Journal Pre-proofs

A Survey on PSO Based Meta-Heuristic Scheduling Mechanism in Cloud

Computing Environment

Arabinda Pradhan, Sukant Kishoro Bisoy, Amardeep Das

PII: S1319-1578(21)00003-3
DOI: https://doi.org/10.1016/j.jksuci.2021.01.003
Reference: JKSUCI 948

To appear in: Journal of King Saud University - Computer and

Information Sciences

Received Date: 20 August 2020

Revised Date: 22 December 2020
Accepted Date: 4 January 2021

Please cite this article as: Pradhan, A., Bisoy, S.K., Das, A., A Survey on PSO Based Meta-Heuristic Scheduling
Mechanism in Cloud Computing Environment, Journal of King Saud University - Computer and Information
Sciences (2021), doi: https://doi.org/10.1016/j.jksuci.2021.01.003

This is a PDF file of an article that has undergone enhancements after acceptance, such as the addition of a cover
page and metadata, and formatting for readability, but it is not yet the definitive version of record. This version
will undergo additional copyediting, typesetting and review before it is published in its final form, but we are
providing this version to give early visibility of the article. Please note that, during the production process, errors
may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

© 2021 Production and hosting by Elsevier B.V. on behalf of King Saud University.
 A Survey on PSO Based Meta-Heuristic
Scheduling Mechanism in Cloud Computing
Environment
Journal of King Saud University Computer and Information Sciences

Arabinda Pradhan
Research Scholar,Department of Computer Science and Engineering,C.V.Raman Global University
Bidya Nagar, Mahura, Janla, Bhubaneswar, Odisha, 752054, India.
Email- [email protected]

Dr. Sukant Kishoro Bisoy

Department of Computer Science and Engineering,C.V.Raman Global University

Bidya Nagar, Mahura, Janla, Bhubaneswar, Odisha, 752054, India.
Email- [email protected]

Amardeep Das
Department of Computer Science and Information Technology,
C.V.Raman Global University
Bidya Nagar, Mahura, Janla, Bhubaneswar, Odisha, 752054, India.
Email- [email protected]

Contact information for the corresponding author

With best regards,

Sukant
------------------------------------------------
Dr.Sukant Kishoro Bisoy
Associate Professor and Head, Department of Computer Science and Engineering,
C.V.Raman Global University
Bidya Nagar, Mahura, Janla, Bhubaneswar, Odisha, 752054, India.
Email- [email protected]
99379103999
 A Survey on PSO Based Meta-Heuristic
Scheduling Mechanism in Cloud Computing
Environment
ABSTRACT
With the increasing of the scale of task or request and dynamic nature of cloud resources, it gives
significant issues of load balancing, resource utilization, task allocation, and system performance
and so on. To solve those problems many researchers have applied different types of scheduling
techniques. But meta-heuristic scheduling is the most accomplish preferred outcomes over
conventional heuristics and hybrid scheduling. Among various meta-heuristics algorithms, PSO
is a famous metaheuristic technique to solved optimization issue. PSO is appropriate for dynamic
task scheduling, workflow scheduling and load balancing. PSO has a strong worldwide searching
capability toward the start of the run and a nearby pursuit close to the furthest limit of the run.
Therefore, it has been generally utilized in different applications and has made incredible
progress. In this paper a systematically reviews is done on different types of particle swarm
optimization (PSO) based scheduling strategy with set of challenges and future direction.
Keywords: Scheduling, Virtual Machine, Cloud computing, Meta-heuristic.

1. Introduction
Cloud computing is a web-based innovation system that handles various request from cloud and
provide a quick service to its clients. It is a computing and processing model, used in different
places across the globe. Cloud computing can be used to rapidly enhance the prediction process
using high-speed computations. In recent COVID-19 pandemic situation, a prediction scheme
based on the machine learning (ML) model can be used in remote cloud nodes for real-time
prediction allowing governments and citizens to respond proactively. Machine learning model
can be run continuously on Cloud Data Centres (CDCs) for accurate spread prediction and
proactive development of strategic response by the government and citizens. In a cloud-based
environment, the government hospitals and private health-centres continuously send their
positive patient count, population density, average and median age, weather conditions, health
facilities etc. are also to be integrated for enhancing the accuracy of the predictions (Tuli et al.,
2020). Due to several concepts joined with cloud computing it becomes most powerful
technology and used in different business sectors or IT industry. It provides convenient and on-
demand access to huge amount of computing resources such as CPU, memory, networks,
servers, storage, applications, etc. Moreover, these resources are frequently assigned to the
clients with minimum service charge. When the number of request increases at a particular time
then it transforms into a troublesome activity to manage the entire request in the most restricted
reaction time and satisfied nature of administration. Work of Cloud Service Provider (CSP) is to
allocate these incoming tasks to a suitable virtual machine (VM) so that no machine is
overloaded and keep the load as balanced in between these machines or resources. Resources are
handled in the incoming tasks that depend on the property of resource information, task
information and proper scheduling algorithm (Kumar et al., 2019). Therefore, a better scheduling
technique is required to avoid the allocation problem in cloud system.

Scheduling is mainly used for sharing the resources at different levels in cloud environment. It is
done by virtualization, where a single physical machine or server is virtually separated into
several virtual machines (VM) in an effective manner. Each and every virtual machine is used to
allocate different task. The primary goal of scheduling is to handle the request made from user
by allocating suitable resources and balance the load among VMs. Also, it is used to upgrading
the nature of administrations (QoS) parameters like task dismissal proportion, resource
utilization, dependability, vitality utilization, execution cost, minimum execution time with limit
make span and boost the throughput (Kumar et al., 2019). Fig. 1 shows how tasks are allocated
to VMs for accomplishing destinations (Masdari et al., 2016).

Fig. 1. Task Allocating in Cloud Computing (Masdari et al. 2016)

Objective of scheduling:

Various objectives are discussed as follows (Vijayalakshmi and Prathibha., 2013).

1. Resource finding: Datacenter broker make a list to discover all available resources and their
information.
2. Resource determination: According to task requirements, a suitable resource is chosen.
3. Assignment accommodation: Allocating the task into selected resources.
In Fig. 2, scheduler or datacenter broker tries to map tasks to the appropriate virtual machines
for fulfilling the user requirement. In (Mansouri et al., 2019) scheduling algorithm can be
isolated into two parts such as: optimal and sub-optimal strategy.
a. Optimal Scheduling Strategy: Scheduler knows all the detail information about the VM
status such as load and capabilities of hardware. If any task arrives to the system the
scheduler put into available VM.
 b. Sub-optimal Scheduling Strategy: Whenever any task is not allocated or there is not
sufficient opportunity to allocate the task then the scheduler picks the sub-optimal
methods.

Fig. 2. Allocation of task in cloud computing (Mansouri et al., 2019)

Problem Statement

The aim of this paper is to review the nature inspired meta-heuristic algorithms for balancing the
load of cloud computing. One of the best meta-heuristic algorithms is PSO based scheduling
algorithm which schedule the upcoming load, applications or task to cloud resources in such a
way that, the client can finish their task in least time and maximize the throughput. Concisely,
the main aids of the paper are as follows:
 Study different state of art PSO based scheduling algorithms in cloud computing.
 Describes meta-heuristic algorithm which focus on to settle the task scheduling and load
balance issues.
 Defines a far-reaching correlation between different PSO scheduling schemes and
described about the different fitness function which have been applied in each scheme.

2. Resource management

Managing resources in cloud environment involves various cloud operations. There are various
approaches to allocate the resources that have been recognized from the survey (Kumar et al.,
2019). Resource management method into cloud computing is classify as shown in Fig. 3 (Singh
and Chana, 2016).
 Fig.3. Resource Management Scheme (Singh and Chana, 2016)

2.1. Resource provisioning

It is the method of empowering the resources in cloud with the help of virtualization technique
for allocation to the incoming task. At the point when cloud controller acknowledges the jobs
from the clients, it found suitable virtual machines (VMs) and distributes that to clients
according to the interest or on demand (Kumar et al., 2019). Resource provisioning means
finding, allocating, deploying and installing various software and hardware resources to achieve
high performance.

2.2. Resource Monitoring

In this method all the information in cloud is maintained by cloud controller and checks how
many numbers of tasks are assigned to VM in a particular time or in a system. The usages of VM
or resources are regularly checked by an administration.

2.2.1. Types of Load Balancing Algorithm

Scheduler or Load balancers execute the particular kind of algorithms to settle down on proper
scheduling or load balancing choices. The choice specifies the remote server which must sent a
new job. Depending upon framework circumstance, the scheduling calculation is to be
categorized in two sections: static approach and dynamic approach (Golchi et al., 2019).Fig.4
shows the categories of load balancing algorithm.

2.2.1.1. Static Algorithm

Static algorithms are suitable for homogeneous environments where it cannot change its attribute
and loads are distributed among the servers are predefined. It shows that all the information
regarding job and resources are known to system in advance. The choice of moving of the load
doesn't rely upon the present condition of framework. Thus, static algorithm is not correct for
distributed computing. Instance of static calculations is First-in-First-out (FIFO), Round Robin
(RR) and Shortest Job First (SJF) and so on (Golchi et al., 2019; Alguliyev et al., 2019).
 2.2.1.2. Dynamic Algorithm

Dynamic algorithms repeatedly check the current workload of system and redistribute the work
load among the resources. This algorithm is used to improve the resource utilization,
performance and reduce response time. No earlier information of system is required. So, it is
superior to static methodology (Acharya et al., 2016). Fig.4 shows two unique methods of
dynamic load balancing algorithm such as distributed and centralized. In the distributed
approach, loads are distributed between all available VM. If any VM is overloaded then the extra
loads are transferred into underloaded VM. In centralized strategy, a central node is responsible
to distribute all jobs between VMs. It controls, co-ordinate and manage entire system (Golchi et
al., 2019; Acharya et al., 2016). Some example of dynamic algorithms is: Ant Colony
Optimization (ACO), Honey Bee Foraging (HBF) and Biased Random Sampling (BRS) and etc.
The comparison between static and dynamic approach is shown in Table 1.

Fig. 4. Different strategies of load balancing algorithm (Golchi et al., 2019)

Table 1
Static and Dynamic Approach.
Static Approach Dynamic Approach
 Requires prior knowledge about the system  Not consider the prior information about
the system.
 Does not depend on the current status of node.  Depends upon the current state of node.
 Decisions are made at request time.  Decisions are made at run time.
 Loads are previously defined.  Loads are not previously defined.
 Less complexity.  More complex.
 Shows best effort where load of a node is very  Extra loads are transferred from
small. overloaded node to underloaded node.
 Easy to implement.  It is difficult to implement.
 Complexity is low  Complexity is high
Ex: SJF, FIFO, RR etc. Ex: ACO, HBF, GA etc.

2.3. Resource scheduling

It is the method through which tasks are allocated to the suitable resources for executing and
provide good quality of service. An effective scheduling technique is required to select best
resources from the physical machines (Kumar et al., 2019). Resource scheduling brings key
benefits such as: decrease the execution period of application, increases the utilization of cloud
resources, and reduces the energy consumption with improved QoS.

3. Types of scheduling method

Due to dynamic nature of task, many researchers focused their researching work on heuristic,
meta-heuristic and hybrid scheduling strategies. Scheduling scheme is divided into three types
such as heuristic, meta-heuristic and hybrid, shown in Fig. 5.

.
Fig. 5. Various Scheduling Algorithms (Kumar et al., 2019)
3.1. Heuristic scheduling algorithm

Heuristic algorithm is used to solve the problem in a quick and an efficient manner. It gives
better solution when classic methods fail. Commonly, heuristic algorithm gives an appropriate
result for specific domain of problem in fixed time period but cannot solve hard optimization
problems (Kenny et al., 2014; Heuristic (computer science), 2019). There are number of heuristic
algorithms have been proposed to deal with the load balancing and task scheduling problem.
Some heuristic algorithms are shown in Table 2.

Table 2
Various heuristic algorithms with favorable circumstances and constraints.
Author Algorithm Techniques Advantage Disadvantage
Mao et al.,  Max-min  Task allocation  Diminishes both  Chances of
(2014) with high speed. reaction time and resource
makespan. imbalanced.
Elzekiet al.,  Improved  Calculate  Reduce Makespan  Resources are
(2012) Max-Min excepted time. not properly
execution time. balanced.
Mondalet al.,  SJF  Select shortest  Reduce execution  Some chances of
(2015) execution of task. time and turn- starvation and
around time. load imbalanced.
Alworafi et  Improved  Computing task  Minimize  Required more
al., SJF length. completion time and focused on
(2016) maximize resources deadline
 utilization. constraint.
Pasare et al.,  DHSJF  Priority of task  Reduce reaction  Difficult to
(2019) size. time and increase predict the burst
the accessibility of time of the task
VMs. execution.
Li and Shi,  FCFS  Front-end  Reduce time  Load imbalanced
(2009) scheduler allots complexity. among the virtual
task to VMs. machines.
Khurma et al.,  Modified  Task scheduling.  Reduce waiting  Short time
(2018) RR time. slicewill cause
switching
problem.
Chawda and  Improved  Underloaded  Decrease the  SLA violation is
Chakraborty, Load resources are makespan and high.
(2016) Balancing more used. increment the
Min-Min resource usage.
Devi and  WRR  Loads are  Improved resource  Workloads are
Uthariaraj, assigned in utilization. not properly
(2016) circular pattern. balanced.
Kanani and  Improved  Tasks are  Minimize makespan  Low priority
Maniyar, Max-Min executed depends and maximize tasks are not
(2015) on priority. resource utilization. balanced.
Pasha et al.,  RR  Check status of  Reduce migration  Less scalability.
(2014) VM. time and increase
resource utilization.

3.2. Meta-heuristic scheduling algorithm

Meta-heuristic is a more significant level technique intended to discover, produce, or select great
answer for an enhancement issue. It has the capacity to solve solving large computational and
complex problems (Milan et al., 2019). Different meta-heuristic is used in cloud environment to
solve NP-hard issue, appeared in Table 3. Some of the meta-heuristic algorithms are available
such as, Artificial Bee Colony (ABC), Particle Swarm Optimization (PSO), Ant Colony
Optimization (ACO), Differential Evolution Algorithm (DEA), Genetic Algorithm (GA),
Simulated Annealing (SA), Bacteria Foraging Optimization (BFO), BAT Optimization, Cat
Optimization, Firefly Optimization, Cuckoo Search CS) and so on. All algorithms are used to
deal with the optimization issue. But among them GA and PSO is most predictable algorithms to
applied in COVID-19 pandemic situation (Babukarthik et al. 2020; Tuli et al. 2020). Most meta-
heuristics accomplish preferred outcomes over conventional heuristics (Xu et al. 2017).Some
useful properties of meta-heuristic algorithms such as: it guide the search problems, find optimal
value within search space, it is not problem specific and solve NP-hard optimization problems
(Metaheuristics, 2020).

Table 3
Various Meta-heuristic algorithm with favorable circumstances and constraints.
Author Algorithm Techniques Advantage Disadvantage
Kumar and PSO-  Adjusting PSO by  Reduce execution time  Less fault
Sharma, COGENT using APSO-VI. and cost but maximize tolerance, less
(2018) the throughput. scalability and
high overhead.
Lin et al., BPSO  Modified sigmoid  Optimize speed ratio,  Less reliability
(2016) transfer function. minimize makespan and take high
and reduce cost. execution time.
Verma et al., ACO  Two level cloud  Minimizes the cost and  Makespan is not
(2017) scheduler approach increase resource reduce.
utilizing ACO utilization.
algorithm.
Tawfeek et ACO  Virtual memory  Minimize memory  Deadline
al., (2014) placement shown in resource wastage. constraint.
VMPACS
algorithm.
Tawfeek et ACO  Task Scheduling  Minimize makespan  Deadline
al., (2015) Constraint.
Nilesh and ACO  Calculate both  Reduce the makespan  Algorithm work
Patel, (2017) expected execution only for
time and transfer independent
time of the task. tasks
Xiang et al., ACO  Tasks are allocated  QoS parameters are  Less secure.
(2017) on their priority. satisfied.
Li et al., LBACO  Task scheduling.  Reduce completion  Less elasticity
(2011) time. and reliability.
Jena, ABC  Multi-target  Reduce cost and  Conflict in
(2017) calculation utilizing increase resource calculation.
ABC. utilization.
Yao and He, ABC  Depend on current  Improve system  Not scalability
(2013) status of resources. throughput. and less
reliability.
Dasguptaet GA  Load balancing  Improve time for both  Not satisfy all
al., (2013) algorithm based on makespan and QoS.
NIC and MIPS. response.
Hamad and GA  TS-GA  Minimize the  Less parameter
Omara, completion time and is taken to
(2016) cost of tasks. achieve QoS.
Wei et al., SAMPGA  MVMP algorithm  Minimize both  VM migration
(2017) using SA. completion time and approach is
degree of load time
imbalance. consuming.
Liu and Liu, SA  Greedy algorithm  Reduce time  Less efficient.
(2016) complexity.

3.3. Hybrid scheduling algorithm

Hybrid scheduling algorithms is the combination of two planning calculations (above two
scheme) to deal with the issue of asset planning for cloud computing. Various hybrid scheduling
is used in cloud environment to solve NP-hard issue, appeared in Table 4.
Table 4
Hybrid algorithm with favorable circumstances and constraints.
Author Algorithm Techniques Advantage Disadvantage
Thanka et al.,  ABPS  ABC and  Limited makespan, cost,  Less security.
(2019) PSO. and expanded resource use.
Raju et al.,  Hybrid  ACO andCS.  Reduce makespan time.  More
(2013) complicacy.
Kaur and  Hybrid  Tasks are  Minimize both makespan  Less
Kaur, allocated to time and cost. reliability.
(2019) VM based on
their priority.
Sobhanayaket  Hybrid  Combination  Minimize makespan time.  Time
al., algorithm of GA complexity
(2018) andBFO.

4. Particle Swarm Optimization

PSO is affected by community conduct of creatures like flock of birds discovering nourishment
source, which is an intelligent evolutionary computing technology. It was introduced by
Kennedy and Eberheart (Kennedy and Eberhart, 1995). It is a computational technique that
upgrades an issue by iteratively attempting to improve an applicant arrangement concerning a
given proportion of value. The aim of PSO is to find the optimal solution through the
collaboration and data sharing among the particle or molecule in a gathering which can be
considered as a population. Particle is a component or part of the population, which is called as
swarm. Swarm is required to fly over the hunt space in order to find promising territories of the
scene. Each particle has a searching space through which they search their food and every
particle initialized randomly and contains velocity as well as position. Every Particle knows their
own best position Pbest and best position among the group of particles Gbest. In each iteration,
velocity and position of each particle is refreshed using following equation (1) and (2). Fig. 6
show the concept of searching position of a particle in PSO and Fig. 7 shows procedure of PSO
(Allaoua et al., 2008; Tran, 2018). Different terms and their meaning that are used in PSO,
appeared in Table 5.

Vkim+ 1 = w ∗ Vkim + c1 ∗ r1 ∗ (Pkbest,im ― Xkim) + c2 ∗ r2 ∗ (Gkbest,im ― Xkim) … (1)

Xkim+ 1 = Xkim + Vkim+ 1 … (2)

 Fig. 6. Searching position of a particle in PSO (Allaoua et al., 2008; Tran, 2018)

Table 5
Denote terms and meaning.
Terms Meaning Terms Meaning
i Particle r Random number
m Dimension V Velocity
k Iteration Xk Current position
k+1
w Weight factor X Modified position
X Position Vk Current velocity
Pbest Personal best Vk + 1 Modified velocity
Gbest Global best VPi best Velocity of Pbest
c Co-efficient VGi best Velocity of Gbest

Fig. 7.The procedures of PSO algorithm (Allaoua et al., 2008; Tran, 2018)
PSO algorithm is iterated when the below stopping criteria is not fulfilling (Chitra et al., 2014):
 Find out fitness value of each particle.
 Calculate Pbest and Gbest and update position and velocity of each particle.

5. Types of PSO Scheduling

Based on PSO techniques, various types of scheduling are proposed in the literature such as
Standard PSO, Jumping PSO, Learning PSO, Bi-objective PSO, Multi-objective PSO, Modified
PSO, Binary PSO, Hybrid PSO and Parallel PSO.

5.1. Standard PSO

In standard PSO, every particle in the swarm is treated as a molecule in a m-dimensional hunt
space and each particle have contained position and velocity. Also, they contain their personal
best (Pbest ) and global best (Gbest). In this section various tasks and load balancing scheduling
are based on common or standard PSO for scheduling. This method has been appropriate to plan
the resources in profitable way and find an optimal result depends on fitness value. Different
researchers have proposed standard PSO appeared in Table 6. Author in (Guo et al., 2012)
proposed a method to optimize task scheduling that show how all the tasks are allocate to
available resources so that total execution time and transfer time is minimized. Proposed
objective is representing in Eq. (3), where Eq. (4) showsexecuting time and Eq. (5) shows the
transferring time. Initial position and velocity of each particle taken in the proposed PSO
algorithm is represent in Eq. (6) and (7) respectively. Finally, the update velocity and position
are representing on Eq. (8) and (9) respectively.

Total(M) = Cexe(M) + Ct(M) … (3)

n m DEik
Cexe(M) = ∑i = 1∑k = 1xik ∗ Pm ∗ Pc
… (4)

n―1 n DTij
Ct(M) = ∑i = 1 ∑j = i + 1yijkl ∗ Bij
… (5)

x1i = xmin + (xmax ― xmin) ∗ rand … (6)

v1i = vmin + (vmax ― vmin) ∗ rand … (7)

vki + 1 = wvki + c1rand1 ∗ (pbesti ― xki) + c2rand2 ∗ (gbest ― xki) … (8)

xki + 1 = xki + vki … (9)

Where,Cexe(M) is execution time, Ct(M) is transferring time, xik means task I is assigned to
processor k, DEik is the amount of data, Pm is the memory, Pc is CPU, yijkl gives both task I and
j are assigned to processor k and l, DTij is the data interchange between task I and j, Bij is the
bandwidth and rand is the random number between 0 and 1. Here, xmax = vmax = 4.0; xmin =
vmin = ―0.4
Author in (Zhang and Zuo, 2013) proposed a hybrid cloud architecture where available resources
are not sufficient to handle all incoming task. By applying Standard-PSO method, the priority of
each task is calculated and sorting in descending order to maximize the profit which is represent
in Eq. (11).
w l w T l n
Profit = ∑j = 1∑v = 1Tjbjvpvrj ― ∑j = 1∑l =j 1∑v = 1∑k = 1yjlkbjvckvrj … (10)

Where,Tj is the number of tasks in jth application, bjv is the use of VM, pv is the cloud provider,
rj is the runtime of jth application, yjlk is the binary decision variable, ckv is the cost and CPk
denotes k number of cloud provider.

In cloud computing load balancing among the resources is a major issue. It can be solved by
using proper task scheduling method. To solve this problem, author in (Yang et al., 2013)
presented a PSO based task scheduling algorithm where the task set is dividing into number of
subtask and each subtask is assigned to the suitable resource in random order that can be depend
upon fitness value. Also, it reduces the makespan time. Fitness value, makespan and total
completion time are representing as in Eq.(11), (12) and (13).

f(Xti) = p(Oti|Xti) … (11)

Makespan = Max(Ttotal_i) … (12)

Ttotal_i = Trece + Texec + Twait … (13)

Where,f(Xti) is the fitness value, Oti is the observation, Xti is the particles, Ttotali is the total
completion time, Trece is the total receiving time, Texec is the total execution time and Twait is the
total waiting time.

Author in (Huang et al., 2013) proposed PSO based workflow scheduling where a tunable fitness
function is used to reduce cost and makespan time which is shown in Eq. (14), (15) and (16).

Fitnessfunction1 = Max(Costtotal(Ri)) … (14)

Fitnessfunction2 = Makespantotal(M) … (15)

Fitnessfunction3 = αCosttotal(M) + (1 ― α)Makespantotal(M) … (16)

Table 6
Standard PSO.
Author Objectives Scheduling Type
Guo et al.,(2012) Minimize execution time and transfer time. Task Scheduling
Zhang and Zuo, (2013) Maximize profit and QoS support. Task scheduling

Yang et al., (2013) Reduce makespan time and increase the Task scheduling
processing capacity
Huang et al., (2013) Limit both cost and makespan time. Workflow Scheduling
 5.2. Jumping PSO

In JPSO, a particle refreshes its present position by bouncing to another position by affecting of
its experience. For their improvement, they can jump to the new event. According to (Masdari et
al., 2017; Ismail and Jeng, 2013) the situation of every particle in Jumping PSO will be refreshed
by Eq. (17).

xt + 1 = (γ1 ∗ xt) + (γ2 ∗ b) +(γ3 ∗ g) … (17)

Where,xt is the current position, xt + 1 is the update position, b is the personal best and g is the
global best. Here, γ1, γ2 and γ3 are the random parameters and having value 1.

Principle downside of standard PSO procedure was local minima problem or slow convergence
rate. It was caused by Gbest or global best among the population. At each iteration, if Gbest value
is not changed then local minima problem arises. To overcome this issue, another method named
jumping PSO (Chitra et al., 2014) has been proposed, here a reasonable workflow schedule
having the knowledge that particles are move from one co-ordinate to other co-ordinate by
utilizing Jump PSO algorithm (JPSO) to enhance load balancing, speedup proportion and
makespan. An improved technique for determination of Gbest is given as Eq. (18).

1 k
Gbest = 𝑘∑i = 1p(i) … (18)

Where, p is the rank obtained personal best and k is the iteration.

5.3. Learning PSO

In this method every particle can learn their position and velocity to their neighbors for achieve
the optimal solution and satisfy all QoS parameters. Table 7 shows the properties of the LPSO-
based scheduling schemes. A resource allocation framework named as self-adaptive learning
PSO (SLPSO) based scheduling approach is proposed by (Zuo et al., 2014). It deals with an
external cloud resources where local resources are not sufficient available. All incoming tasks
are must follow deadline constrained task scheduling to get maximum profit as well as provide
more elasticity to IaaS cloud. For getting the objective this method uses four velocity updating
strategy. Eq. (19) represents initial velocity. Eq. (20), (21), (22) and (23) shows updated velocity
of each particle respectively.

vid(t) = xkd(t) ― xjd(t) … (19)

vid(t + 1) = cvid(t) + c[pid(t) ― xid(t)] … (20)

vid(t + 1) = ωvid(t) + c1r1[xid(t) ― pid(t)] + c2r2[xid(t) ― gd(t)] … (21)

vid(t + 1) = ꞷvid(t) + cr[pkd(t) ― xid(t)] … (22)

vid(t + 1) = ꞷvid(t) +0.5c[pkd(t) ― xid(t) + pid(t) ― xid(t)] … (23)

Where,c is a random value in between 0 to 1. xkd(t) and xjd(t)are randomly selected particle in t
iteration. pkd(t) is the personal best solution. gd(t) is the global best solution.

Author in (Chen and Zhang, 2012) proposed a method named as S-CLPSO to characterize
penalty-based target capacities and utilizations a versatile plan to control the utilization of
existing heuristic technique. S-CLPSO defines three fitness functions to optimize the issues such
as: reliability, makespan, and cost. In order to to improve the system performance, a
comprehensive learning PSO (CLPSO) based on set-based PSO (S-PSO) is proposed to handle
the problem of cloud workflow scheduling. At the beginning of this method, the initial positions
of the M particles are arbitrarily introduced with M feasible solutions. During each cycle of the
S-CLPSO calculation, every particle holds fast to the speed refreshing guideline of CLPSO to
refresh its speed. In the wake of refreshing speeds, every molecule follows the position
refreshing technique to refresh the positions.

Table 7
Objective and scheduling scheme in LPSO.
Author Objectives Scheduling Type
Zuo et al., 2014 Maximize profit and provide elasticity to Task Scheduling
IaaS cloud.
Chen and Zhang, 2012 Reduce makespan and cost. Also, provide Workflow Scheduling
reliability to optimization problem.

5.4. Bi-objective PSO

Bi-objective PSO is a variety of PSO method that streamlines any two target objectives of QoS
parameters. For example, reduce cost and makespan time by applying task scheduling or
workflow scheduling or any other scheduling. Table 8 shows the properties of the Bi-objective
PSO-based schemes. In (Verma and Kaushal, 2014) author proposed Bi-Criteria Priority based
Particle Swarm Optimization (BPSO) algorithm to optimized execution cost and execution time
under the given budget and deadline constraint. Two objective function i.e. execution time is
representing in Eq. (24) and execution cost is representing in Eq. (25) respectively.

ET(i,p) = (Zi ∗ βx) MIPS of rp … (24)

EC(i,p) = μp ∗ ET(i,p) … (25)

Where, ET(i,p) is the execution time of task i on resource p, EC(i,p) is the execution cost, Zi is the
service type, βx is the standard execution time, rp is the resource and μp is the price of resource.

Author in (Beegom and Rajasree, 2014) proposed Integer-PSO based bi-objective method that is
used to smallest position value based PSO technique to optimize the makespan and cost at same
time. Eq. (26) and (27) shows minimize makespan and cost.
k n
Minimize Makespan fn = ∑p = 1∑i = 1Ti ∗ Pfj ∗ xi,j for some j ϵ I … (26)
 k n
Minimize Cost fn = ∑p = 1∑i = 1Cj ∗ Ti ∗ Pfj ∗ xi,j for some j ϵ I … (27)

Where,Pfj is the processing power, Cj is the cost, Ti is the number of tasks, j is the VMs, i is the
VM instance and k is the epoch needed to complete the execution of all tasks.

Table 8
Properties of Bi-objective PSO.
Author Objectives Scheduling Type
Verma and Kaushal, (2014) Minimize total execution time and cost Workflow Scheduling
Beegom and Rajasree, (2014) Minimize cost and makespan Task Scheduling

5.5. Multi-objective PSO

This method is consisting of several objectives to find out the solution for solving optimization
problem where every target works independently. In first methodology, every particle assumed
their best position with one objective function. The subsequent methodology, create a leader
particle having non-dominant best position and other particles follows the leader (Parsopoulos
and Vrahatis, 2008). Properties of the Multi-objective PSO-based scheduling schemes are
exhibited in Table 9. Author in (Alkayal et al., 2016) proposed multi-objective PSO (MOPSO)
task scheduling algorithm that optimizes waiting time and increase throughput. In this algorithm
the scheduler tries to find the best VM and assign the task to the best VM. To achieve this goal,
MOPSO algorithm focused three objectives such as: Processing Time (PT), Task Execution Cost
(TEC) and Task Processing Cost (TPC), all are shown in Eq. (28), (29) and (30).

PT(i,j) = TL(i)/PSV(j) … (28)

TEC(i,j) = TPC((i,j) + IOCost(i) … (29)

TPC(i,j) = PT(i,j) ∗ VMP(j) … (30)

Where, i is the task, j is the VM, IOCost(i) is the cost, TL(i) is the task length, PSV(j) is the
processing speed and VMP(𝑗) is the price of VM.

Author in (Jena, 2015) proposed TSPSO technique to optimize the parameter energy and
processing time or makespan. The main objective of this algorithm is that the entire task must
finish before deadline (𝑑𝑖) and each task can be allocated to only one Datacenter. Proposed
algorithm takes three main objectives to reach the goal such as: Finishing time, Makespan and
Energy Consumption. It is shown in Eq. (31), (32) and (33):

Finish(Tj) = start(Tj) + 𝒯j … (31)

Makespanj = max {Finish(Tj)} … (32)

N
Ej = ∑k = 1(𝒯k ∗ pk) … (33)
Where,Tj is task, 𝒯j is the time require to execute all task, 𝒯k is the time require by processor to
execute all task and pk is the power of processor.

Table 9
Properties of Multi-objective PSO.
Author Objectives Scheduling Type
Alkayal et al., (2016) Minimize waiting time and maximize throughput Task Scheduling
Jena, (2015) Optimize energy and processing time. Task Scheduling

5.6. Modified PSO

Right now, different sorts of enhanced variations of PSO be used within scheduling sectors
which are generally known as modified PSO. In this algorithm, a few upgrades for velocity
refreshing, surpassing limit control, global best optimum and so on (Masdari et al., 2017).The
major characteristics of the Modified PSO-based scheduling are shown in Table 10. Author
in(Zhou et al., 2018) proposed M-PSO task scheduling approach to deal with minimum energy
consumption of the processor, which is represent in Eq. (34). To avoid the slow convergence rate
and local optimum problem of existing PSO algorithm, M-PSO algorithm is used to taking
modified inertia weight method to get the objective, which is represent in Eq. (35). Fitness
function used in M-PSO algorithm is represented in Eq. (36).
n m
E(M) = ∑k = 1Eex(M)k + ∑i = 1Etr(M)i … (34)

ω= (α × tmax
tγ + tmax ) +β … (35)

1
fitness(i) = E(M) … (36)

Where,Eex(M)k is the total execution consumption of resource k, Etr(M)i is the total transfer
consumption between task i, ω is the inertia weight, tmax is the maximum iteration, γ is the
acceleration factor,α and β are constant parameters.

Author in (Abdi et al., 2014) have proposed a new modified PSO method that can be used to
reduce the makespan time of all incoming task. Principle goal of proposed method is to
consolidate to merge smallest Job to fastest processor (SJFP) with PSO algorithm. In SJFP
algorithm, all incoming tasks are arranged their allocation time and existing processors are
arranging that dependent on their processing power. This modified PSO algorithm gives the
balanced mapping with allocate the task to fastest processor. Fig. 8 shows the flowchart of this
modified PSO calculation.
 Fig. 8. Flowchart of modified PSO algorithm (Abdi et al., 2014)

If a greater number of tasks surpasses than the deadline than the cut-off time, at that point task
dismissal extent is more and SLA infringement will happen. Thus, (Kumar and Sharma, 2018)
proposed PSO-COGENT method to improve different QoS characteristics. Therefore, a fitness
function is created to reduce all constraint as shown in Eq. (37):

Fitnessfunctionf(rj) = α ∗ EET𝒯ir𝑗€ϴ𝒯isp + β ∗ EEC𝒯ir𝑗€ϴ𝒯isp + γ ∗ EC𝒯ir𝑗€ϴ𝒯isp … (37)

α+β+γ=1

Where,EET𝒯irj€ϴ𝒯isp is the expected execution time, EEC𝒯ir𝑗€ϴ𝒯isp is the expected execution cost,
EC𝒯ir𝑗€ϴ𝒯isp is the execution cost, α is the execution time, β is the execution cost and γ is the
energy consumption. Author in (Zhao, 2014) has executed a modified PSO calculation that can
be used to minimize both the time for processing as well as cost for processing of all tasks to
reach a global optimal solution. Time for processing all tasks is represented as Eq. (38). Total
cost for processing all tasks is represented as Eq. (39). Proposed algorithm has defined two
fitness functions that is implementing in MPSO shown in Eq. (40) and (41).
m
Tr = ∑i = 1T(r,i) … (38)
m
Costr = ∑r = 1Tr ∗ ERCr … (39)
1
Ft(i) = Ttotal(i),1 ≤ i ≤ S … (40)
1
Fc(i) = Costr(i),1 ≤ i ≤ S … (41)
Where, i is the task, r is the resource, Tr is the total time, Costr is the total cost and Ttotal(i) is the
total time for processing all task.

Author in (Miglani and Sharma, 2019) proposed a modified PSO algorithm that reduce
makespan time and balance the load more efficiently by taking MIPS and Bandwidth factors of
VMs, instead of considering the population randomly.

Table 10
Summary of various Modified PSO.
Author Objectives Scheduling Type
Zhou et al., (2018) Reduce energy consumption Task Scheduling
Abdi et al., (2014) Minimize makespan time Task Scheduling
Kumar and Sharma, Diminishes the execution time, execution cost and PSO-COGENT
(2018) energy consumption scheduling
Zhao, (2014) Limit both the time and cost of processing and increase Task Scheduling
resource utilization
Miglani and Sharma, Reduce makespan time and increase success ratio Task Scheduling
(2019)

5.7. Binary PSO

From previous versions of the PSO was based on continuous space, where the entire particle has
changed their position according to their neighbors. But it shows an optimization problem in
discrete space that can be solved by Binary PSO. Author in (Kennedy and Eberhart, 1997)
proposed the calculation to work on discrete twofold factors. In the twofold form, particles are
changing their situation in the likelihood that can arrange will take the drifting point number in
the middle of zero and one worth. Actually, particles move shown in Eq. (42) and Particle
change their position by Eq. (43).

Vid = Vid + φ(Pid ― Xid) + φ(Pgd ― Xid) … (42)

Xid = Xid + Vid … (43)

Where, φ is a random positive number, vid is the particle velocity and Xid is the position.

Author in (Bansal and Deep, 2012) proposed Modified Binary PSO (MBPSO) algorithm to solve
0-1 Knapsack Problem and Multidimensional Knapsack Problem. Sigmoid function is used to
normalize the velocity of particle and move the particle in discrete space. Table 11 shows the
properties of modified PSO-based techniques.

Table 11
Binary PSO.
Author Parameters Objective
Kennedy and Eberhart, Flexible and robust Discrete binary Version
(1997)
Bansal and Deep, Reliability, cost and quality of Solve Knapsack Problem
(2012) solution.
 5.8. Hybrid PSO

It is the combination of two or more different types of PSO scheduling algorithm to overcome
the issues of previous PSO method. Table 12 describes the properties of the Hybrid PSO-based
technique.

Author in (Jena et al., 2020) proposed QMPSO algorithm which is combine both modified
particle swarm optimization (MPSO) and improved Q-learning algorithm. This algorithm is used
to minimize makespan time i.e. represent in Eq. (44) and (45), increase the machine performance
and maximize the throughput of VMs that based on three fitness function, which is represent in
Eq. (46), (47) and (48). Final fitness function is used in QMPSO algorithm is represented in Eq.
(49).

MS = Max(Tj) … (44)
n
Tj = ∑i = 1Uij × TCij … (45)
p
F1 = ∑j = 0|LHj ― AL| … (46)
m
F2 = EC = ∑j = 1([Tj × αj + (MS ― Tj)βj]) × PSj … (47)
NIPTi
F3 = w1 × MNIPS + w2 × Li … (48)

F = λ1 × F1 + λ2 × F2 + λ3 × F3 … (49)

Where, MS is the makespan, Tj is the execution time for jth VM, Uij is the decision variable, TCij
is the task completion time, LHj is the total load on VM, AL is the average load, EC is the energy
consumption, PSj is the processing time, NIPTi is the number of instruction, MNIPS is the
maximum number of instruction, Li is the delay cost, αj and βj both are joules per millions of
instruction. Here, λ1, λ2 𝑎𝑛𝑑 λ3 are represents weight and the values are 1, 0.5 and 0.5
respectively.

Author in (Jacob and Pradeep, 2019) proposed hybrid algorithm known as Cuckoo Search and
PSO (CPSO) algorithm to decrease both cost and makespan value. Fitness value, makespan time
and cost are represented in Eq. (50), (51) and (52) respectively.

Fitness Function(x) = Minimize∑xMakespan(x), Cost(x) … (50)

Cost(x) = CPU(x) +Memory(x) … (51)

k
Makespan(x) = ∑i = 1Di … (52)

Where, Di is the deadline of task.

Author in (Merly and Jayalekshmi, 2016) proposed Hybrid Discrete Particle Swarm
Optimization (HDPSO) algorithm that can be used to minimize execution cost while meeting the
deadline which is represent in Eq. (53). Also, minimize makespan time, represent in Eq. (54).
HDPSO is a hybrid of DPSO and Min-Min to overcome the local search capability of PSO.

TEC = EC + TC … (53)

Makespan = Max{Fi} … (54)

Where, TEC is the total execution time, EC is the execution cost, TC is the transfer cost and Fi is
the finishing time of task i.

Author in (Huang and Chen, 2009) proposed Shadow Hybrid PSO (SHPSO) algorithm to solve
the flow-shop scheduling problem (FSSP) i.e. known as makespan time. The objective of this
algorithm is used to minimize makespan time, which is represented as in Eq. (55).

C(i, Jj) = max {C(i,Jj ― 1),C(i ― 1,Jj)} +T(i,Jj) … (55)

Where,Jj is the number of jobs, C(i, Jj) is the completion time and T(i,Jj) is the processing time.

Author in (Mansouri et al., 2019) proposed a hybrid task scheduling algorithm named FMPSO
based on Fuzzy system and Modified Particle Swarm Optimization technique to enhance load
balancing and cloud throughput. Objectives of FMPSO strategy are to reduce the execution time
and makespan time which can increase resource efficiency. Eq. (56) and (57) represent execution
time and makespan time.
n TaskLength(i)
execj = ∑i = 1xij × NPE(j) × VMmips(j) … (56)

Makespan = min{Fti} … (57)

Where,NPE(j) is the number of processing element in jth VM, Fti is the finishing time of task i,
VMmips(j) is the CPU speed and xij is the decision variable.

Author in (Sardaraz and Tahir, 2019) proposed a hybrid workflow scheduling algorithm that
reduces the execution time to take higher priority of task within the task list in the first step and
reduce makespan as well as cost in second step. The algorithm also monitors the load balancing
to efficiently utilize cloud resources. Eq. (58) represents makespan time and Eq. (59) represents
cost respectively.

MS = FTni= 1[taskitime] ... (58)

MC = TEC + TTC … (59)

Table 12
Various Hybrid PSO.
Author Objectives Scheduling Type
Jena et al., (2020) Minimize makespan time, increase VM Task Scheduling
 performance and maximize the throughput
Jacob and Pradeep, (2019) Decrease both cost and makespan value Task Scheduling
Merly and Jayalekshmi, Minimize execution cost and makespan time Workflow
(2016) Scheduling
Huang and Chen, (2009) Decrease makespan time Flow-shop
Scheduling
Mansouri et al., (2019) Reduce the execution time and makespan time Task Scheduling
Sardaraz and Tahir, (2019) Reduce, execution time, makespan time and cost Workflow
Scheduling

5.9. Parallel PSO

In view of advances in figuring advances have improved equal calculations, which give a few
focal points contrasted with serial algorithms. The main objective of parallel computing is split
huge problems into simpler one and allocates them into various nodes to find an optimal solution
in quick time. Parallel computing has many advantages than serial computing such as: reduce
time consumption, cost optimize, increase the rate at which complex problems are solved and so
on. Author in (Alkayal, 2018) proposed PPSO algorithm that focused so many objectives such as:
increase the throughput, lessen calculation time, and improve the worldwide hunt that maintain a
strategic distance from to falling in nearby assembly. PPSO handles the neighborhood and
worldwide assembly of the issues to impart between different swarms. The inquiry space of this
calculation is break into littler spaces and running the swarm in various hubs. Once more, a few
hubs to run numerous swarms in equal and the fundamental hub utilized for refreshing the
following cycle at that point chooses the best swarm appeared in Fig. 9.

Fig. 9. Flowchart of PPSO

Numbers of objective functions are defining shown in below:

(a) Task Execution Cost(TEC): Total cost required to finish the task in the VM, shown by
Eq.(60).

TEC(i,j) = (T(i) × costCPU(j)) + (R(i) × costRAM(j)) + (S(i) × CostStorage(j)) +(𝐷(𝑖)

× costB(j)) … (60)

Where,T(i) is the execution time of task, costCPU(j) is the cost of CPU processing, R(i) is the
size of RAM, costRAM(j) is the cost of memory, S(i) is the storage, costStorage(j) is the cost
of storing data, D(i) is the size of task and costB(j) is the cost of transferring data.

(b) Network Delay: Time required moving data among two center points. It appeared in Eq. (61):

NDelay(i,j) = TL(i)/B(j) … (61)

Where,NDelay(i,j) is the delay time, TL(i) is the task length and B(j) is the bandwidth of VM.

(c) Data Center Load: It shows the load of a datacenter and appeared in Eq. (62), where it
focused on host load, show in Eq. (63).
m
DCload = ∑j = 1Hload(j) … (62)

n
Hload(j) = (∑i = 0usedMIPS(i)/TotalMIPS(j))/100 … (63)

Where,DCload is the load of datacenter, Hload(j) is the host load, used MIPS(i) is the MIPS
used by all task, TotalMIPS(j) total MIPS in VM, m is the number of host and n is the number
of VM.

(d) Fitness function is evaluated by using a weighted sum approach, as shown in Eq. (64).

MinF(xi) = 0.4 ∗ TEC(i) +0.3 ∗ NDelay +0.3 ∗ DCload … (64)

6. Discussions

This section shows the concluding remark achieved from comparative analysis of different PSO
scheduling approaches in cloud computing. Essential objectives are analysed in existing
approaches are shown in Table 13. Various task scheduling and workflow scheduling are used in
proactive based PSO scheduling are presented in Fig. 10. It is clear that most of researcher gave
more emphasis on task scheduling rather than workflow scheduling. The comparison of various
objectives of PSO-based scheduling algorithms is shown in Fig. 11. It is clear that most of
researcher give priority to reduce cost of services that user can get from cloud. From Fig. 12, it is
clearly showing that most of the reactive approaches fall under meta-heuristic scheduling which
contribute to 48.3%, followed by heuristic and hybrid approaches, each contributes 37.9% and
13.8% respectively.

Table 13
Various PSO based meta-heuristic scheduling algorithm.

Execution Time
Makespan time

Throughput

Scalability
Utilization

Reliability
Resource
Author

Energy
Cost
Guo et al., 2012 √ √
Zhang and Zuo, 2013 √
Yang et al., 2013 √ √
Huang et al., 2013 √ √
Zuo et al., 2014 √ √ √ √ √
Chen and Zhang, 2012 √ √ √ √
Verma and Kaushal, 2014 √ √
Beegom and Rajasree, 2014 √ √
Alkayal et al., 2016 √ √
Jena, 2015 √ √ √
Chitra et al., 2014 √ √
Zhou et al., 2018 √ √
Zhao, 2014 √ √ √ √
Abdi et al., 2014 √ √
Kumar and Sharma, 2018 √ √ √ √ √ √
Bansal and Deep, 2012 √ √ √
Jena et al., 2020 √ √
Jacob and Pradeep, 2019 √ √
Merly and Jayalekshmi, 2016 √ √
Mansouri et al., 2019 √ √
Alkayal, 2018 √ √ √ √
Sardaraz and Tahir, 2019 √ √ √
Miglani and Sharma, 2019 √

Fig. 10.Shows PSO scheduling schemes

 Fig. 11.Objectives of PSO meta-heuristic scheduling schemes

Fig. 12. Percentage based on state of scheduling types

In this paragraph we discuss implementation aspects of PSO algorithms in cloud computing

environments. Let us take an example, how to calculate makespan and resource utilization
parameters that based on PSO algorithm in cloud computing. In this taks, the author (Pradhan
and Bisoy, 2020) scheduled six tasks on three VMs as shown in Table 14. All tasks are randomly
assigned to VMs, additionally shows the execution time of each task on VM. Table 15 and 16
shows two task scheduling examples. According to author in (Pradhan and Bisoy, 2020),
makespan is the highest execution time of the task that allocated among all VMs and resource
utilization is the total execution time of all task with respect to highest execution time or
makespan. As shown in Table 15, the makespan is 14 seconds and utilization of VM1, VM2 and
VM3 are 1, 0.64 and 0.93 respectively. In Table 16, the makespan is 13 seconds and utilization
of VM1, VM2 and VM3 is 1, 0.62 and 1 respectively. It tends to be presumed that all tasks are
distributed to VMs so that makespan is decreased, resource utilization is expanded and the
throughput of our system will be increased.
Table 14
Execution time of tasks on VMs (Pradhan and Bisoy, 2020).
Machines Task 1 Task 2 Task 3 Task 4 Task 5 Task 6
VM 1 9 10 7 8 5 6
VM 2 6 2 6 10 11 7
VM 3 5 10 7 6 4 8

Table 15
Task scheduling example1(Pradhan and Bisoy, 2020).
Machines Assigned Task Execution Time (ET) Resource Utilization (𝑹𝒖)
VM 1 T 1, T5 9+5=14 14/14=1
VM 2 T2, T6 2+7=9 9/14=0.64
VM 3 T3, T4 7+6=13 13/14=0.93

Table 16
Task scheduling example2 (Pradhan and Bisoy, 2020).
Machines Assigned Task Execution Time (ET) Resource Utilization (𝑹𝒖)
VM 1 T4, T5 8+5=13 13/13=1
VM 2 T2, T3 2+6=8 8/13=0.62
VM 3 T1, T6 5+8=13 13/13=1

In this paragraph we have presented research issues and challenges of cloud computing. Cloud
computing is a branch of innovation, that can be utilized by people and business advancement.
Like all other innovation, cloud computing also has weaknesses. For cloud computing
environments, we need an effective methods to: (a) fulfill the requirements of cloud service
providers; (b) control the performance of program; (c) completely use cloud computing benefits;
so that the life span of cloud computing is more increased. Moreover, by investigating the
present strategies, it has been fund that desired goals of optimization which includes energy
efficiency, minimize cost, reduce makespan, maximize throughput and reliability of cloud
computing network is achieved. Due to dynamic nature of cloud computing environment there
are couple of issues still required to be researched. Specifically, it is important to lead top to
bottom investigation in measuring the energy consumption, carbon emission, cost of service,
renewable energy-based cloud computing, security, flexibility, handling long running jobs and
migration. Each of the above terms is explained below.

Conclusion

In this paper a systematic survey of resource management and concept of scheduling in cloud
computing environment is presented. A thorough investigation of the most well-known PSO
based meta-heuristic swarm intelligent optimization algorithms for cloud computing
environment is done. These days, swarm intelligent optimization algorithms are viewed as a
productive way to deal with improve network limit, load balancing, resource utilization and
achieve all QoS parameters under heavy data traffic conditions. To increase the system
performance in cloud computing environment, author moved to PSO approaches. In this paper
we have made a thorough audit and investigation on PSO based meta-heuristic swarm
optimization algorithm in cloud computing.

Because of dynamic and distributed nature of cloud computing environment, different authors
focused on modified and hybrid PSO algorithm to achieve their optimal solution. Among various
meta-heuristics approach studied, GA and PSO is the most important techniques which gives
strong optimal result than others meta-heuristic methods. However, PSO is easy to understand
implement and more computationally efficient than GA technique. As per our observation, PSO
has the following advantages over GA: simple concept, easily programmable, faster in
convergence and mostly provides better solution.

References

Abdi, S., Motamedi, S. A., Sharifian, S., 2014: Task Scheduling using Modified PSO Algorithm
in Cloud Computing Environment. International Conference on Machine Learning, Electrical
and Mechanical Engineering (ICMLEME), pp. 37-41. http:// dx.doi.org/10.15242/
IIE.E0114078.
Acharya, J., Mehta, M., & Saini, B., 2016. Particle swarm optimization-based load balancing in
cloud computing. International Conference on Communication and Electronics Systems
(ICCES). https://doi:10.1109/cesys.2016.7889943.
Alguliyev, R. M., Imamverdiyev, Y. N., Abdullayeva, F. J., 2019. PSO-based Load Balancing
Method in Cloud Computing. Automatic Control and Computer Sciences. Vol. 53, No. 1, pp.
45–55.
Alkayal, E., 2018. Optimizing resource allocation using multi-objective particle swarm
optimization in cloud computing systems.Electronics & Computer Science. Doctoral Thesis, pp.
1-234.
Alkayal, E. S., Jennings, N. R., Abulkhair, M. F., 2016. Efficient Task Scheduling Multi-
Objectives Particle Swarm Optimization in Cloud Computing. 41st Conference on Local
Computer Networks Workshops, IEEE.
Allaoua, B., Abderrahmani, A., Brahim, G., Nasri, A., 2008. The Efficiency of Particle Swarm
Optimization Applied on Fuzzy Logic DC Motor Speed Control. Serbian Journal of Electrical
Engineering. Vol.5, pp.247-262.
Alworafi, M., Dhari, A., Al-Hashm, A., A., Darem, A., B., Suresha., 2016. An Improved SJF
Scheduling Algorithm in Cloud Computing Environment. International Conference on
Electrical, Electronics, Communication, Computer and Optimization Techniques
(ICEECCOT), pp. 208-212.
Babukarthik, R. G., Adiga, V. A. K., Sambasivam, G., Chandramohan, D., Amudhavel, J., 2020.
Prediction of COVID-19 Using Genetic Deep Learning Convolutional Neural Network
(GDCNN). IEEE Access, Vol. 8, pp. 177647-177666.
Bansal, J.C., Deep, K., 2012. A Modified Binary Particle Swarm Optimization for Knapsack
Problems. Applied Mathematics and Computation, Elsevier, pp. 11042–11061.
Beegom, A.A., Rajasree, M., 2014. A Particle Swarm Optimization Based Pareto Optimal Task
Scheduling in Cloud Computing. In: Advances in Swarm Intelligence. Springer, pp. 79–86.
Chawda, P., Chakraborty, P., S., 2016. An Improved Min-Min Task Scheduling Algorithm for
Load Balancing in Cloud Computing. International Journal on Recent and Innovation Trends
in Computing and Communication (IJRITCC). Vol. 4, Issue 4, pp. 60 – 64.
Chen, W. N., Zhang, J., 2012. A Set-Based Discrete PSO for Cloud Workflow Scheduling with
User-Defined QoS Constraints. International Conference on Systems, Man, and Cybernetics.
IEEE, pp. 773–778.
Chitra, S., Madhusudhanan, B., Sakthidharan, G., Saravanan, P., 2014. Local Minima Jump PSO
for Workflow Scheduling in Cloud Computing Environments. Advanced in Computer Science
and Its Applications. Springer, pp. 1225–1234. https://doi.org/10.1007/978-3-642-41674-
3_170.
Dasgupta, K., Mandal, B., Dutta, P., Mondal, J. K., Dam, S., 2013. A Genetic Algorithm (GA)
based Load Balancing Strategy for Cloud Computing. Procedia Technology 10, 340-347.
Devi, D.C., Uthariaraj, V.R., 2016. Load Balancing in Cloud Computing Environment Using
Improved Weighted Round Robin Algorithm for Non preemptive Dependent Tasks. The
Scientific World Journal. Volume 2016, Article ID 3896065, pp. 1-14.
Elzeki, O., M., Reshad, M., Z., Elsoud, M., A., 2012. Improved Max-Min Algorithm in Cloud
Computing. International Journal of Computer Applications. Vol.50, No.12, pp. 22-27.
Golchi, M. M., Saraeian, S., Heydari, M., 2019. A hybrid of firefly and improved particle swarm
optimization algorithms for load balancing in cloud environments: Performance evaluation.
Computer Networks. Elsevier, Vol. 162, pp 1-15.
Guo, L., Zhao, S., Shen, S., Jiang, C., 2012. Task scheduling optimization in cloud computing
based on heuristic algorithm. Journal of Networks, No. 7, pp.547–553.
Hamad, S. A., Omara, F. A., 2016. Genetic-Based Task Scheduling Algorithm in Cloud
Computing Environment. International Journal of Advanced Computer Science and
Applications. Vol. 7, No. 4, pp. 550-556.
Heuristic (computer science), 2019. Retrieved from https://en.wikipedia.org/wiki/Heuristic_
(computer science).
Huang, J., Wu, K., Leong, L. K., Ma, S., Moh, M., 2013. A Tunable Workflow Scheduling
Algorithm Based on Particle Swarm Optimization for Cloud Computing. The International
Journal of Soft Computing and Software Engineering [JSCSE], Vol. 3, No. 3, pp. 351-358.
DOI: 10.7321/jscse. v3.n3.53.
Huang, S.Y., Chen, C. L., 2009. A Hybrid Particle Swarm Optimization Algorithm with
Diversity for Flow-Shop Scheduling Problem. Fourth International Conference on Innovative
Computing, Information and Control, pp. 864-867.
Ismail, A., Jeng, D., 2013. SEANN: a self-evolving neural network based on PSO and JPSO
algorithms. International Journal of Civil and Environmental Engineering. Vol. 6, No. 9,
pp.723–729.
Jacob, T. P., Pradeep, K., 2019. A Multi-objective Optimal Task Scheduling in Cloud
Environment Using Cuckoo Particle Swarm Optimization. Wireless Personal
Communications. Springer. https://doi.org/10.1007/s11277-019-06566-w.

Jena, R. K., 2017. Task scheduling in cloud environment: A multi-objective ABC framework.
Journal of Information & Optimization Sciences. Taylor & Francis, Vol. 38, No. 1, pp. 1–19.
https://doi.org/10.1080/02522667.2016.1250460.
Jena R. K., 2015. Multi objective Task Scheduling in Cloud Environment Using Nested PSO
Framework. Procedia Computer Science. Elsevier, Vol. 57, pp.1219 -1227.
Jena, U. K., Das, P. K., Kabat, M. R., 2020. Hybridization of meta-heuristic algorithm for load
balancing in cloud computing environment. Journal of King Saud University- Computer and
Information Sciences. Elsevier, pp 1-11. https://doi.org/10.1016/j.jksuci.2020.01.012.
Kanani, B., Maniyar, B., 2015. Review on Max-Min Task scheduling Algorithm for Cloud
Computing. Journal of Emerging Technologies and Innovative Research (JETIR). Vol.2, No.3,
pp. 781-784.
Kaur, A., Kaur, B., 2019. Load balancing optimization based on hybrid Heuristic- Metaheuristic
techniques in cloud environment. Journal of King Saud University- Computer and Information
Sciences. Elsevier, pp 1-12. https:// doi.org/10.1016/ jksuci.2019.02.010.
Kennedy, J., Eberhart, R.C., 1995. Particle Swarm Optimization. In Proceedings of the IEEE
International Conference on Neural Networks. Piscataway, NJ, USA, pp. 1942–1948.
Kennedy, J., Eberhart, R.C., 1997. A discrete binary version of the particle swarm algorithm.
Proc. IEEE Int. Conf. Syst. Man Cybern. Vol.5, pp.4104–4108.
Kenny, V., Nathal, M., Saldana, S.: Heuristic algorithms, 2014. Retrieved
fromhttps://optimization.mccormick.northwestern.edu/index.php/Heuristic_algorithms.
Khurma, R., A., Harahsheh, H., A., Sharieh, A., 2018. Task Scheduling Algorithm in Cloud
Computing Based on Modified Round Robin Algorithm, Journal of Theoretical and Applied
Information Technology.Vol. 96, No.17, pp.5869-5888.
Kumar, M., Sharma, S.C., 2018. PSO-COGENT: Cost and energy efficient scheduling in cloud
environment with deadline constraint. In Sustainable Computing. Informatics and Systems 19,
147–164.
Kumar, M., Sharma, S.C., Goel, A., Singh, S. P., 2019. A comprehensive survey for scheduling
techniques in cloud computing. Journal of Network and Computer Applications. Elsevier, Vol.
143, pp 1–33.
Li, K., Xu, G., Zaho, G., Dong, Y., 2011. Cloud Task Scheduling Based on Load Balancing Ant
Colony Optimization. Sixth Annual ChinaGrid Conference. pp. 3-9.
Li, W., Shi, H., Dec. 2009. Dynamic Load Balancing Algorithm Based on FCFS. In: 4th
International Conference on Innovative Computing, Information and Control (ICICIC). pp.
1528–1531. https://doi.org/10.1109/ICICIC.2009.182.
Lin, J.C.-W., Yang, L., Fournier-Viger, P., Hong, T.-P., Voznak, M., 2016. A binary pso
approach to mine high-utility itemsets. Soft Computing. Springer, pp.1–19.
Liu, X., Liu, J., 2016. A Task Scheduling Based on Simulated Annealing Algorithm in Cloud
Computing, International Journal of Hybrid Information Technology. Vol. 9, No.6, pp. 403-
412.
Mansouri, N., Zade, B. M. H., Javidi, M. M., 2019. Hybrid task scheduling strategy for cloud
computing by modified particle swarm optimization and fuzzy theory. Computers & Industrial
Engineering. Elsevier, Vol. 130, pp.597-633.
Mao, Y., Chen, X., Li, X., 2014. Max–min task scheduling algorithm for load balance in cloud
computing. In: Proceedings of International Conference on Computer Science and Information
Technology. Springer, vol. 255, pp. 457–465.
Masdari, M., Salehi, F., Jalali, M., Bidaki, M., 2016. A Survey of PSO-Based Scheduling
Algorithms in Cloud Computing. Journal of Network and Systems Management. Vol. 25, No.
1, pp. 122–158.
Merly, M., Jayalekshmi, S., 2016. Improved Scheduling of Scientific Workflows Using HDPSO
IARJSET, Vol. 3, Special Issue 3, pp. 42-46. DOI 10.17148/IARJSET.
Metaheuristics, 2020. Retrieved fromhttps://en.wikipedia.org/wiki/Metaheuristic.
Miglani, N., Sharma, G., 2019. Modified Particle Swarm Optimization based upon Task
categorization in Cloud Environment. IJEAT, Vol. 8, Issue 4C, pp. 67-72.
Milan, S. T., Rajabion, L., Ranjbar, H., Navimipour, N. J., 2019. Review Nature inspired meta-
heuristic algorithms for solving the load-balancing problem in cloud environments. Computers
and Operations Research, Vol.110, pp.159-187.
Mondal, R., K., Nandi, E., Sarddar, D., 2015. Load Balancing Scheduling with Shortest Load
First. International Journal of Grid Distribution Computing. Vol. 8, No.4, pp. 171-178.
Nilesh, A. M., Patel, C. A., 2017. Load Balancing in Cloud Computing Using Ant Colony
Optimization. International Journal of Computer Engineering & Technology (IJCET). Vol.8,
No. 6, pp. 54–59.
Parsopoulos, K. E., Vrahatis, M. N., 2008. Chapter 2: Multi-Objective Particles Swarm
Optimization Approaches. Retrieved from
https://bee22.com/resources/PARSOPOULOS%202008.pdf.
Pasare, G., Gade, A., Bhat, R., 2019. Enhance Dynamic Heterogeneous Shortest Job first
(DHSJF): A Task Scheduling Approach for Heterogeneous Cloud Computing Systems. IOSR
Journal of Engineering (IOSRJEN). Vol. 17, pp.68-72.
Pasha, N., Agarwal, A., Rastogi, R., 2014. Round robin approach for VM load balancing
algorithm in cloud computing environment. International Journal of Advanced Research in
Computer Science and Software Engineering. Vol. 4, No.5, pp. 34–39.
Pradhan, A., Bisoy S.K., 2020. A Novel Load Balancing Technique for Cloud Computing
Platform based on PSO, Journal of King Saud University - Computer and Information Sciences,
doi: https://doi.org/10.1016/j.jksuci.2020.10.016

Raju, R., Babukarthik, R. G., Dhavachelvan, p., 2013. Hybrid Ant Colony Optimization and
Cuckoo Search Algorithm for Job Scheduling.Proceedings of the Second International
Conference on Advances in Computing and Information Technology (ACITY).Vol.2, pp.491-
501. DOI: 10.1007/978-3-642-31552-7_50.
Sardaraz, M., Tahir, M., 2019. A Hybrid Algorithm for Scheduling Scientific Workflows in
Cloud Computing. IEEE Access, Vol 7, pp. 186137-186146.
Singh, S., Chana, I., 2016. Cloud resource provisioning: survey, status and future research
directions. Knowledge and Information Systems. Vol. 49, No.3, pp. 1005-1069.
Sobhanayak, S., Turuk, A. K., Sahoo, B., 2018. Task scheduling for cloud computing using
multi-objective hybrid bacteria foraging algorithm. Future Computing and Informatics
Journal. pp. 1-28. doi: 10.1016/j.fcij.2018.03.004.
Tawfeek, M. A., El-Sisi, A. B., Keshk, A. E., Torkey, F. A., 2014. Virtual Machine Placement
Based on Ant Colony Optimization for Minimizing Resource Wastage. Advanced Machine
Learning Technologies and Applications. Springer, pp.153-164.
Tawfeek, M. A., El-Sisi, A. B., Keshk, A. E., Torkey, F. A., 2015. Cloud Task Scheduling Based
on Ant Colony Optimization, The International Arab Journal of Information Technology. Vol.
12, No. 2, pp 64-69.
Thanka, M. R., Maheswari, P. U., Edwin, E. B., 2019. A hybrid algorithm for efficient task
scheduling in cloud computing environment. Int. J. Reasoning-based Intelligent Systems. Vol.
11, No. 2, pp. 134-140.
Tran, T. A., 2018. The Optimization of Marine Diesel Engine Rotational Speed Control Process
by Fuzzy Logic Control Based on Particle Swarm Optimization Algorithm. Future Internet.
Vol.10, No. 99, pp. 1-18. https://doi.org/10.3390/fi10100099.
Tuli, S., Tuli, S., Tuli, R., Gill, S. S., 2020. Predicting the growth and trend of COVID-19
pandemic using Machine Learning and Cloud Computing. pp. 1-13. doi:
https://doi.org/10.1101/2020.05.06.20091900.
Verma, A., Kaushal, S., 2014. Bi-Criteria priority based particle swarm optimization workflow
scheduling algorithm for cloud. Recent Advances in Engineering and Computational Sciences
(RAECS), pp.1–6. https://doi.org/10.1109/RAECS.2014.6799614.
Verma, P., Shrivastava, S., Pateriya, R. K., 2017. Enhancing Load Balancing in Cloud
Computing by Ant Colony Optimization Method. International Journal of Computer
Engineering In Research Trends. Vol.4, Issue 6, pp. 277-284.
Vijayalakshmi, R., Prathibha, S., 2013. A novel approach for task scheduling in cloud. Fourth
International Conference on Computing Communications and Networking Technologies
(ICCCNT). https://doi.org/10.1109/ICCCNT.2013.6726717.
Wei, X. J., Bei, W., Jun, L., 2017. SAMPGA task scheduling algorithm in cloud computing. 36th
Chinese Control Conference (CCC). pp 5633-5637. doi:10.23919/chicc.2017.8028252.
Xiang, B., Zhang, B., Zhang, L., 2017. Greedy-Ant: Ant Colony System-Inspired Workflow
Scheduling for Heterogeneous Computing. IEEE Access. Vol 5,pp. 11404-11412.
Xu, M., Tian, W., Buyya, R., 2017. A survey on load balancing algorithms for virtual machines
placement in cloud computing.Concurrency Computat: Practice and Experience, 29(12),
e4123, pp. 1-16.
Yang, Z., Qin, X., Li, W., Yang, Y., 2013. Optimized task scheduling and resource allocation in
cloud computing using PSO based fitness function. Information Technology Journal. Vol.12,
Issue. 23, pp.7090–7095.
Yao, J., He, J., 2013. Load balancing strategy of cloud computing based on adaptive artificial
bee colony algorithm. Journal of Computer Applications, 32(9), 2448–2450.
Zhang, G., Zuo, X., 2013. Deadline constrained task scheduling based on standard-PSO in a
hybrid cloud. In: Advances in Swarm Intelligence. Springer, pp. 200–209.
Zhao, G., 2014. Cost-aware scheduling algorithm based on PSO in cloud computing
environment. International Journal of Grid & Distributed Computing. Vol.7, No.1, pp.33-42.
Zhou, Z., Chang, J., Hu, Z., Yu, J., Li, F., 2018. A modified PSO algorithm for task scheduling
optimization in cloud computing.Concurrency Computat Pract Experiment, Wiley, pp 1-11.
Zuo, X., Zhang, G., Tan, W., 2014. Self-adaptive learning PSO-based deadline constrained task
scheduling for hybrid IaaS cloud. IEEE Transactions on Automation Science and Engineering.
Vol.11, No. 2, pp.564–573.