A novel meta-heuristic approach for load balancing in cloud computing

Sagarika Reddi
Undergraduate Student
Computer Science and Engineering
Lovely Professional University, Phagwara

Abstract - A crucial task in cloud computing is Because the workload in the cloud is heterogeneous and
scheduling. The method used to assign work to the dynamic, scheduling is one of the core issues in cloud
available resources determines how well they perform computing. As a result, resource provisioning became
overall and in terms of consumption. Conventional necessary. Task scheduling and process scheduling are two
scheduling methods are simpler to use, but they don't categories under which scheduling may be classified. Task
work well for issues of increasing complexity. To impart scheduling is the process of allocating virtualized resources
effective scheduling approaches, meta-heuristic to a certain task for a specified period of time. Assigning the
algorithms rely on the use of operators like transition, sources to the interdependent tasks within the dispensed
assessment, and resolution. Meta-heuristic algorithms sources is the definition of workflow scheduling. The
are evaluated only on the basis of computing time, following metrics in cloud computing can influence
resource consumption, response rate, and scheduling scheduling: These are throughput, memory usage, CPU
rate. We provide a brief evaluation of current meta- usage, wait time, response time, and turnaround time. While
heuristic methods in this study. We also provide an cloud computing scheduling is an NP-hard problem, in this
approach for scheduling tasks. The proposed scheduling study we have explored almost all the different meta-
approach is similar to a scheduling algorithm that relies heuristic scheduling algorithms (both task scheduling and
mostly on Particle Swarm Optimization. workflow scheduling) because they have a quicker
convergence to the global solution. The remainder of the
Keywords –Cloud computing, Scheduling, Genetic algorithms, essay is structured as follows: Section 2 provides information
Scheduling algorithms, Task analysis, Computational modelling, on the literature review; Section 3 provides information on
particle swarm optimization, meta-heuristic approach, load balancing,
cloud computing, resource utilization, particle swarm optimization, motivation; Section 4 provides information on the objective;
Meta-heuristic scheduling algorithms . Section 5 provides information on the proposed algorithm;
Section 6 provides information on the simulation
environment and results; and Section 7 provides information
on the conclusion and future work.
I. INTRODUCTION

NIST defines cloud computing as "an architecture for

enabling widespread, convenient, on-demand community
access to a collective group of programmable computer
resources." Cloud sources may be made available to
customers as pay-as-you-go products. Three categories may
be used to group the cloud's main features. IaaS, PaaS, and
SaaS are these.

IaaS (Infrastructure-as-a-Service): By the usage of virtualized

frames, cloud customers in this service may be granted
access to resources like computation, storage, and network on Figure-1
demand. Amazon AWS is the key player in the market for
IaaS services. By using this service provider, consumers may Classification of scheduling in cloud computing.
save upfront costs for the physical infrastructure .

PaaS (Platform-as-a-Service): With the aid of the cloud

platform offered by this service, cloud customers will be able
to use the PaaS service to create new apps. Like Java
and.NET.

SaaS (Software-as-a-Service): We will be able to use the

software that the cloud platform provides on this service. Ex:
Email and Google Docs.
 II. Literature Survey Optimization and Particle Swarm Optimization. The criteria
on which they have concentrated in this set of rules are CPU,
Mezmaz M. The Genetic Algorithm, a scheduling approach
disc, and memory consumption. Also, based entirely on the
that focuses on minimizing makespan and power
records of the requests made with the help of the users, it
consumption inside the parallel applications in
forecasts the workload of any future requests.
Heterogeneous computing systems, is proposed as a bijective
model. The authors of this research employed a bijective Abdullahi M, Md AsriNgadi. provided a set of guidelines
island parallel model and ma ulti-begin parallel version that concentrate on cutting the makespan and improving
tominimizee makespan and caloric consumption. machine usage. They have created an entirely new set of
PSO-based guidelines for symbiotic organisms. This method
Pandey S. provided a set of scheduling guidelines that use the
has been used in the cloudSim toolbox.
PSO algorithm to schedule the cloud applications while
taking the cost of computation and data transfer into account.
Delavar A, and Aryan Y. provided a set of guidelines for
The authors applied this set of guidelines on workflows,
compared their results with the best resource selection scheduling cloud-based processes. This algorithm's
method, and found that the cloud provided outstanding goals are to shorten the makespan, balance the device's
workload distribution and computational value reduction. burden, and increase speedup ratio. In order to get the
best results for lowering the makespan and load
Kaur S, and Verma A. developed a novel Genetic Algorithm balancing among the machine assets, the set of rules
that may be utilized to schedule tasks in the cloud efficiently
was modeled after the Genetic Algorithm (GA) and
by reducing execution time and execution cost. They utilized
the JGAP to bring these rules into effect (Java primarily used mutation and crossover parameters.
based Genetic Algorithm Package).
Verma A and Kaushal S. suggested a genetic algorithm that
Rodriguez MA, Buyya R. provided a heuristic set of maps the process to cloud resources using a heuristic-based
guidelines that combine aids for workload provisioning and approach. This algorithm's goal is to map the process to
scheduling in IaaS clouds. The execution price and makespan cloud resources by reducing the execution cost while taking
were the measures that the writers focused on. This paper's the closing date limitations into account.
goal is to keep execution costs as low as possible while still Yassa S. offered a multi-objective version to plan the cloud-
adhering to deadline restrictions. They employed PSO as the based workflow applications while reducing power
optimization strategy and built this set of rules using the consumption. Dynamic Voltage Frequency Scaling (DVFS),
CloudSim toolkit to mimic this process. which runs numerous voltages by forgoing the clock
frequencies, is a technique used to cut down on the amount of
Tawfeek MA. a scheduling technique that shortens the power needed while still maximaximizingkflow scheduling.
duration of the provided commitments in the cloud
environment has been suggested. This set of guidelines was Mandal T, Acharyya S. To reduce the total processing time
modeled using the ACO (Ant Colony Optimization) method of virtual machines, annealing, the Firefly method, and the
and put into practice using the cloudSim toolbox. cuckoo search algorithm were all simulated. The authors
discovered that the Firefly Algorithm delivers the best overall
Li K. suggested a scheduling technique that focuses on performance in scheduling out of the three simulations.
system load balancing and shortens the makespan of the
specified task collection. The technique was applied to the Devipriya S and Ramesh C. suggested a RASA-stepped-
cloudSim toolkit and was modeled using an Ant Colony forward ward min-max algorithm. The suggested set of rules
Based Total Optimization approach. As compared to focuses on one metric, namely Makespan, whereas the real
traditional algorithms like FCFS and ACO, the suggested min-max method prioritizes overall execution time.
Load Balancing Ant Colony Optimization approach
effectively balances the whole load of the device by treating Bittencourt LF, Madeira ERM. presented a new hybrid set of
the load as simultaneous impartial duties. guidelines for cloud computing workflow scheduling. The
main goal of the authors is to decrease the processing costs in
Shojafar M. By considering the factors of execution time and
hybrid clouds by choosing how many resources to admit into
price, Shojafar M. suggested a hybrid model that combines
the private cloud and then effectively scheduling the
fuzzy concepts and evolutionary algorithms to stabilize the
resources in the public cloud by adhering to deadline limits.
weight and shorten the makespan within the cloud. Based on
standard genetic algorithms and fuzzy logic, this method was
III. Motivation
modified. The clouds toolbox was used to execute this
algorithm.
Many logical studies on distributed computing have focused
Cho KM. suggested a hybrid approach that effectively on the task planning skill of execution. Task scheduling
schedules the cloud's resources using both Ant Colony focuses on accurately matching duties to the right resources.
In distributed computing, finding a perfect connection is seen
as an NP-complete task. Every scheduling computation is Round Robin (RR): The round-robin scheduling technique is
reliant on at least one method. The most important methods a preemptive first come, first serve CPU. Scheduling method
or ends usually used are time, cost, energy, quality of service that cycles over each process in the ready state to give it the
(QoS), and adaptability to internal failure. For cloud CPU for a set amount of time (turn by turn). The first
frameworks, a chaotic scheduling architecture based on scheduling method, it is mostly employed for multitasking.
heuristic algorithms, such as Min-Min, Max-Min, and
Heterogeneous Earliest Finish Time (HEFT) computations,
has been developed. Moreover, incredible metaheuristic task
scheduling algorithms that generate optimal schedules have
been developed. They include the scheduling algorithms
based on the Genetic Algorithm (GA), Molecule Swarm
Optimization (PSO), and Ant Colony Enhancement (ACO).
Regular task scheduling algorithms like Shortest-Job-First
IV. Objective
(SJF), Round Robin (RR), and First-Come-First-Serve
(FCFS), Multilevel line booking (MQ), Max-Min, and Min-
Min achieved outstanding results over years in various PC Assignment scheduling for the computing quarter was
framework types, but they consistently suffer from serious prepared using a variety of meta-heuristic techniques. The
issues like increased waiting times in RR and FCFS and viability of the meta-heuristic algorithm is proven using two
starvation in SJF and Max-Min. One of the major challenges techniques. The primary system is called "exploitation," and
in cloud task scheduling is the starvation problem, which it takes advantage of the unusual arrangement from a few
causes a project to sit tight for at least one of its declared previous impacts. The second method is "exploration," which
assets for an extended period of time. Starvation can be looks into new areas of the arranging area. Genetic
purposefully caused by using tactics for a rejection of- Algorithms (GA), Particle Swarm Optimization (PSO), Ant
corporation ambush, asset spills, and scheduling counted Colony Optimization (ACO), League Championship
omissions, among other methods. For instance, if a poorly Algorithm (LCA), and a few other algorithms are just a few
structured device for doing multiple jobs dependably examples of the algorithms that make up this category. Part 2
alternates between the initial two activities but the third task already has all the information on existing meta-heuristic
never has the opportunity to run, then indicates that the third methods, i.e., the literature survey. Nevertheless, the Lion
task is voracious. Optimization Algorithm, a different set of advancement
Similar to other algorithms, the major considerations in the principles, has been offered here to schedule other tasks. The
case of meta-heuristics are the execution time, effective way in which lions live and their social relationships are
resource use, and connections to several other performance factors. The "Lion Pride Optimizer" method was developed
measures. In this study, we are driven by nature and proposed by Wang [26] and was inspired by a number of lion-related
a new task scheduling method for cloud environments using characteristics. It depends on the lions fighting and mating.
the lion optimization algorithm (LOA), which was previously The "Lion's Algorithm that Rajkumar proposed was a series
proposed as a nature-driven algorithm. This algorithm is of laws that rely on mating, neighbourhood defense, and
reliant on the unusual lifestyle and pleasing behaviors of close-by takeover . The Lion Optimization Calculation
lions. (LOA), which was predicted by Yazdani and Jolai [27], is no
longer as enticing as the previous technique. It was fueled by
Genetic Algorithm (GA): One evolutionary method used to mimicking the solitary lifestyle and amicable behaviors of
swiftly solve difficult problems is the genetic algorithm lions, such as chasing, area stamping, and mobility, as well
(GA). In order to schedule tasks on cloud computing with a as the distinctive lifestyles of travelers and, more
variety of priority queues, this research suggests a parallel
GA using a MapReduce architecture.. importantly, resident lions, such as mating and combat .

The main goal of this investigation is to design a task-

Molecule Swarm Optimization (PSO): A stochastic
scheduling approach for distributed computing using the
optimization approach called PSO is based on how swarms
LOA idea in order to reduce the total completion time of the
move and function. The idea of social interaction is utilized
project at the cloud assets (make span) and to effectively
in PSO to resolve a conflict. It employs a swarm of particles
utilize the resources in cloud environments.
(agents) that move about the search space in quest of the best
answer. V. Proposed Algorithm

Ant Colony Enhancement (ACO): Ant colony optimization

(ACO) is an optimization algorithm that uses a probabilistic
approach to solve computation-related issues and locate the
best path using graphs.
A. Existing Lion Optimization Algorithm:
2) LOA Process:
1) Inspiration:
The most socially inclined of all naturally occurring 1. Parameterize
feline species, lions exhibit abnormally high levels of
involvement and opposition. Lions are incredibly convincing
animals, which is a result of their strong sexual Set value of parameters, Number of Lions.
dimorphbehaviorhaviour and appearance. The lion is a wild
felid with two different social groups: wanderers and Create an arbitrary solution for each Lion.
inhabitants. People reside in a community known as pride. A
pride of lions normally consists of five or more females, their
young of both sexes and at least one or two adult males. As Set up Prides and Nomad lions.
they get explicitly wider, boys who are too young for the
world are prohibited from making their entrance.
2. For every Pride
The second authoritative characterized behavior, known as
wanderers, is characterized by occasional movement, either
in groups or on their alone. When related males are spared A number of females are elected arbitrarily for
from their maternal pride, sets tend to be more reasonable. hunting.
Keep in mind that lions can change their lifestyles;
inhabitants can become travelers and vice versa. The LOA
Remained females shifted coastwards to the
was developed based on simulations of lion behaviors such as
preferred place of the region.
pursuing, mating, and barrier. Two authoritative customs are
followed by lions: occupant cobehaviorsnduct and
trtravelingonduct. Peopltravelinge Liv lives in groups known Each male roams in %R of territory.
as points of pride. A resident lion might change into a
traveler, and vice versa. In the LOA, the underlying populous
is generated haphazardly throughout the space of collections, %M of female mate with one or extra resident adult
each of which is referred to as a "Lion." The remaining lions males.
in the population are residents, and (%N) of them are
randomly selected to be wandering lions. Residents are The weakest males are driven out of pride and
randomly divided into (P) pride. Every pride of lions has become nomads.
(%S) female lions and (%S) male lions. For nomad lions,
however, this fraction is the opposite. The best-visited
position for each lion is the one that has been visited the most 3. For every Nomad lion
times in previous cycles, and during the streamlined progress,
this position is regularly updated. A pride area in LOA is a Both males and females travel haphazardly in the
region that has every outstanding tourist attraction. Several search space
randomly selected females from each pride begin chasing. To
encircle and capture their target, hunters advance toward
them. The remaining ladies advance in different directions %M of female’s mate with only one male.
around the area. Male lions roam the region in pride. In
pride, females’ mate with one or more resident males. As Nomad males hit feelings of pride.
young men reach adulthood and reach a capability that is not
quite that of resident males, they are rejected from their
maternal pride and become travelers. Moreover, a male or 4. For every pride
female explorer lion wanders subjectively in the area of the
pursuit to find a better location (arrangement). In the unlikely
%I of females roam from pride and turn into
event that the strong traveling male attacks the resident male,
nomads.
the traveler lion drives the resident male outside of the pride.
The male tourist transforms into the resident lion.
Throughout their evolution, some inhabitant female lions 5. Perform
switch from one pride to the next, change their way of life to
become travellers, or do the opposite and join pride of
wandering female lions. The weakest lion will perish or be Every gender of nomad lion is looked after based on
killed due to a variety of factors, including malnutrition and their fitness cost.
competition. The aforementioned process continues until the
halting condition is met.
 The finest women are elected and allotted to pride B. Problem Statement with Proposed Algorithm
filling vacant spaces. (MLOA):
By researching task scheduling in the cloud using meta-
Nomad lions with the least health cost might be heuristic optimization algorithms, we discovered certain
eliminated based on the max authorized quantity of problems with task execution time. From that study, we
every gender. discovered LOA, a novel optimization technique that
provides better results and provides the best value or
6. If the execution criterion is not fulfilled, then. solution. As a result, we were inspired and attempted to
modify the optimized method to provide a more optimal
result. The ultimate goal of the suggested technique is to
Go to step 2. schedule tasks with a minimum Makespan and efficient
standard resource usage. In order to demonstrate the
superiority of our proposed method, we also compared it
Return the best solution.
against PSO, the most well-known meta-heuristic work
scheduling algorithm used in cloud computing.
3) Flow Chart of LOA Process: The Modified Lion Optimization Algorithm (MLOA), a
proposed job-scheduling algorithm based on the principle of
LOA, goes through the following steps:
1. MLOA Process:

Input: Number of tasks, Number of machines.

Output: Optimized solution/best solution w.r.t

Makespan and Average Resource Utilization.

Steps:

Parameterize

Set the value of parameters, machines, and

Iterations.

pd =10; % Problem dimension (number of decision

variables)

N=20; % population size

AP= 0.1; % Awareness probability

fl= 2; % Flight length (fl)

[x l u] =init(N,pd); % Initialization
Figure-2
%Position of the Lion in the space
Flow chart of LOA Process.
Set an upper limit and a lower limit.

Calculate the Fitness function with respect to

initialized parameters.
 Do memory initialization. VI. Simulation Environment and Result

Calculate the Fitness of memory positions. A. MATLAB:

A high-performance language for technical computing is
Generate random candidate lions by taking called MATLAB. It combines computation, concept, and
maximum iterations. (tmax =5000) programming in a user-friendly environment where issues are
expressed using well-known mathematical notation.
Arithmetic and computing, algorithm development,
for t=1: tmax modeling, simulation, and prototyping, data analysis,
exploration, and visualization, scientific and technical
num = ceil (N*rand(1, N)); graphics, and application development, including graphical
user interface creation, are examples of typical usage.
The fundamental data element of MATLAB, an
Evaluate the fitness of new solutions. interactive structure, is an array that doesn't need to be
dimensioned. In comparison to the time it would take to build
Update the memory and position. a program in a scalar non-interactive language like C or
FORTRAN, this enables you to tackle many technical
Found the best solution. computing issues, particularly ones involving matrix and
vector formulations.
For both running our algorithm and determining
Evaluate.
outcomes, we used MATLAB.

Sort all the values till we get our best optimized B. Flow Chart of Simulation Environment:
final value.

Terminate once we get our best value Otherwise go

to step 2.

2. Flow Chart of MLOA Process:

C. Experimental Metrics:
Make span and average consumption are the performance
characteristics considered for the planned job scheduling.

1.Makespan:

By displaying the ultimate task's completion point,

Makespan determines the fastest culminating time. The most
well-known improvement requirement for work scheduling is
limiting the make span. The subsequent equation allows for
its intentional use.
Figure 4:
Makespan comparison of PSO and MLOA
Makespan=maxtask1(FnTime)

Where Fn Time shows the concluding time of job i.

3. Average Utilization:
Another important criterion for cloud providers is asset
utilization, which involves keeping assets as busy as is
reasonably possible to maximize benefits. The average usage
is calculated using the formula below:

Average utilization=∑ni=1Execution time of resource 

____________________________
iMakespan∗n

Where n is the number of resources.

D. Experimental Result:
Figures 4 and 5 show how the PSO and MLOA are
compared. We have found from the comparative analysis that
MLOA takes less time to execute than PSO, and the
suggested MLOA procedure is more resource efficient than
PSO. By combining multiple heuristics and meta-heuristics
algorithms, hybrid techniques may be created. We have taken
the Average response time (ms) metric into consideration
when comparing many GA hybrid algorithms.

Figure 5
Average Utilization comparison of PSO and MLOA
Table I Comparison of Average Response Time of 3. Keshavarzi, A., Khodabakhshi, M., & Keshavarzi, M.
Different Algorithms (2020). A novel load balancing approach based on the firefly
algorithm in cloud computing. Journal of Ambient
Intelligence and Humanized Computing, 11(6), 2347-2356.
4 .Elshinawy, H. M., El-Desouky, A. I., & Abdel-Badeeh, M.
S. (2018). A novel load balancing algorithm for cloud
computing based on cuckoo search optimization. Journal of
Ambient Intelligence and Humanized Computing, 9(5),
1627-1640.
5. Shoukath, R., & Ramakrishnan, S. (2021). A novel meta-
heuristic approach for load balancing in cloud computing
using ant colony optimization. Computers & Electrical
Engineering, 92, 107001 .
6. M. Young, The Technical Writer’s Handbook, Mill
Valley, CA:University Science, 1989.
7. Y. Yorozu, M. Hirano, K. Oka and Y. Tagawa,
"Electron spectroscopy studies on magneto-optical
media and plastic substrate interface", IEEE Translated
J. Magn. Japan, vol. 2, pp. 740-741, August 1987, 1982.
VII. Conclusion and Future Scope
8. H. Simpson, Dumb Robots, Springfield:UOS Press,
The undertaking scheduling method has been expanded pp. 6-9, 2004.
using several meta-heuristic techniques. A new cloud 9. M. King and B. Zhu, "Gaming strategies", Path
assignment-scheduling algorithm was proposed in our paper,
Planning to the West, vol. II, pp. 158-176, 1998.
mostly based on the concept of LOA, a freshly developed set
of guidelines based on the lions' way of life. The suggested 10. B. Simpson et al., Title of paper goes here if known.
collection of rules was contrasted with the set of PSO meta- 11. J.-G. Lu, "Title of paper with only the first word
heuristic rules, which were mostly focused on performance capitalized", J. Name Stand. Abbrev..
indicators. We compared the MLOA and PSO schemes and
12. H Krishnaveni and Sinthu Janita Prakash V,
analyzed the effects of various work sizes and device
diversity. We found that MLOA performed better in terms of "Execution time based sufferage algorithm for static task
make span and general usage when compared to PSO. This scheduling in cloud", Advances in Big Data and Cloud
suggested algorithm may also be improved or changed to Computing, pp. 61-70, 2019, [online] Available:
produce more ideal results in the future.
In the future, we intend to enhance the suggested https://doi.org/10.1007/978-981-13-1882-5_5.
approach by combining it with other meta-heuristic 13. N. Almezeini and A. Hafez, Task scheduling in
techniques and performance measures. Also, we are cloud computing using lion optimization algorithm,
concentrating on lowering the cost of employing resources to algorithms, 2017.
complete tasks by creating a pricing prototype.
15. Maziar Yazdani and Fariborz Jolai, "Lion
VIII. References optimization algorithm (LOA): a nature-inspired
metaheuristic algorithm", Journal of computational
1. Liu, X., Huang, Y., & Zhang, Y. (2020). A novel meta- design and engineering, vol. 3, no. 1, pp. 24-36, 2016.
heuristic approach for load balancing in cloud computing. 16. M. King, B. Zhu and S. Tang, "Optimal path
IEEE Access, 8, 107401-107408. planning", Mobile Robots, vol. 8, no. 2, pp. 520-531,
2. Xu, Y., He, X., Zhou, H., & Zhang, Y. (2020). A novel March 2001
load
balancing approach based on improved particle swarm 17. S Abdi, Motamedi SA and S. Sharifian, "Task
optimization algorithm in cloud computing. Journal of
scheduling using modified PSO algorithm in cloud
Ambient Intelligence and Humanized Computing, 11(11),
computing environment", International Conference on
5045-5054.
Machine Learning Electrical and Mechanical
Engineering, pp. 1-5, 2014.
18. S Zhan and H. Huo, "Improved PSO-based task
scheduling algorithm in cloud computing", Journal of
Information and Computational Science, vol. 9, no. 13,
pp. 3821-9, 2012.
19. Xue SJ and W. Wu, "Scheduling workflow in cloud
computing based on hybrid Particle Swarm Algorithm",
Indonesian Journal of Electrical Engineering and
Computer Science, vol. 10, no. 7, pp. 1560-6, 2012,
[online] Available:
https://doi.org/10.11591/telkomnika.v10i7.1452.
20. I. Gupta, "A hybrid Metaheuristic approach for load
balanced workflow scheduling in IaaS cloud",
International Conference on Distributed Computing and
Internet Technology, pp. 73-89, 2019, [online]
Available: https://doi.org/10.1007/978-3-030-05366-
6_6.