IAES International Journal of Artificial Intelligence (IJ-AI)

Vol. 13, No. 4, December 2024, pp. 4186∼4194

ISSN: 2252-8938, DOI: 10.11591/ijai.v13.i4.pp4186-4194 ❒ 4186

A new approach based on genetic algorithm for

computation offloading optimization in multi-access edge
computing networks
Marouane Myyara, Oussama Lagnfdi, Anouar Darif, Abderrazak Farchane
Laboratory of Innovation in Mathematics, Applications, and Information Technology, Department of Computer Science,
Faculty of Polydisciplinary, Sultan Moulay Slimane University, Beni Mellal, Morocco

Article Info ABSTRACT

Article history: The proliferation of smart devices and the increasing demand for resource-
intensive applications present significant challenges in terms of computational
Received Dec 29, 2023
efficiency, leading to surge in data traffic. While cloud computing offers partial
Revised Feb 11, 2024 solutions, its centralized architecture raises concerns about latency. Multi-access
Accepted Mar 21, 2024 edge computing (MEC) emerges as promising alternative by deploying servers
at the network edge to bring computations closer to user devices. However, op-
Keywords: timizing computation offloading in the dynamic MEC environment remains a
complex challenge. This paper introduces novel genetic algorithm-based ap-
Cloud computing proach for efficient computation offloading in MEC, considering processing
Computation offloading and transmission delays, user preferences, and system constraints. The pro-
Genetic algorithms posed approach integrates computation offloading and resource allocation al-
Multi-access edge computing gorithm based on evolutionary principles, combined with a greedy strategy to
Resource optimization maximize overall system performance. By utilizing genetic algorithms, the pro-
Service time minimization posed method enables dynamic adaptation to changing conditions, eliminating
the need for intricate mathematical models and providing an appealing solution
to the complexities inherent in MEC. The urgency of this research arises from
the critical need to enhance mobile application performance. Simulation re-
sults demonstrate the robustness and efficacy of our approach in achieving near-
optimal solutions while efficiently balancing computation offloading, minimiz-
ing latency, and maximizing resource utilization. Our approach offers flexibility
and adaptability, contributing to advancement of MEC networks and addressing
the requirements of latency-sensitive applications.

This is an open access article under the CC BY-SA license.

Corresponding Author:
Marouane Myyara
Laboratory of Innovation in Mathematics, Applications, and Information Technology
Department of Computer Science, Faculty of Polydisciplinary, Sultan Moulay Slimane University
Beni Mellal 23000, Morocco
Email: marouane.myyara@usms.ac.ma

1. INTRODUCTION
In recent years, the rapid proliferation of smart devices has led to an increased demand for appli-
cations such as surveillance, augmented reality, voice recognition, automated transportation, healthcare, and
autonomous driving. These applications generate a large amount of data traffic, especially from tasks that re-
quire low latency and involve complex computations [1]. However, smart devices often have limited processing
power and storage capacity, which makes it challenging to perform computation-intensive tasks locally [2]. To

Journal homepage: http://ijai.iaescore.com

Int J Artif Intell ISSN: 2252-8938 ❒ 4187

address these challenges, cloud computing has emerged as a solution by utilizing the computing and storage
resources of cloud servers over a wide area network for task processing [3]. While cloud computing can ef-
fectively handle resource-intensive applications and overcome certain limitations of user devices [4], it faces
difficulties with latency-sensitive applications due to congestion and packet losses caused by the increasing
data generated by user devices [5].

To overcome these challenges, multi-access edge computing (MEC) proposes the deployment of edge
servers at the network’s periphery to improve support for low-latency applications and services while ensuring
quality of service (QoS) for end-users [6]. The MEC framework consists of three levels [7]: the user device
level, the edge level, and the cloud level. At the user device level, devices communicate with the local edge
server through WLAN. The edge level consists of multiple interconnected MEC servers via a metropolitan area
network (MAN), while the cloud level involves global cloud servers using a wide area network. Computation
offloading allows devices to offload tasks to the edge servers, contributing to overall performance enhancement.
Various studies prioritize computation offloading in MEC networks, focusing on the full offloading of tasks to
an edge server for each user device request [8]. However, selecting the optimal offloading server is a complex
task that requires adaptation to fluctuating capacities and task diversity [9]. Traditional offline optimization is
unsuitable due to uncertainties in offloading requests and changing network and computing conditions. The
use of MEC for task offloading represents a promising strategy for reducing network latency [10]. In particular,
MEC addresses the computational offloading needs of user devices by processing tasks near the edge, rather
than relying solely on a central cloud [11]. However, the task offloading problem is recognized as an NP-hard
problem [12], and finding a solution poses significant challenges.

Existing research has explored computation offloading in edge computing and offered various ap-
proaches to minimize latency in MEC systems. For example, Silva et al. [13] introduced a hybrid cloud-based
offloading scheme. In contrast, Maleki and Mashayekhy [14] considered the runtime and application move-
ment impact for offloading in mobile edge computing. Hossain et al. [15] proposed a task-offloading model
that balances energy consumption and user satisfaction. Heuristic techniques, such as genetic algorithms, have
also been utilized for quick offloading decisions, despite potentially suboptimal outcomes. Genetic algorithms
have shown effectiveness in heterogeneous systems, as demonstrated by Neema et al. [16], who proposed
genetic techniques for dynamic scheduling and load balancing in cloud environments, and Li et al. [17], who
utilized genetic algorithms to address task offloading proportion, channel bandwidth, and computing resources
of mobile edge servers. While these studies provide valuable insights, a comprehensive approach is needed that
considers system constraints, user preferences, and dynamic MEC conditions.

In this research, we propose a genetic algorithm-based approach for offloading in MEC, leveraging a
dynamic genetic algorithm and a greedy strategy for performance maximization. The adaptability of genetic
algorithms makes them ideal for the dynamic environment of MEC, offering near-optimal solutions without
complex models [18]. Our research methodology introduces the computation offloading challenges in MEC
networks, identifies gaps in task-offloading approaches through a literature review, and acknowledges the suc-
cess of genetic algorithms in MEC. We detail the MEC system modeling, integrating task offloading compu-
tations and user preferences. We incorporate task and computation models representing the MEC system to
ensure precision. The problem formulation centers on computation offloading optimization, to minimize the
total task service time. We lay the groundwork for our genetic algorithm design, which is explained in the
subsequent section. Our offloading strategy based on genetic algorithms balances exploration and exploitation
for efficient convergence without sacrificing diversity. An experimental methodology outlines metric selection
and test scenarios, configuring simulation parameters. We assess the performance of our computation offload-
ing approach using metrics such as total service time, task failure rate, and computation resource utilization.
Results are discussed and compared with existing approaches for validation. In conclusion, we summarize con-
tributions, exploring future research directions, including genetic algorithm enhancements in real environments
and integration into operational MEC systems.

The remainder of this paper is organized as follows, section 2 provides details of our system model
and problem formulation, section 3 explains our offloading strategy using genetic algorithms, section 4 presents
the performance evaluation of our approach, and discuss the results, and finally, section 5 concludes the paper
and explore future research directions.

A new approach based on genetic algorithm for computation offloading optimization in ... (Marouane Myyara)
4188 ❒ ISSN: 2252-8938

2. SYSTEM MODEL AND PROBLEM FORMULATION

The system model in this paper is based on a mobile/MEC/cloud architecture. When a smart mobile
device (SMD) intends to execute a resource-intensive task, it can offload it to an MEC or cloud server. The MEC
system comprises three levels, as shown in Figure 1. Mobile users as clients utilizing the system’s resources,
MEC nodes/servers near users, and the centralized cloud server with more substantial resources. MEC servers
connect to the orchestrator, forming a hierarchical network. When an SMD offloads a task, the MEC server
covering its location attempts to execute it. If resources are insufficient, the orchestrator delegates the remaining
tasks to other servers. Each computation task is defined by length, deadline, and resource consumption. Each
MEC node has virtual machines (VMs) with maximum computing capacity, including CPU, memory, and
storage, allocated for task execution. Two execution modes exist for a computation task: local and remote.
In the local mode, the mobile device performs the task computation, generating the result. This method is
simple and efficient, requiring no data transmission to a remote server. In the remote mode, the SMD sends
an offloading request to the orchestrator via the MEC node. The MEC server allocates necessary computing
resources based on requirements, executes the task, and sends the result back to the SMD.
As illustrated in the MEC system framework in Figure 1, let ES = {S1 , S2 , . . . , Sp , Sp+1 } be the set
of MEC and cloud servers, where p is the total number of servers in the MEC architecture.
Let VMSi = {vmi,1 , vmi,2 , . . . , vmi,n } be the set of VMs, where n is the total number of VMs in each MEC
server Si . To distinguish between VMs in MEC and the cloud server, let VMC = {vmC,1 , vmC,2 , . . . , vmC,m }
be the set of VMs in the cloud server Sp+1 , where m is the total number of VMs.

Figure 1. MEC system model with mobile/MEC/cloud architecture

2.1. Task model

Unlike other works [19], [20], each SMD in this study isn’t confined to a single task but rather has a set
of tasks. Equipped with a set of applications, each SMD’s tasks are denoted by ASMD t
i i
= {τt,1 i
, τt,2 i
, . . . , τt,j }.
Here, SMDi is the smart mobile device i, t is the application type, and j is the total tasks in application t.
For simplicity, each task is an atomic unit of input data, not fragmented. Each task has three parameters:
task length Lτt,j
i represents the task data amount transferred to the MEC server or the cloud, indicated by
instructions. Delay sensitivity DτSi specifies task sensitivity to delay, close to 1 indicates high sensitivity.
t,j
Maximum delay demand Dτmax i indicates the maximum acceptable delay for chosen server processing.
t,j

2.2. Computation model

The task computation model incorporates considerations for both local and remote execution times.
i
Specifically, the local execution time of a task represents the time required for the execution of task τt,j on the
SMDi . This parameter captures the duration needed for the local processing of the task on the specific SMDi :
Lτt,j
i
SMDi
TS,τ i = (1)
t,j FSMDi
i
Here, Lτt,j
i represents the amount of data for task τt,j , and FSMDi is the processing speed of device SMDi
Off,S
in million instructions per second (MIPS). The remote execution time of a task, denoted TS,τ i p , is the time
t,j
required to execute the task on a MEC or cloud server. It is given by (2):
Off,Sp Trans,Sp Process,Sp Up,Sp Dw,Sp Process,Sp
TS,τ i = Tτ i + Tτ i = Tτ i + Tτ i + Tτ i (2)
t,j t,j t,j t,j t,j t,j

Int J Artif Intell, Vol. 13, No. 4, December 2024: 4186–4194

Int J Artif Intell ISSN: 2252-8938 ❒ 4189
Trans,Sp i
Here, Tτ i denotes the transmission time of task τt,j to server Sp , representing the time required to transfer
t,j
task data from the mobile device to the MEC server or the cloud. The upload and download data transfer times
Up,S Dw,S
are denoted by Tτ i p and Tτ i p for the respective transfers of task input data from the mobile device to
t,j t,j
the target server (Sp ) and output data from the computing server to the source mobile device. These delays are
calculated using a network model integrated into the EdgeCloudSim simulator [21]. The simulator dynamically
evaluates data transmission delays, considering the locations and number of mobile devices, traffic, and network
Process,Sp
types. The processing time, denoted Tτ i , is calculated using (3):
t,j

Process,Sp
Lτt,j
i
Tτ i = (3)
t,j FSp ,n
Here, the processing time represents the time required to execute the task on the MEC or cloud server, and FSp ,n
represents the processing speed expressed in MIPS. This parameter characterizes the computational capability
of the n-th VM on the server Sp , providing a measure of its efficiency in processing instructions.
2.3. Problem formulation
In this section, the computation offloading problem in the MEC environment is formulated. The
objective is to assign a set of tasks to MEC or cloud servers to meet the QoS requirements of mobile applications
and the constraints of computing resources while minimizing the total processing and transmission time of these
i
tasks. The computation offloading decision for a task τt,j is represented by the binary variable ξi . If ξi = 0,
the task is executed locally, and if ξi = 1, it is offloaded to a MEC or cloud server. The problem can be
mathematically formulated as follows:
NT
Off,Sp
X
SMDi
minimize T= (1 − ξi ) × TS,τ i + ξi × TS,τ i (P 1)
t,j t,j
i=1

subjectto :
i
C1 : ξi ∈ {0, 1}; ∀τt,j ∈ ASMD
t
i
, ∀i ∈ {1, . . . , NT }
 
SMDi Off,S
C2 : max TS,τ i , TS,τ i
p
≤ DτSi ; ∀i ∈ {1, . . . , NT }, ∀Sp ∈ ES
t,j t,j t,j
vmi,n
CReq,τt,j
i CT arget vm
i,n
C3 : vmi,n ≤1 & vmi,n > 1 & CT arget ≤ T hSp ; ∀vmi,n ∈ VMSi , ∀Sp ∈ ES
CT arget CSelect
The problem formulation (P1) is designed to minimize the total service time of tasks, considering both
local and remote execution times weighted by the parameter ξi . The binary variable ξi decides whether task
i
τt,j is offloaded (ξi = 1) or processed locally (ξi = 0). Constraints (C1 -C3 ) define the conditions and limits
of the problem. C1 ensures that the binary variable ξi is within the interval (0,1), indicating whether the task is
offloaded or processed locally. C2 guarantees that the processing and transmission time of each task does not
exceed its latency sensitivity DτSi . Constraint C3 encompasses conditions on computing resources, ensuring
t,j
that the demand and capacity of virtual resources meet predefined criteria. It concurrently regulates the load of
VMson MEC servers, preventing it from exceeding the limit defined by T hSp .

3. COMPUTATION OFFLOADING OPTIMISATION STRATEGY BASED ON GENETIC ALGO-

RITHMS
The genetic algorithm is an evolutionary approach that simulates the process of natural evolution [22].
Applied to solve complex problems [23], genetic algorithms find extensive use in addressing NP-complete
problems, including load balancing in cloud computing [24]. Their implementation involves a population of
solutions, where each solution is represented by chromosomes. These algorithms utilize operators inspired by
biological processes, facilitating convergence towards optimal solutions while avoiding local optima. Genetic
algorithms efficiently find solutions close to optimal within a reasonable timeframe, making them a promising
approach for addressing the computation offloading problem in dynamic MEC environments. Given the chal-
lenge of solving (P1) using traditional methods, genetic algorithms are employed, and the steps are outlined as
follows:

A new approach based on genetic algorithm for computation offloading optimization in ... (Marouane Myyara)
4190 ❒ ISSN: 2252-8938

– Step 1: Encoding and generation. This involves creating an initial population of potential solutions
represented by chromosomes. Each chromosome corresponds to a computation offloading scheme for an
application task, with genes representing possible execution locations.
– Step 2: Evaluation. To obtain an optimal assignment of tasks to computing servers, an adaptation function
measures the optimality of offloading decisions. After evaluating the chromosomes, they are ranked
based on their fitness, where the fitness of a chromosome x is calculated as in (4):
1
f (x) = SMDi Off,Sp
(4)
(1 − ξi ) × TS,τ i + ξi × TS,τ i
t,j t,j

This fitness function represents the inverse of the total service time of tasks, considering both local and
remote execution times weighted by the binary variable ξi . The chromosomes are then ranked based on
their fitness values, with higher values indicating better solutions.
– Step 3: Selection operation. The tournament selection method is used to choose the fittest chromosomes,
which are retained for reproduction, while less performing ones are eliminated.
– Step 4: Crossover operation. The crossover operator combines the features of two parents to create new
offspring, gradually improving the quality of solutions.
– Step 5: Mutation operation. Mutation modifies one or more genes of a chromosome, preventing prema-
ture convergence to a local optimum. Single-point mutation, altering a random gene, is used.
– Step 6: Replacement. The new offspring solutions replace the worst solutions in the current population,
maintaining the population’s quality and avoiding premature convergence.
– Step 7: Termination and iteration. Steps 2 through 6 are iteratively executed for a designated number
of generations. The algorithm terminates upon reaching the maximum specified iterations or when the
overall fitness function converges to its minimum value.
The resolution method based on these steps is described in Algorithm 1. The proposed genetic algo-
rithm optimizes computation offloading in MEC through an evolutionary approach. In the initialization step,
a population is created with chromosomes representing distinct computation offloading schemes onto MEC
and cloud servers. The algorithm iterates until convergence or a preset limit. Fitness values, based on the
inverse of total service time, guide chromosome selection, crossover, and mutation. A combined population
of parents and offspring is formed, and the top-performing chromosomes advance to the next iteration. The
algorithm concludes by selecting the optimal solution from the final population, offering a versatile framework
for efficient computation offloading optimization in dynamic MEC environments.

Algorithm 1 Genetic algorithm for computation offloading optimization

Require: τt,ji : Incoming Task, E : Set of MEC and Cloud Servers
S
1: Initialization
i ,E )
2: P ← Generate an initial population of feasible solutions(τt,j S
3: repeat
4: Evaluation
5: for all x ∈ P do
6: f (x) ← Calculate fitness function as in (4)
7: end for
8: Psorted ← Sort chromosomes based on their fitness
9: Pselected ← Select top-performing chromosomes from Psorted
10: while the number of generations is not exhausted do
11: Crossover
12: Poffspring ← Generate offspring by applying crossover to Pselected
13: Mutation
14: Poffspring ← Apply mutation on Poffspring
15: Evaluation
16: for all x ∈ Poffspring do
17: f (x) ← Calculate fitness function(x)
18: end for
19: Pcombined ← Pselected ∪ Poffspring
20: Pselected ← Select top-performing chromosomes from Pcombined
21: end while
22: until Convergence
Ensure: x : The best chromosome from the final population(Pselected )

Int J Artif Intell, Vol. 13, No. 4, December 2024: 4186–4194

Int J Artif Intell ISSN: 2252-8938 ❒ 4191

4. PERFORMANCE EVALUATION AND SIMULATION RESULTS

This section is dedicated to the thorough evaluation of the genetic algorithm for computation offload-
ing optimization (GA-COO) algorithm’s performance through simulations. We comprehensively consider a
range of parameters, and we benchmark our results against various offloading schemes and algorithms within
the MEC context. The objective is to provide a detailed analysis of how the proposed GA-COO algorithm
compares and performs under diverse scenarios, shedding light on its effectiveness in addressing computation
offloading challenges in MEC networks.

4.1. Experimental evaluation

The GA-COO algorithm’s experimental assessment aims to minimize task response time and failure
rates, considering mobile application constraints and MEC/cloud server utilization. Metrics analyzed include
task response time (processing and transmission time), computational task failure rates, and resource utilization.
Results are compared with four algorithms: i) DCOA-ST [25]: a greedy approach minimizing task service time
while adhering to sensitivity constraints, ii) Random [26]: random task assignment to available VM, serving as
a reference, iii) Utilization-based [27]: allocation based on real-time VM utilization with predefined thresholds,
iv) Network-based [27]: allocation considering network characteristics, including bandwidth, and v) Hybrid
[27]: an approach combining real-time VM utilization and network characteristics.

4.2. Simulation environment

To assess GA-COO, EdgeCloudSim [21] was chosen for its realistic simulation of decentralized envi-
ronments, precise computing architecture modeling, and accurate workload generation. Task parameters, gen-
erated randomly using EdgeCloudSim’s workload generator, defined simulated applications (e.g., augmented
reality, health, infotainment, and heavy computation) with specific characteristics. These applications varied
in upload/download sizes, task lengths, and VM utilization. Delay sensitivity ranged from 0.9% to 0.1%, and
maximum delay requirements varied from 0.5 seconds to 2 seconds. The MEC architecture included 14 MEC
servers, 1 cloud server, and a variable number of SMDs. Each MEC server had 8 VMs per host, the cloud server
had 4 VMs per host, and SMDs had 1 core per VM. VM CPU speed was specified in GIPS, with values of 10
for MEC servers, 100 for the cloud server, and 4 for SMDs. Experiments were conducted using EdgeCloudSim
in JAVA on a computer with 4 Intel Core i7-9600U processors at 2.59 GHz and 8 GB of RAM.

4.3. Simulation results

With the specified parameters, we evaluate our genetic algorithm’s performance. Figure 2 displays
computation task service times relative to the number of SMDs, showcasing GA-COO’s efficiency. Results
reveal a linear increase in service time with growing SMD numbers, highlighting challenges associated with
limited shared computing resources. GA-COO maintains competitive service times, ranging from 1.1 seconds
for 100 SMDs to 1.95 seconds for 2300 SMDs. In comparison, the DCOA-ST algorithm exhibits notable
performance, ranging from 0.8 seconds to 2.5 seconds for the same SMD values.
Simulation results of failure rates in Figure 3 provide insights into the robustness of different ap-
proaches concerning SMD numbers. GA-COO maintains a low and stable failure rate up to 2100 SMDs, then
experiences a notable increase to 12% for 2300 SMDs. DCOA-ST maintains a low failure rate, slightly in-
creased to 20% for 2300 SMDs. In contrast, Random exhibits a rapid growth in the failure rate, reaching 61%
for 2300 SMDs. Hybrid shows a progressive increase to 28%; Network-based maintains stability up to 2100
SMDs, then increases to 18% for 2300 SMDs. Utilization-based demonstrates a significant increase to 46% for
2300 SMDs.
Analyzing Figure 4 reveals trends in average VM utilization for MEC servers. GA-COO maintains
stability at 25% for 1900 SMDs, persisting at 24% for 2100 and 2300 SMDs. DCOA-ST’s performance remains
consistent, slightly increasing to 19% for 2300 SMDs. In contrast, Random experiences rapid growth, reaching
61% for 2300 SMDs. Hybrid progresses to 69.5%, indicating sensitivity to high loads. Network-based stays
stable up to 1900 SMDs, then increases to 44% for 2300 SMDs. Utilization-based increases to 59% for 1900
SMDs, slightly decreasing to 49% for 2300 SMDs.
In terms of cloud VM utilization, depicted in Figure 5, GA-COO maintains gradual usage, ranging
from 0.1% to 3.9% for 100 to 2300 SMDs. DCOA-ST increases from 0.1% to 8.1% for 100 to 2300 SMDs.
Random shows a slight increase, from 0.3% to 1% for 100 to 2300 SMDs. Hybrid maintains minimal usage,
from 0.01% to 1.25% for 100 to 2300 SMDs. Network-based displays lower usage, from 0.3% to 2.1% for 100
to 2300 SMDs.

A new approach based on genetic algorithm for computation offloading optimization in ... (Marouane Myyara)
4192 ❒ ISSN: 2252-8938

4.5 70

UTILIZATION BASED UTILIZATION BASED

NETWORK BASED NETWORK BASED
4 60 HYBRID
HYBRID
RANDOM RANDOM
DCOA−ST DCOA−ST
3.5 GA-COO
GA-COO 50

3
Service Time (sec)

Failed Tasks (%)

40

2.5

30
2

20
1.5

10
1

0.5 0
100 300 500 700 900 1100 1300 1500 1700 1900 2100 2300 100 300 500 700 900 1100 1300 1500 1700 1900 2100 2300
Number of SMD Number of SMD

Figure 2. Comparison of average service times for Figure 3. Comparison of failure rates among different
different offloading approaches with various SMDs offloading approaches for various SMDs

70 9

UTILIZATION BASED
NETWORK BASED 8
60 HYBRID
RANDOM
7
Average VM Utilization of MEC server (%)

DCOA−ST
Average VM Utilization of Cloud (%)

50 GA-COO
6

UTILIZATION BASED
40 5 NETWORK BASED
HYBRID
RANDOM
30 4 DCOA−ST
A−S
GA-COO
3
20
2

10
1

0 0
100 300 500 700 900 1100 1300 1500 1700 1900 2100 2300 100 300 500 700 900 1100 1300 1500 1700 1900 2100 2300
Number of SMD Number of SMD

Figure 4. Average VM MEC servers utilization as a Figure 5. Average VM cloud utilization as a function
function of the number of SMDs of the number of SMDs

The simulation results confirm the stability and effectiveness of the GA-COO approach for compu-
tation offloading in MEC networks. GA-COO maintains consistent performance under increasing workloads,
with a linear increase in task service times and a low, stable task failure rate even under high loads. The
utilization of VMs in MEC servers remains stable, indicating efficient resource management in the presence
of significant workloads. In response to the identified challenges, GA-COO demonstrates competitive per-
formance with short task service times, low failure rates, and stable resource utilization, indicating efficient
resource management, particularly under heavy workloads.

5. CONCLUSION
The proposed GA-COO employs a genetic algorithm to optimize computation offloading in MEC
networks. Comparative evaluations, involving various reference offloading schemes, unequivocally establish
GA-COO’s effectiveness in minimizing task service times and failure rates, all while considering the constraints

Int J Artif Intell, Vol. 13, No. 4, December 2024: 4186–4194

Int J Artif Intell ISSN: 2252-8938 ❒ 4193

of mobile applications and computational resource utilization. GA-COO consistently outperforms traditional
methods and certain reference algorithms, especially in scenarios with a substantial number of SMD. This
robust performance underscores the significance of tailored offloading strategies that account for application
specifics and system constraints. This work not only introduces a novel computation offloading approach in
MEC but also aligns with broader goals of optimizing edge computing systems for diverse applications in an
increasingly interconnected and data-intensive landscape. Future research prospects involve not only enhancing
GA-COO’s performance in real-world environments but also considering specific factors such as quality of
experience and mobility. Furthermore, crucial investigations into GA-COO’s impact on other aspects of MEC
networks, including energy consumption and security, are imperative for a comprehensive understanding of its
implications.

REFERENCES
[1] N. E. Nwogbaga, R. Latip, L. S. Affendey, and A. R. A. Rahiman, ”Attribute reduction based scheduling algorithm with enhanced
hybrid genetic algorithm and particle swarm optimization for optimal device selection,” in Journal of Cloud Computing, vol. 11, no.
15, 2022, doi: 10.1186/s13677-022-00288-4.
[2] Q. Zhang, L. Gui, S. Zhu, and X. Lang, ”Task offloading and resource scheduling in hybrid edge-cloud networks,” IEEE Access,
vol. 9, pp. 85350-85366, 2021, doi: 10.1109/ACCESS.2021.3088124.
[3] J. Fang, J. Shi, S. Lu, M. Zhang, and Z. Ye, ”An efficient computation offloading strategy with mobile edge computing for IoT,”
Micromachines, vol. 12, no. 2, 2021, doi: 10.3390/mi12020204.
[4] J. Zhang, W. Zhu, X. Wu, and T. Ma, ”Traffic state detection based on multidimensional data fusion system of internet of things,”
Wireless Communications and Mobile Computing, vol. 2021, pp. 1-12, 2021, doi: 10.1155/2021/1374186.
[5] L. Long, Z. Liu, Y. Zhou, L. Liu, J. Shi, and Q. Sun, ”Delay optimized computation offloading and resource allocation for mobile
edge computing,” in 2019 IEEE 90th Vehicular Technology Conference (VTC2019-Fall), Sep. 2019, pp. 1-5, doi: 10.1109/VTC-
Fall.2019.8891166.
[6] F. Hoseiny, S. Azizi, M. Shojafar, and R. Tafazolli, ”Joint QoS-aware and cost-efficient task scheduling for fog-cloud resources in a
volunteer computing system,” ACM Transactions on Internet Technology (TOIT), vol. 21, no. 4, pp. 1–21, 2021, ACM New York,
doi:10.1145/3418501.
[7] C. Sonmez, A. Ozgovde, and C. Ersoy, ”Fuzzy workload orchestration for edge computing,” IEEE Transactions on Network and
Service Management, vol. 16, no. 2, pp. 769-782, 2019, doi: 10.1109/TNSM.2019.2901346.
[8] U. Saleem, Y. Liu, S. Jangsher, Y. Li, and T. Jiang, ”Mobility-aware joint task scheduling and resource allocation for co-
operative mobile edge computing,” IEEE Transactions on Wireless Communications, vol. 20, no. 1, pp. 360-374, 2021, doi:
10.1109/TWC.2020.3024538.
[9] C. Li, M. Song, M. Zhang, and Y. Luo, ”Effective replica management for improving reliability and availability in edge-cloud com-
puting environment,” Journal of Parallel and Distributed Computing, vol. 143, pp. 107-128, 2020, doi: 10.1016/j.jpdc.2020.04.012.
[10] D. Lim and I. A. Joe, ”DRL-based task offloading scheme for server decision-making in multi-access edge computing,” Electronics,
vol. 12, 2023, doi: 10.3390/electronics12183882.
[11] A. Samanta and Z. Chang, ”Adaptive service offloading for revenue maximization in mobile edge computing with delay-constraint,”
IEEE Internet of Things Journal, vol. 6, no. 2, pp. 3864-3872, 2019, IEEE, doi: 10.1109/JIOT.2019.2892398.
[12] B. Wang, Y. Song, J. Cao, X. Cui, and L. Zhang, ”Improving task scheduling with parallelism awareness in heterogeneous compu-
tational environments,” Future Generation Computer Systems, vol. 94, pp. 419-429, 2019, doi: 10.1016/j.future.2018.11.012.
[13] J. Silva, E. R. Marques, L. M. Lopes, and F. Silva, ”Jay: adaptive computation offloading for hybrid cloud environ-
ments,” in 2020 Fifth International Conference on Fog and Mobile Edge Computing (FMEC), pp. 54-61, 2020, doi:
10.1109/FMEC49853.2020.9144950.
[14] E. F. Maleki and L. Mashayekhy, ”Mobility-aware computation offloading in edge computing using prediction,” in 2020 IEEE 4th
International Conference on Fog and Edge Computing (ICFEC), pp. 69-74, 2020, doi: 10.1109/ICFEC50348.2020.00015.
[15] M. D. Hossain, L. N. T. Huynh, T. Sultana, T. D. T. Nguyen, J. H. Park, C. S. Hong, and E. N. Huh, ”Collaborative task offloading for
overloaded mobile edge computing in small-cell networks,” in 2020 International Conference on Information Networking (ICOIN),
pp. 717-722, 2020, doi: 10.1109/ICOIN48656.2020.9016452.
[16] G. Neema, A. B. Kadan, and V. P. Vijayan, ”Multi-objective load balancing in cloud infrastructure through fuzzy based decision
making and genetic algorithm based optimization,” IAES International Journal of Artificial Intelligence (IJ-AI), vol. 12, no. 2, pp.
678-685, 2023, doi: 10.11591/ijai.v12.i2.pp678-685.
[17] Z. Li and Q. Zhu, ”Genetic algorithm-based optimization of offloading and resource allocation in mobile-edge computing,” Infor-
mation, vol. 11, no. 2, 2020, doi: 10.3390/info11020083.
[18] K. Wang, X. Wang, and X. Liu, ”Sustainable internet of vehicles system: A task offloading strategy based on improved genetic
algorithm,” Sustainability, vol. 15, no. 9, 2023, doi: 10.3390/su15097506.
[19] H. Gu, M. Zhang, W. Li, and Y. Pan, ”Task offloading and resource allocation based on DL-GA in mobile edge computing,” Turkish
Journal of Electrical Engineering and Computer Sciences, vol. 31, no. 3, pp. 498-515, 2023, doi: 10.55730/1300-0632.3998
[20] M. Zhao and K. Zhou, ”Selective offloading by exploiting ARIMA-BP for energy optimization in mobile edge computing networks,”
Algorithms, vol. 12, no. 2, 2019, doi: 10.3390/a12020048.
[21] C. Sonmez, A. Ozgovde, and C. Ersoy, ”Edgecloudsim: An environment for performance evaluation of edge computing systems,”
Transactions on Emerging Telecommunications Technologies, vol. 29, no. 11, 2018, doi: 10.1002/ett.3493.
[22] A. Ulah, N. M. Nawi, J. Uddin, and S. Baseer, ”Artificial bee colony algorithm used for load balancing in cloud computing: Review,”
IAES International Journal of Artificial Intelligence, vol. 8, no. 2, pp. 156-167, 2019, doi: 10.11591/ijai.v8.i2.pp156-167.
[23] B. Keshanchi, A. Souri, and N. J. Navimipour, ”An improved genetic algorithm for task scheduling in the cloud environments using

A new approach based on genetic algorithm for computation offloading optimization in ... (Marouane Myyara)
4194 ❒ ISSN: 2252-8938

the priority queues: formal verification, simulation, and statistical testing,” Journal of Systems and Software, vol. 124, pp. 1-21,
2017, doi: 10.1016/j.jss.2016.07.006.
[24] Z. Zhou, F. Li, H. Zhu, H. Xie, J. H. Abawajy, and M. U. Chowdhury, ”An improved genetic algorithm using greedy strategy
toward task scheduling optimization in cloud environments,” Neural Computing and Applications, vol. 32, pp. 1531-1541, 2020,
doi: 10.1007/s00521-019-04119-7.
[25] M. Myyara, A. Darif, and A. Farchane, ”Task deadline-based computation offloading algorithm for service time minimization in
mobile edge computing,” in International Conference on Artificial Intelligence and Green Computing, pp. 149-162, 2023, doi:
10.1007/978-3-031-46584-0 12.
[26] S. Fu, C. Ding, and P. Jiang, ”Computational offloading of service workflow in mobile edge computing,” Information, vol. 13, no.
7, 2022, doi: 10.3390/info13070348.
[27] V. Nguyen, T. T. Khanh, T. D. Nguyen, C. S. Hong, and E. N. Huh, ”Flexible computation offloading in a fuzzy-based mobile edge
orchestrator for IoT applications,” Journal of Cloud Computing, vol. 9, no. 1, pp. 1-18, 2020, doi: 10.1186/s13677-020-00211-9.

BIOGRAPHIES OF AUTHORS

Marouane Myyara received his B.Sc. in Electronic and Telecommunication Engineering

in 2019 and M.Sc. in Telecommunication Systems and Computer Networks in 2021 from Sultan
Moulay Slimane University, Beni Mellal, Morocco. He is currently a Ph.D. candidate at the Labora-
tory of Innovation in Mathematics, Applications, and Information Technology (LIMATI), Faculty of
Polydisciplinary, Sultan Moulay Slimane University, Morocco. His current research focuses on im-
proving the performance of multi-access edge computing networks, cloud computing, computation
offloading, and internet of things. He can be contacted at email: marouane.myyara@usms.ac.ma.

Oussama Lagnfdi received his B.Sc. in Physical Matter Science in 2020 and M.Sc. in
Telecommunications Systems and Computer Networks in 2022 from Sultan Moulay Slimane Uni-
versity, Beni Mellal, Morocco. Currently, he is a Ph.D. candidate at the Laboratoire d’Innovation
en Mathématiques et Applications et Technologies de l’Information (LIMATI), Faculty of Polydisci-
plinary, Sultan Moulay Slimane University, Morocco. His ongoing research is focused on enhancing
the performance of internet of things and mobile edge computing, artificial intelligence, deep learn-
ing, and fuzzy logic. He can be contacted at email: lagnfdi.o@gmail.com.

Anouar Darif received the bachelor in Informatique Electrothecnique, Electronique and

Automatique (IEEA) from Dhar El Mahraz, Faculty of Sciences at Mohamed Ben Abdellah Univer-
sity Fez, Morocco in 2005. He received the Diplôme d’Etudes Supérieurs Approfondies in Computer
Sciences and Telecommunications from Faculty of Sciences Rabat in 2007. He received the Ph.D.
degree in Computer Sciences and Telecommunications from Faculty of Sciences of Rabat in 2015.
He is currently a Research and a Teaching Associate in Faculty of Multidisciplinary at University of
Sultan Moulay Slimane Beni Mellal, Morocco. His research interests include wireless sensor net-
work (WSN), mobile edge computing (MEC), internet of things (IoT), cloud computing, and neural
networks. He can be contacted at email: anouar.darif@gmail.com.

Abderrazak Farchane received his B.Sc. in Computer Science and Engineering in June
2001 and M.Sc. in Computer Science and Telecommunication from the University of Mohammed V
Agdal, Rabat, Morocco, in 2003. He obtained his Ph.D. in Computer Science and Engineering at EN-
SIAS, Rabat, Morocco. He is currently an Associate Professor of Computer Science in the Faculty of
Polydisciplinary at Sultan Moulay Slimane University, Morocco. His areas of interest are information
coding theory, cryptography, and security. He can be contacted at email: a.farchane@gmail.com.

Int J Artif Intell, Vol. 13, No. 4, December 2024: 4186–4194