(IJACSA) International Journal of Advanced Computer Science and Applications,

Vol. 15, No. 7, 2024

Metaheuristic Optimization for Dynamic Task

Scheduling in Cloud Computing Environments
Longyang Du*, Qingxuan Wang
School of Artificial Intelligence, Jiaozuo University, Jiaozuo 454000, Henan, China

Abstract—Cloud computing enables the sharing of resources B. Problem Statement

across the Internet in a highly adaptable and quantifiable way. Cloud computing is a paradigm that enables universal,
This technology allows users to access customizable distributed
flexible, and immediate access to various configurable
resources and offers various services for resource allocation,
scientific operations, and service computing via virtualization.
computing resources in the form of services, applications,
Effectively allocating tasks to available resources is essential to storage, servers, and networks, easily delivered and released
providing reliable consumer performance. Task scheduling in without much service provider interaction or management effort
cloud computing models presents substantial challenges as it [7]. It serves as a solution with several advantages to overcome
necessitates an efficient scheduler to map multiple tasks from economic and technological challenges. The cloud computing
numerous sources and dynamically distribute resources to users model offers lower total costs and allows companies to
based on their requirements. This study presents a metaheuristic concentrate on their primary tasks and functions without
optimization methodology that integrates load balancing by concerning themselves with infrastructure issues or the
dynamically distributing tasks across available resources based on availability and flexibility of resources [8].
current load conditions. This ensures an even distribution of
workloads, preventing resource bottlenecks and enhancing overall Furthermore, the amalgamation of cloud services, including
system performance. The suggested method is suitable for both computation, infrastructure, and storage, into the utility model
constant and variable activities. Our technique was compared of cloud computing presents an exceptionally appealing
with established metaheuristic methods, including HDD-PLB, environment for scientists to conduct their experiments [9].
HG-GSA, and CAAH. The proposed method demonstrated Cloud computing provides various service models tailored to
superior performance due to its adaptive load balancing meet distinct customer requirements. Cloud service models can
mechanism and efficient resource utilization, reducing task be classified as Platform as a Service (PaaS), Software as a
completion times and improving overall system throughput. Service (SaaS), or Infrastructure as a Service (IaaS) [10]. IaaS
offers virtual computing resources over the Internet. It allows
Keywords—Dynamic task scheduling; cloud computing; users to manage and operate applications without needing to
metaheuristic optimization; load balancing; task allocation; handle physical hardware complexities by combining virtual
resource utilization machines, storage, and networks [11]. PaaS allows customers to
develop, run, and manage applications independently of the
I. INTRODUCTION
underlying infrastructure [12]. It includes development
A. Context frameworks, databases, and tools. SaaS provides subscription-
Cloud computing is a rapidly evolving technology, marking based access to software applications over the Internet [13].
its place as the next generation in IT and business landscapes C. Motivation
[1]. It offers a spectrum of services, including reliable software
In recent years, the issue of task scheduling within a
and hardware, accessible through the Internet and remote data
distributed environment has become a focal point for
centres [2]. With its architecture, cloud services efficiently
researchers. Task scheduling is regarded as a critical concern in
manage diverse computing tasks on a large scale, covering
the cloud computing domain, taking into account various factors
multiple IT functions such as storage, computation, database,
such as completion time, overall cost of executing users' tasks,
and application services [3]. The increasing demand for storage,
resource utilization, power consumption, and fault tolerance
processing, and analysis of extensive datasets has propelled
[14]. The challenge arises in attaining the optimal equilibrium
organizations and individuals to embrace cloud computing [4].
between the time required to complete a task and the amount of
Scientific applications, notably those requiring significant
energy consumed for a parallel application bound by
computational resources for extensive experiments, have found
precedence, resulting in a problem of bi-objective optimization.
refuge in cloud deployments due to limitations in local server
The resolution to this problem yields a collection of Pareto
facilities [5]. Reduced capital costs, immense data generation,
points, where Pareto solutions indicate that enhancing one target
and consumption growth from these experiments have driven
requires making concessions in at least one other objective.
this shift. Moreover, cloud service providers are now
Therefore, the resolution to a bi-objective issue comprises a
incorporating data parallelism capabilities into their offerings,
collection of Pareto points rather than a single answer.
empowering users to leverage cloud resources and execute their
workflows more effectively [6].

590 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 15, No. 7, 2024

Task scheduling in cloud computing environments is methodologies have been explored for optimizing the allocation
commonly known as an NP-complete and multi-objective of tasks to virtual machines, enhancing system efficiency,
optimization issue [15]. It deals with the allocation of user- reducing execution times, and maximizing resource utilization.
defined tasks on the existing cloud virtual machines. The main Table I compares various cloud computing task scheduling
goal of any task scheduling strategy is to minimize total approaches.
execution time. An effective solution to this challenge may be
achieved by integrating multiple approaches to enhance task Yang, et al. [16] proposed a task scheduling algorithm
execution and optimize the utilization of resources. This can be derived from game theory in their research. This paper presents
achieved by optimizing task placement, task scheduling, and three significant contributions tailored to the features of cloud
task execution. Additionally, task scheduling algorithms should computing. Primarily, leveraging game theory enhances the
be adaptive and capable of continuously optimizing their coordination between task distribution and energy allocation.
operations in response to changing workloads and resource Secondly, the paper offers a task-scheduling framework to
availability. handle big data through a mathematical formulation.
Verification by experiment in this research attests to both stable
D. Contribution states and optimal computational efficiency.
The current investigation centers on implementing and Chaudhary and Kumar [17] proposed a novel load
comparing metaheuristic optimization techniques for task scheduling technique named Hybrid Genetic-Gravitational
scheduling. We compare such methods with conventional Search Algorithm (HG-GSA) with the aim of reducing the
heuristics, addressing the problem of scheduling static tasks overall computational burden, encompassing both execution and
independently in cloud infrastructure contexts. Experiments are transfer costs. HG-GSA employs a hybrid crossover mechanism
conducted in both uniform and uneven environments. In the to explore the optimal arrangement of particles in the search
uniform scenario, virtual machine characteristics remain space. The calculated force is then utilized to determine an
constant, whereas the asymmetric environment involves a optimal particle position. The performance of HG-GSA is
random selection of virtual machines based on diverse features evaluated against alternative methods using the CloudSim
like MIPS, Bandwidth, and RAM. Despite the simplicity of the simulator. Through convergence analysis and quantitative
symmetric scheduling approach, it fails to fully exploit the assessments, the proposed HG-GSA approach significantly
potential offered by the asymmetric characteristics of virtual reduces the total computation cost over existing algorithms such
machines. Section I and Section II provide an overview of as PSO, Cloudy-GSA, and LIGSA-C.
various conventional metaheuristic task scheduling approaches
along with their inherent limitations. Section III gives a Imene, et al. [18] applied the Non-dominated Sorting
comprehensive description of the proposed optimization Genetic Algorithm (NSGA-III), a third-generation multi-
strategy. Section IV describes the simulation setup and outlines objective optimization strategy, for scheduling cloud computing
diverse experiments conducted, all grounded in the proposed tasks. They introduced an innovative multi-objective adaptation
technique. Finally, Section V articulates the paper's conclusions process designed to optimize three crucial factors: cost, power
and suggests potential avenues for future research enhancements consumption, and runtime. Further, the study conducted a
applicable to the proposed optimization technique. comparative analysis between NSGA-III and its precursor,
NSGA-II, revealing that NSGA-III outperformed NSGA-II.
II. RELATED WORK
This section discusses existing research efforts addressing
task scheduling challenges in cloud computing contexts. Several
TABLE I. AN OVERVIEW OF THE RECENT CLOUD TASK SCHEDULING APPROACHES

References Algorithm Contributions Evaluation metrics

Mathematical model for big data task scheduling Equilibrium states and computational
[16] Game theory-based task scheduling
and experimental verification efficiency
Convergence analysis, statistical
[17] Hybrid genetic-gravitational search algorithm Novel hybrid crossover mechanism assessments, and computation cost
reduction
[18] Non-dominated sorting genetic algorithm Novel multi-objective adaptation function Runtime, power consumption, and cost
Hybrid deadline-constrained, dynamic VM Hybridization of heuristic techniques with
[19] Makespan, cost, and VM utilization
provisioning and load balancing metaheuristic
Context-aware adaptive heuristic-based Context-aware adaptive heuristic-based solution Performance efficiency and energy
[20]
mechanism and significant performance improvements savings
Task completion time, execution cost, and
[21] Adaptive ant colony optimization algorithm Pheromone adaptive update mechanism
balance degree
Moth search algorithm with differential
[22] Strong exploration and exploitation Makespan
evolution
Integration of chemical reaction optimization and Execution time, makespan, cost, and
[23] Chemical reaction partial swarm optimization
partial swarm optimization energy
Makespan, SLA violation, and energy
[24] Deep Q-learning network Utilization of deep Q-learning network
consumption

591 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 15, No. 7, 2024

Kaur and Kaur [19] proposed a hybrid delay-constrained heuristic and meta-heuristic algorithms, utilizing both synthetic
dynamic virtual machine provisioning and load balancing and real-world data.
approach called HDD-PLB. The primary goal of HDD-PLB is
to enhance VM utilization by achieving uniform load Dubey and Sharma [23] introduced a pioneering task
distribution. This optimization strategy relies on combining scheduling approach, termed Chemical Reaction Partial Swarm
heuristics with metaheuristics to attain optimum performance, Optimization (CRPSO), to allocate several independent tasks to
focusing on metrics such as cost and makespan. Within the available virtual machines. This innovative method combines
HDD-PLB methodology, two heuristics are proposed: hybrid partial swarm optimization and chemical reaction optimization,
heterogeneous earliest finish time heuristic with Ant Colony amalgamating their features to sequence the optimal task
Optimization (ACO) algorithm and hybrid predicted earliest schedule based on demand and deadlines. The aim is to enhance
finish time heuristic with ACO algorithm. A comprehensive quality across various factors such as cost, energy, and
analysis and comparison of these approaches is conducted to makespan. Their simulation experiments, conducted via the
determine their superiority within the proposed HDD-PLB CloudSim toolkit, confirm the performance of the proposed
model. algorithm. Comparative tests, varying the number of tasks and
virtual machines, demonstrate an average reduction in execution
Kulkarni and Annappa [20] proposed an effective context- time ranging between 1% to 6%, exceeding 10% in certain
aware adaptive heuristic-based (CAAH) methodology tailored scenarios. The makespan results also exhibit an effectiveness
for virtual machine allocation in diverse and heterogeneous enhancement between 5% to 12% and a total cost reduction
cloud data centers. CAAH accounts for both the inherent between 2% to 10%, while the energy consumption rates show
properties of physical machines and the varying load conditions an improvement of 1% to 9%.
(moderate or high) within heterogeneous data centers. The
primary objective is to augment performance efficiency and Mangalampalli, et al. [24] utilized a multi-dimensional deep
facilitate power savings for operators managing data centers. learning algorithm to manage the cloud task scheduling issue,
Through experimental assessments employing both genuine conducting extensive simulations through the Cloudsim toolkit.
The simulations were executed in two phases: first utilizing
cloud workloads and synthetic workloads, noteworthy
enhancements in performance and energy conservation were randomly generated workloads and then incorporating HPC2N
observed with CAAH in comparison to a widely recognized and NASA workloads to assess the efficiency of the suggested
adaptive heuristic-based technique. algorithm. The proposed scheduler was compared against
conventional schedulers like Earliest Deadline First, RR, and
Liu [21] proposed a dynamic task-scheduling technique FCFS.
designed for cloud computing and based on the ACO algorithm.
Their proposed approach enhances the standard ACO by III. PROPOSED METHOD
integrating pheromone adaptive updating to expedite The client provides a set of tasks, aiming to generate an
convergence while effectively circumventing local optima. This optimal task execution plan using a metaheuristic method based
enhanced algorithm generates a distribution scheme that offers on optimization techniques. One vital aspect of any meta-
reduced processing time, minimized costs, and well-balanced heuristic algorithm in achieving an optimal solution is the
task loads based on user-submitted tasks. By conducting selection of a seed arrangement. Arrangements of seeds serve as
experiments on a cloud computing platform, the traditional ACO initial feasible solutions to the problem, aiding optimization
is compared against the enhanced adaptive ACO algorithm. The algorithms in the quest for an optimal solution. These
empirical data illustrates that the improved adaptive ACO arrangements play a critical role in the rapid convergence of any
efficiently identifies optimal solutions for cloud computing optimization-based solution. Researchers have employed
resource scheduling issues, resulting in reduced task completion various strategies to generate seed arrangements, depending on
times, decreased execution costs, and maintaining a balanced the nature of the problem. These strategies encompass selecting
load across the cloud system. a complex arrangement, a logical configuration based on a
Abd Elaziz, et al. [22] introduced an innovative approach to particular problem model, or a heuristic-based arrangement.
solving the cloud task scheduling challenge with a primary focus Each approach has its own advantages and drawbacks.
on minimizing the amount of time needed to schedule diverse However, in many cases, an uneven seed arrangement is utilized
tasks across distinct virtual machines. The proposed to generate a seed arrangement in the absence of a proper
methodology incorporates the Differential Evolution (DE) heuristic.
technique into the Moth Search Algorithm (MSA). The MSA The proposed algorithm aims to minimize execution time
draws inspiration from moth navigation toward a light source, a and meet deadlines while preserving task dependencies across
natural process, leveraging Levy flights and phototaxis to various users. It operates based on a Directed Acyclic Graph
emulate exploitation and exploration capabilities. While the (DAG) representing the task set (T) and task dependencies
MSA exhibits robust exploration abilities, its exploitation facet through edges denoting data transmission time. These
requires enhancement, prompting the integration of DE as a relationships establish entry-exit dependencies between child
local search technique. Three experiments were performed to and parent nodes, forming the basis of the task behavior. The
measure the effectiveness of the newly introduced MSDE proposed algorithm operates as a population-based approach,
algorithm. The initial test compares the performance between aligning with swarm intelligence behavior to provide an
the classic MSA and the modified algorithm across twenty optimized solution to complex data. It functions as a meta-
global optimization problems. In the subsequent two testing heuristic technique in comparison to other algorithms. The
phases, the proposed algorithm was benchmarked with various algorithm encompasses two phases: scheduling jobs with static

592 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 15, No. 7, 2024

constraints and dynamic constraints. These phases aim to handle In Eq. (2), Pi represents the property of the ith job, n defines
task positions in the schedule based on execution time and the number of job properties, m denotes the number of jobs, and
deadlines, with a primary goal of minimizing the makespan. The C refers to the cost of executing the function. Job costs are
algorithm considers the constraints stated by the user and adjusts considered when computing fitness values for potential VM
the tasks accordingly, ensuring dependencies are maintained clusters, and these values are used to pick the most appropriate
throughout the sorting process. VM cluster. Eq. (3) details the fitness function calculation.
𝑛,𝑚,𝑝
During the task scheduling process, the algorithm utilizes 𝐹(𝑆) = 𝑚𝑖𝑛 ∑𝑖=1,𝑗=1,𝑘=1(𝑃1 , 𝑃2 , … , 𝑃𝑥 ) + 𝑇𝑗 + 𝑀𝑘 (3)
swarm behavior, mimicking the distribution and interaction
patterns of particles for improved optimization of multi- Where F(S) represents the fitness value of the respective
objective tasks. Its adaptability helps in solving a wide range of cluster, p indicates the available machines within the respective
NP-hard-level tasks, effectively handling the scheduling of cluster, M is the machine, m signifies the number of tasks, T
various tasks on different machines. The algorithm's efficiency refers to the corresponding task, x reflects the job's property, and
is further enhanced by its ability to detect all scheduled tasks, n indicates the number of jobs. The proposed algorithm defines
ensuring effective outcomes. Employing a random phase, the the static scheduling of tasks, as outlined in the Algorithm. 1.
algorithm aims to optimize the scheduling of cloudlets for
execution on Virtual Machines (VMs). Particle fitness, Algorithm. 1. Pseudocode for the proposed static task
bandwidth, MIPS, flow time, response time, resource usage, scheduling
throughput time, and imbalance degree guide the selection of Function ProposedAlgorithm(J, P, M):
particles in the pursuit space, ensuring improved wellness values Initialize BestFitCluster as empty
and effective execution outcomes. For Each Task Ti in Job J:
For Each VM Cluster Mk in Set of VM Clusters:
The aim is to allocate a set of tasks (𝑇 = 𝑇1 , 𝑇2 , 𝑇3 , … , 𝑇𝑛 ) Calculate Cost(C) for placing Ti in Mk based on properties
onto a designated group of processors (𝑃 = 𝑃1 , 𝑃2 , 𝑃3 , … , 𝑃𝑛 ) P
within a cluster of VMs. This task allocation, called solution S, Select VM Cluster M with minimum Cost(C)
follows predetermined measures and constraints within the If BestFitCluster is empty or Cost(C) < Cost(BestFitCluster):
cloud environment.
Update BestFitCluster with M
𝐹(𝑆) = 𝑚𝑖𝑛 ∑𝑡,𝑚
𝑖=1,𝑗=1 𝐶𝑡 (1) Return BestFitCluster
End Function
In Eq. (1), F(S) represents the fitness function of the
solutions, m corresponds to the total number of available Dynamic task scheduling involves the possibility of
machines, t stands for the entire number of tasks submitted by unspecified task properties, which necessitates the service
the user, and Ct denotes the completion time of all tasks. Fitness provider to establish a minimal set of parameters and their
function values vary with the type of job. Jobs can be assigned weights. The user provides these variables during the
categorized as either dynamic or static. Static jobs possess first task submission, which determines the allocation of
predefined properties, such as a fixed total data amount, data required machines. The quantity of machines in operation can
flow within the system, and time constraints. Conversely, be modified according to the workload duration of the task. The
dynamic jobs encompass undefined job properties, like data cost of task execution is determined by the highest value that the
bursting and indeterminate data types. Users are required to user is willing to pay for the first setup, as specified in Eq. (4).
specify whether the job properties are static or dynamic when 𝐶 = 𝑚𝑎𝑥 ∑𝑛,𝑚
𝑖=1,𝑗=1 𝑃𝑖 (4)
submitting the data.
For static job scheduling, where job properties are Where Pi represents the maximum value of the property,
determined by the cloud service provider, denoted as while n signifies the number of job properties, and m represents
(𝑃1 , 𝑃2 , 𝑃3 , … , 𝑃𝑛 ). The user-provided variables are used to the task count. Throughout runtime, the highest property
characterize the present infrastructure utilization and determine estimates gathered from the job's tasks are logged. The average
the needed virtual machine cluster. Various categories of VM value of the property is considered when allocating a new VM
clusters are evaluated, and their corresponding fitness scores are cluster for arriving tasks carrying varying properties, determined
computed. By employing the optimal fit algorithm, the VM by Eq. (5) and Eq. (6).
cluster that is most appropriate is chosen, thereby accomplishing (∑𝑛𝑖=1 𝑉𝑖 )⁄
load balancing. The cost of task execution is determined by the 𝑃𝑖(𝑇) = 𝑛 (5)
user's given property values, as per Eq. (2). Fig. 1 depicts the 𝑛,𝑚,𝑝
𝐹(𝑆) = 𝑚𝑖𝑛(∑𝑖=1,𝑗=1,𝑘=1(𝑃1 , 𝑃2 , … , 𝑃𝑥 ) + 𝑇𝑗 + 𝑀𝑘 ) <
system architecture of the suggested approach.
𝑚𝑎𝑥 ∑𝑛,𝑚
𝑖=1,𝑗=1 𝑃𝑖 (6)
𝐶 = ∑𝑛,𝑚
𝑖=1,𝑗=1 𝑃𝑖 (2)

593 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 15, No. 7, 2024

Fig. 1. System architecture.

Eq. (5) defines Pi(T) as the average property value of the job,
Vi as the specific property value, and n as the total number of IV. RESULTS AND DISCUSSION
property values obtained from the operations. The fitness value The efficiency of the suggested technique has been evaluated
of the VM cluster is denoted by F(S). In this equation, n refers using the CloudSim simulator. Table II provides a concise
to the number of jobs, x stands for the value of property P for overview of the specifications for key components of a cloud
the job, T is the task, m signifies the number of tasks, M platform, including virtual machines (VMs), cloudlets, data
represents the machine, and p refers to the number of machines centers, and clients. These parameters are regarded as minimal
within the VM cluster. The computed value should be the for dynamic simulation and constant for static simulation,
minimum among all fitness values lower than the maximum allowing for changes during execution. The suggested method is
value the user is willing to accept for task execution. The method evaluated against three techniques described in Section II,
proposed in Algorithm 2 defines the dynamic scheduling of specifically HDD-PLB [19], HG-GSA [17], and CAAH [20].
tasks. The performance metrics considered include VM utilization
ratio, execution time, response time, and makespan time. VM
Algorithm 2. Pseudocode for the proposed dynamic task utilization ratio denotes the number of VMs deployed relative to
scheduling the total VMs available, execution time indicates the duration of
Start task completion in the VMs, response time signifies the
Input: Job with set of Tasks J(T1, T2, ….. Tj); scheduler's time taken to schedule tasks, and makespan
Set of task properties (P1, P2, ……Pn); represents the finishing time of the last task.
Set of VM cluster properties (M1, M2, …… Mk);
Set of Task property values V; TABLE II. SIMULATION PARAMETER SPECIFICATIONS
Output: Bestfit VM cluster F(S) Components Attributes Values
Initialize BestfitCluster to null Task Total number 250-2000
For each task T in J Cloudlet Length 50,000
Calculate Cost C for each VM cluster using task properties P Total number 500
and values V Host Bandwidth 10 GB/s
For i = 1 to n & j = 1 to m Storage 100 GB
Calculate Fitness F(S) for each VM cluster RAM 2 GB
end For Total number 3
Find the VM cluster with the minimum Fitness (min F(S)) VM Total number 100
If BestfitCluster is null or F(S) < Fitness of BestfitCluster OS Linux
MIPS 10000
Update BestfitCluster with the current VM cluster
Processor 2.4 GHz
End For SSD 100 GB
Return BestfitCluster RAM 2 GB
End Data center Total number 2

594 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 15, No. 7, 2024

data presented in the figure proves that our technique, which

14
HDD-PLB supports both active and passive property-based tasks,
HG-GSA outperforms other present ones through optimized VM
12 CAAH
Proposed method placement, resulting in improved execution time. By efficiently
allocating fewer complex tasks to VMs, our technique reduces
Number of VMs

10
the overall execution time.
8
In Fig. 4, the proposed algorithm is compared to other
6 algorithms for response time. This comparison shows that our
algorithm converges faster and delivers VM placement
4 scheduling for the given tasks more rapidly. It can promptly
allocate VMs by prioritizing static property values for simpler
2 tasks, enhancing its ability to effectively handle tasks based on
dynamic properties.
0
400 600 800 1000 1200 1400 1600 1800 2000
25000
Number of tasks HDD-PLB
HG-GSA
Fig. 2. VM utilization ratio comparison. 20000 CAAH

Average makespan time

Proposed method
2e+5
15000
HDD-PLB
HG-GSA
CAAH
2e+5 10000
Average execution time

Proposed method

5000
1e+5

0
400 600 800 1000 1200 1400 1600 1800 2000
5e+4
Number of tasks

Fig. 4. Response time comparison.

0
400 600 800 1000 1200 1400 1600 1800 2000
25000
Number of tasks HDD-PLB
HG-GSA
Fig. 3. Execution time comparison. 20000 CAAH
Average makespan time

Proposed method
Fig. 2 compares the suggested strategy with existing
alternatives in terms of the VM utilization ratio. This indicator 15000

is crucial for evaluating the efficiency and performance of

resource allocation in a cloud computing environment. It offers
10000
valuable information on the efficient utilization of resources by
virtual machines, enabling the discovery of virtual machines that
are either underutilized or overutilized while making appropriate 5000
modifications to achieve optimal performance. The figure
illustrates that the proposed algorithm schedules tasks optimally,
0
using a lower number of VMs compared to the comparative 400 600 800 1000 1200 1400 1600 1800 2000
algorithms. Specifically, for tasks based on static properties, our
algorithm minimizes the number of VMs required, Number of tasks
outperforming other methods. The performance comparison
indicates that our algorithm optimizes VM scheduling and Fig. 5. Makespan time comparison.
achieves load balancing, enabling the cloud service provider to
handle more tasks from different users in real-time. However, a Fig. 5 compares the proposed algorithm with existing
drawback of the proposed approach is its tendency to utilize methods measured by the makespan metric. The figure indicates
more VM placements when dealing with dynamically changing that our algorithm offers optimal placement of VMs regardless
tasks, making it challenging to predict future VM usage of increasing workloads compared to the comparative
accurately. algorithms. The algorithm considers task properties to place
Fig. 3 illustrates a comparative analysis of our approach and VMs optimally.
other algorithms based on their respective execution times. The

595 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 15, No. 7, 2024

Despite all these improvements in task scheduling and deeper into exploring the scalability of the proposed method for
resource allocation, the proposed method has some limitations. larger and more diverse cloud environments. Moreover,
One primary limitation is the overhead resulting from more VM considering real-world deployment and testing on varied cloud
placements to accommodate dynamically changing tasks. While platforms could further validate the applicability and robustness
the method is superior in terms of efficiency in carrying out tasks of the approach. Overall, the proposed technique stands as a
characterized by static properties of resources, it can over- promising step towards addressing the challenges in task
allocate VMs in such a task with dynamic properties. This scheduling within cloud infrastructure services, offering a
tendency can bring about some issues, including reduced potential avenue for further advancements and practical
efficiency and increased operation costs since forecasting the implementations in the field.
future usage of VMs becomes complicated. As such, the method
does not guarantee the selection of an inexpensive solution that REFERENCES
would be advantageous in systems with volatile traffic loads. [1] V. Hayyolalam, B. Pourghebleh, A. A. P. Kazem, and A. Ghaffari,
"Exploring the state-of-the-art service composition approaches in cloud
Another limitation is that complexity can result from manufacturing systems to enhance upcoming techniques," The
handling scheduling algorithms employed in the examination International Journal of Advanced Manufacturing Technology, vol. 105,
process. Although load balancing coupled with metaheuristic no. 1-4, pp. 471-498, 2019.
optimization techniques offers a solution for resolving [2] K. Saidi and D. Bardou, "Task scheduling and VM placement to resource
inefficiencies, it is accompanied by increased computational allocation in Cloud computing: challenges and opportunities," Cluster
Computing, vol. 26, no. 5, pp. 3069-3087, 2023.
complexity. This complexity might affect the method's usability
[3] W. Wang and Z. Liu, "Cloud Service Composition using Firefly
when scaling up and applying it to larger and more diverse cloud Optimization Algorithm and Fuzzy Logic," International Journal of
environments. Further, due to the dependency of the algorithm Advanced Computer Science and Applications, vol. 14, no. 3, 2023.
on chosen performance parameters and simulation parameters, [4] S. Zhao, J. Miao, J. Zhao, and N. Naghshbandi, "A comprehensive and
it is possible to infer that the efficiency of the algorithm can be systematic review of the banking systems based on pay-as-you-go
different from that of another Cloud platform and from that of a payment fashion and cloud computing in the pandemic era," Information
real-world scenario. Mitigating these limitations involves an Systems and e-Business Management, pp. 1-29, 2023.
enhancement of the method and its application with the aim of [5] X. Liu and Y. Deng, "A new QoS-aware service discovery technique in
the Internet of Things using whale optimization and genetic algorithms,"
achieving uniformity and efficiency in various settings and Journal of Engineering and Applied Science, vol. 71, no. 1, p. 4, 2024.
projects. [6] B. Kruekaew and W. Kimpan, "Multi-objective task scheduling
optimization for load balancing in cloud computing environment using
V. CONCLUSION hybrid artificial bee colony algorithm with reinforcement learning," IEEE
Cloud computing has gained popularity due to its flexible Access, vol. 10, pp. 17803-17818, 2022.
and resourceful nature, providing adaptable resources on a [7] B. Pourghebleh, A. A. Anvigh, A. R. Ramtin, and B. Mohammadi, "The
shared infrastructure. This technology provides a framework for importance of nature-inspired meta-heuristic algorithms for solving
virtual machine consolidation problem in cloud environments," Cluster
various services, from scientific operations to service Computing, pp. 1-24, 2021.
computing, highlighting the critical role of efficient task [8] A. G. Gad, E. H. Houssein, M. Zhou, P. N. Suganthan, and Y. M. Wazery,
scheduling in ensuring optimal resource allocation and "Damping-assisted evolutionary swarm intelligence for industrial iot task
performance. In this study, we introduced a novel approach for scheduling in cloud computing," IEEE Internet of Things Journal, 2023.
task scheduling in cloud infrastructure services, addressing the [9] M.-L. Chiang, H.-C. Hsieh, Y.-H. Cheng, W.-L. Lin, and B.-H. Zeng,
significant challenge of mapping tasks to available resources "Improvement of tasks scheduling algorithm based on load balancing
while minimizing execution plan objectives. The proposed candidate method under cloud computing environment," Expert Systems
with Applications, vol. 212, p. 118714, 2023.
technique leverages a metaheuristic optimization method along
[10] M. Yadav and A. Mishra, "An enhanced ordinal optimization with lower
with load distribution, designed to optimize cloud computing scheduling overhead based novel approach for task scheduling in cloud
service providers' overall performance and effectively alleviate computing environment," Journal of Cloud Computing, vol. 12, no. 1, p.
scheduling issues. One of the key strengths of our proposed 8, 2023.
approach is its adaptability to both static and dynamic task [11] H. Godhrawala and R. Sridaran, "Apriori Algorithm Based Approach for
conditions. In static scenarios, where VM parameters are fixed, Improving QoS and SLA Guarantee in IaaS Clouds Using Pattern-Based
and in dynamic conditions, where parameters are adjusted in Service-Oriented Architecture," SN Computer Science, vol. 4, no. 5, p.
700, 2023.
real-time, our method exhibits efficacy and flexibility.
[12] F. Thabit, O. Can, R. U. Z. Wani, M. A. Qasem, S. Thorat, and H. A.
Simulation results unequivocally demonstrated the superiority Alkhzaimi, "Data security techniques in cloud computing based on
of our proposed technique over existing methods, showcasing machine learning algorithms and cryptographic algorithms: Lightweight
notable improvements in key performance metrics such as algorithms and genetics algorithms," Concurrency and Computation:
makespan, response time, and execution time. Practice and Experience, p. e7691, 2023.
[13] P. A. Malla and S. Sheikh, "Analysis of QoS aware energy‐efficient
The outcomes of our study underscore the practical viability resource provisioning techniques in cloud computing," International
and potency of our proposed metaheuristic optimization Journal of Communication Systems, vol. 36, no. 1, p. e5359, 2023.
approach with load balancing in addressing the complexities of [14] I. Behera and S. Sobhanayak, "Task scheduling optimization in
task scheduling in cloud infrastructure services. It not only heterogeneous cloud computing environments: A hybrid GA-GWO
optimizes resource utilization but also contributes significantly approach," Journal of Parallel and Distributed Computing, vol. 183, p.
104766, 2024.
to enhancing user experience by ensuring guaranteed
performance and efficient resource allocation. While this [15] P. V. Reddy and K. G. Reddy, "An energy efficient RL based workflow
scheduling in cloud computing," Expert Systems with Applications, vol.
research presents promising results, future work could delve 234, p. 121038, 2023.

596 | P a g e
www.ijacsa.thesai.org
 (IJACSA) International Journal of Advanced Computer Science and Applications,
Vol. 15, No. 7, 2024

[16] J. Yang, B. Jiang, Z. Lv, and K.-K. R. Choo, "A task scheduling algorithm [20] A. K. Kulkarni and B. Annappa, "Context aware VM placement
considering game theory designed for energy management in cloud optimization technique for heterogeneous IaaS cloud," IEEE access, vol.
computing," Future Generation computer systems, vol. 105, pp. 985-992, 7, pp. 89702-89713, 2019.
2020. [21] H. Liu, "Research on cloud computing adaptive task scheduling based on
[17] D. Chaudhary and B. Kumar, "Cost optimized hybrid genetic- ant colony algorithm," Optik, vol. 258, p. 168677, 2022.
gravitational search algorithm for load scheduling in cloud computing," [22] M. Abd Elaziz, S. Xiong, K. Jayasena, and L. Li, "Task scheduling in
Applied Soft Computing, vol. 83, p. 105627, 2019. cloud computing based on hybrid moth search algorithm and differential
[18] L. Imene, S. Sihem, K. Okba, and B. Mohamed, "A third generation evolution," Knowledge-Based Systems, vol. 169, pp. 39-52, 2019.
genetic algorithm NSGAIII for task scheduling in cloud computing," [23] K. Dubey and S. C. Sharma, "A novel multi-objective CR-PSO task
Journal of King Saud University-Computer and Information Sciences, scheduling algorithm with deadline constraint in cloud computing,"
vol. 34, no. 9, pp. 7515-7529, 2022. Sustainable Computing: Informatics and Systems, vol. 32, p. 100605,
[19] A. Kaur and B. Kaur, "Load balancing optimization based on hybrid 2021.
Heuristic-Metaheuristic techniques in cloud environment," Journal of [24] S. Mangalampalli, G. R. Karri, M. Kumar, O. I. Khalaf, C. A. T. Romero,
King Saud University-Computer and Information Sciences, vol. 34, no. 3, and G. A. Sahib, "DRLBTSA: Deep reinforcement learning based task-
pp. 813-824, 2022. scheduling algorithm in cloud computing," Multimedia Tools and
Applications, pp. 1-29, 2023.

597 | P a g e
www.ijacsa.thesai.org