A Greedy Dynamic Task Offloading and Service Management in V2V

Networks

Hanyang Xing

Institute of Communication Engineering, University of Electronic Science and Technology of China,

2006 Xiyuan Avenue, High-tech Zone (West), Chengdu, China

Keywords: Vehicle-to-Vehicle Communication, Knapsack Algorithm, Game Theory, Decision Tree.

Abstract: This paper introduces an advanced framework for optimizing task offloading and service caching in Vehicle-

to-Vehicle (V2V) communication networks. The proposed approach leverages a greedy algorithm to address

key challenges such as offloading latency, energy consumption, and system overhead. By incorporating

practical factors such as task size, server storage capacity, and task popularity, the framework efficiently

allocates tasks, thereby reducing computational delays and enhancing network performance. The

effectiveness of the algorithm is validated through comprehensive simulations that demonstrate significant

improvements in both time efficiency and resource utilization compared to existing methodologies. The

results underscore the potential for future advancements in V2V networks, particularly in enhancing network

stability under high-speed conditions and developing robust communication systems that maximize the use

of roadside computational resources.

1 INTRODUCTION

The computational demands for information

transmission are rapidly increasing with the

proliferation of the Internet of Things (IoT). While the

existing cloud server infrastructure remains an

efficient and cost-effective solution for handling most

user requests, the growing number of local devices

with stringent computational needs—including

augmented reality, virtual reality, autonomous

vehicles, and vehicle communication—has created

new challenges. These applications require ultra-

reliable and low-latency communications (URLLC) to

ensure low latency, reduced power consumption, and

adequate computational resources without

compromising performance (Zeng, Zhang, Wang, Liu,

Wang, 2020). In this context, Mobile Edge Computing

(MEC) is becoming a critical component of modern

networks, serving as a distributed computational

infrastructure embedded within hosts (He, Li, Chen,

Wang, 2019). Tasks can be offloaded to these edge

servers, a process known as Mobile Edge Computing

Offloading (MECO) (He et al, 2019), allowing highly

demanding tasks to be processed and subsequently

downloaded to local devices.

https://orcid.org/0009-0008-5534-0930

However, current URLLC devices and services

face several significant challenges. Firstly, many

communication networks are burdened with a

substantial number of computation-intensive tasks,

which strains the computational capacity of nodes or

hosts. This strain, coupled with the network's

intermittent availability, can result in inaccurate

communication processes (Zeng et al, 2020).

Secondly, the network often operates near full

capacity, making immediate transmission and

processing of data nearly impossible due to

congestion (Zeng et al, 2020; He et al, 2019). Thirdly,

although the existing capacity allocation algorithms

within MEC networks are sophisticated, they fall

short in accounting for long-term effectiveness,

leading to suboptimal resource allocation over

extended periods (He et al, 2019).

Vehicle-to-vehicle (V2V) communication, which

is crucial for the advancement of driverless vehicles

and the improvement of traffic conditions, is

particularly affected by these challenges. V2V

communication requires stable connections between

vehicles, which function as both hosts and nodes

while in motion (Liu, Wang Chen, Bensaou 2021).

Current vehicular ad hoc networks rely on cellular

Xing, H.

A Greedy Dynamic Task Ofﬂoading and Service Management in V2V Networks.

DOI: 10.5220/0013474100003941

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 11th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2025), pages 653-659

ISBN: 978-989-758-745-0; ISSN: 2184-495X

653

networks for direct data transfer. However, this

approach is limited, often leading to inefficiencies

and wasted computational resources within the

network (Pachat, Chen, Chen, 2020).

To address these challenges and integrate mobile

edge networks into vehicular systems, our research

proposes a robust framework for task offloading and

service caching within both edge and cloud networks.

This framework is designed to reduce latency and

energy consumption, thereby minimizing time delays

and resource demands during information

transmission. Based on computational capabilities,

data packets will dynamically choose the most

optimal processing route—whether locally, at the

edge server, or within the cloud—correlating

computational power with transmission time. The

proposed framework incorporates a non-cooperative

game theory-based algorithm to efficiently distribute

packets across various endpoints. Additionally,

considering the limited energy resources at the edge

servers, the framework leverages a 0-1 knapsack

algorithm to refine the initial selection process,

implementing dynamic service caching based on task

popularity (i.e., frequency of service requests) to

ensure that the endpoint can handle the required data

volume. The final offloading decision is derived from

a comprehensive analysis of both service caching and

the original selection.

1.1 Accessibility

Vehicle-to-Vehicle (V2V) communication is a form

of Mobile Edge Computing (MEC) that leverages

resources from vehicles and roadside infrastructure to

collect and disseminate traffic information, thereby

supporting self-driving vehicles and enhancing traffic

conditions. Currently, V2V communication demands

low latency and energy-efficient transmission, but

existing algorithms struggle to fully meet these

requirements. However, through the optimization and

refinement of these algorithms, V2V communication

can be made faster and more energy-efficient.

To improve the accessibility of V2V

communication, task offloading should be prioritized

based on the popularity or importance of the tasks,

ensuring a more rational processing by the algorithm.

Additionally, the algorithm must operate within

reasonable computational requirements to reduce

processing time and further minimize latency.

1.2 Inserting V2V Concepts

V2V communication, short for Vehicle-to-Vehicle

communication, is a subset of Mobile Edge

Computing (MEC) that offers an alternative to

Mobile Cloud Communication (MCC). An MEC

system comprises not only a cloud center but also

various edge devices, such as base stations and

mobile phones. Tasks requiring computation are

partitioned, with some processed locally while others,

particularly those that are computation-intensive, are

offloaded to MEC servers or cloud servers to utilize

higher computational power. Key factors in

determining the optimal task partitioning include the

offloading ratio, CPU-cycle frequency, and

transmission power. This approach results in lower

latency, reduced energy consumption, and shorter

transmission distances. Consequently, the system is

transformed into an information-centric architecture

capable of partitioning and offloading tasks more

efficiently, rather than rigidly transferring all tasks

from one end to another.

2 RELATED WORK

Vehicle-to-Vehicle (V2V) communication

addressing specific traffic issues has been extensively

studied in various contexts. Dey et al. explored

wireless communication within a heterogeneous

network (Het-Net) environment, utilizing Wi-Fi,

DSRC, and LTE for the transmission of accident

information between vehicles. Their research

demonstrated that this system effectively reduces the

dependency on infrastructure communication and

establishes a stable connection between rapidly

moving vehicles (Dey, Ding, Zheng, 2016). Navas et

al. developed a device equipped with an adaptive

cruise control (ACC) system that can be easily

installed on vehicles to mitigate poor traffic

conditions (Navas, Milanés, 2019). Notably, this

device remains effective even when preceding

vehicles are not equipped with it, enhancing its

acceptability and widespread adop8tion (Navas et al,

2019).

In 2017, Perfecto et al. proposed a framework for

beam alignment in millimeter-wave V2V networks,

which improves millimeter-wave communication by

addressing issues related to directionality, blockage,

and alignment delay (Perfecto, Del Ser, Bennis 2017).

This framework has proven effective under complex

conditions in high-density, multi-lane highway

scenarios (Perfecto et al, 2017). Ahmad et al.

recommended a validation method for congestion

control and performance in V2V systems through

vehicle-level testing, which enables congestion

detection within a 1.5-meter range and achieves a

transmission latency of 600ms (Ahmad,

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

654

Bakhshizadeh, Yilmaz, 2019). Bian et al. introduced

a linear feedback controller designed to reduce time

headway in V2V communication under various

conditions. Their approach includes a novel definition

of the inter-vehicle distance, which helps prevent

undesirable intermittent fluctuations in distance

detection (Bian, Wu, Zheng 2019). By increasing the

number of predecessors, their method ensures internal

asymptotic stability and meets string stability

requirements (Bian et al, 2019). Gao et al. proposed

an enhanced GPSR-based wireless routing scheme

that incorporates newly introduced parameters, such

as a prediction mechanism and computed weights, to

improve wireless communication stability (Gao,

Zhao, Yin, 2021). Their approach demonstrated

superior performance in terms of packet delivery

ratio, wireless hops, and time delay (Gao et al, 2021).

Bazzi et al. conducted a comparison between the

IEEE 802.11 standard and the proposed V2V

communication network, revealing that the proposed

network outperforms IEEE 802.11 in terms of

maximum awareness range and vehicle density, given

identical inputs and external conditions (Bazzi,

Wong, 2017).

3 ALGORITHMS

In this part, it shows the part of original code of the

purposed algorithm. The principle of this part of code

illustrates how the decision is made based on the task

size and capacity requirement. In Q1, it is a 2-

dimensional array which contains size and capacity

requirement information about the task.

In this section, we present a segment of the

original code implementing the proposed algorithm.

This code illustrates the core decision-making

mechanism, which is based on the task size and the

corresponding capacity requirements. The algorithm

operates on a two-dimensional array, denoted as Q1,

which encapsulates the size and capacity

requirements of the tasks. The iterative nature of the

algorithm allows for dynamic adjustments to be made

as tasks are processed.

To evaluate the proposed framework, we consider

a V2V communication system comprising vehicles

and roadside edge servers. Vehicles are assumed to

move at varying speeds with densities ranging from

sparse rural areas to dense urban environments. In this

model, each vehicle generates tasks with specific

computational requirements, including CPU cycles

and storage needs. Edge servers are equipped limited

storage and processing capacities which influencing

task offloading decisions. Moreover, we built a traffic

model simulates real-world scenarios where vehicles

dynamically interact with edge servers, enabling

comprehensive analysis of system performance.

Our proposed algorithm incorporates three main

modules: Task Offloading, Dynamic Service

Caching, and Adjustment of Task Offloading. These

modules work together to optimize computational

resource utilization. The algorithm begins by

evaluating task size and capacity requirements

through a two-dimensional array. Probabilities are

calculated to determine task selection, and a random

selection mechanism ensures balanced resource

allocation.

As Algorithm 1 shown, this algorithm begins by

resetting the two-dimensional array Q1 through the

Restart function, ensuring that all task-related data is

cleared before processing. The main decision-making

process occurs in the UpdateAction1 function, where

tasks are evaluated based on their size and capacity

requirements stored in Q1.

Probabilities are calculated to determine which

tasks are likely to be selected for execution, and a

random selection mechanism is employed to choose

the next task, ensuring a balanced distribution of

computational resources. This approach enables the

algorithm to adapt dynamically to varying task

demands, optimizing performance in a V2V

communication environment.

Data: task size and capacity requirements

Result: how to select the next task;

while not iterate over all the tasks do

read the task index;

if the loaded task is not current task then

calculate the resource ratio;

compute the probability for task selection

update the probability list

else

select a new task

end

Algorithm 1: How to select a new task

4 METHODOLOGY

This section outlines the methodology of the

proposed framework, including the algorithm's

processing flow and structure, which encompasses:

1. Task Offloading: Using non-cooperative

game theory to minimize latency.

2. Dynamic Service Caching: Employing the

0-1 knapsack algorithm to cache tasks based

on popularity and storage capacity.

A Greedy Dynamic Task Ofﬂoading and Service Management in V2V Networks

655

3. Adjustment of Task Offloading: Refining

initial decisions to optimize resource

utilization.

4.1 Algorithm Process

The principles and procedures of the framework are

briefly introduced below with corresponding

diagrams: The process begins with tasks that are

candidates for offloading to an edge server. The IoT

device first identifies and confirms the edge server

before initiating task transmission. Upon receiving

the notification, the edge server evaluates the service

request based on its frequency of occurrence. A

higher frequency increases the likelihood that the task

will be offloaded to that particular edge server. The

edge server then downloads services from the cloud

server to cache them, considering both the popularity

of the services and the storage limitations of the edge

server. If the requested service for a task is already

cached in the edge server, the task can be offloaded

to it. Otherwise, the task will need to be offloaded to

another endpoint.

4.2 Algorithm Structure

The algorithm first identifies the optimal terminal for

task offloading and then caches the task accordingly.

The objective is to minimize energy consumption

while maximizing task popularity, using the 0-1

knapsack algorithm and incorporating game theory.

Based on these factors, the algorithm provides the

most effective strategy for task offloading.

4.2.1 Algorithm Explanation

In this algorithm, five IoT devices are assumed to

handle local processing. Additionally, both edge

servers and cloud servers are available for task

offloading. The values 0, 1, and -1 represent local

processing, edge server processing, and cloud server

processing, respectively, for each task.

The algorithm is composed of three main modules:

Task Offloading, Dynamic Service Caching, and

Adjustment of Task Offloading. The Task Offloading

module employs a non-cooperative game theory

approach to make initial task offloading decisions,

aiming to minimize system overhead. The Dynamic

Service Caching module uses the 0-1 knapsack

algorithm to maximize the popularity of task-

requested services by implementing dynamic service

caching in edge servers. Finally, the Adjustment of

Task Offloading module fine-tunes the initial

offloading decisions based on the caching outcomes at

the edge server.

The following algorithms are used for

comparative experiments:

• The TOCS algorithm, which

comprehensively addresses task offloading

and service caching within a mobile edge

computing network without relying on cloud

server assistance.

• The TO algorithm, which focuses solely on

the offloading problem within a mobile edge

computing network.

• The LP algorithm, where all tasks are

processed on local IoT devices.

5 EXPERIMENT

This section presents two experiments designed to

validate the algorithm under different conditions and

scenarios. It begins with an introduction to the

parameters and key components of the algorithm and

the experimental setup. Following this, the specific

steps of the experiments are detailed, along with an

analysis of the results.

Two experiments were conducted to evaluate the

proposed framework under different conditions. Key

parameters include CPU capacity, task size, and

server storage. The experiments simulate task

generation and offloading in both sparse and dense

traffic scenarios.

5.1 Experiment Introduction

This subsection describes two experiments conducted

under varying conditions, along with the primary

results. Figure 1 illustrates the algorithm's flowchart

and provides an overview of the experimental

procedure. Initially, devices and vehicles request

services from nearby edge servers, such as base

stations. These tasks are subsequently aggregated

Figure 1: Flowchart of the Algorithm.

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

656

at the base station, where they are ranked based on

their popularity, size, and the storage capacity of the

base station. The algorithm subsequently generates a

decision based on this ranking. The appropriate

services are then downloaded from the cloud server

and processed on different hosts, ensuring that tasks

are executed where energy consumption and

processing time are minimized.

The parameters used in the experiments are listed

in Table 1 below. Each parameter plays a critical role

in influencing task offloading latency. For example,

the Task CPU parameter affects the processing time

and directly impacts the task offloading decision,

especially since the edge server may not provide

sufficient computational capacity for certain tasks.

The Task Size parameter indicates the storage

requirement for hosting the task. Task Popularity

reflects the importance of each task, determining

which tasks should be prioritized in the scheduling

process.

Table 1: Factors to select the best server of task.

Factor Name Definition

Task CPU Re

uired CPU ca

acit

for tas

Task Size Re

uired server size for storin

Task Popularit

Importance of tas

Server CPU CPU ca

acit

offered b

serve

Server Stora

e Stora

e size offered b

serve

5.2 The Capacity Experiment

In the first experiment, the algorithm's performance is

evaluated by varying the CPU capacity for task

requests. By adjusting the CPU value, different task

offloading decisions are observed. Here, 3 types of

extant method, that are introduced in chapter 4.2.1,

are imported to compare with the proposed algorithm.

Initially, the algorithm is run to determine a

selection that minimizes system overhead. In this

phase, the capacities of the cloud server, edge server,

and local devices are predefined. Here, "Task CPU"

refers to the required processing capacity, while

"Task Size" indicates the storage requirement for

each task. The algorithm employs a non-cooperative

game theory approach, making decisions based on

single-oriented connections. To assess the impact of

CPU requirements, the fifth task is treated as a

variable, with its required CPU capacity adjusted

from 0.1GHz to 1GHz.

After iterating the selection process three times,

the resulting task allocation is 0-0-1-0-0, indicating

that the third task is offloaded to the edge server while

the others are processed locally.

Following the initial result, the algorithm

considers task popularity and the storage capacity of

edge servers in the Dynamic Services Caching phase.

The outcome of this phase is 0-1-1-0-0, meaning that

tasks 2 and 3 cache their services on the edge server,

while the others do not.

Finally, combining the initial result with the

caching outcome, the final task offloading decision is

made, yielding the selection 0-0-1-0-0, where only

the third task is offloaded to the edge server, and the

remaining tasks are processed locally. As depicted in

Figure 2, when the CPU requirement is not

excessively high, the algorithm achieves the lowest

task offloading cost compared to other methods.

However, when the CPU requirement exceeds a

certain threshold, the cost becomes constrained by the

limited storage capacity of the edge server, leading to

no further cost reduction.

Table 2: Factors to select the best server of task.

Method

Storage

Propose TOSC TO LP

0.1 3.75602 3.75602 5.0203 6.48231

0.2 4.04102 4.04102 5.3053 6.76731

0.3 4.32602 4.32602 5.5903 7.05231

0.4 4.38102 4.61102 5.8753 7.33731

0.5 4.38102 4.89602 6.1603 7.62231

0.6 4.38102 5.18102 6.4453 7.90731

0.7 4.38102 5.46602 6.7303 8.19231

0.8 4.38102 5.75102 7.0153 8.47731

0.9 4.38102 6.03602 7.3003 8.76231

1.0 4.38102 6.32102 7.5853 9.04731

Figure 2: The cost of the task offloading with the change of

CPU requirement.

A Greedy Dynamic Task Ofﬂoading and Service Management in V2V Networks

657

5.3 The Storage Experiment

In the second experiment, the performance of the

algorithm is assessed by varying the storage capacity

of the edge servers. The process for determining the

final task offloading decision follows the same steps

as in the first experiment, but with a key difference:

the storage capacity of the edge servers is altered in

each iteration. This adjustment is crucial, as the

storage capacity directly impacts the number of tasks

that can be offloaded, as well as the popularity of

those tasks within the network.

For this experiment, the storage unit is set to 3,

meaning that each edge server has the capacity to

accommodate three tasks. Each task has a uniform

size of 1, while their respective values are assigned as

1, 2, 0, 0, and 1. These values represent the

importance or priority of the tasks in the system,

which influences the offloading decisions.

The experiment begins with the execution of the

Dynamic Service Caching algorithm. This step

determines which tasks should be cached on the edge

servers based on their popularity and the available

storage space. The outcome of this step guides the

subsequent task offloading process by ensuring that

high-priority tasks are positioned where they can be

processed most efficiently.

Following the caching decisions, the Adjustment

of Task Offloading phase is conducted. In this phase,

the algorithm refines the initial offloading decisions,

taking into account the caching results and the storage

constraints of the edge servers. The final offloading

decisions, under these conditions, are determined to

be 0-0-0-1-1. This result indicates that the last two

tasks will be offloaded to the edge server, while the

first three tasks will remain for local processing.

The experiment also explores how different

storage conditions, combined with a fixed CPU

capacity of 0.3, influence the overall cost associated

with each offloading decision. The analysis reveals

that as the storage capacity and task values vary, the

cost of offloading decisions fluctuates accordingly.

This relationship underscores the importance of

optimizing both storage and processing capacities to

minimize costs and maximize efficiency in the V2V

communication system.

This experiment highlights the critical role of

edge server storage in the task offloading process. By

adjusting storage capacity and analyzing its impact on

task allocation, the algorithm demonstrates its ability

to adapt to varying network conditions, thereby

ensuring efficient resource utilization and optimal

performance across different scenarios.

Table 3: Factors to select the best server of task.

Storage

Methods

1 2 3 4

Proposed 7.68642 7.59442 7.40793 7.40793

TOSC 8.8587 7.59442 7.40793 7.40793

TO 8.86992 8.8587 8.6722 7.40793

LP 10.3207 10.3207 10.3207 10.3207

5.4 Analysis

In the first experiment, the results, as illustrated in

Figure 2, demonstrate the impact of varying CPU

capacity in the edge server using different data

transfer methods. The proposed algorithm

consistently achieves the lowest transmission cost

compared to a singular processing method.

Furthermore, the cost of task offloading peaks when

the CPU capacity is sufficiently high, as the CPU can

easily handle the tasks without requiring additional

resources.

In the second experiment, the results, presented in

Figure 3, focus on the influence and effectiveness of

server storage. Under identical storage conditions, the

proposed algorithm outperforms other algorithms by

minimizing the cost of task offloading. Similar to the

first experiment, when the storage capacity is

sufficiently large, the cost stabilizes because the tasks

no longer fully occupy the available server space.

These experiments collectively demonstrate the

validity and effectiveness of the proposed algorithm

in optimizing task offloading in V2V communication

systems. The algorithm's ability to minimize costs

while effectively utilizing CPU capacity and server

storage underscores its potential for practical

applications in edge computing environments.

Figure 3: The cost of the task offloading with the change of

server storage.

VEHITS 2025 - 11th International Conference on Vehicle Technology and Intelligent Transport Systems

658

However, sufficient edge computing capacity is

currently a challenge and the relative movement

between vehicles and vehicles causes the challenge of

stable information connection and further influences

the entire communication network of V2V.

6 CONCLUSIONS

This paper presents an effective algorithm and

framework for task offloading and service

downloading, inspired by multi-predecessor

investigations and research. The algorithm considers

various factors, including task size, server storage,

and task popularity, to optimize task offloading

decisions. These considerations allow for intelligent

service downloading decisions based on the specific

conditions of the network. As demonstrated through

experimental simulation and validation, the proposed

algorithm can effectively minimize both time delays

and energy consumption by making informed,

strategic decisions.

Looking ahead, further improvements could be

made to enhance the stability of V2V networks,

particularly in scenarios where vehicles are moving at

high speeds. Additionally, the development of

vehicle-to-infrastructure communication holds

promise for fully leveraging the computational

resources of roadside infrastructures.

REFERENCES

Zeng, J., Zhang, H., Wang, H., Liu, Y., & Wang, W. (2020).

Mobile edge communications, computing, and caching

(MEC3) technology in the maritime communication

network. China Communications, 17(5), 223-234.

He, Y., Li, Y., Chen, Y., & Wang, H. (2019). D2D

communications meet mobile edge computing for

enhanced computation capacity in cellular networks.

IEEE Transactions on Wireless Communications,

18(3), 1750-1763.

Liu, L., Wang, T., Chen, Y., & Bensaou, B. (2021).

Vehicular edge computing and networking: A survey.

Mobile Networks and Applications, 26(3), 1145-1168.

Pachat, J., Chen, Y., & Chen, Y. (2020). Index coding in

vehicle to vehicle communication. IEEE Transactions

on Vehicular Technology, 69(10), 11926-11936.

Dey, K. C., Ding, Y., & Zheng, Y. (2016). Vehicle-to-vehicle

(V2V) and vehicle-to-infrastructure (V2I)

communication in a heterogeneous wireless network –

Performance evaluation. Transportation Research Part

C: Emerging Technologies, 68, 168-184.

Navas, F., & Milanés, V. (2019). Mixing V2V- and non-

V2V-equipped vehicles in car following.

Transportation Research Part C: Emerging

Technologies, 108, 167-181.

Perfecto, C., Del Ser, J., & Bennis, M. (2017). Millimeter-

wave V2V communications: Distributed association

and beam alignment. IEEE Journal on Selected Areas

in Communications, 35(9), 2148-2162.

Ahmad, S. A., Bakhshizadeh, A., & Yilmaz, M. (2019). V2V

system congestion control validation and performance.

IEEE Transactions on Vehicular Technology, 68(3),

2102-2110.

Bian, Y., Wu, C., & Zheng, Y. (2019). Reducing time

headway for platooning of connected vehicles via V2V

communication. Transportation Research Part C:

Emerging Technologies, 102, 87-105.

Gao, H., Zhao, J., & Yin, Y. (2021). V2VR: Reliable hybrid-

network-oriented V2V data transmission and routing

considering RSUs and connectivity probability. IEEE

Transactions on Intelligent Transportation Systems,

22(6), 3533-3546.

Bazzi, A., & Wong, K. W. (2017). On the performance of

IEEE 802.11p and LTE-V2V for the cooperative

awareness of connected vehicles. IEEE Transactions on

Vehicular Technology, 66(11), 10419-10432.

A Greedy Dynamic Task Ofﬂoading and Service Management in V2V Networks

659