Load Balancing Techniques in Cloud Computing: Extensive Review

Abstract

It has become difficult to handle traditional networks because of extensive network developments and an increase in the number of network users, and also because of new technologies like cloud computing and big data. Traditional networks are experiencing an increase in VM load and in the time taken for processing tasks. Hence, it has become essential to modify the traditional network architecture. A notion called Load balancing techniques that increases the conformance of network management was presented recently to deal with this problem. The critical need for load balancing emerges due to network resources limitations and requirements fulfillment that facilitates traffic distribution through various resources to enhance the efficiency and reliability of network resources. This task has been carried out by several researchers before, who have presented various algorithms with their benefits and shortcomings. The focus of this research is on the notion of cloud computing load balancing and on the advantages and disadvantages of a chosen load balancing algorithm. Furthermore, it examines the metrics and issues of these algorithms.

Full text

Advances in Science, Technology and Engineering Systems Journal
Vol. 6, No. 2, 860-870 (2021)
www.astesj.com
Special Issue on Multidisciplinary Sciences and Engineering
ASTES Journal
ISSN: 2415-6698
Load Balancing Techniques in Cloud Computing: Extensive Review
Ahmad AA Alkhatib*, Abeer Alsabbagh,Randa Maraqa, Shadi Alzubi
Alzaytoonah University of Jordan, Amman, 11183, Jordan
ARTICLE INFO ABSTRACT
Article history:
Received: 21 December, 2020
Accepted: 24 March, 2021
Online: 04 April, 2021
Keywords:
Cloud computing
Load Balancing
Software Defined Network
Static Algorithms
Dynamic Algorithm
It has become difficult to handle traditional networks because of extensive network developments
and an increase in the number of network users, and also because of new technologies like cloud
computing and big data. Traditional networks are experiencing an increase in VM load and in
the time taken for processing tasks. Hence, it has become essential to modify the traditional
network architecture. A notion called Load balancing techniques that increases the conformance
of network management was presented recently to deal with this problem. The critical need for
load balancing emerges due to network resources limitations and requirements fulfillment that
facilitates traffic distribution through various resources to enhance the efficiency and reliability
of network resources. This task has been carried out by several researchers before, who have
presented various algorithms with their benefits and shortcomings. The focus of this research is
on the notion of cloud computing load balancing and on the advantages and disadvantages of
a chosen load balancing algorithm. Furthermore, it examines the metrics and issues of these
algorithms.
1 Introduction
Cloud Computing has universally greater interest in web technolo-
gies currently. With the increasing demands of the cloud, popular
website’s servers are getting overloaded, in order to fulfill users
requirement load balancing is one of the promising solutions. Load
balancing is the procedure of sharing the load between multiple
processors in a distributed environment to minimize the turnaround
time taken by the servers to cater service requests and make better
utilization of the available resources. cloud computing incorporates
and enhances the advantages of a few existing distributed computing
technologies. Network computing refers to a “model of distributed
computing that employs geographically distant resources and hence,
allows users to gain access to computers and data and to handle
their accounts from varied locations”. Virtualization is another tech-
nology that hides the physical properties of computing resources
to obscure complexity when dealing with applications, systems or
end-users (1)–(3).
Cloud computing initially represented online business applica-
tions, referred to as application service provision (ASP). Later on
the term was used more extensively following the distribution of
services by the company, where the name given to each service
provider was based on the service they offered to the customer (4).
Since its introduction in late 2006, several definitions of cloud
computing have been formulated. It can essentially be defined as a
framework for allowing on-demand network access to a shared set
of online configurable computing resources (like network, storage,
server, services and applications) (5),(6).
The definition of cloud computing provided by HP is as fol-
lows: “Everything as a Service” (
7
). On the other hand, Microsoft
considers cloud computing to signify “Cloud and the Client” (
8
).
According to T-Systems, cloud computing refers to “renting infras-
tructure, software and bandwidths under specified service conditions.
It should be possible to modify these components routinely on the
basis of the customer requirements and should be widely available
and secure. Furthermore, there are 2-end service level agreements
(SLAs) and use-dependent service invoices that are part of cloud
computing” (9).
It is possible to distinguish cloud computing into three mod-
els, i.e. SaaS (SOFTWARE–AS-A SERVICE), which provides
a user interface to the user which they can access through a
browser or desktop application; PaaS (PLATFORM-AS-A SER-
VICE, which offers applications development tools and a program-
ming language execution environment to the user; and finally IaaS
(INFRASTRUCTURE- AS-A-SERVICE), which offers virtual com-
puterized resources to the user that they can manage in accordance
with their requirements. These models can also be classified in
accordance with the services they offer (10), (11).
•Database as a Service (DaaS)
•Storage as a Service (SaaS)
•Network as a Service (NaaS)
*Corresponding Author: Ahmad AA Alkhatib, Ahmad.Alkhatib@zuj.edu.jo
www.astesj.com
https://dx.doi.org/10.25046/aj060299
860

A. Alkhatib et al. /Advances in Science, Technology and Engineering Systems Journal Vol. 6, No. 2, 860-870 (2021)
•Expert as a Service (EaaS)
•Communication as a Service (CaaS)
•Security as a Service (SECaaS)
•Monitoring as a Service (Maas)
•Testing as a Service (TaaS)
The Cloud Computing categories are shown in Figure 1.
Figure 1: Cloud Computing categories
Cloud computing is becoming quite popular; hence, there has
been a significant increase in the amount of processing being carried
out in the clouds. However, cloud computing is experiencing several
issues in providing the required services (11), including:
•Load Balancing
•Performance analysis and modeling
•Throughput and response time
•Security and privacy
•Resource management
•QoS
The structure of the rest of the paper is as follows: the Load
Balancing Concept and Challenges described in Section 2. Section
3 presents the load balancing techniques classification . A few static
and dynamic load balancing algorithms are discussed in Section 4.
Section 5 Discussion and comparison with the latest studies in this
field are compared and Lastly,section 6 gives a conclusion of the
study .
2 Load Balancing
There are a series of nodes in the cloud that are connected to one
another. Here, one of the nodes is either selected randomly or on
the basis of an algorithm to fulfill requests made by users. There
may be different data and hence, there is different load on every
node, and each node in a cloud can have unequal load of tasks based
on the quantity of work requested by clients. Load balancing is a
very significant issue that made it imperative for cloud computing
to distribute load on the resources available so as to achieve lower
response time. This makes sure that no VMs in the system are over-
loaded at any time (
12
). In addition, balance workload continues to
be an issue in the cloud as they are unable to determine the quantity
of demand that is obtained in the cloud (13).
Load balancing concepts face several challenges because there
are various physical and logical issues that may have an impact on
the technique being used. A few of these issues are listed in Table I
(11):
Table 1: LOAD BALANCING ISSUES
Graphical
Distribution
node
There is geographical distribution of data centers
in clouds. On the other hand, distributed nodes are
considered as a single system without taking into
account various factors like communication delay,
networking delay and distances between nodes,
resources and users.
Virtual
Machine
Migration
Multiple VM are allowed by the VM concept on
the same Physical Machine. Due to the distinct
VM structure, the physical machine may
become overloaded.
Algorithm
Complexity
To ensure that it does not affect the efficiency
of the cloud, load balancing algorithm should
be simple and concise.
Heterogeneous
nodes
In these times, users have different requirements.
It is essential to have heterogeneous nodes to
fulfill users’ need for services. The load balancing
decision is influenced by the heterogeneity of the
nodes.
Single
point of
Failure
In general, load balancing algorithm is typically
carried out on a central node to assign tasks. The
entire computing fails in case the central node fails.
Load
Balancer
Scalability
Computing power, topology, storage, etc.
determine the response time of load balancing
3 Load Balancing Techniques Classifica-
tion
The classification of the algorithms depends on the existing status
of the system and involves two categories (
14
), (
11
): static algo-
rithms that obtain information pertaining to the system and identify
the existing resources for using before commencing. The load is
distributed on the existing VM till the work concludes, which is
preferably used when the VM capabilities are similar to one another.
This type of algorithm has various techniques, for example round
www.astesj.com 861

A. Alkhatib et al. /Advances in Science, Technology and Engineering Systems Journal Vol. 6, No. 2, 860-870 (2021)
robin, Min-Min, weight, Max-Min, honey bee foraging, throttled,
etc. The issue faced with static algorithms is that they depend on
static information which is unable to effectively demonstrate dy-
namic load variations taking place within the VMs (
15
). Dynamic
algorithms are the second kind of algorithms and are different from
static algorithms in the sense that they exhibit flexibility. These
algorithms work on the basis of rescheduling the tasks allocated to
them over the available VM while carrying out the work. Better per-
formance is exhibited by static algorithms in contrast to the dynamic
algorithm, whereas with respect to the competitive ratio, dynamic
algorithms have a larger ratio compared to the static algorithm (
16
).
The typical structure of load balancing in cloud environments is
demonstrated in Figure 2, whereas Figure 3 shows the classification
of the load balancing methods. Several parameters are used to assess
load balancing algorithms, which are regulated by various policies.
The metrics are presented in Table 3 (15), (17), while an outline of
the load balancing policies is given in Table 4.
Table 2: Load Balancing Metrics
Metrics Description
Performance It refers to ensuring the consistency and
efficiency of the algorithms carried out
Response
time
It refers to the time taken by the system
to give a reaction to the user request.
he more rapid the response time, the
higher the satisfaction obtained
by the user.
Throughput The number of tasks carried out for
each unit of time.
Scalability
The ability of the algorithm to handle
the tasks of the user in an efficient and
effective way.
Fault
Tolerance
The robustness of the algorithm to
solve issues and errors so that its
performance can be regained
Migration
Time
The time needed to shift the load from
a certain VM to another in accordance
with the VM loading during situations
of overload or under-load.
Resource
Usage
Determines the optimum use of
computing resources in the data center
that should be employed by the algorithm
Makespan Time needed to carry out and treat a
given set of tasks
Figure 2: General Structure of Load balancing in Cloud Environment (18)
Figure 3: Load balancing techniques classification
Table 3: Load Balancing Policies
Selection
Policy
The required tasks that need to be transferred from
a certain VM to another are specified in this policy.
This is done depending on the extent of overhead
that is to be transferred
Location
policy
In this policy, the tasks are sent to the free, under-
loaded and available VMs so that they can be
fulfilled. The desired VM is determined in this
policy on the basis of the availability of the
required services so that the task can be transferred
in accordance with the available technique, like
Negotiation, Random and Probing.
Transfer
policy
In this policy, the conditions for shifting the task
from a local VM to another local or remote VM
are determined. Two types of tasks are used in this
regard: the existing tasks and the last task received.
Information
policy
This policy deals with keeping information
pertaining to the resources safe in the system to
ensure that other policies can benefit from them
while making decisions.
4 Load Balancing Common Algorithms
A few static and dynamic load balancing algorithms will be pre-
sented in this section, with benefits and drawbacks for each algo-
rithm:
4.1 Round Robin Algorithm
This static algorithm is one of the simplest methods used following
the selection of the existing VMs. One of these VMs is randomly
chosen by the data center unit to commence its operations. In ad-
dition, it is organized in a circular manner. After this, every VM
that gets a request is shifted towards the final part of the list (
19
),
(20)(21),.
www.astesj.com 862

A. Alkhatib et al. /Advances in Science, Technology and Engineering Systems Journal Vol. 6, No. 2, 860-870 (2021)
There is, however, no interaction between this algorithm and
the distinct abilities of the VMs because the tasks are equally di-
vided on the VMs, even though they have distinct abilities (Figure
4) (
15
). Hence, Weight Round Robin algorithm is used to enhance
this algorithm, which assigns weight for each VM on the basis of its
abilities, and then offers the ability to each node to have a specific
number of tasks on the basis of the weights allocated to each VM.
This algorithm is considered similar to Round Robin algorithm from
the perspective of time division because it distributes the time as a
circular. However, the distinction is that the tasks are offered to VM
on the basis of particular restrictions, for example checking weights
(22), (23).
Figure 4: Round Robin Load Balancer (19)
4.2 Throttled Load Balancer (TLB)
A table is generated in this algorithm that includes the virtual ma-
chines as well as the existing state (available/busy). If a specific task
is allocated to a virtual machine, a request is made to the control unit
within the data center, which will look for the ideal VM suit with
respect to their abilities to achieve the required task (
24
). The load
balancer will send -1 back to the data center if an appropriate VM
is not available (
15
). Figure 5 presents a demonstration of Throttled
Load Balancer.
Figure 5: Throttled Load Balancer
In contrast, the process of looking for the ideal virtual machines
always takes place from the start of the table each time; therefore,
certain VMs are not employed. A modified throttled algorithm was
put forward by (
25
), which works by changing the process of VM
selection. After receiving the subsequent request, the VM at index
adjacent to the VM already allocated is selected based on the state of
the VM. In addition, (
22
) presented an efficient throttled algorithm,
which consisted of 3 algorithms, i.e. throttled algorithm, ESCE
(Equally Spread Current Execution algorithm) and Round Robin.
This algorithm is considered as an advancement of the main throt-
tled algorithm, where the enhanced algorithm employed the data
structure to maintain information regarding the VMs. In contrast,
Hash Map index is used to search for VM and allocated tasks where
it works at a faster pace compared to the throttled algorithm.
Furthermore, Divide-and-Conquer and Throttled algorithm
(DCBT) approach is taken to be the hybrid approach because it
gives preference to the order when allocating the VM. This is done
to achieve the highest usage of resources by reducing the overall
time taken to execute the task. In addition, this algorithm seeks to
update the table by dividing the independent jobs equally among all
the VMs. Hence, it works faster than the throttled algorithm (26).
4.3 Min-Min Load Balancing Algorithm
This algorithm is easy to use and works at a faster pace (
27
). In
addition, it improves performance and consists of a series of tasks.
The time taken to execute the task is computed and allocated to
VMs on the basis of the smallest completion time for the existing
tasks. The process will continue till it is ensured that each task has
been allocated to VM (
16
). Because of the existence of a greater
number of smaller tasks, this algorithm performs better compared
to if there were bigger tasks. However, this will lead to starvation
because of giving priority to smaller tasks and deferring the bigger
tasks (21).
4.4 Max Min Load Balancing Algorithm
As stated by (
28
), (
29
), this algorithm is quite similar to the Min-Min
Load Balancing, based on the calculation time. In this algorithm,
all existing tasks are sent to the system, after which the calculation
is carried out for determining the least time to complete each of
the given tasks. The selected task then has the maximum time to
be completed will be allocated to the relevant machine (
11
). A
comparison of the performance of this algorithm with the Min-Min
algorithm shows that the Max-Min algorithm is better because there
is just one large task in the set, which means that the Max-Min
algorithm will carry out the shorter tasks alongside the larger task
(15).
4.5 Opportunistic Load Balancing Algorithm
This algorithm is a type of static algorithm that is not capable of
describing the existing workload of the VM; therefore, it allocates
the tasks randomly to all nodes in the system to ensure they are
all working (
30
), (
31
), and (
32
). Tasks are accomplished at a slow
pace through this algorithm as it does not compute the existing
implementation time (
33
). Hence, it provides incorrect outcomes
for the load balance (15).
www.astesj.com 863

A. Alkhatib et al. /Advances in Science, Technology and Engineering Systems Journal Vol. 6, No. 2, 860-870 (2021)
4.6 Ant Colony Optimization
This algorithm that is based on ACO allows for distributing the
workload efficiently between the nodes of a cloud. The working
of this algorithm is influenced by the Ant concept, where it looks
for a new pathway if it comes across an obstacle and allocates a
new route between the nodes (
34
). The algorithm functions by
first initializing the table, carrying out the data flow and achieving
the threshold level for the required nodes. When the flow passes
through the nodes, the algorithm examines the node; if it is under
load, then it uses the highest trailing pheromone (TP) that provides
the route for the under-loaded node. The table is then updated till
the node achieves the threshold limit. Nonetheless, if the threshold
limit is attained, it uses the Foraging Pheromone (FP) to examine
new sources of food and update the table till it attains an under-load,
after which it reassigns resources. This cycle keeps on occurring till
the completion of the process.
4.7 Honey Bee Algorithm
The working of this algorithm is influenced by the way bees behave
when looking for honey, because its main objective is to divide the
workload on the VM, considering the lack of excessive resource
utilization and lack of under-usage of resources. This algorithm
works by choosing a VM that fulfills two main requirements. Fewer
tasks are allocated to this VM compared to those assigned to other
VM machines. The time taken by VM to process falls within the
average processing time taken by all other VMs (35), (23).
Figure 6: Active Monitoring Load Balancerr
4.8 Active Monitoring Load Balancer (AMLB)
It is a type of dynamic load technology (
21
). This technology ob-
tains information relevant to each VM and to the number of requests
that are presently allocated to each of them (
24
). The Data Center
Controller (DCC) scans the VM index table after receiving a new
request to determine the VM that is least loaded or idle. First-come-
first serve concept is employed by this algorithm to allocate load
to the VM that has the smallest index number for more than two
servers (21).
The VM ID is sent back by the AMLB algorithm to the DCC
which then sends the request to the VM represented by that ID. The
AMLB is informed about the new allocation by the DCC and it is
sent the cloudlet (24).
Once the task is completed, the information is sent to the DCC
and the VM index table is reduced. When a new request is received,
it goes over the table again using load balancer and then the process
allocation occurs. Figure 6 presents an illustration of AMLB.
4.9 Genetic Algorithm
In the process of genetic algorithms, scheduling takes place on the
basis of the biological notion of population generation. The pur-
pose of using this algorithm was to enhance the distribution of load
within the cloud, where the algorithm starts functioning using the
preliminary population procedure. The initial population consists
of the set of all individuals who are involved in determining the
optimal solution. Each solution in the population is referred to as
an individual, and each individual is described as a chromosome
to make it appropriate for carrying out genetic operations (
36
). It
was presumed by the authors in (
37
) that a comparison of the newly
developed population will be carried out with the previous one. The
solutions are then chosen to obtain solutions (offspring) on the basis
of its fitness function (
38
). The fitness function is used to deter-
mine the quality of individuals in the population with respect to the
specified optimization goal (37).
For every chromosome, the fitness function is computed, after
which the most appropriate results are chosen to be used as parents.
The operation then commences with a crossover process; a part of
each parent is used to create a new child; this child is then improved
by employing a mutation process. This process keeps on occurring
till the best results are obtained (
39
). The basis of genetic algorithms
is randomness; however, it is not the same as the random search in
that it approves the most suitable individuals within a population.
The crossover rate and mutation probability values have a significant
impact on the performance of the genetic algorithm (38).
The general scheme of the genetic algorithm is shown in Figure
7.
Figure 7: General scheme of genetic algorithm (40)
4.10 First Come First Serve
This algorithm works by distributing new tasks to resources that
have the least waiting time (i.e. resources that have the least number
of tasks). Here, there is sequential execution of the tasks, with the
work commencing from the first task till its completion, after which
the subsequent task from the queue is carried out. This algorithm is
used as it allocates tasks to the virtual machines without considering
the specifications of the virtual machines available and also the time
that will be taken by the tasks in queue when allocating new tasks
(41).
www.astesj.com 864

A. Alkhatib et al. /Advances in Science, Technology and Engineering Systems Journal Vol. 6, No. 2, 860-870 (2021)
4.11 Generalized Priority Algorithm
The mechanism of action of this algorithm is to give priority to the
tasks according to their size (Higher missions size take higher prior-
ity), and the default machines are given priority based on the power
of the processor, then the appropriate virtual machine is chosen for
the priority of the required task (
41
). An example on GPA suppose
you have 6 VMs represented by their Id and processor speed VMs
=0, 850 , 1, 1000, [5], 3, 150, 4,250, 5,750. Here the priority goes
to VM 4 will because it has the highest speed, then priority is given
to VM 2 and then VM 6 and so on (
41
). This algorithm works as in
Figure (pseudo):
Figure 8: K. Generalized Priority Algorithm Pseudo code
4.12 NBST Algorithm
Initially, this algorithm assumes that the number of instructions
required for the tasks is on the waiting list and their length is known,
then it arranges all available virtual devices according to their im-
plementation speed MIPS (Million instructions per second) and
arranging the cloudlets according to their length. Where this algo-
rithm depends on dividing the number of VM and cloudlets into two
parts after arranging them in descending order until access to the
VM or cloudlets one at most, then assigning the VM groups to the
cloudlets groups (42).
4.13
Load Balancing Technique based on Cuckoo
Search and Firefly
The cuckoo search is based on the brooding behavior of cuckoo bird,
where this bird laid their eggs on other bird’s nest, the best quality
eggs are carried forward to the next generation and the worst nests
are abandoned (
43
). Cuckoo Search with Firefly is a hybrid algo-
rithm is used for loud balancing in a cloud computing environment;
this algorithm schedules the tasks and allocates the overloaded VMs
tasks to the under-loaded VMs. This algorithm finds the best VM in
less time and improves the loud balance efficiency and avoids the
task imbalanced situations in the entire system (
44
). The LB tech-
nique starts with scheduling the tasks using Round Robin method,
then calculate the capacity and the load of each VM, then it iden-
tifies the overloaded and under-loaded tasks using Cuckoo Search
with Firefly to start the migration of tasks from overloaded VMs
to under-loaded VMs (
44
). It avoids the task imbalanced situations
in the entire system. Also this method has the advantages of high
interchange rate and less number of tuning parameter. This method
has the advantages of high interchange rate and less number of tun-
ing parameter. The proposed method migrates 2 task, while existing
method requires 7 task for migration.
4.14
predicted load balancing algorithm based on the
dynamic exponential smoothing
This algorithm predicts the load of the VMs by using the dynamic
exponential smoothing model. Where the dynamic balancing algo-
rithms focus on the fastest response speed algorithm and the least
connection algorithm (45).
Exponential smoothing is one of prediction algorithms based
on time series, which takes advantage of all historical data, and
differentiates them through the smoothing factor to allow recent
data make a greater impact on the analytical value than long-term
data, (46).
The allocated node has the shortest corresponding time, when
a new request arrives to the server node that has a little connection
numbers, the dynamic algorithm takes the load characteristics of
the server node as the reference, makes the adjustment with every
weighted factor, and reflects the real-time load under the support
of weighted value, by that the algorithm helps to find the corre-
sponding smoothing coefficient with the VM load time series of
current phrase, and helps to make prediction with the load value at
the next moment of this VM(
45
). The load predicted balancing al-
gorithm developed by the improvement that permits to construct the
dynamic smoothing exponent and to obtain the load prediction with
a higher accuracy through the comparison of short load time series.
Also it helps to provide more accurate service demand for users and
enhance the resource utilization ratio of the server node to a higher
level (
45
). The accuracy of the model prediction can be affected,
because of the selection standard of
α
exponential smoothing is not
very clear (45).
5 Discussion
A comparison of the load balancing algorithms is presented in this
section. Each load balancing algorithm is described in specific
identifiers with its main properties and limitations.
5.1 Round Robin
•Algorithm Name: [Round Robin].
•Algorithm Type: [Static].
•Algorithm Overhead: [No Overhead].
•Algorithm Degree of complexity: [Simple].
•Algorithm Strength points :
–Simplicity
–Easy to install
•Algorithm Limitations:
–Cannot be improved any more
–No multi-tasking capabilities
www.astesj.com 865

A. Alkhatib et al. /Advances in Science, Technology and Engineering Systems Journal Vol. 6, No. 2, 860-870 (2021)
–Could lead to over head
–
Does not take capacity importance and task duration
into account
•Algorithm Performance: [Low Performance].
•Algorithm Throughput: [Low Throughput].
•Algorithm Fault tolerance: [No FT].
5.2 Throttled
•Algorithm Name: [Throttled].
•Algorithm Type: [Static].
•Algorithm Overhead: [No Overhead].
•Algorithm Degree of complexity: [Medium].
•Algorithm Strength points :
–
The data center searches for the best VM that fits its
capabilities with the task required
–Maximization of resource usage
–Increase average of execution time
•Algorithm Limitations:
–Does not define time limits
–Could result in idle VMs
•Algorithm Performance: [medium Performance].
•Algorithm Throughput: [high Throughput].
•Algorithm Fault tolerance: [FT].
5.3 Min-Min
•Algorithm Name: [Min-Min].
•Algorithm Type: [Static].
•Algorithm Overhead: [Medium Overhead].
•Algorithm Degree of complexity: [Simple].
•Algorithm Strength points :
–Quick
–Easy to install
•Algorithm Limitations:
–Leads to starvation
•
Algorithm Performance: [Increased performance for smaller
tasks].
•Algorithm Throughput: [Improved Throughput].
•Algorithm Fault tolerance: [No FT].
5.4 Max-Min
•Algorithm Name: [Max-Min].
•Algorithm Type: [Static].
•Algorithm Overhead: [Medium Overhead].
•Algorithm Degree of complexity: [Simple].
•Algorithm Strength points :
–Execute the large tasks in parallel with short tasks
•Algorithm Limitations:
–Leads to starvation.
•
Algorithm Performance: [Better performance than min-min].
•Algorithm Throughput: [Improved Throughput].
•Algorithm Fault tolerance: [No FT].
5.5 Opportunistic
•Algorithm Name: [Opportunistic].
•Algorithm Type: [Static].
•Algorithm Overhead: [Low Overhead].
•Algorithm Degree of complexity: [Simple].
•Algorithm Strength points :
–The advantage is quite simple and reach load balance
•Algorithm Limitations:
–No fairness in task assignment
–
Tasks are planed slowly because it does not determine
the nodes current execution time
–the whole completion time (Make span) is very poo
•Algorithm Performance: [poor Performance].
•Algorithm Throughput: [Limited Throughput].
•Algorithm Fault tolerance: [No FT].
www.astesj.com 866

A. Alkhatib et al. /Advances in Science, Technology and Engineering Systems Journal Vol. 6, No. 2, 860-870 (2021)
5.6 Ant Colony Optimization
•Algorithm Name: [Ant Colony Optimization].
•Algorithm Type: [Dynamic].
•Algorithm Overhead: [High Overhead].
•Algorithm Degree of complexity: [High complexity].
•Algorithm Strength points :
–The style of living ants depends on the
–transition from VM to another VM.
–Better performance
–Reduce operation time
–Can be combined with heuristic search algorithm
•Algorithm Limitations:
–Complex and load on the system unit.
–Untrusted in reality
•Algorithm Performance: [High Performance].
•Algorithm Throughput: [High Throughput].
•Algorithm Fault tolerance: [FT].
5.7 Honey Bee
•Algorithm Name: [Honey Bee].
•Algorithm Type: [Dynamic].
•Algorithm Overhead: [High Overhead].
•Algorithm Degree of complexity: [High complexity].
•Algorithm Strength points :
–
it is chosen the best VM by comparing the cost of exe-
cuting a task on one VM with the other available VMs.
–
Takes into account excessive resources and under used
ones
–
Number of assigned tasks less than number of handled
by other VMs to reduce execution time
•Algorithm Limitations:
–Complexity
–
Increase the machine size does not increase throughput
•Algorithm Performance: [High Performance].
•Algorithm Throughput: [Low Throughput].
•Algorithm Fault tolerance: [FT].
5.8 AMLB
•Algorithm Name: [AMLB].
•Algorithm Type: [Static].
•Algorithm Overhead: [High Overhead].
•Algorithm Degree of complexity: [medium complexity].
•Algorithm Strength points :
–It identifies the least loaded VM
–Increase performance
–Update index table every time task executed
•Algorithm Limitations:
–Time consuming in index table calculation
–
algorithm gives better results when there is low varia-
tion in workload
•Algorithm Performance: [High Performance].
•Algorithm Throughput: [High Throughput].
•Algorithm Fault tolerance: [FT].
5.9 Genetic Algorithm
•Algorithm Name: [Genetic Algorithm].
•Algorithm Type: [Dynamic].
•Algorithm Overhead: [High Overhead].
•Algorithm Degree of complexity: [High complexity].
•Algorithm Strength points :
–Executed the cloudlets in less time
–Improve load distribution in the cloud
–Better resource usage
–
Better approve the best fitted population indexing in
VMs
•Algorithm Limitations:
–Complexity
–
High resources required for population generation and
calculation
–
Crossover mutation is highly effected by probability
parameters
•Algorithm Performance: [High Performance].
•Algorithm Throughput: [High Throughput].
•Algorithm Fault tolerance: [FT].
www.astesj.com 867

A. Alkhatib et al. /Advances in Science, Technology and Engineering Systems Journal Vol. 6, No. 2, 860-870 (2021)
5.10 First come first serve
•Algorithm Name: [First come first serve].
•Algorithm Type: [Dynamic].
•Algorithm Overhead: [There is Overhead].
•Algorithm Degree of complexity: [Simple complexity].
•Algorithm Strength points :
–Distribute tasks to all available VMs
•Algorithm Limitations:
–Not preventive
–
assigning tasks to all virtual machines without paying
attention to the specifications of the available virtual
machine
–
not paying attention to the time that the tasks in the
queue will take when distributing the new tasks
•Algorithm Performance: [Medium Performance].
•Algorithm Throughput: [Medium Throughput].
•Algorithm Fault tolerance: [No FT].
5.11 Generalized Priority Algorithm
•Algorithm Name: [Generalized Priority Algorithm].
•Algorithm Type: [Dynamic].
•Algorithm Overhead: [No Overhead].
•Algorithm Degree of complexity: [medium complexity].
•Algorithm Strength points :
–
give priority to the tasks according to their size and the
machine processor power
•Algorithm Limitations:
–take more execution time
–could lead to starvation
•Algorithm Performance: [Increased Performance].
•Algorithm Throughput: [High Throughput].
•Algorithm Fault tolerance: [FT].
5.12 NBST Algorithm
•Algorithm Name: [NBST Algorithm].
•Algorithm Type: [Dynamic].
•Algorithm Overhead: [Yes the is Overhead].
•Algorithm Degree of complexity: [medium complexity].
•Algorithm Strength points :
–Distribute all task on all available VM
•Algorithm Limitations:
–take more time in an assign VMs for cloudlets
•Algorithm Performance: [increased Performance].
•Algorithm Throughput: [High Throughput].
•Algorithm Fault tolerance: [FT].
5.13 Load Balancing based on CSF
•Algorithm Name: [Load Balancing based on CSF].
•Algorithm Type: [Dynamic].
•Algorithm Overhead: [medium Overhead].
•Algorithm Degree of complexity: [medium complexity].
•Algorithm Strength points :
–
schedules the tasks and allocates the overloaded VMs
tasks to the under-loaded VMs
–
avoids the task imbalanced situations in the entire sys-
tem
–
high interchange rate and less number of tuning param-
eter
•Algorithm Limitations:
–multiple algorithms have to be followed
•Algorithm Performance: [Improved Performance].
•Algorithm Throughput: [High Throughput].
•Algorithm Fault tolerance: [FT].
5.14
predicted load balancing algorithm based on the
dynamic exponential smoothing
•
Algorithm Name: [predicted load balancing algorithm based
on the dynamic exponential smoothing].
•Algorithm Type: [Dynamic].
•Algorithm Overhead: [Yes there is Overhead].
•Algorithm Degree of complexity: [High complexity].
www.astesj.com 868

A. Alkhatib et al. /Advances in Science, Technology and Engineering Systems Journal Vol. 6, No. 2, 860-870 (2021)
•Algorithm Strength points :
–
provide more accurate service demand for users and
improve the resource utilization of the VMs
•Algorithm Limitations:
–
the accuracy of the model prediction can be affected, be-
cause of the selection standard of
α
exponential smooth-
ing is not very clear
•Algorithm Performance: [Improved Performance].
•Algorithm Throughput: [Medium Throughput].
•Algorithm Fault tolerance: [No FT].
6 Conclusion and Future Work
This paper presents an extensive Review for load balancing that
carry out load distribution among the VM in a different way. con-
sidering the lack of excessive resource utilization and not keeping
the VMs idle. As a service that is carried out over the network is
known as cloud computing, where a lot of significance is given to
load balancing issues. The performance will decrease with an over-
loaded system. Hence, smart load balancing algorithm is needed to
maintain the position of QoS. a description for algorithm techniques
has been explained and followed by a comparison between them.
Conflict of Interest The authors declare no conflict of interest.
Acknowledgment
The Authors Thanks The Alzaytoonah Univer-
sity of Jordan for their support.
References
[1]
A. A. AlKhatib, T. Sawalha, S. AlZu’bi, “Load Balancing Techniques in
Software-Defined Cloud Computing: an overview,” in 2020 Seventh Inter-
national Conference on Software Defined Systems (SDS), 240–244, IEEE,
2020,doi: 10.1109/SDS49854.2020.9143874.
[2]
N. A. Sultan, “Reaching for the “cloud”: How SMEs can manage,” In-
ternational journal of information management,
31
(3), 272–278, 2011,doi:
https://doi.org/10.1016/j.ijinfomgt.2010.08.001.
[3]
S. AlZu’bi, Y. Jararweh, H. Al-Zoubi, M. Elbes, T. Kanan, B. Gupta, “Multi-
orientation geometric medical volumes segmentation using 3d multiresolution
analysis,” Multimedia Tools and Applications,
78
(17), 24223–24248, 2019,
doi:https://doi.org/10.1007/s11042-018-7003-4.
[4]
A. Susarla, A. Barua, A. B. Whinston, “Understanding the ‘service’component
of application service provision: an empirical analysis of satisfaction with ASP
services,” in Information Systems Outsourcing, 481–521, Springer, 2006, doi:
https://doi.org/10.1007/978-3-540-34877-1 17.
[5]
G. Lu, W. H. Zeng, “Cloud computing survey,” in Applied Mechanics
and Materials, volume 530, 650–661, Trans Tech Publ, 2014, doi: DOI:
10.4028/www.scientific.net/AMM.530-531.650.
[6]
A. Abusukhon, M. N. Anwar, Z. Mohammad, B. Alghannam, “A hybrid
network security algorithm based on Diffie Hellman and Text-to-Image Encryp-
tion algorithm,” Journal of Discrete Mathematical Sciences and Cryptography,
22(1), 65–81, 2019, https://doi.org/10.1080/09720529.2019.1569821.
[7] S. Robison, “Everything-as-a-Service: A blue sky view of the cloud,” 2009.
[8]
L. Xin, C. Song, “Cloud-based innovation of Internet long tail,” in 2011 In-
ternational Conference on Product Innovation Management (ICPIM 2011),
603–607, IEEE, 2011, doi: https://doi.org/10.1007/s11042-019-7367-0.
[9]
R. F. El-Gazzar, “A literature review on cloud computing adoption issues in
enterprises,” in International Working Conference on Transfer and Diffusion of
IT, 214–242, Springer, 2014, doi: 10.1007/978-3-662-43459-8 14.
[10] V. Paul, S. Pandita, M. Randiva, “CLOUD COMPUTING REVIEW,” 2018.
[11]
P. Kumar, R. Kumar, “Issues and challenges of load balancing techniques in
cloud computing: a survey,” ACM Computing Surveys (CSUR),
51
(6), 1–35,
2019, doi:https://doi.org/10.1145/3281010.
[12]
E. J. Ghomi, A. M. Rahmani, N. N. Qader, “Load-balancing algorithms in
cloud computing: A survey,” Journal of Network and Computer Applications,
88, 50–71, 2017, doi: https://doi.org/10.1016/j.jnca.2017.04.007.
[13]
M. R. Mesbahi, M. Hashemi, A. M. Rahmani, “Performance evaluation and
analysis of load balancing algorithms in cloud computing environments,” in
2016 Second International Conference on Web Research (ICWR), 145–151,
IEEE, 2016, doi:10.1109/ICWR.2016.7498459.
[14]
K. Al Nuaimi, N. Mohamed, M. Al Nuaimi, J. Al-Jaroodi, “A survey of load
balancing in cloud computing: Challenges and algorithms,” in 2012 second
symposium on network cloud computing and applications, 137–142, IEEE,
2012, doi:10.1109/NCCA.2012.29.
[15]
S. K. Upadhyay, A. Bhattacharya, S. Arya, T. Singh, “Load optimization in
cloud computing using clustering: a survey,” Int. Res. J. Eng. Technol,
5
(4),
2455–2459, 2018.
[16]
L. Xu, M. Xu, J. Wang, R. Semmes, Q. Wang, H. Mu, S. Gui, H. Yu, W. Tian,
R. Buyya, “A Data-driven Approach to Identify Resource Bottlenecks for
Multiple Service Interactions in Cloud Computing Environments,” .
[17]
S. Afzal, G. Kavitha, “Load balancing in cloud computing–A hierarchical
taxonomical classification,” Journal of Cloud Computing,
8
(1), 22, 2019, doi:
https://doi.org/10.1186/s13677-019-0146-7.
[18]
M. A. Shahid, N. Islam, M. M. Alam, M. M. Su’ud, S. Musa, “A Comprehen-
sive Study of Load Balancing Approaches in the Cloud Computing Environ-
ment and a Novel Fault Tolerance Approach,” IEEE Access,
8
, 130500–130526,
2020, doi:10.1109/ACCESS.2020.3009184.
[19]
R. Sajjan, B. R. Yashwantrao, “Load balancing and its algorithms in cloud com-
puting: a survey,” International Journal of Computer Sciences and Engineering,
5(1), 95–100, 2017.
[20]
M. M. D. Shah, M. Kariyani, M. Agrawal, “Allocation of virtual machines
in cloud computing using load balancing algorithm,” International Journal of
Computer Science and Information Technology & Security (IJCSITS),
3
(1),
2249–9555, 2013.
[21]
N. R. Tadapaneni, “A Survey Of Various Load Balancing Algorithms In Cloud
Computing,” 2020.
[22]
G. Megharaj, K. Mohan, “A survey on load balancing techniques in cloud com-
puting,” IOSR Journal of Computer Engineering (IOSR-JCE),
18
(2), 55–61,
2016, doi: 10.9790/0661-18215561.
[23]
V. Mathur, “A Comparative Study of Load Balancing Techniques in Distributed
Systems,” .
[24]
J. James, B. Verma, “Efficient VM load balancing algorithm for a cloud comput-
ing environment,” International Journal on Computer Science and Engineering,
4(9), 1658, 2012.
[25]
S. G. Domanal, G. R. M. Reddy, “Load balancing in cloud computing us-
ing modified throttled algorithm,” in 2013 IEEE International Conference
on Cloud Computing in Emerging Markets (CCEM), 1–5, IEEE, 2013, doi:
10.1109/CCEM.2013.6684434.
[26]
S. G. Domanal, G. R. M. Reddy, “Load balancing in cloud environment using
a novel hybrid scheduling algorithm,” in 2015 IEEE International Conference
on Cloud Computing in Emerging Markets (CCEM), 37–42, IEEE, 2015, doi :
10.1109/CCEM.2015.31.
www.astesj.com 869

A. Alkhatib et al. /Advances in Science, Technology and Engineering Systems Journal Vol. 6, No. 2, 860-870 (2021)
[27]
V. Arulkumar, N. Bhalaji, “Performance analysis of nature inspired load bal-
ancing algorithm in cloud environment,” Journal of Ambient Intelligence and
Humanized Computing, 1–8, 2020, doi: https://doi.org/10.1007/s12652-019-
01655-x.
[28]
S. Kaur, T. Sharma, “Efficient load balancing using improved central load bal-
ancing technique,” in 2018 2nd International Conference on Inventive Systems
and Control (ICISC), 1–5, IEEE, 2018, doi: 10.1109/ICISC.2018.8398857.
[29]
D. Saranya, L. S. Maheswari, “Load balancing algorithms in cloud computing:
a review,” International Journal of Advanced Research in Computer Science
and Software Engineering, 5(7), 1107–1111, 2015.
[30]
A. A. Prakash, V. Arul, A. Jagannathan, “A look at of efficient and more suit-
able load balancing algorithms in cloud computing,” Int. J. Eng. Res. Comput.
Sci. Eng., 5(4), 7, 2018.
[31]
P. Prajapati, A. K. Sariya, “A review: Methods of load balanc-
ing on cloud computing,” Int. J. Res. Anal. Rev.,
6
, 2019, doi:
10.6084/m9.doi.one.IJRAR19J2346.
[32]
S.-C. Wang, K.-Q. Yan, W.-P. Liao, S.-S. Wang, “Towards a load balancing in
a three-level cloud computing network,” in 2010 3rd international conference
on computer science and information technology, volume 1, 108–113, IEEE,
2010, doi:10.1109/ICCSIT.2010.5563889.
[33]
W. Duan, X. Tang, J. Zhou, J. Wang, G. Zhou, “Load balancing opportunistic
routing for cognitive radio ad hoc networks,” Wireless Communications and
Mobile Computing, 2018, 2018, doi: https://doi.org/10.1155/2018/9412782.
[34]
K. Nishant, P. Sharma, V. Krishna, C. Gupta, K. P. Singh, R. Rastogi, et al.,
“Load balancing of nodes in cloud using ant colony optimization,” in 2012
UKSim 14th international conference on computer modelling and simulation,
3–8, IEEE, 2012, doi: 10.1109/UKSim.2012.11.
[35]
W. Hashem, H. Nashaat, R. Rizk, “Honey bee based load balancing in cloud
computing.” KSII Transactions on Internet & Information Systems,
11
(12, doi:
10.3837/tiis.2017.12.001), 2017.
[36]
J. Lim, D. Lee, “A Load Balancing Algorithm for Mobile Devices in
Edge Cloud Computing Environments,” Electronics,
9
(4), 686, 2020,
doi:https://doi.org/10.3390/electronics9040686.
[37]
P. Kumar, A. Verma, “Scheduling using improved genetic algorithm in cloud
computing for independent tasks,” in Proceedings of the international confer-
ence on advances in computing, communications and informatics, 137–142,
2012, doi: 10.1145/2345396.2345420.
[38]
J. AnantKumar, S. PremChand, “Multi-Agent Genetic Algorithm for Efficient
Load Balancing in Cloud Computing,” in International Journal of Innovative
Technology and Exploring Engineering (IJITEE), 45–51, 2020.
[39]
R. Panwar, B. Mallick, “Load balancing in cloud computing using dy-
namic load management algorithm,” in 2015 International Conference on
Green Computing and Internet of Things (ICGCIoT), 773–778, IEEE, 2015,
doi:10.1109/ICGCIoT.2015.7380567.
[40]
G. Joshi, S. K Verma, “Load balancing approach in cloud computing using
improvised genetic algorithm: a soft computing approach,” in IJCA, volume
122, 24–28, 2015, doi: 10.5120/21729-4894.
[41]
D. Agarwal, S. Jain, et al., “Efficient optimal algorithm of task scheduling
in cloud computing environment,” arXiv preprint arXiv:1404.2076, 2014,
doi:10.14445/22312803/IJCTT-V9P163.
[42]
S. Sidana, N. Tiwari, A. Gupta, I. S. Kushwaha, “NBST algorithm: A load
balancing algorithm in cloud computing,” in 2016 International Conference
on Computing, Communication and Automation (ICCCA), 1178–1181, IEEE,
2016. doi: 10.1109/CCAA.2016.7813914.
[43]
M. Yakhchi, S. M. Ghafari, S. Yakhchi, M. Fazeli, A. Patooghi, “Proposing a
load balancing method based on Cuckoo Optimization Algorithm for energy
management in cloud computing infrastructures,” in 2015 6th International
Conference on Modeling, Simulation, and Applied Optimization (ICMSAO),
1–5, IEEE, 2015, doi: 10.1109/ICMSAO.2015.7152209.
[44]
K. Kumar, T. Ragunathan, D. Vasumathi, P. Prasad, “An Efficient Load Balanc-
ing Technique based on Cuckoo Search and Firefly Algorithm in Cloud,” Int. J.
Intell. Eng. Syst, 13(3), 422–432, 2020, doi: 10.22266/ijies2020.0630.38.
[45]
L. Yan, X. Liu, “The predicted load balancing algorithm based on the dy-
namic exponential smoothing,” Open Physics,
18
(1), 439–447, 2020, doi:
10.1515/phys-2020-0108.
[46]
X. Ren, R. Lin, H. Zou, “A dynamic load balancing strategy for cloud comput-
ing platform based on exponential smoothing forecast,” in 2011 IEEE inter-
national conference on cloud computing and intelligence systems, 220–224,
IEEE, 2011, doi: 10.1109/CCIS.2011.6045063.
www.astesj.com 870