Optimizing Workload Scheduling in Cloud Paradigm using Robust
Neutrosophic C-Means Clustering Boosted with Fish School Search
S. YUVARAJ GANDHI, T. REVATHI
Department of Computer Science,
PSG College of Arts and Science, Coimbatore, Tamilnadu,
INDIA
Abstract:- In the present internet world, accessing cloud resources for a low cost, according to their needs, is
available to all users. Sharing resources is becoming increasingly necessary as people complete their activities
in the cloud. It becomes essential for distributed workloads to be optimized to perform efficient workload
scheduling and progressing resource utilization in a cloud environment. Scheduling cloud resources
considerably benefits from the invention of machine learning and metaheuristic models to address this scenario.
Though many existing algorithms are developed in cloud-based task scheduling using unsupervised clustering
methods, the problem of unknown task requirements or resource availability in adverse conditions is still
challenging. In this study, an uncertainty-based unsupervised technique is constructed to group incoming tasks
according to the required resources, and it is scheduled to the most suitable resources more prominently. This
paper introduced a Robust Neutrosophic C-Means Clustering boosted with the fish school search algorithm
(RNCM-FSSA) for clustering the incoming tasks and the resources based on their requirement and availability.
With the degree of indeterminacy, neutrosophic C-means discriminating the deterministic and indeterministic
schemes and scheduling them to the optimal resources more effectively. Using the fitness value computed by
FFSA, the potential cluster centroids are utilized for clustering, thus avoiding the early convergence in the
grouping process. The simulation results explore that the robustness of the proposed RCNM-SSA achieves
better resource utilization, the degree of imbalance is minimal, and computation complexity is also
considerably decreased compared with other unsupervised models.
Key-Words: - Work Scheduling, Uncertainty, Neutrosophic C-Means clustering, Fish School Search Algorithm,
indeterminacy.
Received: May 14, 2022. Revised: August 17, 2023. Accepted: September 19, 2023. Available online: November 14, 2023.
1 Introduction
Cloud computing has ushered in a transformative
era in computing, offering a dynamic and scalable
environment that combines a multitude of resources
to meet the ever-growing demand for services.
However, at the heart of this cloud paradigm lies a
formidable challenge – efficient resource allocation.
The relentless surge in service requests collides with
the limited availability of resources, creating a
pivotal concern, [1]. Consequently, a pressing need
exists for the establishment of a robust mechanism
to distribute task workloads effectively among the
available resources.
As this challenge looms, and in response to the
needs of cloud users, this study embarks on a
journey to create a robust load-balancing-based
resource scheduling policy. The overarching goals
include a substantial reduction in task execution
response times and the optimization of resource
utilization, [2], [3]. Although various resource
scheduling strategies are currently available, they
are often tailored for homogeneous resource types,
which can lead to inefficiencies when dealing with
heterogeneous environments. This homogeneity-
centric approach inevitably results in the
cumbersome task of scanning the entire list of
virtual machines for each incoming work request,
[4].
In contrast to these challenges, our work
harnesses the power of clustering techniques,
designed to meet cloud service requirements
effectively. Through the intelligent grouping of
incoming jobs and virtual machines based on their
capacity, this technique efficiently mitigates the
overhead linked to the screening process.
Consequently, it offers an effective solution to the
challenges associated with resource scheduling in a
heterogeneous environment and ensures the efficient
distribution of workloads, [5]. Conventional
clustering algorithms are often plagued by issues
such as local optima and premature convergence
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
8
Volume 23, 2024
due to suboptimal centroid selection. In response to
these shortcomings, our current research places a
strong emphasis on the imperative goal of
optimizing load balancing. Achieving a well-
balanced distribution of workloads across suitable
cloud resources, however, becomes increasingly
complex due to the inherent inconsistencies and
indeterminacies associated with incoming task
requests and the fluctuating availability of cloud
resources, a common reality in real-time cloud
applications, [6].
In addition to these novel approaches, our study
provides several key benefits. It excels in optimizing
load balancing, significantly reducing task execution
times, and maximizing resource utilization.
Furthermore, it effectively addresses the unique
challenges posed by heterogeneous resource
environments, streamlining workload distribution.
Hence, in this paper, the knowledge of uncertainty is
acquired by devising a robust model known as the
Neutrosophic C-Means clustering model for
grouping similar types of task requests and the
available resource capacity by computing the degree
of membership using truthiness, falsity, and
indeterminacy. The centroid selection is
accomplished by inducing the metaheuristic
algorithm called fish school searching. The
assignment of resources in an optimized way is
explained in the following sections.
2 Related Work
The authors in, [6], devised a modified heterogenous
dominant sequence clustering by ranking tasks
based on their priority and balanced load
distribution. Handling outliers and noisy task
requests are ignored; thus, it is not feasible in a
dynamic environment.
Fuzzy C-means clustering, which uses a
streamlined scanning procedure, and was used by,
[7]. to conduct a clustering-based load balancing.
The workload associated with screening the set of
accessible virtual machines is clustered according to
their capabilities, and the resource demands for the
issue are more than met. But quick convergence is
the outcome of local search optima.
The authors in, [8], conducted a detailed survey
on scheduling resources using various clustering
techniques, and the author stated multiple problems
faced during the clustering process load balancing,
selection of appropriate architecture for scheduling,
assignment of resources for the specific job request,
etc.
The authors in, [9], devised a distributed inert
fog-based scheduling of cloud resources with a time-
based restriction model. They focused on
constructing adaptability and scalability-based
resource utilization architecture with noteworthy
methods. However, the problem of handling noise or
outliers is not considered during workload
scheduling.
The authors in, [10], constructed a task
scheduling algorithm using the K-Means clustering
concept. The two parameters, virtual machine
capacity and task length, are used for computation.
The task is clustered using its size, and the resources
are grouped by their processing ability. Finally, tasks
are assigned with the respective cluster resource
type.
The authors in, [11], developed a chicken swarm
optimization-oriented evolutionary model with the
mutation and crossover operator in vehicular
networks. The Brownian motion-based bacteria
foraging optimization is used for selecting the
features of vehicles to be clustered. Depending on
the resource availability, the scheduling is carried out
in the cloud.
The authors in, [12], introduced a resource
provisioning method using a decision-making
algorithm that uses distance measures to cluster
tasks. The tasks are clustered depending on their
need, and the time series prediction is used for
energy saving to schedule the lesson with the
appropriate sources in the cloud.
This literature review analysis identifies a
particular part of workload scheduling and
clustering, [6], [7], [8], [9], [10], [11], [12]. Our
proposed research has taken various key points from
the above-proposed model research article. For
improving the cloud environment-based cluster, this
proposed research study was inspired by the fish
school search method for improving the efficiency of
the cloud environment. This fish school search
method helps to reduce the risks in a cloud
environment. The above literature review analysis
also motivates us to ameliorate the voraciousness,
lack of precision, and completeness in workload
scheduling and clustering. This helps to make better
extreme distribution of cloud resources. The system's
performance was enhanced with the help of
optimization process resources like cloud resource
scheduling.
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
9
Volume 23, 2024
3 Robust Neutrosophic C-Means
Clustering (RNCM)
Implementing the Robust Neutrosophic C-Means
(RNCM) clustering algorithm involves several steps
to handle uncertainties and noise in the workload
data and efficiently group cloud tasks into clusters.
Before applying the RNCM clustering algorithm,
workload data is pre-processed to address any
missing values or outliers. And then performed data
normalization to ensure that all features were on a
similar scale, [13].
Fig. 1: Robust Neutrosophic C-Means Clustering
Figure 1 discusses the parameters required for
the RNCM algorithm, including the number of
clusters (c), the fuzziness parameter (m), and the
robustness threshold (θ). The robustness threshold θ
determines the sensitivity to outliers in the data, and
it should be carefully chosen based on the dataset's
characteristics. Next, Calculate the membership
degree of each cloud task to each cluster using the
RNCM membership function. The membership
function accounts for uncertainty and partially
allows a task to belong to multiple groups. We
computed the robustness measures for each
collection to assess the influence of noisy and
uncertain data. It was updated. The cluster centers
are based on the computed membership degrees and
the robustness measures. By evaluating the
difference between the current and previous cluster
centers, convergence is checked. (If the difference is
below a predefined threshold, stop the iterations;
otherwise, repeat steps 3 to 5 until convergence.)
After the algorithm converges, assign each task to
the cluster with the highest membership degree, [14]
and finally, perform post-processing, such as
analyzing the results for further insights.
3.1 Cloud-based Machine Learning
Cloud-based machine learning offers optimized load
balancing by leveraging its inherent scalability,
flexibility, and resource management capabilities.
Load balancing in cloud computing refers to the
efficient distribution of incoming workload across
available computing resources to ensure optimal
resource utilization, minimize response time, and
avoid overloading any single help. Cloud platforms
provide auto-scaling capabilities that automatically
adjust the number of computing resources based on
the current workload demand. Machine learning
models for load balancing can be designed to
monitor the current workload and trigger auto-
scaling actions when certain thresholds are reached.
When the workload increases, additional resources
are automatically provisioned to handle the load,
and when the workload decreases, unnecessary
resources are scaled down, optimizing resource
allocation.
Figure 2 discusses the resource management
needs such as business intelligence, business
reporting, capacity planning, integration, resource
forecasting, resource planning, and resource
scheduling, access to the services is very important.
In that scenario, cloud providers get access to the
resource management feature by using Cloud-based
machine learning systems. This access helps to
allocate and reallocate tasks even the different kinds
of resources on the workload scheduling. Hence the
resource management features reduce the overload
on the load balancing with the help of Machine
learning algorithms. In addition to this,
simultaneous processing is also processed for task
distribution with the help of resource management
features to improve cloud computing-specific
actions like task distribution across multiple nodes.
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
10
Volume 23, 2024
Fig. 2: Cloud-based machine learning to optimize load balancing
This proposed scenario helps to manage and
maximize resource utilization by improving the
load-balancing decisions, and workload distribution
in an efficient manner. Simultaneously, our
suggested system helps to compute the used
resources and workload by using the Cloud-based
machine learning models. Hence, indicated
optimized resource allocation are step-up and
increment as the workload distribution task by
load-balancing. The real-time performance metrics
and resource feedback, are formulated by machine
learning. Even the load-balancing decisions are also
improved by the criteria like predictive analytics.
Finally, the resource management features are
developed according to load balancing workload
load distribution.
Cloud providers often have data centers located
in different geographical regions. Load balancing
algorithms can exploit this geographic distribution
by directing workload to the nearest available data
center with sufficient resources. This reduces
latency and improves response times for end-users.
Machine learning models for load balancing can
consider service level agreements defined for
different workloads. By considering SLAs, the
algorithm can prioritize critical workloads and
allocate resources accordingly, ensuring that
performance targets are met. By combining cloud
computing capabilities with machine learning
algorithms, cloud-based machine learning can
efficiently and dynamically distribute workloads to
suitable resources in the cloud, achieving optimized
load balancing and maximizing the utilization of
cloud resources, [15].
3.2 Workload Scheduling Problem in Cloud
Computing
A novel approach to tackle the workload-
scheduling problem in cloud computing using the
combined power of two advanced techniques, [16].
This article proposes a novel approach to tackle the
workload-scheduling problem in cloud computing
using the combined power of two advanced
techniques:
Fig. 3: Flow of load balancing in cloud computing
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
11
Volume 23, 2024
Fig. 4: Robust Neutrosophic Fuzzy C-Means Clustering fused with fish school search optimization for effective
workload scheduling and allocation of resources in a cloud
Figure 3 discusses the Robust Neutrosophic C-
Means (RNCM) clustering and Fish School Search
(FSS) algorithm flow. By integrating these
methodologies, the project aims to achieve an
enhanced and optimized workload scheduling
solution that can adapt to changing cloud
conditions while ensuring efficient resource
allocation and improved system performance. In
this work, the process of workload scheduling in
the cloud is accomplished by considering the type
of resource required in a vague and indeterministic
environment by developing a robust multivalued
uncertainty theory known as Neutrosophic logic,
which introduces the degree of indeterminacy to
cope with the inconsistent and ambiguous
information about the workload distribution and
resource availability, [17]. It becomes necessary to
schedule work evenly among the cloud resources to
use them more effectively when there is a strong
demand for work requests. Neutrosophic C-Means
with Fishing school search behavior provides a
resilient model with a Quality of Service at a
reduced computation complexity with the best
possible use of cloud resources.
Figure 4 shows the suggested robust
neutrosophic C-Means enhanced with fish school
searching-based clustering. Cloud computing offers
a flexible and scalable infrastructure that plays a
crucial role in the efficient implementation and
execution of the proposed workload scheduling
algorithm. The incoming task requests from cloud
users are grouped according to their storage,
bandwidth, and processing speed needs, [18].
Additionally, the cloud groups the virtual machines
according to their configuration and availability. To
exhibit the uncertainty features associated with
clustering, each request and resource are
characterized in the triplet components of
membership degree of truthiness (T), indeterminacy
(I), and falsehood (F). Assigning tasks and the most
potential cloud resources as centroids requires
understanding the fish school searching method.
Search refinement enables the cloud resource
scheduling policy by preventing chaotic cluster
centroid selection and inducing the fish school's
prey-seeking behavior.
3.3 The preamble of Neutrosophic Fuzzy C-
Means Clustering
The Neutrosophic concept is a versatile framework
that encompasses various types of logic, such as
Intuitionistic fuzzy, paraconsistent, ambiguous, and
multivalued logic. In Neutrosophic, each element is
described in terms of three membership degrees:
truthiness (T), falsity (F), and indeterminacy (I),
[19], [20].
Step 1: Set the incoming task requests and resource
allocation using Neutrosophic.
Step 2: Check the uncertain demands with the
resource availability except for the Vague
information
Step 3: Create independent membership degrees T,
F, and I with the ranges among (0 to 1)
Step 4: 0 ↔ absence and 1 ↔ presence || (vague
nuances, uncertain information)
Step 5: Execute the Neutrosophic C-Means
Clustering into the clusters.
Step 6: Create a deterministic cluster for on-task
requests
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
12
Volume 23, 2024
Step 7: Create an indeterministic cluster for
resource requirements
Step 8: Calculate the resource needs using an
objective function (OJ) of the Neutrosophic C-
Means Clustering
Step 9: Calculate the membership degrees and
weight factors.
The objective function is expressed as follows:
(1)
NT, NI, and NF refer to the degree of
membership of truthiness, indeterminacy, and
falsity of the resource/task 'Zi.' Nc denotes the
number of classes, wti = 1,2,3 belongs to the weight
factor, and P is the number of work requests or
resources available in the cloud. refers to the
cluster centroid. The control parameter ϑ is
introduced to manage outlier elements within the
clustering process. It plays a role in determining
which data points may be outliers and need special
consideration. The mean value of the first two
most significant clusters is represented by
. It
is computed as the average between two cluster
centroids (Ai and Bi), which are determined based
on the maximum membership degree of truthiness
().
(2)
Ai =
; Bi =
(3)
Clusters are formed based on the degrees of
truthiness, falsity, and indeterminacy. The
classification helps in grouping tasks and resources
effectively, taking into account the varying degrees
of clarity in their requirements. By employing the
Neutrosophic Fuzzy C-Means Clustering technique
with the detailed components and equations
explained above, the research aims to create more
efficient and flexible resource allocation strategies
in cloud computing.
Algorithm: Preamble of Neutrosophic Fuzzy C-
Means Clustering
Input: Z= {z1, z2, ..., zN}
Output: Return final cluster centroids clj,
membership matrices NT, NI, NF
Initialization: no, of data points- N, no. of clusters-
C, fuzzifier parameter-m, cluster centers V
Procedure:
Begin
Step 1. for j = 1 to C do:
Randomly initialize clj
End for
Step 2. for iteration = 1 to maxIterations do
for each data point, zi do
for each cluster center, clj do
Calculate membership
degree
End for
End for
End for
Step 3. For each cluster center, clj do
Update cluster center
End for
End
3.4 Fish School Search Algorithm (FFSA)
The fish school searching model (FFS) is based on
the feeding behavior of fish and uses the
contraction and expansion of fish during their
feeding cycles, [21]. In n-dimensional space,
maximizing the procedure of seeking approach is
carried out depending on the agent/fish location,
and the primary metric used for evaluating the
search for the solution is accomplished by
employing the weight variable, [22]. The primary
tasks of FFS include consuming food and motion,
[23].
A collection of essential responding agents,
known as fish, conducts the search procedure in the
FSS. The aquatic area is used as a search space for
the search agents. A location within the search
space, zi(t), and weight wti (t) represent each fish.
According to how much food is found in the tank,
the weight of the fish is revised iteratively.
3.4.1 Movement Operator of FFSA
To perform the movement of fish, individual,
collective-volatile, and collective instinctive are
three operators used. Individuals are moved
randomly using the operator individual as depicted
in the equation.
(4)
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
13
Volume 23, 2024
Where is the position of the jth fish after and
before movement concerning
a random
vector assigned during each dimension, iteration,
and its value is predefined using a uniform
probability distribution.
3.4.2 Feeding Operator of FFSA
After applying the movement operation, next, the
feeding operator is implied by searching the
neighbor based on the computation of their fitness
function. If the neighbor's location is better than its
present location, it moves to the new position or
remains in the same situation. When the condition
ff () > ) is satisfied, then only the new
location will be agreed. Otherwise, the
fish remain in the same location so that its
succeeding location will not be updated
= . The average movement of all the fish Z in
the school is used for commuting the collective,
instinctive part of the movement. The biased mean
of shifts for discrete vector It is
represented as:
, (5)
Where M signifies displacement of the fish from
one location to another location, s is the size of the
fish school. After the computation of M, each
individual will move towards the new location as
mathematically modeled.
= + M (6)
The collective-violent component controls fish
school predation or probing during the search
process. It begins by calculating the fish school's
barycenter as shown in the below section
concerning the fish's position Zi of the fish and its
weight wgti
(7)
While the total school weight
If iteration
l to l+1 is improved, the search agents (fishes)
move toward the barycentre
, (8)
(t) is a hyperparameter, its value lies between
1 to . The initial value of each weight is
2.
Algorithm: Fish School Search Algorithm
(FFSA)
Initialization: fish positions-Zi , weights-wgti , step
size-α, c1 and c2 -collective-instinctive and
collective-volitive movement
Procedure:
Begin
Step 1. for iteration = 1 to maxIterations do
for each fish, i do
Generate random vector Ri
Update position
End for
Step 2. For each fish, i do the following:
Compute fitness value
if ff(Zi + α * Ri) > ffi then:
= + M
Else =
End if
End for
Step 3. Compute barycenter
Step 4. Update weight
,
Step 5. Compute mean
displacement
,
Step 6. Update position
= + M
End for
End
3.5 Robust Neutrosophic C-Means
Clustering boosted with Fish School
Search Algorithm
The finding was comprehensively compared to
standard neutrosophic clustering, where the cluster
centroids are selected randomly depending on their
neighborhood. As a result, clustering begins to
converge early, and because local optima are
compromised, its effectiveness will be significantly
reduced in ambiguous and inconsistent
circumstances. Hence, in this robust Neutrosophic
C-Means clustering, the centroid selection is
boosted by acquiring the intelligence of the fish
school search algorithm, which works in a specific
problem and achieves the highest optimal solution.
The centroid selection among the available
resources and the workload scheduling is made by
computing their fitness value, and the best resource
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
14
Volume 23, 2024
is selected for the specific workload with the
characteristic of neutrosophic values, [24].
Algorithm: Robust Neutrosophic C-Means
Clustering boosted with Fish school search
Algorithm for Workload Scheduling among
available Cloud Resources
Input: Cloud User Incoming Task {Itsk}, Resource
Available {RA}
Output: Atsk load distribution with optimum
resource usage
Procedure:
Begin
Step 1. For each Itxk(i=1…n) incoming task request do
Pri-value = LEN(Itski) * PR(Itski ) *
DL(Itski) * CST(Itski)
Where LEN = Length, PR =
priority, DL = deadline , CST = cost
End for
Step 2. Apply C-Means clustering
Step 3. Determine potential centroid
Step 4. For each RA(i = 1….m) do
Discover RAi(parameters)
//Capacity of RAM, Bandwidth, Memory,
MIPS
end for
Step 5. Cluster low_resources = [],
medium_resources = [], high_resources = []
Step 6. For resource_index in range (len (data)), do
resource_objective_value
=ObjectiveFunction (data
[resource_index])
if (resource_objective_value <=
low_threshold) then
low_resources.append
(resource_index)
elif( low_threshold <
resource_objective_value <=
high_threshold) then
medium_resources.append
(resource_index)
else
high_resources.append(resource_i
ndex)
end if
end for
Step 7. for each Itsk(i=1…n) do
Allocate the corresponding
resources relative to the concern cluster
model
End for
End
4 Results and Discussions
This part discusses the evaluation of the proposed
model Robust Neutrosophic C-Means Clustering
boosted with Fish school search algorithm for
optimized cloud resource scheduling by uniformly
distributing the cloud users' workload. The
proposed model RNCM-FFSA is simulated using a
cloudsim simulator. The task ranges from 250 to
1000. The evaluation metrics used for comparison
are Makespan, degree of imbalance, resource
utilization, and Execution Time.
Table 1 presents a comparison of the execution
times for four different clustering algorithms,
namely K-Means Clustering (KMC), Fuzzy C-
Means Clustering (FCM), Neutrosophic C-Means
Clustering (NCM), and Robust Neutrosophic C-
Means Clustering boosted with the fish school
search algorithm (RNCM-FSSA), as the number of
tasks increases. In task scheduling, Makespan
represents the time to execute all assignments and
achieve a balanced workload distribution among
available resources, [25].
Table 1. Makespan Comparison over Small Number of Tasks
Methods
Makespan over an increasing number of tasks
250
500
750
1000
KMC
105
140
275
440
FCM
95
120
230
345
NCM
75
100
190
285
Proposed RNCM-FSSA
55
78
125
148
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
15
Volume 23, 2024
Fig. 5: Evaluation based on Makespan
Table 2. Makespan Comparison over a Large Number of Tasks
Methods
Makespan over a large number of tasks
1250
1500
1750
2000
KMC
1350
1590
1760
1960
FCM
1280
1550
1790
1930
NCM
1220
1450
1690
1860
Proposed RNCM-FSSA
1150
1380
1650
1800
Figure 5 examines the effectiveness of four
clustering methods based on how long it takes to
assign a workload to the available cloud resources.
KMC exhibits makespan values of 105, 140, 275,
and 440 units for 250, 500, 750, and 1000 tasks,
respectively. FCM performs slightly better with
makespan values of 95, 120, 230, and 345 units for
the corresponding task numbers. NCM further
improves the Makespan, achieving 75, 100, 190, and
285 units. However, the Proposed RNCM-FSSA
outperforms all other methods significantly,
showcasing the lowest makespan values of 55, 78,
125, and 148 units, respectively. By adopting the
fish school search utilized for initial centroid
selection for clustering and assigning workload
based on the resource available, the RNCM-FFSA
has a relatively short makespan. K-means clustering
makes use of predetermined centroids that are
picked at random. Additionally, centroids are
chosen by FCM at random, and works are
distributed according to the clusters' membership
grades, in conventional NCM using, deterministic
and indeterministic clustering for workload
allocation. However, NCM still chooses the initial
centroids at random, and when re-clustering, all
three traditional methods merely employ the
distance measure to assign workload to particular
clusters, [26].
Table 2 presents the Makespan values for four
different clustering methods as the number of tasks
increases. The methods evaluated are K-Means
Clustering (KMC), Fuzzy C-Means Clustering
(FCM), Neutrosophic C-Means Clustering (NCM),
and the proposed Robust Neutrosophic C-Means
Clustering with the fish school search algorithm
(RNCM-FSSA).
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
16
Volume 23, 2024
Fig. 6: Evaluation based on Makespan
Table 3. Success rate Comparison over the increasing number of Tasks
Methods
Success rate over an increasing number of tasks
250
500
750
1000
KMC
66
68
66
70
FCM
74
72
76
77
NCM
76
79
77
83
Proposed RNCM-FSSA
86
88
91
94
As the task counts grow from 1250 to 2000, the
Makespan values for each method reflect the time
required to complete tasks under varying workloads
shown in Figure 6. KMC demonstrates Makespan
values ranging from 1350 to 1960, showing an
increase in task execution times with higher task
counts. FCM exhibits slightly better performance
with Makespan values ranging from 1280 to 1930.
NCM further reduces the Makespan, achieving
values from 1220 to 1860. Notably, the proposed
RNCM-FSSA consistently outperforms the other
methods, boasting the lowest Makespan values in
all task count scenarios, ranging from 1150 to
1800. These results highlight the efficiency of
RNCM-FSSA in distributing workloads and
minimizing Makespan, making it a promising
approach for optimizing cloud resource scheduling
over a large number of tasks.
Table 3 displays the success rates of four
different clustering methods as the number of tasks
increases. The methods evaluated are K-Means
Clustering (KMC), Fuzzy C-Means Clustering
(FCM), Neutrosophic C-Means Clustering (NCM),
and the proposed Robust Neutrosophic C-Means
Clustering with the fish school search algorithm
(RNCM-FSSA).
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
17
Volume 23, 2024
Fig. 7: Evaluation based on Success rate
Table 4. Resource Utilization Analysis of Different Methods
Methods
Resource Utilization over an increasing number of tasks
250
250
250
250
KMC
48
54
66
70
FCM
62
68
71
76
NCM
67
75
78
82
RNCM-FSSA
78
84
89
93
The success rate measures the effectiveness of
these clustering methods in achieving a balanced
workload distribution among available resources as
the workload grows shown in Figure 7. KMC
exhibits success rates that vary between 66% and
70% as the number of tasks increases from 250 to
1000. FCM's performance ranges from 72% to
77%. NCM showcases success rates from 76% to
83%. In contrast, the proposed RNCM-FSSA
consistently outperforms the other methods, with
success rates increasing from 86% to 94% across
the task count scenarios. These results demonstrate
that RNCM-FSSA excels in effectively balancing
workloads, resulting in higher success rates. It
offers a promising approach for optimizing cloud
resource scheduling as the number of tasks grows,
ensuring better utilization of available resources
and timely task execution.
Table 4 presents the Makespan, which
represents the completion time of tasks for four
different methods. Resource utilization measures
the efficiency of resource usage in a system. Cloud
computing refers to how effectively the available
cloud resources are utilized to execute tasks, [27].
Higher resource utilization indicates that resources
are fully used, while lower utilization suggests
underutilization or idle resources. The equation for
Resource Utilization is as follows,
Resource Utilization (%) =
(
)*100
(9)
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
18
Volume 23, 2024
Fig. 8: Evaluation based on Resource Utilization
Table 5. Degree of Imbalance Evaluation over Increasing Number of Tasks
Methods
Degree of Imbalance over the increasing number of tasks
250
250
250
250
KMC
0.39
0.48
0.59
0.66
FCM
0.38
0.45
0.56
0.60
NCM
0.38
0.35
0.42
0.57
RNCM-FSSA
0.28
0.21
0.19
0.10
As incoming task requests are scheduled using
four distinct clustering algorithms, a performance
comparison concerning cloud resource utilization is
shown in Figure 8. For traditional clustering
models such as k-means, FCM, and NCM, the
imprecision of the task requirement is more
complicated, and cloud resource availability is also
tough to predict. For KMC, the resource utilization
starts at 48 units and gradually increases to 54%,
66%, and 70% as the number of tasks rises. FCM
exhibits resource utilization values of 62%, 68%,
71%, and 76% for the corresponding task
increments. NCM shows an escalating resource
utilization trend with discounts of 67%, 75%, 78%,
and 82%. Finally, RNCM-FSSA displays the
highest resource utilization among the methods,
starting at 78% and reaching 84%, 89%, and 93%
as the number of tasks increases. Based on the
requirements, the RNCM-FFSA groups jobs into
high, low, and medium categories. When virtual
machines are chosen with the assistance of the fish
school search algorithm in the proposed RNCM-
FFSA, incoming works are distributed evenly, and
resource utilization is increased more noticeably
than with the other three clustering models, [28].
Table 5 presents the Degree of Imbalance
results for four methods (KMC, FCM, NCM,
RNCM-FSSA) as task counts increase. It measures
how evenly tasks are distributed among resources.
RNCM-FSSA consistently demonstrates the lowest
imbalance, showcasing its superior workload
balancing. KMC exhibits the highest imbalance,
indicating its inefficiency. More commentary in the
text regarding Table 5 is needed to provide a
comprehensive understanding of the results. The
degree of imbalance measures how evenly or
unevenly the workload is distributed among
resources. The equation for the Degree of
Imbalance is as follows,
Degree of Imbalance =
)
(10)
Max Load is the maximum load among all
resources, and Min Load is the minimum load
among all resources.
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
19
Volume 23, 2024
Fig. 9: Evaluation based on the Degree of Imbalance
Table 6. Execution Time Assessment over Increasing Number of Tasks
Methods
Execution time over an increasing number of tasks
250
500
750
1000
KMC
3.4
5.6
8.7
12
FCM
3.2
4.9
7.8
10.2
NCM
2.8
3.5
6.2
7.5
RNCM-FSSA
1.4
1.7
2.3
2.8
Figure 9 shows how four clustering models—
K-means, FCM, NCM, and RNCM-FFSA- handle
the degree of imbalance parameter to distribute
load among computing resources fairly. KMC
consistently shows the highest degree of imbalance,
indicating its limitations in addressing class
imbalance issues, with values ranging from 0.39 to
0.66 as the number of tasks increases. FCM and
NCM exhibit relatively better results but still
demonstrate notable imbalance, with values
ranging from 0.38 to 0.60 and 0.35 to 0.57,
respectively. In contrast, RNCM-FSSA stands out
as the most effective method, consistently
achieving the lowest degree of imbalance across all
tasks, with values ranging from 0.28 to 0.10. The
RNCM-FFSA intelligently handles load balancing
between virtual servers in the cloud by expressing
every work request regarding the degree of
truthiness, falsity, and indeterminacy to combat
outliers, [29] [30], and adopting the Fish school
search algorithm to improve the uncertain condition
in resource selection to prevent overloading and
local optimum in search of essential resources. The
other three conventional clustering are due to the
random selection of initial centroids and searching
for resource availability using local search results
in early convergence.
Table 6 displays the execution time of four
different methods over increasing tasks. In
workload scheduling, execution time denotes the
time the scheduling algorithm takes to assign tasks
to resources and create an optimal schedule [31].
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
20
Volume 23, 2024
Fig. 10: Evaluation based on Execution Time
Figure 10 proves that the RNCM-FFSA takes
relatively little time to accomplish the work
schedule compared to the other three cluster-based
workload distributions in the cloud. As the number
of tasks increases, the execution times for the
clustering algorithms rise accordingly. From Figure
10, We can see that, At 250 tasks, RNCM-FSSA
exhibits the shortest execution time (1.4s), followed
by NCM (2.8s), FCM (3.2s), and KMC (3.4s).
However, as the task count reaches 1000, RNCM-
FSSA remains the fastest (2.8s), with NCM (7.5s),
FCM (10.2s), and KMC (12s) showing longer
execution times. Overall, RNCM-FSSA
consistently outperforms the other algorithms,
offering the most efficient clustering solution as the
number of tasks increases. This is because the
algorithm strategically selects initial cluster centers
using a measure of uncertainty, called membership
degree of indeterminacy. This means it's better at
handling tasks and resources with unclear or
changing requirements in the dynamic cloud
environment. The fish school search algorithm, a
part of RNCM-FFSA, significantly enhances the
clustering process, making it quicker and more
efficient compared to other algorithms like KMC,
FCM, and NCM.
5 Conclusion
To handle the issue of heterogenous environment-
based resource scheduling and load balancing, the
clustering technique can able to meet the demand
for cloud resources and reduce the screening
process-related overhead by forming clusters of
both the incoming job and virtual machines based
on their capacity in this paper a robust NCM-FSSA
algorithm is developed. The indeterminacy of the
incoming task request is very challenging for
conventional clustering. Hence in this work, the
generalization of the uncertainty theories known as
neutrosophic logic is used for clustering the outliers
in the resource scheduling scheme. The second
factor, in FCM, KCM, and NCM, the centroids are
selected arbitrarily, and the clustering process
begins, which results in early convergence in
scheduling, and their performance is directly
degraded. To overcome it, this work adopts the
metaheuristic model of fish schooling searching,
and its searching behavior is utilized for centroid
selection. The assessment of the proposed model
RNCM-FFSA provides the highest resource
utilization with the slightest degree of imbalance
and execution time in the workload scheduling of
cloud resources.
References:
[1] Gawali, M.B., Shinde, S.K. Task scheduling
and resource allocation in cloud computing
using a heuristic approach. J Cloud Comp 7,
4 (2018).
[2] Mohammad S Aslanpour, Adel N Toosi, Raj
Gaire, and Muhammad Aamir Cheema.,
Auto-scaling of Web Applications in Clouds:
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
21
Volume 23, 2024
A Tail Latency Evaluation. In 2020
IEEE/ACM 13th International Conference on
Utility and Cloud Computing, pp.186-195
[3] Alhaidari F, Balharith TZ, Enhanced Round-
Robin Algorithm in the Cloud Computing
Environment for Optimal Task Scheduling.
Computers. 2021; 10(5):63.
[4] Zheng, T., Wan, J, Zhang, J, Deep
Reinforcement Learning-Based Workload
Scheduling for Edge Computing. J Cloud
Comp 11, 3 (2022).
[5] Bezdan T., Zivkovic, M., Bacanin, N.,
Strumberger, I., Tuba, E., Tuba, M, Multi-
objective task scheduling in cloud computing
environment by hybridized bat algorithm,
Journal of Intelligent and Fuzzy Systems,
Volume 42, Issue 1, 2022, pp.411-423
[6] Amer Al Rahayfeh, Saleh Atiewi, Abdullah
Abuhussein, MuderAlmiani, Novel Approach
to Task Scheduling and Load Balancing
Using the Dominant Sequence Clustering and
Mean Shift Clustering Algorithm, Future
Internet, vol. 11, no. 109, pp.1-15, 2019.
[7] Geetha Megharaj, Dr. Mohan G. Kabadi,
Rajani, Deepa M, "FCM-LB: Fuzzy C Means
Cluster Based Load Balancing in Cloud",
International Journal of Innovative Research
in Science, Engineering and Technology, vol.
7, Special Issue 6, 2018.
[8] Khallouli, W., Huang J, Cluster resource
scheduling in cloud computing: literature
review and research challenges. J
Supercomput 78, pp.6898–6943 (2022).
[9] Hajvali, M., Adabi, S., Rezaee, A,
Decentralized and scalable hybrid
scheduling-clustering method for real-time
applications in volatile and dynamic Fog-
Cloud Environments. J Cloud Comp 12, 66
(2023).
[10] Allimuthu, U. "BAU FAM: biometric-
blacklisting anonymous users using fictitious
and adroit manager." J. Adv. Res. Dyn.
Control Syst., Volume 9 (2021) No. 12-
Special, pp.722-738
[11] R. Ruhin Kouser, T. Manikandan, A novel
clustering and optimal resource scheduling in
vehicular cloud networks using MKMA and
the CM-CSO algorithm, International
Journal of Communication Systems, Vol. 36,
Issue 5, 2023
[12] B. Fei, Elastic Resource Provisioning Using
Data Clustering in Cloud Service Platform,
Services Computing, vol. 15, no. 03,
pp.1578-1591, 2022.
[13] N. Akhtar and M. V. Ahmad, "A Modified
Fuzzy C Means Clustering Using
Neutrosophic Logic," 2015 Fifth
International Conference on Communication
Systems and Network Technologies, Gwalior,
India, 2015, pp.1124-1128.
doi: 10.1109/CSNT.2015.164
[14] H. A. Nugroho, M. Rahmawaty, Y. Triyani
and I. Ardiyanto, "Neutrosophic and fuzzy C-
means clustering for breast ultrasound image
segmentation," 2017 9th International
Conference on Information Technology and
Electrical Engineering (ICITEE), Phuket,
Thailand, 2017, pp.1-5.
doi: 10.1109/ICITEED.2017.8250453
[15] E. Rashno, A. Rashno and S. Fadaei, "Fluid
Segmentation in Neutrosophic Domain,"
2019 5th Iranian Conference on Signal
Processing and Intelligent Systems (ICSPIS),
Shahrood, Iran, 2019, pp.1-5.
[16] Gaber, Tarek, Gehad Ismail, Ahmed Anter,
Mona Soliman, Mona Ali, Noura Semary,
Aboul Ella Hassanien, and Vaclav Snasel.
"Thermogram breast cancer prediction
approach based on neutrosophic sets and
fuzzy c-means algorithm." In 2015 37th
Annual International Conference of the IEEE
Engineering in Medicine and Biology Society
(EMBC), pp.4254-4257. Ieee, 2015.
doi: 10.1109/EMBC.2015.7319334
[17] X. Lei, X. Yang and F. -X. Wu, "Artificial
Fish Swarm Optimization Based Method to
Identify Essential Proteins," in IEEE/ACM
Transactions on Computational Biology and
Bioinformatics, vol. 17, no. 2, pp.495-505, 1
March-April 2020.
doi: 10.1109/TCBB.2018.2865567
[18] M. Abdel-Basset, G. Manogaran, A. Gamal,
F. Smarandache, A group decision making
framework based on neutrosophic TOPSIS
approach for smart medical device selection.
Journal of medical systems, vol.43(2) (2019),
pp.38-54.
[19] Saha, Abhijit, Smarandache, Florentin, Said,
Broumi, Neutrosophic Soft Sets Applied on
Incomplete Data, Neutrosophic Sets and
Systems, Vol.32, pp.282-293, 2020.
[20] Bova V, Kuliev, E, Rodzin S, Prediction in
Intellectual Assistant Systems Based on Fish
School Search Algorithm. Izv. Sfedu Eng.
Sci, 2, pp.34-47, (2019).
[21] Alhaqbani, Amjaad Kurdi, Heba Hosny,
Manar. Fish-Inspired Heuristics: A Survey of
the State-of-the-Art Methods. Archives of
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
22
Volume 23, 2024
Computational Methods in Engineering. 29,
(2022).
[22] M. Habibi, A. Ayatollahi, N. Dallalazar and
A. Kermani, "Lumen boundary detection
using neutrosophic c-means in IVOCT
images," 2019 5th Conference on Knowledge
Based Engineering and Innovation (KBEI),
Tehran, Iran, 2019, pp.599-604.
doi: 10.1109/KBEI.2019.8734946
[23] X. Wang and J. Zhao, "A Novel
Neutrosophic Image Segmentation Based on
Improved Particle Swarm Optimization
Fuzzy C-Means Algorithm," 2019 18th
International Symposium on Distributed
Computing and Applications for Business
Engineering and Science (DCABES), Wuhan,
China, 2019, pp.80-83.
doi: 10.1109/DCABES48411.2019.00027
[24] S. Ashour, C. Du, Y. Guo, A. R. Hawas, Y.
Lai and F. Smarandache, "A Novel
Neutrosophic Subsets Definition for
Dermoscopic Image Segmentation," in IEEE
Access, vol. 7, pp. 151047-151053, 2019.
doi: 10.1109/ACCESS.2019.2946762
[25] J. M. Mathew, Surya S.R and P. Simon,
"Automatic cloud detection based on
neutrosophic set in satellite images," 2013
International Conference on Control
Communication and Computing (ICCC),
Thiruvananthapuram, India, 2013, pp.210-
215. doi: 10.1109/ICCC.2013.6731652
[26] R. Sivanandan and J. Jayakumari, "An
Improved Ultrasound Tumor Segmentation
Using CNN Activation Map Clustering and
Active Contours," 2020 IEEE 5th
International Conference on Computing
Communication and Automation (ICCCA),
Greater Noida, India, 2020, pp.263-268.
doi: 10.1109/ICCCA49541.2020.9250909
[27] W. F. Halawa, S. M. Darwish and A. A.
Elzoghabi, "Cotton Warehousing
Improvement for Bale Management System
Based on Neutrosophic Classifier," in IEEE
Access, vol. 9, pp. 159413-159420, 2021.
doi: 10.1109/ACCESS.2021.3126790
[28] Y. Zhu and H. Gao, "Improved Binary
Artificial Fish Swarm Algorithm and Fast
Constraint Processing for Large Scale Unit
Commitment," in IEEE Access, vol. 8,
pp.152081-152092, 2020. doi:
10.1109/ACCESS.2020.3015585
[29] C. Wu, X. Fu, J. Pei and Z. Dong, "A Novel
Sparrow Search Algorithm for the Traveling
Salesman Problem," in IEEE Access, vol. 9,
pp.153456-153471, 2021.
doi: 10.1109/ACCESS.2021.3128433
[30] C. J. A. Bastos-Filho, W. A. S. Silva and L.
R. M. Lira, "Comparing Meta-heuristics for
AdaBoost Training Applied to Platelets
Detection," in IEEE Latin America
Transactions, vol. 12, no. 5, pp.942-950,
Aug. 2014. doi: 10.1109/TLA.2014.6872910
[31] S. Tuo, H. Chen and H. Liu, "A Survey on
Swarm Intelligence Search Methods
Dedicated to Detection of High-Order SNP
Interactions," in IEEE Access, vol. 7, pp.
162229-162244, 2019.
doi: 10.1109/ACCESS.2019.2951700.
Contribution of Individual Authors to the
Creation of a Scientific Article (Ghostwriting
Policy)
S. Yuvaraj Gandhi, carried out the article writing
and executing the Optimizing Workload
Scheduling in Cloud Paradigm. T. Revathi has
implemented and took survey about Workload
Scheduling.
S. Yuvaraj Gandhi and T. Revathi has organized
and executed the comptre expriments of Section 4.
Hence the The authors equally contributed in the
present research, at all stages from the formulation
of the problem to the final findings and solution.
Sources of Funding for Research Presented in a
Scientific Article or Scientific Article Itself
No funding was received for conducting this study.
Conflict of Interest
The authors have no conflicts of interest to declare.
Creative Commons Attribution License 4.0
(Attribution 4.0 International, CC BY 4.0)
This article is published under the terms of the
Creative Commons Attribution License 4.0
https://creativecommons.org/licenses/by/4.0/deed.e
n_US
WSEAS TRANSACTIONS on SYSTEMS
DOI: 10.37394/23202.2024.23.2
S. Yuvaraj Gandhi, T. Revathi
E-ISSN: 2224-2678
23
Volume 23, 2024
