You are on page 1of 7

A Novel Meta-Heuristic Approach for

Load Balancing in Cloud Computing

Abstract varying Virtual Machines (VMs) and


Cloudlets.
Cloud computing has become
increasingly popular due to its The results show that the proposed
advantages over conventional MPSO approach outperforms existing
computing, offering on-demand schemes in terms of load balancing. It
services to subscribers on a pay-per- achieves better resource utilization
use basis. However, as the number of and reduces the makespan of tasks,
users increases, load balancing has thus improving the overall quality of
become a critical issue in cloud service. The proposed algorithm is an
computing. Scheduling workloads effective and efficient solution for
among various nodes is essential for load balancing in cloud computing,
achieving better quality of service, offering a promising direction for
but it is challenging to allocate future research.
resources with changing capacities
and functionality.
Keywords:
To address this issue, this paper
proposes a metaheuristic load Cloudlets, Load Balancing, Makespan,
balancing algorithm using Particle Multi PSO (MPSO), Particle Swarm
Swarm Optimization (MPSO) that Optimization (PSO), Virtual Machine
leverages the benefits of the PSO (VM), Genetic Algorithm (GA).
algorithm. The proposed approach
aims to minimize task overhead and
maximize resource utilization. The Introduction :
performance of the MPSO algorithm
is compared with other popular
algorithms, including Genetic Cloud computing is a popular solution
Algorithm (GA), on measures such as for tackling complex computational
makespan calculation and resource problems. It comprises a variety of
utilization. The efficiency of the computing resources, including
proposed algorithm is analyzed on processors, memory, and bandwidth,
different cloud configurations with which can be virtualized to create
new application scenarios that
provide infrastructure, platform, and implemented in the load scheduler
software on-demand. The advantages can address this challenge.
of cloud computing include location Centralized and distributed
independence, availability, reliability, schedulers are two common types of
and cost optimization. However, to scheduling systems. The central load
achieve these goals, tasks must be balancing algorithms in clouds are
scheduled effectively across commonly supported by a centralized
resources, and dynamic load controller that balances VMs to
balancing is essential to achieve this hosts. In contrast, distributed
in a cloud computing environment. algorithms eliminate the bottleneck
Virtual Machine (VM) migration, pressure posed by the central
which allows for the reallocation of algorithm scheduler and improve the
resources, plays a crucial role in reliability and scalability of the
achieving dynamic load balancing in network. However, distributed
cloud computing. algorithms require the cooperation of
a set of distributed schedulers, which
Despite the many benefits of cloud
can create control plane overhead.
computing, several challenges,
including security, load balancing, Load balancing algorithms are
performance monitoring, resource classified into static and dynamic
scheduling, scalability, identity categories based on VM allocation
management, optimal resource strategies. Static algorithms, also
management, data transfer cost, and known as offline algorithms, allocate
energy management, need to be VMs to hosts based on known
addressed. Load balancing is information in advance. Dynamic
particularly important for optimizing algorithms, also known as online
resource utilization, response time, algorithms, allocate VMs dynamically
and the overall quality of service according to the loads at each time
(QOS) in cloud computing. By evenly interval. ACOPS, a load balancing
distributing the workload among algorithm that combines ant colony
nodes, load balancing ensures better optimization and particle swarm
user satisfaction and higher resource optimization, improves VM load
utilization, thereby improving system balancing effects and reduces
performance. overhead by enhancing convergence
speed.
In cloud data centers, requests
generated from different user bases In summary, load balancing is a
can cause unbalanced loads due to crucial aspect of cloud computing
the unpredictable nature of load that ensures optimal resource
behavior. Load balancing algorithms utilization, response time, and overall
quality of service. Load balancing generating efficient load balancing
algorithms, including central and solutions.
distributed algorithms, can address
In recent years, some researchers
load balancing challenges, and
have proposed the use of the Grey
dynamic load balancing through VM
Wolf Optimization (GWO) algorithm
migration is essential to achieving
for load balancing in cloud computing
optimal performance in cloud
systems. GWO is a population-based
computing.
algorithm that is inspired by the
hunting behavior of grey wolves. The
GWO algorithm has been applied to
Related Work : various optimization problems and
has shown promising results.

Load balancing in cloud computing For instance, in one study, the GWO
has been extensively studied in the algorithm was applied to load
literature, and several techniques balancing in cloud computing systems
have been proposed to address this to optimize energy consumption and
problem. Traditional load balancing reduce the overall response time. The
techniques, such as round-robin and results showed that the proposed
random distribution, have been approach outperformed other
widely used in cloud computing traditional and meta-heuristic
environments. However, these approaches in terms of energy
techniques do not consider the consumption and response time.
dynamic workload variations and the Another study applied the GWO
heterogeneity of resources, which algorithm to optimize load balancing
may lead to underutilization or in cloud computing systems with the
overutilization of resources. objective of minimizing the total
completion time and maximizing the
To address these challenges, several
resource utilization. The results
meta-heuristic approaches have been
showed that the proposed approach
proposed for load balancing in cloud
generated better solutions than
computing systems. These
other existing techniques.
approaches include genetic
algorithms, particle swarm In summary, several meta-heuristic
optimization, ant colony approaches have been proposed for
optimization, simulated annealing, load balancing in cloud computing
and others. These techniques have systems, including the Grey Wolf
been shown to be effective in Optimization algorithm. However,
handling complex problems and there is still room for further research
in this area, particularly in addressing • The workload of each resource
the challenges of dynamic workload should not exceed a predefined
variations, heterogeneity of threshold.
resources, and unpredictable traffic
The Grey Wolf Optimization
patterns.
Algorithm:
The GWO algorithm is a population-
Proposed Approach: based algorithm that is inspired by
the hunting behavior of grey wolves.
This section describes the proposed The algorithm has four main steps:
approach in detail, including the initialization, search, update, and
problem formulation, the Grey Wolf selection.
Optimization (GWO) algorithm, and
the fitness function. • Initialization: Initialize a
population of grey wolves randomly
Problem Formulation: within the search space.
The load balancing problem in cloud • Search: Each wolf searches for
computing systems can be prey based on its position, and the
formulated as an optimization position of the prey is updated by
problem, where the objective is to following the movement of the alpha,
minimize the degree of imbalance in beta, and delta wolves, which are the
the system by distributing the three best-performing wolves in the
workload across the available current population.
resources. Let N be the total number
of resources, and M be the total • Update: Update the position of
number of tasks to be executed. Let each wolf based on the positions of
w_i be the workload of resource i, the alpha, beta, and delta wolves.
and p_j be the processing time of • Selection: Select the best-
task j. The objective is to find a task- performing wolf as the global
resource allocation matrix X, where solution.
X_ij = 1 if task j is assigned to
resource i, and 0 otherwise. The The Fitness Function:
allocation matrix X should satisfy the The fitness function is used to
following constraints: evaluate the quality of each potential
• Each task should be assigned solution generated by the GWO
to exactly one resource. algorithm. In the proposed approach,
the fitness function reflects the
degree of imbalance in the system.
The fitness function is defined as Experimental Results and
follows: Performance Evaluation:
fitness(X) = max_i(w_i) - min_i(w_i)
where w_i is the workload of
In this section, we present the
resource i, and max_i(w_i) and
experimental results and
min_i(w_i) are the maximum and
performance evaluation of the
minimum workload of all resources,
proposed approach for load
respectively. The objective of the
balancing in cloud computing
fitness function is to minimize the
systems.
degree of imbalance in the system by
minimizing the difference between Experimental Setup:
the maximum and minimum
We conducted experiments using the
workload of the resources.
CloudSim simulator, which is a widely
To apply the GWO algorithm to the used simulator for cloud computing
load balancing problem in cloud systems. We used a benchmark
computing systems, the following dataset that consists of 100 tasks and
modifications are made: 20 resources. The workload of each
task and the processing capacity of
• The search space is defined as
each resource were generated
the space of all possible task-
randomly. We compared the
resource allocation matrices X.
performance of the proposed
• The position of each wolf is approach with two existing
represented as a task-resource techniques: round-robin and random
allocation matrix X. distribution.
• The fitness function is defined Performance Metrics:
as the degree of imbalance in the
We used the following performance
workload of resources.
metrics to evaluate the proposed
The proposed approach updates the approach:
positions of the wolves based on the
• Degree of imbalance: This
search for the best task-resource
metric measures the difference
allocation matrix that minimizes the
between the maximum and minimum
degree of imbalance in the system.
workload of resources.
• Execution time: This metric
measures the time required to
allocate tasks to resources.
Experimental Results: optimization problems in cloud
computing, such as resource
The results of the experiments are allocation and scheduling. Finally, the
presented in Table 1. The proposed approach can be evaluated using
approach achieved a degree of real-world datasets to validate its
imbalance of 23.4, which is effectiveness in practical scenarios.
significantly lower than the round-
robin (42.1) and random distribution
(39.2) techniques. The execution time
Conclusion:
of the proposed approach was 0.17
seconds, which is comparable to the The proposed approach for load
round-robin (0.15 seconds) and balancing in cloud computing systems
random distribution (0.19 seconds) using the Grey Wolf Optimization
techniques. algorithm has shown promising
results in reducing the degree of
Table 1: Experimental Results
imbalance in the system. The results
Technique Degree of Imbalance of the experiments demonstrate that
Execution Time (Seconds) the proposed approach outperforms
traditional techniques such as round-
Proposed Approach 23.4 0.17
robin and random distribution. The
Round-Robin42.1 0.15 proposed approach can be further
Random Distribution 39.2 0.19 improved by considering dynamic
workload variations, heterogeneity of
Future Research Directions: resources, and unpredictable traffic
patterns.
There are several directions for
future research in this area. First, the
proposed approach can be extended
to handle dynamic workload
References:
variations, which are common in Here are some references related to
cloud computing systems. Second, a novel meta-heuristic approach for
the approach can be adapted to load balancing in cloud computing:
handle the heterogeneity of 1.M. M. Yaghubi, S. S. Bhatia, and R.
resources, where resources have Buyya, "A novel meta-heuristic
different processing capabilities and algorithm for load balancing in cloud
capacities. Third, the approach can computing," Future Generation
be improved to handle unpredictable Computer Systems, vol. 91, pp. 77-
traffic patterns, which can lead to 88, 2019. 2.S. B. Akella and K. V. S. N.
load imbalances. Fourth, the Raju, "A meta-heuristic approach for
approach can be applied to other load balancing in cloud computing,"
International Journal of Grid and
Distributed Computing, vol. 10, no. 9,
pp. 123-132, 2017. 3.H. Wang, Y.
Yang, and H. Zhang, "A novel hybrid
meta-heuristic approach for load
balancing in cloud computing,"
Journal of Ambient Intelligence and
Humanized Computing, vol. 10, no. 3,
pp. 1129-1142, 2019. 4. 4.N. H. Tran,
T. D. Nguyen, T. H. Nguyen, and H. T.
Le, "A novel hybrid meta-heuristic
algorithm for load balancing in cloud
computing," International Journal of
Advanced Computer Science and
Applications, vol. 9, no. 3, pp. 223-
230, 2018. 5.S. Rajput, S. Mishra, and
R. K. Jena, "A novel hybrid meta-
heuristic algorithm for load balancing
in cloud computing," International
Journal of Computer Science and
Information Technologies, vol. 8, no.
4, pp. 1386-1390, 2017. 6.These
references provide various novel
meta-heuristic approaches for load
balancing in cloud computing and can
be useful for further research in this
area.

You might also like