A Review of Particle Swarm Optimization in Cloud Computing

Abstract

To meet the ever-growing demand for the online computational resources, it is mandatory to have the best resource allocation algorithm to allocate the resources to its end users. For most of the Internet of Things applications, the destination for the generated data is the cloud. The data may be processed instantaneously, or it may be done afterward depending on the type of data and the applications which generated that data or depending on applications that consume this data and produce some analysis result. Many algorithms have been proposed in this area, few of them are the linear method, few used heuristics, few using artificial intelligence and machine learning, and few used the meta-heuristic approach. Out of all the available methods, meta-heuristic algorithms stand out. Particle swarm optimization (PSO) is a meta-heuristic powerful technique of optimization technique that concerns the finding of maxima or minima of functions in the possible region. This chapter provides a review and discussion on different variations of the PSO algorithm and also compares different PSO optimization algorithms.

Full text

93© Springer Nature Switzerland AG 2022
M. Moh et al. (eds.), Smart IoT for Research and Industry, EAI/Springer
Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-71485-7_5
Chapter 5
A Review ofParticle Swarm Optimization
inCloud Computing
DevarajVermaC, HarshvardhanTiwari, andMadhumalaRB
5.1 Introduction
Leasing the computational resources on demand is nothing new. It started in the late
1990s and is growing exponentially year by year. The resource allocation problem
is an NP-hard problem, and the time taken for allocating the resources plays a vital
role in dening efciency. Linear methods are good in allocating the resources, but
the time taken to allocate will increase exponentially as the number of clients
requesting for the resources increase. Though the articial intelligence and machine
learning algorithms are the best t, they suffer from the requirement that we need to
have a huge amount of computational power and internal memory to efciently
allocate the resources, and hence they are economically not a feasible solution.
They are even time efcient and require less computational power and internal
memory. The only disadvantage is they may converge at a local optimum, i.e., the
solution may not be the best. Here in this chapter, we discuss particle swarm intel-
ligence and its applications.
Cloud computing is a technology where the cloud user is able to get the demanded
cloud resources over the Internet, used to minimize the cost of computing [1]. Cloud
computing is an on-demand technology for proving the quality service to the end
users. Cloud computing is one of the popular options for people and business for
number of reasons such as cost savings, resource management, increased productiv-
ity, increased in energy savings, and also in speed. Cloud computing provides ser-
vices by which we can access the customized applications over the Internet and
allows users to learn how to congure applications. Cloud users can access resources
D.V. C · M. RB (*)
Department of Computer Science Engineering, Jain University, Bangalore, India
e-mail: rb.madhumala@jainuniversity.ac.in
H. Tiwari
CIIRC, Jyothy Institute of Technology, Bangalore, India

94
through the Internet from anywhere anytime, as long as they need resources without
worrying any maintenance of resources. The cloud computing refers to a network of
nodes that provide services to end users on demand over the Internet. In other words,
cloud computing means providing the resources at remote location. Cloud can
provide services on WAN, LAN, or VPN.
Users customized applications such as e-mail, video conferencing, and customer
relationship management (CRM) database applications. Cloud computing manipu-
lates, congures, and accesses the applications online. As shown in Fig.5.1, cloud
computing offers on-demand resource infrastructure platform for its users.
5.1.1 Cloud Computing
Cloud technology is a combination of both hardware and software and database
applications; users need not worry about the maintenance of the resources because
it is completely taken cared of by the cloud providers,; users can just apply their
technologies and they can work based on the demanded resources. A huge number
of applications are available over cloud.
Many service providers are providing maximum resources over cloud with QoS
policies. Security concerns are very high for all customer relationship management
applications, irrespective of the domain; it may be research domain, medical
domain, agricultural, military applications, and many more. Due to these
Organizations
Servers
Storage
Applications
Services
Individual
s
Individuals
Fig. 5.1 Cloud computing architecture
D.V.C et al.

95
tremendous applications over the cloud data center, optimization of the resources
over the cloud data center is a great challenge; one such key technology is the cloud
virtualization technology, where virtual machines are mapped to the physical
machines.
Many optimization techniques are proposed by many researchers, based on the
different methodologies, few used heuristic, and few used meta-heuristic where the
algorithms must be in a position to solve both single and multidimensional prob-
lems. These algorithms also solve continuous and discrete mathematical problems.
Figure5.2 shows the basic architecture of virtualization over the cloud data centers.
Infrastructure as a service (IaaS) is the platform where it provides infrastructure as
a service to the end users based on their need. Cloud virtualization allows the user-
customized applications. Cloud computing serves both software and hardware
applications based on the user demands. Only few services feasible cloud comput-
ing services and accessible to end users with less amount of time, cloud working
models. IaaS is the delivery of resources over the cloud to provide infrastructure
based on demand scalable service of the customer. IaaS provides fundamental
resources such as physical machines, virtual machines, storage, CPU, etc., usually
billed depending on the usage based on the type of service model used [2]. Platform
as a service (PaaS) provides the dynamic environment for applications, develop-
ment and deployment tools, etc. PaaS provides needed facilities which are required
to support the life cycle of delivering web-based applications over the cloud data
center services via the Internet.
Virtual
Machine
(VM)
APP A
Host
Operating
System(OS)
Virtual
Machine
(VM)
APP B
Host
Operating
System(OS)
Virtual
Machine
(VM)
APP C
Host
Operating
System(OS)
Hypervisor
Infrastructure
Fig. 5.2 Virtualization architecture
5 A Review ofParticle Swarm Optimization inCloud Computing

96
5.1.2 Virtual Machines
Optimization is a technique that allows a part of our day-to-day life. In the sense, it
can be called as an art of choosing the best alternative solutions among an available
set of options.
Optimization is needed to improve the precopy approach based on time series
[3]. In the olden decades, many global optimization methods have been proposed
due to the nature-inspired environment. These are common population meta-
heuristics called generic algorithms because of their applicability in a wide area of
problems. Population-based global optimization problems are fast tools that over-
come the limitations [4].
5.1.3 Virtual Machine Placement Problems Dened
Virtual machine placement is a method of mapping virtual machines to host
machines. As virtualization is a process of cloud computing, VM selection is a vital
approach for improving resources, energy efciency, and maximum utilization in
cloud services. In cloud computing, virtual machine (VM) placement problem is a
critical task which includes the VM migration and intended to nd the customized
physical machine (PM) to host the applications where there is co-relation between
the patterns to nd the specic group of desktop [5]. It has an effect on the overall
performance, maximizing resource utilization and reducing power consumption
over the data centers and reducing maintenance cost. Many VM placement tech-
niques are proposed for VM placement in the data centers to improve the resource
utilization.
Let us consider the situation. We have four servers, which contain a quad-core
processor, having a capacity of executing two VMs. The system is hosting four
customized applications 1–4.
1. For every server, calculate the resource requirement using the available resources
needed by each server for certain period of time, for example, 5 weeks.
2. Select a target server which has the compatible software, required CPU, net-
work, and also the storage.
3. Place the primary virtual machine on the rst server. Try to place the second
virtual machine on the same old server, and check whether it can satisfy resource
demands or not. If not, then add the new physical machine; accommodate the
VM on this newly created machine. Continue this step so that all the VMs have
been placed and add a new physical machine when it is required.
4. The resulting host machines comprise the set of the consolidated servers. Finally
the total number of servers required will be reduced from four to two. If only two
active VMs are running on the same server, then the CPU utilization of server is
calculated as the total of the CPU utilizations over the two VMs and also the
same is applicable for utilization of memory resources over the cloud data
centers.
D.V.C et al.

97
For example, consider an example of VMPM pair such as the following: (i)
(30%, 40%) is a pair of the CPU cycles and memory of a VM and (45%, 50%) for
the second VM.Then, the resultant server accommodates the two VMs having the
conguration as (65%, 80%), i.e., vector sum. To avoid CPU and memory usage
from reaching 100%, we need to have an upper bound on resources for single server
by considering the threshold value. With this it is possible to achieve 100% utiliza-
tion of cloud resources which can reduce the performance. VM live migration tech-
nology consumes fewer amounts of CPU cycles during the migration multi-cloud
database (MCDB) [6] (Fig.5.3).
The above gure shows the different types of virtualization techniques and how
best we can employ different virtualization techniques efciently and effectively
[7]. Desktop virtualization is a technology where it allows the users to simulate their
services remotely or to the devices that are connected to the workstation. This con-
tributes the separation of desktop workstation from the client applications. This
increases the simplicity of the work and the productivity of the work also reduces
the downtime of the workers. Server virtualization is to divide the server into unique
virtual servers based on the demand of the users and provides the virtual version of
the demanded resources. Each server runs on its own operating system indepen-
dently. It allows the server to increase the server ability, reduce cost in the user-
customized applications, and also increase the performance ratio.
Storage virtualization is the abstraction of physical resources and housed in a
central server as and when required will be allocated, i.e., based on the customer
demand the virtualized resources will be allocated. It creates an abstraction layer
Desktop
Virtualization
Virtualization
Server
Virtualization Network
Virtualization
Storage
Virtualization
Application
Virtualization
Fig. 5.3 Virtualizationin cloud environment
5 A Review ofParticle Swarm Optimization inCloud Computing

98
between the physical storage and the operating system which means logical space is
created to manage the metadata. Network virtualization is the abstraction of net-
work resources, creates a overlay, and runs the virtual network. Application virtual-
ization creates a virtual version of the application and interacts directly with the
operating system, and with this invention it is possible to increase the dynamic
application which is available on the host system.
The main objectives of the virtual machine and physical machine mapping are to
(i) reduce energy consumption by reducing the number of running physical machines
and to do dynamic resource allocation, (ii) improve resource utilization and to mini-
mize cost of a data center, and (iii) improve SLA and reduce number of VM migra-
tions (Fig.5.4).
5.1.4 Resource Management inCloud Computing
Resource allocation is one of the biggest challenges in cloud data center due to het-
erogeneity of resources; it is very difcult to manage cloud resource allocation. The
above diagram shows the glimpse of the methods used in resource allocation.
Primary factor here is to detect the type of resource which is in demand based on the
Users
Request Generator
Datacenter Controller
Load Balancer (Level 1)
PM
VM VM VM VM VM VM VM VM
PM
Load Balancer (Level 2)
PM
Fig. 5.4 VM placement in Cloud datacenter
D.V.C et al.

99
demand the resources will be allocated. Resource provisioning plays a very impor-
tant role where it mainly concentrates on the quality of the data and time required to
complete the task without compromising in quality of resources. Resource mapping
depends on the availability of the resources and the demanded resources, and based
on this ratio resource allocation will be done. Resource monitoring cloud data cen-
ter is in turn providing the needed data to the cloud users depending on the cloud
environment resources that will be allocated. Novel taxonomies must be proposed
to balance the resources over cloud [8] (Fig.5.5).
Many researchers explained the concept in a strategic way; the main idea here is
to monitor the ratio of available and demanded resources. Once it is matched with
the need, then automatically resource load balancing can be tackled. So resource
monitoring and load balancing are both issues that can be solved very easily.
Cloud resource monitoring is the basic necessity before thinking about the qual-
ity of services, which mainly concentrated on the availability in all the directions.
The data or the resource is not from one direction, i.e., data or information may
come from different dimensions, and balancing is the major task; we need an ef-
cient load balancing algorithm to manage the resources in an efcient manner.
5.2 Particle Swarm Optimization
Particle swarm optimization is (PSO) one of the nature-inspired population-based
algorithms which utilize the swarm intelligence to nd a better solution in the com-
plete problem search space. PSO was introduced by Kennedy and Eberhart in 1995
[8], and it is a meta-heuristic powerful technique of optimization that concerns the
nding of maxima or minima of functions in the possible region [32]. The number
Resource
Management
Resource
Provisioning
Resource
Detection
Resource
Selection
Resource
Mapping
Resource
Allocation
Resource
Monitor
Load
Balancing
Resource
Scheduling
Fig. 5.5 Resource management in cloud data center
5 A Review ofParticle Swarm Optimization inCloud Computing

100
of people using cloud resources is increasing at an exponential rate; this necessitates
efcient algorithms for resource sharing and allocation, and many researchers
worked in this area to bring out optimization in cloud resource sharing and alloca-
tion. A tree diagram is given below to summarize the work done in this area.
Optimization is a trial and error method where classication is done based on the
type of the problem, on the constraints used, on the problem size, and on the mea-
surement using the criteria used in solving the problem [9]. The problem with the
existing algorithms is either they looked at single-dimensional optimization or the
optimization process taking away too much of cloud resources and time. Main
objective is to study all the existing algorithms and analyze algorithm for multidi-
mensional optimization in less amount of time and to reduce the number of running
physical systems thereby increasing the power efciency of the whole data center.
5.2.1 Parameters ofPSO
To solve the said problem, we use PSO to where each particle maintains a local best
and a global best solutions, and after n number of iterations, the global best solution
is the selected physical machine where the virtual machine will be placed. The main
objective is to place the requested VMs in such a way to reduce the number of active
physical machines and the total power consumption of the data center. Being an
approximation algorithm, PSO performs better when there are a lot of VM instances
to be allocated on an active PM while satisfying the given objective.
The particle swarm optimization algorithm is a population-based optimization
technique for solving global optimization problems, based on the social behavior. In
a PSO multiple candidate solutions exist and collaborate continuously, solution is
named as a particle and ies in the given problem space searching for the optimal
position. A particle modies its position as it moves from one place to another place.
Its position changes according to its own experience and the experience of its neigh-
bors. PSO is a combination of local search and the global search methods based on
the particle experience in the problem search space, i.e., each iteration particle tries
to update its best position. A particle status is represented by two factors: particle’s
position and its velocity. The new velocity and the particle’s position will be updated
in each iteration (Fig.5.6).
PSO is one of the powerful optimization techniques where only few parameters
are to adjust when compared to other heuristic algorithms. PSO has been applied to
a wide range of applications where nding optimal solution is abundant. Datacenter
has considerable amount of interest from the nature-inspired community computing
that has seen too many offers which inuence solving the optimization problems in
multidimensional search spaces.
D.V.C et al.

101
5.2.2 PSO Algorithm
The PSO algorithm is having four main components which will decide the ef-
ciency of the given algorithm, namely, initial position, velocity, weight parameters,
and the tness function. Here in this paper, we will discuss how to set the initial
position and initial velocity so that the candidate solution obtained is the best one.
To verify the authenticity of the arrived solution, we use the tness function, and the
tness function ensures the PSO is optimized for the parameters we intended to.
Algorithm: Particle Swarm Optimization
1: procedure PSO
2: swarm ← Initialize Particles(no. of particles)
3: for each particle in swarm do
4: afterFlight(swarm)
5: for all itrations do
6: for each particle in swarm do
7: monitor ← Select Optimal Particle(swarm)
8: Post(particle, monitor)
9: change(particle)
10: afterFlight(particle)
11: procedure afterFlight(particle)
12: Check updated Particle (particle)
13: Compute Fitness (particle)
14: Update Local optimal (particle)
Fig. 5.6 Bird ocking
behavior
5 A Review ofParticle Swarm Optimization inCloud Computing

102
5.2.3 Modications oftheOriginal PSO
Particle swarm optimization has been developed by Kennedy and Eberhart in 1995
[9]. After this many researchers modied the original PSO to improve the searching
capability of particles over the problem space. Many hybrid versions are introduced.
Few versions are discussed here. There are many new PSO algorithms to help
improve the performance of original PSO, thereby enabling the application of PSO
to various optimization problems which includes constrained optimization, multi-
objective optimization, and unconstrained optimization. The applications of PSO
vary in complexity and cover a wide range of areas (Fig.5.7).
The basic PSO algorithm simulates bird ocking behavior. The ight of bird
ocks is simulated with good accuracy by maintaining distance between the differ-
ent birds. The distance depends on size. The bird is treated as a particle and each
particle is assigned a parameter called tness value that is evaluated by a function
which is optimized and has a speed corresponding to the ying of the particle.
5.3 PSO Variants
Classications are formed depending on domain where it is applied, on attributes
selected, and some other criteria. In this section, some of the different classications
of PSO algorithm along with the mathematics behind each algorithm are studied.
Standard
PSO
Bi-objective
PSO
Learning
PSO
Modified
PSO
Jumping
PSO
Hybrid
PSO
Binary
PSO
PSO-based
Scheduling
Algorithms in
Cloud
Computing
Fig. 5.7 Modications of PSO
D.V.C et al.

103
5.3.1 Continuous PSO Algorithm Techniques
Mathematics behind some of the standard PSO algorithm and its variants is sum-
marized in Table5.1.
5.3.2 Discrete PSO Algorithm Techniques
Initially PSO algorithm was stated as a continuous valued problem after the advance-
ment in research and technology there came up many PSO algorithms that see it as
a discrete valued problem. There are many different discrete PSO algorithms as
summarized in Table5.2.
One of the classications of PSO is done by viewing how the particles in a group
are associated, depending on some principles such as proximity, quality, and adapt-
ability. The approaches that are mainly considered for differentiating PSO are along
with its subclassication as summarized in Table5.3.
5.3.3 PSO Analysis inStock Market
There are many new PSO algorithms to help improve the performance of original
PSO thereby enabling the application of PSO to various optimization problems
which include constrained optimization, multi-objective optimization, and uncon-
strained optimization. The application of PSO varies in complexity and covers a
wide range of areas. The basic PSO algorithm simulates bird ocking behavior. The
ight of bird ocks is simulated with good accuracy by maintaining distance
between the different birds. The distance depends on size. The bird is treated as a
particle and each particle is assigned a parameter called tness value that is evalu-
ated by a function which is optimized and has a speed corresponding to the ying of
the particle.
One of the applications of PSO is stock market where a huge amount of analysis
is required for predicting the current stock value. In order to predict the stock val-
ues, the historic data and its relation to the market are required. This relationship is
used to predict the feature stock value. The various prediction methods include the
following:
• Technical analysis
• Fundamental analysis
• Traditional time series of prediction
• Machine learning methods
Technical analysis This involves predicting the appropriate time for buying or
selling the stocks. The principle behind the technical analysis [15] is that share
5 A Review ofParticle Swarm Optimization inCloud Computing

104
Table 5.1 Different continuous PSO algorithm
Sl
no. Algorithm Description Parameters Conclusion
1. Standard
algorithm [10]
Applies element-by-
element vector
multiplication
Iteration k, velocity
vk, best positions
This is basic standard
algorithm used to derive
specic algorithm
depending on
application
2. One-
dimensional
algorithm [11]
Standard algorithm
reduced for analysis
purposes to t for
one-dimensional case
Iteration k, velocity
vk, best positions
Useful for one-
dimensional case
3. Deterministic
algorithm [12]
Relationship between the
random and the
deterministic versions of
the algorithm is
established
Iteration k, velocity
vk, best positions
along with
attraction
coefcient b
The deterministic
version is obtained by
setting the random
numbers to their
expected values
4. Algorithm with
d=1 [13]
Velocity can be
eliminated from standard
algorithm
Iteration k, best
positions along
with attraction
coefcient b
Objective function only
depends on x
5. Algorithm with
c=1 [14]
Population of particles
merges the optimum
location found so far
a=1, c=1 True velocity which is
difference between two
successive particle
positions is found
Table 5.2 Different discrete PSO algorithm
Sl
no. Algorithm Description Parameters Conclusion
1. Binary particle
swarm
optimization [21]
Velocities are mapped to
a scalar value using a
sigmoidal transformation
functions
Velocity Particle positions are
binary strings, while the
velocities exist in
continuous space
2. Probability binary
particle swarm
optimization [22]
The pseudo-probability
is transformed to a
binary position vector,
uses linear
transformation
Velocity,
position
Used for multidimensional
knapsack problem
3. Extended
probability binary
particle swarm
optimization [23]
Includes mutation
operator
Velocity,
position,
mutation
4. Catsh binary
particle swarm
optimization [24]
Considers catsh
particle’s position
Number of
dimensions of
the search
space
Poor performing particles
to move out of local
optima for better
performance
5. Set-based particle
swarm
optimization [25]
Velocity as mathematical
sets
Velocity,
inertia weight
Generic set-based
algorithm, cannot be
applied to many discrete
optimization problems
D.V.C et al.

105
prices change according to the trends indicated by changing attributes of investors.
Technical data includes price, volume, and highest and lowest prices in the trading
period which is used to predict the feature stock values. The price charts are used to
detect trends which are based on supply and demand issues.
Table 5.3 Different PSO algorithms based on topology
Sl
no. Approach Subtype Description
1. Topology
[26–29]
Circle topology Local best or ring topology
Wheel topology
Star topology Also called as global best topology which is
the fastest communication topology
Pyramid topology
Von Neumann topology
2 Social
concepts [30]
Human result interaction
intelligence from social
Learning from experience
Adapt to the environment
Determine optimal patterns of behavior and
attitudes
Culture and cognition Mutual social learning Allows individuals to
move toward adaptive patterns of behavior
3 Swarm
intelligence
principles [31]
Proximity principle,
quality principle, diverse
principle response
stability, principle
stability, principle
adaptability
Simple space and time computations. Must
respond quality factors in the environment not
constrained to excessively narrow channels.
Must not change when the environment
changes, and when it is worth the
computational price
5 A Review ofParticle Swarm Optimization inCloud Computing

106
Fundamental analysis This involves applying the principles of foundation theory
for selecting the individual stocks. The analyst [16] uses this method to have a clear
picture of industry or market where they want to invest their wealth for gaining
prot. The analysts consider parameters such as growth, dividend payout rate of
interest, risk associated with investment, sales achieved, and tax rates. The main
objective is to calculate the asset value. If the value of the asset is higher than the
market value, then invest in it. This is helpful in predicting the market on a long-
term basis.
Traditional time series of prediction This involves analyzing historical data and
determines future values of a time series as a [17] linear combination. There are two
types of time series forecasting:
1. Univariate traditional time series of prediction
2. Multivariate traditional time series of prediction
Which are regression models? This involves identifying a set of factors that inu-
ence the series under prediction. Univariate is based on a single variable [18],
whereas multivariate depends its prediction on multiple variable [19] values.
Machine learning methods There are several methods in this category. All these
methods use a set of samples for generating an approximation function [20] to gen-
erate the training data. The aim is to nd conclusion from the samples in a way that
when a new data is presented to the model it is possible to identify the variable that
is used for representing the data.
5.4 Conclusion
The Internet of Things (IoT) involves the Internet-connected devices where we use
to perform the processes and access the services. This massive information needs to
be normalized. Virtualization is the key concept to balance the heterogeneous data
and in turn support our way of life. Cloud computing is a technology where the
cloud user can get the demanded cloud resources over the Internet. Cloud comput-
ing is an on-demand technology for proving quality service to the end users. Virtual
Machine optimization is the key concept for maximum utilization of cloud resources
over the cloud data center, and one such optimization algorithm is particle swarm
optimization. The resource allocation problem is an NP-hard problem, and the time
taken for allocating the resources plays a vital role in dening efciency. Applying
PSO for MSA belongs to the fourth sequence alignment approach. In this chapter
we discussed the modications of particle swarm optimization algorithm along with
the parameter considered in each. This chapter also gives a bird’s eye view on the
mathematical formulae associated with each algorithm. PSO has high global con-
vergence performance and few parameters to adjust and reduced memory and per-
forms at an improved computation speed. These are the important reasons for PSO
D.V.C et al.

107
to be popular. The implementation of PSO is an effective solution for several prob-
lems and tons of applications in various elds including bioinformatics, and other
optimization problems.
References
1. Amanatullah, Y., Lim, C., Ipung, H.P., & Juliandri, A. (2013). Toward cloud computing refer-
ence architecture: Cloud service management perspective. In International conference on ICT
for smart society (pp.1–4). Jakarta. https://doi.org/10.1109/ICTSS.2013.6588059.
2. Yu, W.D., Joshi, B., & Chandola, P. (2011). A service modeling approach to service require-
ments in SOA and cloud computing– Using a u-Healthcare system case. In 2011 IEEE 13th
international conference on e-Health networking, applications and services (pp.233–236).
Columbia. https://doi.org/10.1109/HEALTH.2011.6026754.
3. Johnson, J.A. (2013). Optimization of migration downtime of virtual machines in cloud. In
2013 Fourth international conference on computing, communications and networking tech-
nologies (ICCCNT) (pp.1–5). Tiruchengode. https://doi.org/10.1109/ICCCNT.2013.6726508.
4. Zhou, C., & He, G. (2008). A global convergence algorithm with stochastic search for con-
strained optimization problems. In 2008 Second international conference on genetic and evo-
lutionary computing (pp.75–78). Hubei. https://doi.org/10.1109/WGEC.2008.40.
5. Le Thanh Man, C., & Kayashima, M. (2011). Virtual machine placement algorithm for virtual-
ized desktop infrastructure. In 2011 IEEE international conference on cloud computing and
intelligence systems (pp.333–337). Beijing. https://doi.org/10.1109/CCIS.2011.6045085.
6. Gupta, S., & Sharma, K.P. (2020). A review on applying tier in multi cloud database (MCDB)
for security and service availability. In 2020 International conference on computer science,
engineering and applications (ICCSEA) (pp. 1–4). Gunupur, India. https://doi.org/10.1109/
ICCSEA49143.2020.9132931.
7. Abdul-Rahman, O., Munetomo, M., & Akama, K. (2011). Multi-level autonomic architecture
for the management of virtualized application environments in cloud platforms. In 2011 IEEE
4th international conference on cloud computing (pp.754–755). Washington, DC. https://doi.
org/10.1109/CLOUD.2011.58.
8. López-Pires, F. (2016). Many-objective resource allocation in cloud computing datacenters. In
2016 IEEE international conference on cloud engineering workshop (IC2EW) (pp.213–215).
Berlin. https://doi.org/10.1109/IC2EW.2016.32.
9. Kennedy, J., & Eberhart, R. (1995). Particle swarm optimization. International Conference on
Neural Networks, 1995.
10. Madhumala, R.B., & Tiwari, H. (2020). Analysis of virtual machine placement and optimiza-
tion using swarm intelligence algorithms. In A.Haldorai, A.Ramu, & S.Khan (Eds.), Business
intelligence for enterprise internet of things. EAI/Springer innovations in communication and
computing. Cham: Springer.
11. Jiang, Y., et al. (2007). An improved particle swarm optimization algorithm. Applied
Mathematics and Computation, 193(1), 231–239.
12. Blondin, J. (2009). Particle swarm optimization: A tutorial. Retrieved from http://cs.armstrong.
edu/saad/csci8100/pso tutorial.pdf
13. Clerc, M. (1999). The swarm and the queen: Towards a deterministic and adaptive particle
swarm optimization. In Proceedings of the 1999 Congress on Evolutionary Computation-
CEC99 (Cat. No. 99TH8406).
14. Patro, S., et al. (2015). Technical analysis on nancial forecasting. arXiv preprint
arXiv:1503.03011.
15. Bagheri, A., Peyhani, H.M., & Akbari, M. (2014). Financial forecasting using ANFIS net-
works with quantum-behaved particle swarm optimization. Expert Systems with Applications,
41(14), 6235–6250.
5 A Review ofParticle Swarm Optimization inCloud Computing

108
16. Zhiqiang, G., Huaiqing, W., & Quan, L. (2013). Financial time series forecasting using LPP
and SVM optimized by PSO. Soft Computing, 17(5), 805–818.
17. Hu, M., etal. (2019). Modern machine learning techniques for univariate tunnel settlement
forecasting: A comparative study. Mathematical Problems in Engineering, 2019.
18. Han, M., et al. (2018). Multivariate chaotic time series online prediction based on improved
kernel recursive least squares algorithm. IEEE Transactions on Cybernetics, 49(4), 1160–1172.
19. Hegazy, O., Soliman, O.S., & Salam, M. A. (2015). Comparative study between FPA, BA,
MCS, ABC, and PSO algorithms in training and optimizing of LS-SVM for stock market pre-
diction. International Journal of Advanced Computer Research, 5(18), 35–45.
20. Khanesar, M.A., Teshnehlab, M., & Shoorehdeli, M.A. (2007). A novel binary particle swarm
optimization. In 2007 Mediterranean conference on control & automation. IEEE Digital
Library.
21. Wang, L., etal. (2008). A novel probability binary particle swarm optimization algorithm and
its application. Journal of Software, 3(9), 28–35.
22. Liu, J., & Fan, X. (2009). The analysis and improvement of binary particle swarm optimiza-
tion. In International conference on computational intelligence and security. IEEE
23. Chuang, L.-Y., Tsai, S.-W., & Yang, C.-H. (2011). Catsh binary particle swarm optimization
for feature selection. In International Conference on Machine Learning and Computing IPCSIT.
24. Chou, S.-K., Jiau, M.-K., & Huang, S.-C. (2016). Stochastic set-based particle swarm optimi-
zation based on local exploration for solving the carpool service problem. IEEE Transactions
on Cybernetics, 46(8), 1771–1783.
25. Goswami, L., Kaushik, M. K., Sikka, R., Anand, V., Prasad Sharma, K., & Singh Solanki,
M. (2020). IOT based fault detection of underground cables through node MCU module. In
2020 International conference on computer science, engineering and applications (ICCSEA)
(pp.1–6). Gunupur, India. https://doi.org/10.1109/ICCSEA49143.2020.9132893.
26. Wang, Y.-X., & Xiang, Q.-L. (2008). Particle swarms with dynamic ring topology. In 2008
IEEE Congress on Evolutionary Computation (IEEE World Congress on Computational
Intelligence). Hindawi: IEEE.
27. Figueiredo, E.M., & Ludermir, T.B. (2012). Effect of the PSO Topologies on the Performance
of the PSO-ELM.In 2012 Brazilian symposium on neural networks. IEEE Digital Library.
28. Madhumala, R.B., Tiwari, H., & Devaraj Verma, C. (2021). A reliable frame work for vir-
tual machine selection in cloud datacenter using particles warm optimization. International
Journal of Mathematics and Computer Science (IJMCS), 16(2021). ISSN 1814-0432.
29. Ni, Q., & Deng, J. (2013). A new logistic dynamic particle swarm optimization algorithm
based on random topology. The Scientic World Journal, 2013.
30. Gupta, S., Vyas, S., & Sharma, K.P. (2020). A survey on security for IoT via machine learn-
ing. In 2020 International conference on computer science, engineering and applications
(ICCSEA) (pp.1–5). Gunupur, India. https://doi.org/10.1109/ICCSEA49143.2020.9132898.
31. Del Valle, Y., etal. (2008). Particle swarm optimization: Basic concepts, variants and applica-
tions in power systems. IEEE Transactions on Evolutionary Computation, 12(2), 171–195.
32. Agrawal, R. (2020). Perspectives and foundations of swarm intelligence and its application.
Swarm Intelligence Optimization: Algorithms and Applications, 41–48.
D.V.C et al.