cooperation (SCOCC) in which a CV
selects WVs among surrounding vehicles and finds a sched- and location. Then, it assigns tasks to other vehicles. Com-
ule that allocates application’s tasks in a DAG to the WVs. munication cost, total execution time, and network stability
SCOCC allocates tasks sequentially in a dynamic allocation vary depending on the schedule map X . Also, in a wireless
manner. A CV searches for nearby vehicles by broadcasting communication system, it is very important to minimize the
a request message, and chooses some vehicles as WVs to be energy and time required for data transmission. Furthermore,
offloaded tasks. In order to allocate tasks to the selected WVs, in a vehicle environment, the distance between vehicles is
the distance from the CV, average computing overhead of easy to change due to rapid mobility. This makes it difficult
tasks, and transmission speed of WVs are computed based on to perform reliable cooperative computing and increases the
the information returned by the WVs. Rather than allocating cost, so a CV should select vehicles that are least likely to go
all tasks at once, it assigns only tasks that are likely able out of the CV’s communication range as WVs.
to be executed in parallel soon. This is to reduce the time The proposed SCOCC builds a DAG with tasks of an
interval between the decision time for task assignment and the application, and sequentially schedules the task execution
actual execution time. Processing tasks together with WVs, considering the dependency between tasks. Initially, SCOCC
the CV completes the application by sequentially allocating in the CV starts the assignment process for the source
unassigned tasks along the DAG. task. Calculating the EFT of the source task when per-
formed in each WV, and according to the EFTs, it selects
A. STEPWISE COMPUTATION OFFLOADING FOR a WV and assigns the source task. After the source task
COST-EFFICIENT COOPERATION assignment, SCOCC only assigns tasks whose i-predecessors
This section explains in detail SCOCC, in which a CV have started running. For this assignment, SCOCC newly
requests WVs to compute tasks cooperatively. This coopera- searches for neighbor vehicles and transfers succeeding tasks
tive computing in vehicular clouds consists of two steps. First, to them. SCOCC repeats the search and allocation process
a CV searches for nearby vehicles. It requests cooperation whenever a new task starts running, which can be a draw-
from nearby vehicles and forms a vehicular cloud with the back compared to previous methods. However, the searching
vehicles that return information on their computing resources processes overhead may be negligible since the offloading

VOLUME 11, 2023 3863

 M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

request and response messages are just a few hundred bytes,

while offloaded data and task execution results are tens of
megabytes. The energy consumption overhead is also low,
since the energy needed to transmit is proportional to the data
size according to (13).
Algorithm 1 describes the process of SCOCC. Line 1 resets
Texe , the set of tasks that have started execution and Line 2
initializes Tready , the set of tasks that are ready to be exe-
cuted, with the source task. Then the loop that determines
the scheduling of tasks in T begins (Line 3). The client
vehicle broadcasts a message to search for nearby vehicles in FIGURE 2. Example of DAG.

Line 4. The i-successors of tasks in Texe are stored in Tready

(Lines 5-7). At first, no task is in Texe and the source task
has already been put into Tready in Line 2. In Line 9, only η
vehicles are selected as WVs in the order closest to the CV
among the searched vehicles. The CV itself also participates
in task processing. The value of η is determined later by
experiments.
In order to determine task priorities, we compute the
b-level of all tasks in Tready . The b-level of a task indicates
how far away the task is from the completion time of the
task that will be finally performed. This means that the exe-
cution of any task must begin at least its b-level before the
application deadline T , in order to complete the whole work
successfully. The b-level is calculated as follows. FIGURE 3. Timeline of task allocation by SCOCC.

b-leveli = max {b-levelk + tri } + wi , (17)

k∈succ(ti ) TABLE 2. Execution time of tasks on vehicles.

where wi is the average execution time of ti . If succ(ti ) = ∅,

which means ti is one of the final tasks without any succes-
sors, then b-level is set to just its execution time wi . The higher
the b-level, the earlier the task must be started. Thus, each task
is given priority in descending order of the b-level (Line 11).
Line 12 starts a loop to assign each task in Tready to a
vehicle in V w . In every round, the loop takes the first task
in Tready , which has the highest b-level among Tready ’s tasks.
Then it calculates EFT for the task when executed in each vj
in V w . The vehicle that shows the smallest EFT for the task is In this example, tasks are scheduled with three rounds.
marked as the worker for the task in X (Lines 12-17). Then, In the first round, only the source task t0 is assigned. Three
according to X , each ready task is assigned to the associated vehicles v1 , v2 , and v3 are found through the search mech-
worker (Lines 18-21). After that, the CV waits for the arrival anism, and the EFT of t0 in each vehicle is calculated as
of a message saying that any assigned task has started running shown in Table 4. Since v1 has the smallest EFT, t0 is
(Line 22). This message is sent by a WV that has started the assigned to v1 .
execution of an assigned task. When receiving this message, Right after starting t0 , v1 sends vc a message about it. Then
the CV puts the task into Texe and performs Lines 4-23 again. vc begins the second round to assign the successor tasks t1 , t2 ,
This is repeated until the entire application is completed. and t3 . A new search process discovers the same v1 , v2 , and
While searching for nearby vehicles in Line 4, a message v3 as before. Priority is given in the order t1 , t3 , t2 according
stating the start of a task may arrive. This task is also put to their b-levels. The EFTs of t1 , t2 , t3 in v1 , v2 , v3 are given
into Texe . in Table 5. According to this result, t1 , t3 , t2 are assigned to
v1 , vc , v2 respectively, since these vehicles have the smallest
B. NUMERICAL EXAMPLE EFT for each task. When t1 in v1 starts, the successor tasks
Figs. 2 and 3 illustrate an example of cooperative computing t4 and t5 can be scheduled. In this third round, vehicles v4 and
using SCOCC. A CV vc executes the application represented v5 are additionally discovered, and t4 is given a higher priority
by the DAG in Fig. 2 with the help of nearby vehicles. Table 2 than t5 according to the b-level. The EFTs calculated for all
shows the execution time of each task in each WV. The WVs are shown in Table 6, so t4 and t5 are assigned to v1 and
b-levels calculated using the execution time are in Table 3. v5 , respectively.

3864 VOLUME 11, 2023

M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

TABLE 3. b-level of tasks. TABLE 5. EFTs of t1 , t3 , and t2 on each WV.

TABLE 6. EFTs of t4 and t5 on each WV.

TABLE 4. EFT of t0 on each WV.

sum of the running time of all tasks divided by the schedule

V. PERFORMANCE EVALUATION length. This value is a performance improvement through
The performance of SCOCC was evaluated through simula- computation offloading, and the higher this value, the better
tion. We used Veins [16] to consider the practical situation of the algorithm.
the road and the movement of vehicles. In this simulation, P
vehicles move at random speeds on a straight road in the i wi
Speedup = (18)
Manhattan Street network, which is a typical topology for max{FTi } − ST0
various VANET simulations. A new vehicle is generated on
the road every 2 seconds and moves at random speed between 2) DATA TRANSFER ENERGY
25 and 30 km/h. To take into account the change of the inter- The data transfer energy Etr , defined in (14), is the total
vehicle distance, each vehicle speed is increased by 5 km/h amount of energy consumed for data transmission while
after 5 seconds and then reduced back to the original speed an application is executed through computation offloading.
after 5 seconds. All vehicles have arbitrary computing power, Energy is one of the most important resources for elec-
which follows a normal distribution with a mean of 400 and tric vehicles, and the largest part of the energy addition-
a standard deviation of 100. The data transmission speed ally needed for computation offloading is data transmission
between vehicles is set to 25 Mbps. For the link and physical energy. The lower the energy, the more likely the offloading
layer, we rely on DSRC/IEEE 802.11p and cellular networks. algorithm will be adopted in practice.
They provide collision control using CSMA/CA and channel
allocation. The data transmission speed between vehicles is 3) pout
set to 25 Mbps, since the DSRC data rate is known to be up In SCOCC, vehicles use direct V2V communication by
to 27 Mbps. The performances are compared with Heteroge- default for sending offloading tasks and the results. However,
neous Earliest-Finish Time (HEFT) [26], Greedy-Based Task the distance between cooperating vehicles may be outside
Scheduling Algorithm (GBTSA) [31], and a random schedul- the direct communication range while executing assigned
ing. The target applications include a randomly generated tasks, so the task results may not be delivered through V2V
DAG, a DAG of Gaussian Elimination, and a DAG of a Fast communication. In this case, the results should be transmitted
Fourier Transform. over cellular networks instead. In (15), pout was defined as the
We do not take congestion into consideration by making probability that a vehicle is outside the direct communication
vehicles sequentially offload their data, because this simula- range of other cooperating vehicles at the completion time
tion focuses on how well each algorithm distribute tasks to of its assigned task. Cellular networks are generally more
other vehicles. The congestion is not directly related to the expensive than V2V communications, so the lower the pout ,
decision what task should be given to a worker vehicle. Mean- the lower the data transmission cost. We will obtain the
while, some literature such as [6] consider the case where expected value of pout through simulation.
vehicles simultaneously send packets to study load balancing
of data offloading. Also, as with most previous studies on B. SETTING THE VALUE OF η
computation offloading, we do not consider the case where The value of η, the number of WVs selected from searched
WVs go offline or do not return execution results on purpose ones, is important to the performance of SCOCC. Intuitively,
because it is irrelevant for the following performance metrics. the number of WVs required will be closely related to the
number of tasks to be assigned, nt . Thus, we repeated the
A. PERFORMANCE METRICS simulation with η = nt + k and k = 0, . . . , 6, to find
1) SPEEDUP the best η.
The schedule length means the time consumed to complete Fig. 4 presents Speedup, data transfer energy, and pout
an application through cooperative computing. Speedup is the against k. Speedup did not change against k, but the data

VOLUME 11, 2023 3865

 M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

FIGURE 4. Performance for Fast Fourier Transform (FFT) applications depending on various η values.

TABLE 7. Choice of options.

FIGURE 6. DAG with a depth of 6 and a width of 3 or 4.

FIGURE 5. DAG with a depth of 3 and a width of 4 to 8.

FIGURE 7. Speedup against the number of tasks in a DAG with depth =

transfer energy and pout were the best when k = 3. Based on 3 and width = 4 ∼ 8.
this result, η was set to nt + 3 in all subsequent simulations.

C. RANDOMLY GENERATED DAGs

We randomly generated two types of DAGs. One is shallow
and wide, and the other is deep and narrow. The first type has
a depth of 3 and each level, except for the root, consists of
4 to 8 tasks at random (Fig. 5). On the other hand, the second
type of DAG has a depth of 6 and each level, except for the
root, consists of 3 or 4 tasks (Fig. 6). While the former is for
applications with a high degree of parallelism in tasks, and
the latter is for applications with long dependencies between
tasks. The number of tasks in each level follows a uniform
distribution and each task has n incoming edges from its FIGURE 8. Etr against the number of tasks in a DAG with depth = 3 and
predecessors following the probability width = 4 ∼ 8.

(
(2/3)(1/3)n−1 n = 1, 2, 3
P(n) = 3
. (19) Figs. 7∼9 compare the performances of four schedul-
(1/3) n=4
ing methods in the shallow and wide DAGs. First, Fig. 7

3866 VOLUME 11, 2023

M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

FIGURE 9. pout against the number of tasks in a DAG with depth = 3 and
FIGURE 10. Speedup against the number of tasks in a DAG with depth =
width = 4 ∼ 8.
6 and width = 3 ∼ 4.

is Speedup for each algorithm according to the number of

tasks. While GBTSA and HEFT achieved speed improve-
ments 3.31 times and 2.89 times, respectively, SCOCC
accelerated 2.74 times. The random scheduling showed
the lowest improvement, which was just 2.59. The rea-
son why the Speedup value of SCOCC is lower than that
of GBTSA and HEFT is that SCOCC intentionally limits
parallelism to reduce problems caused by fast mobility
of vehicles. This reduces the possibility that the distance
between vehicles becomes greater than the V2V communi-
cation range, thus reducing data transmission costs. The next FIGURE 11. Etr against the number of tasks in a DAG with depth = 6 and
two figures clearly show the effectiveness of this SCOCC’s width = 3 ∼ 4.
philosophy.
Fig. 8 shows the energy used for data transmission accord-
ing to the number of tasks in the DAG. While GBTSA and
the random scheduling used 41.3 kJ and 28.4 kJ on average,
HEFT consumed relatively smaller energy of 17.4 kJ for data
transmission. On the other hand, SCOCC used just 5.4 kJ for
data transmission, which is much smaller than the other algo-
rithms. SCOCC tries to reduce the average distance between
cooperating vehicles by taking only η vehicles as WVs in the
order of proximity to CV, and this is definitely effective in
reducing energy consumption.
Fig. 9 compares pout , the probability of vehicles out of
FIGURE 12. pout against the number of tasks in a DAG with depth =
V2V communication range from each other. While pout val- 6 and width = 3 ∼ 4.
ues of GBTSA, random scheduling, and HEFT were 0.184,
0.181, and 0.11 respectively, it was only 0.003 when using
SCOCC, which means WVs rarely went outside the transmis- are a small number of tasks at one level. As for the energy
sion range. Speedup of SCOCC was 17.2% and 5.2% lower consumption, GBTSA, the random scheduling, and HEFT
than that of GBTSA and HEFT, but it consumed only 13.1% consumed 56.9 kJ, 47.3 kJ, and 26.7 kJ respectively (Fig. 11).
and 31.2% of the energy that GBTSA and HEFT used for data On the other hand, SCOCC used 12.4 kJ, which is only 21.8%
transmission. This proves the efficiency of SCOCC relative to compared to GBTSA.
the amount of energy used. Finally, Fig. 12 compares pout . Whereas pout were 0.190,
Figs. 10∼12 illustrate the performances in the deep and 0.191, and 0.163 for GBTSA, random scheduling, and HEFT
narrow DAGs. Fig. 10 is Speedup according to the number respectively, it was just 0.032 for SCOCC, which is 16.8%
of tasks. The Speedup values were 2.86, 2.47, 2.44, and compared to GBTSA. For deep DAGs with a depth of 6,
2.15 in the order of GBTSA, HEFT, SCOCC, and the ran- Speedup of SCOCC was 13.6% and 1.1% lower than that of
dom scheduling, which were not much different from one GBTSA and HEFT, but SCOCC saved as much as 53.8% and
another. The performance gap between SCOCC and GBTSA 78.3% of energy compared to GBTSA and HEFT. This sub-
or HEFT was reduced compared to the case of the shallow stantiates that as the DAG depth increases, SCOCC becomes
and wide DAGs, as the constraint on the number of WVs in more and more energy efficient than others without lagging
SCOCC does not significantly harm parallelism when there in performance.

VOLUME 11, 2023 3867

 M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

FIGURE 13. DAG of Gaussian Elimination (GE) with a matrix size of 5.

FIGURE 14. Speedup for the GE DAG against ccr.

D. GAUSSIAN ELIMINATION
In order to evaluate the performance with actual applications,
we adopted two applications. The first is Gaussian elimina-
tion (GE). The GE is widely used to compute matrices for
various machine learning algorithms, thus it will be likely
used for various vehicular AI applications as well. Fig. 13
depicts the DAG of GE with a matrix size of 5. For a more
fair comparison, we used the concept of communication-to-
calculation ratio (ccr) [35], which is the communication cost
divided by the computation cost on a specific machine. The
ccr changes from 0.2 to 1.0 in units of 0.1.
Fig. 14 shows Speedup against ccr for each algorithm.
When the ccr was 0.2, Speedup of GBTSA was 1.82, higher
FIGURE 15. Etr for the GE DAG against ccr.
than 1.73 of HEFT and 1.72 of SCOCC. The random schedul-
ing has the lowest Speedup of 1.51. However, as the ccr
increased, Speedup of GBTSA continuously degraded, and
even lower than all the others when the ccr reached 0.6.
This is because GBTSA seeks to improve performance by
copying data to all succeeding tasks. The advantage of data
duplication was offset by increased energy costs. Actually,
HEFT and SCOCC showed the best Speedup after the ccr
exceeded 0.3. As the ccr increased to 1.0, Speedup of all the
methods decreased to 1.0, which means we cannot expect any
speedup through computation offloading.
Fig. 15 illustrates the energy consumed for data transmis-
sion against the ccr. The random scheduling consumed the
FIGURE 16. pout for the GE DAG against ccr.
most energy on average, but when the ccr was greater than
0.8, GBTSA used the most. As for HEFT, its data transfer
energy continued to slightly decreased as the ccr increased
because the number of offloaded operations was gradually
reduced. Overall, SCOCC consumed the smallest amount of
energy for data transmission. It was only about 36.8% of
GBTSA, 35.5% of HEFT and about 20.6% of the random
scheduling.
Fig. 16 shows pout against the ccr. For all ccrs, pout of the
random scheduling was the highest at 0.198, and SCOCC
consistently achieved the lowest at 0.021. The averages of
GBTSA and HEFT were around 0.10, about 5 times greater
FIGURE 17. DAG of Fast Fourier Transform (FFT) with k = 2.
than that of SCOCC. While pout of GBTSA increased rapidly
with the ccr, HEFT slightly reduced the probability as the ccr
increased.
algorithms. Since FFT is a basic module for all wireless
E. FAST FOURIER TRANSFORM communication, it will be used in vehicular networks as well.
The second real application to test is Fast Fourier Transform FFT performs computation on N = 2k points, then we set
(FFT), which is also widely used to evaluate scheduling k = 2 as shown in Fig. 17.

3868 VOLUME 11, 2023

M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

TABLE 8. Feature comparison.

FIGURE 18. Speedup for the FFT DAG against ccr. FIGURE 20. pout for the FFT DAG against ccr.

VI. DISCUSSION
We performed simulations on DAGs of randomly generated
DAGs, GE, and FFT to compare the proposed SCOCC with
HEFT, GBTSA, and a random scheduling. The simulation
results show that SCOCC can save data transfer energy by
70% compared to the other algorithms and reduce the propor-
tion of cooperating vehicles outside the V2V communication
range from each other by as much as 84%, although its
speedup improvement is 9% lower than the others. This result
proves the outstanding energy efficiency of SCOCC and the
FIGURE 19. Etr for the FFT DAG against ccr. high stability of VANETs built by SCOCC.
On the other hand, this study has clear limitations. The
larger the ccr, the lower the performance of cooperative
Fig. 18 shows Speedup for each algorithm against the computing. Additionally, as vehicle speed increases and
ccr. When the ccr was 0.2, the Speedup values for SCOCC, direction of movement becomes diversified, the stability of
GBTSA, and HEFT were 2.32, 2.31, and 2.27. Then when VANET clouds deteriorates. Also, SCOCC cannot maximize
the ccr was greater than 0.4, the four algorithms showed its advantages for DAGs with shallow depths. Therefore,
almost the same performance. When the ccr was 1.0, Speedup SCOCC may not be a good choice if the execution speed
approached 1. Therefore, there is no motivation to use task of shallow DAG applications is the top priority. Otherwise,
offloading. SCOCC is energy efficient and reliable with sufficient per-
Fig. 19, the energy consumption for transmission, formance improvements for most applications. The features
illustrates a similar pattern to that of GE. On average, a ran- of each algorithm are summarized in Table 8.
dom scheduling consumed the most energy, and the energy Lastly, SCOCC does not provide a specific MAC protocol
consumption of GBTSA increased sharply when ccr became that fits best for its operation. The cellular and DSRC/IEEE
0.6 or higher. The energy consumption of HEFT and SCOCC 802.11p MAC themselves provide error recovery using FEC
did not change much with the increase of ccr. SCOCC used and ACK mechanisms, but the 802.11p MAC cannot pre-
6.8 kJ on average for FFT, while it was just 3.8 kJ for GE. vent the hidden terminal problem, which can deteriorate the
This is because the communication overhead increased with proposed method performance. We will extend the proposed
the increased number of edges. method by adding a MAC like Enhanced Distributed Channel
Lastly, Fig. 20 depicts pout for the FFT application. Over- Access (EDCA) to reduce the offloading failure probability in
all, pout of all the other algorithms was high around 0.15, but future work. Also, the advanced SCOCC will be tested on a
SCOCC achieved very low probabilities of less than 0.04 for variety of road topologies in addition to the Manhattan Street
all ccr. network.

VOLUME 11, 2023 3869

 M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

VII. CONCLUSION [15] D. Fernandez-Baca, ‘‘Allocating modules to processors in a distributed

This paper proposed SCOCC to establish a stable VANET system,’’ IEEE Trans. Softw. Eng., vol. 15, no. 11, pp. 1427–1436,
Nov. 1989.
cloud and perform an energy-efficient computation offload- [16] C. Sommer, D. Eckhoff, A. Brummer, D. S. Buse, F. Hagenauer, S. Joerer,
ing between vehicles. To achieve the goal, SCOCC i) delays and M. Segata, ‘‘Veins: The open source vehicular network simula-
task assignment as much as possible and ii) considers the tion framework,’’ in Recent Advances in Network Simulation. Cham,
Switzerland: Springer, 2019, pp. 215–252.
distance from a client vehicle when selecting worker vehicles. [17] K. Kumar, J. Liu, Y.-H. Lu, and B. Bhargava, ‘‘A survey of computa-
The reduction of the time interval between task assignment tion offloading for mobile systems,’’ Mobile Netw. Appl., vol. 18, no. 1,
and its execution enhanced the VANET stability, and the pp. 129–140, Feb. 2013.
[18] A. Haydari and Y. Yilmaz, ‘‘Deep reinforcement learning for intelli-
reduction of average inter-vehicle distance was helpful for gent transportation systems: A survey,’’ IEEE Trans. Intell. Transp. Syst.,
both the stability and energy consumption. vol. 23, no. 1, pp. 11–32, Jan. 2022.
Unlike the static algorithms such as HEFT and GBTSA, [19] L. Guevara and F. A. Cheein, ‘‘The role of 5G technologies: Challenges in
smart cities and intelligent transportation systems,’’ Sustainability, vol. 12,
SCOCC does not try to optimize execution speed of no. 16, p. 6469, Aug. 2020.
entire tasks at once, so it does not show the better per- [20] F. Sun, F. Hou, N. Cheng, M. Wang, H. Zhou, L. Gui, and X. Shen,
formance in terms of execution time. However, SCOCC ‘‘Cooperative task scheduling for computation offloading in vehicular
cloud,’’ IEEE Trans. Veh. Technol., vol. 67, no. 11, pp. 11049–11061,
achieves essential stability in a fast moving vehicle envi- Nov. 2018.
ronment and minimization of energy consumed for wireless [21] X. Xu, Y. Xue, L. Qi, Y. Yuan, X. Zhang, T. Umer, and S. Wan, ‘‘An edge
communication. computing-enabled computation offloading method with privacy preser-
vation for Internet of Connected Vehicles,’’ Future Gener. Comput. Syst.,
vol. 96, pp. 89–100, Jul. 2019.
REFERENCES [22] P. Zhou, T. Braud, A. Zavodovski, Z. Liu, X. Chen, P. Hui, and
[1] T. do Vale Saraiva, C. A. V. Campos, R. D. R. Fontes, C. E. Rothenberg, J. Kangasharju, ‘‘Edge-facilitated augmented vision in vehicle-to-
S. Sorour, and S. Valaee, ‘‘An application-driven framework for intelligent everything networks,’’ IEEE Trans. Veh. Technol., vol. 69, no. 10,
transportation systems using 5G network slicing,’’ IEEE Trans. Intell. pp. 12187–12201, Oct. 2020.
Transp. Syst., vol. 22, no. 8, pp. 5247–5260, Aug. 2021. [23] T. L. Adam, K. M. Chandy, and J. R. Dickson, ‘‘A comparison of list
[2] G. Dimitrakopoulos and P. Demestichas, ‘‘Intelligent transportation sys- schedules for parallel processing systems,’’ Commun. ACM, vol. 17, no. 12,
tems,’’ IEEE Veh. Technol. Mag., vol. 5, no. 1, pp. 77–84, Mar. 2010. pp. 685–690, Dec. 1974.
[3] L. Zhu, F. R. Yu, Y. Wang, B. Ning, and T. Tang, ‘‘Big data analytics in [24] H. Arabnejad and J. G. Barbosa, ‘‘List scheduling algorithm for heteroge-
intelligent transportation systems: A survey,’’ IEEE Trans. Intell. Trans. neous systems by an optimistic cost table,’’ IEEE Trans. Parallel Distrib.
Syst., vol. 20, no. 1, pp. 383–398, Jan. 2018. Syst., vol. 25, no. 3, pp. 682–694, Mar. 2014.
[4] A. K. Haghighat, V. Ravichandra-Mouli, P. Chakraborty, Y. Esfandiari, [25] H. Wang and O. Sinnen, ‘‘List-scheduling versus cluster-scheduling,’’
S. Arabi, and A. Sharma, ‘‘Applications of deep learning in intelli- IEEE Trans. Parallel Distrib. Syst., vol. 29, no. 3, pp. 1736–1749,
gent transportation systems,’’ J. Big Data Anal. Transp., vol. 2, no. 2, Aug. 2018.
pp. 115–145, 2020. [26] H. Topcuoglu, S. Hariri, and M.-Y. Wu, ‘‘Performance-effective and
[5] A. Moubayed, A. Shami, P. Heidari, A. Larabi, and R. Brunner, low-complexity task scheduling for heterogeneous computing,’’
‘‘Edge-enabled V2X service placement for intelligent transportation sys- IEEE Trans. Parallel Distrib. Syst., vol. 13, no. 3, pp. 260–274,
tems,’’ IEEE Trans. Mobile Comput., vol. 20, no. 4, pp. 1380–1392, Mar. 2002.
Apr. 2021. [27] Y.-K. Kwok and I. Ahmad, ‘‘Static scheduling algorithms for allocating
[6] Y. Dai, D. Xu, S. Maharjan, and Y. Zhang, ‘‘Joint load balancing and directed task graphs to multiprocessors,’’ ACM Comput. Surv., vol. 31,
offloading in vehicular edge computing and networks,’’ IEEE Internet no. 4, Dec. 1999, Art. no. 406471.
Things J., vol. 6, no. 3, pp. 4377–4387, Jun. 2018. [28] R. Bajaj and D. P. Agrawal, ‘‘Improving scheduling of tasks in a hetero-
[7] Y. Mao, J. Zhang, Z. Chen, and K. B. Letaief, ‘‘Dynamic compu- geneous environment,’’ IEEE Trans. Parallel Distrib. Syst., vol. 15, no. 2,
tation offloading for mobile-edge computing with energy harvesting pp. 107–118, Feb. 2004.
devices,’’ IEEE J. Sel. Areas Commun., vol. 34, no. 12, pp. 3590–3605, [29] K. He, X. Meng, Z. Pan, L. Yuan, and P. Zhou, ‘‘A novel task-
Dec. 2016. duplication based clustering algorithm for heterogeneous computing envi-
[8] D. Hahn, A. Munir, and V. Behzadan, ‘‘Security and privacy issues ronments,’’ IEEE Trans. Parallel Distrib. Syst., vol. 30, no. 1, pp. 2–14,
in intelligent transportation systems: Classification and challenges,’’ Jan. 2019.
IEEE Intell. Transp. Syst. Mag., vol. 13, no. 1, pp. 181–196, [30] X. Geng, Y. Mao, M. Xiong, and Y. Liu, ‘‘An improved task scheduling
May 2021. algorithm for scientific workflow in cloud computing environment,’’ Clus-
[9] R. Hussain, S. H. Bouk, N. Javaid, A. M. Khan, and J. Lee, ‘‘Realization ter Comput., vol. 22, no. S3, pp. 7539–7548, May 2019.
of VANET-based cloud services through named data networking,’’ IEEE [31] S. Hu, G. Li, and W. Shi, ‘‘LARS: A latency-aware and real-
Commun. Mag., vol. 56, no. 8, pp. 168–175, Aug. 2018. time scheduling framework for edge-enabled Internet of Vehicles,’’
[10] P. Arthurs, L. Gillam, P. Krause, N. Wang, K. Halder, and A. Mouzakitis, IEEE Trans. Services Comput., early access, Aug. 19, 2021, doi:
‘‘A taxonomy and survey of edge cloud computing for intelligent trans- 10.1109/TSC.2021.3106260.
portation systems and connected vehicles,’’ IEEE Trans. Intell. Transp. [32] H. Chen, J. Wen, W. Pedrycz, and G. Wu, ‘‘Big data processing workflows
Syst., vol. 23, no. 7, pp. 6206–6221, Jul. 2022. oriented real-time scheduling algorithm using task-duplication in geo-
[11] K. Abboud, H. A. Omar, and W. Zhuang, ‘‘Interworking of DSRC distributed clouds,’’ IEEE Trans. Big Data, vol. 6, no. 1, pp. 131–144,
and cellular network technologies for V2X communications: A sur- Mar. 2020.
vey,’’ IEEE Trans. Veh. Technol., vol. 65, no. 12, pp. 9457–9470, [33] X. Liu and J. Wu, ‘‘A method for energy balance and data transmission
Dec. 2016. optimal routing in wireless sensor networks,’’ Sensors, vol. 19, no. 13,
[12] R. Molina-Masegosa and J. Gozalvez, ‘‘LTE-V for sidelink 5G V2X p. 3017, Jul. 2019.
vehicular communications: A new 5G technology for short-range vehicle- [34] L. Zou, A. Javed, and G.-M. Muntean, ‘‘Smart mobile device power
to-everything communications,’’ IEEE Veh. Technol. Mag., vol. 12, no. 4, consumption measurement for video streaming in wireless environments:
pp. 30–39, Dec. 2017. WiFi vs. LTE,’’ in Proc. IEEE Int. Symp. Broadband Multimedia Syst.
[13] K. Sehla, T. M. T. Nguyen, G. Pujolle, and P. B. Velloso, ‘‘Resource Broadcast. (BMSB), Jun. 2017, pp. 1–6.
allocation modes in C-V2X: From LTE-V2X to 5G-V2X,’’ IEEE Internet [35] M. Crovella, R. Bianchini, T. LeBlanc, E. Markatos, and R. Wisniewski,
Things J., vol. 9, no. 11, pp. 8291–8314, Jun. 2022. ‘‘Using communication-to-computation ratio in parallel program design
[14] M. R. Garey and D. S. Johnson, Computers and Intractability, vol. 29. and performance prediction,’’ in Proc. 4th IEEE Symp. Parallel Distrib.
New York, NY, USA: Freeman, 2002. Process., Dec. 1992, pp. 238–245.

3870 VOLUME 11, 2023

M. Gong et al.: VC Forming and Task Scheduling for Energy-Efficient Cooperative Computing

MINYEONG GONG received the B.S. degree in SANGHYUN AHN (Member, IEEE) received
electrical and computer engineering from the Uni- the B.S. and M.S. degrees in computer engi-
versity of Seoul, South Korea, in 2016, where he neering from Seoul National University, Seoul,
is currently pursuing the Ph.D. degree in electri- South Korea, in 1986 and 1988, respectively, and
cal and computer engineering. His research inter- the Ph.D. degree in computer science from the
ests include vehicle networks, the IoT, and cloud University of Minnesota, in 1993.
computing. She is currently a Professor with the Department
of Computer Science and Engineering, University
of Seoul, Seoul. Her research interests include
wireless ad hoc, sensor and vehicular networks,
internet protocols, routing protocols, and the IoT.

YOUNGHWAN YOO (Member, IEEE) received

the B.S. and M.S. degrees in computer engineering
and the Ph.D. degree in electrical engineering and
computer science from Seoul National University,
Seoul, South Korea, in 1996, 1998, and 2004,
respectively.
From 2004 to 2006, he was a Research
Employee at the OBR Center for Distributed
and Mobile Computing, University of Cincinnati,
USA. He is currently a Professor with the
School of Computer Science and Engineering, Pusan National University,
South Korea. He is also a Visiting Professor with the Department of Electri-
cal and Computer Engineering, The University of British Columbia, Canada.
His research interests include 5G/6G, wireless communications, and AI in
wireless networks.
Dr. Yoo served as a program committee member or a reviewer for a variety
of journals and conferences.

VOLUME 11, 2023 3871