Content uploaded by Moses Abiodun
Author content
All content in this area was uploaded by Moses Abiodun on Nov 04, 2021
Content may be subject to copyright.
Full Terms & Conditions of access and use can be found at
https://www.tandfonline.com/action/journalInformation?journalCode=tcom20
International Journal of Computer Mathematics:
Computer Systems Theory
ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/tcom20
Applicability of MMRR load balancing algorithm in
cloud computing
Abiodun Kazeem Moses, Awotunde Joseph Bamidele, Ogundokun Roseline
Oluwaseun, Sanjay Misra & Adeniyi Abidemi Emmanuel
To cite this article: Abiodun Kazeem Moses, Awotunde Joseph Bamidele, Ogundokun Roseline
Oluwaseun, Sanjay Misra & Adeniyi Abidemi Emmanuel (2021) Applicability of MMRR load
balancing algorithm in cloud computing, International Journal of Computer Mathematics: Computer
Systems Theory, 6:1, 7-20, DOI: 10.1080/23799927.2020.1854864
To link to this article: https://doi.org/10.1080/23799927.2020.1854864
Published online: 14 Dec 2020.
Submit your article to this journal
Article views: 19
View related articles
View Crossmark data
INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS: COMPUTER SYSTEMS THEORY
2021, VOL. 6, NO. 1, 7–20
https://doi.org/10.1080/23799927.2020.1854864
Applicability of MMRR load balancing algorithm
in cloud computing
Abiodun Kazeem Moses a, Awotunde Joseph Bamidele b, Ogundokun Roseline
Oluwaseun a, Sanjay Misra cand Adeniyi Abidemi Emmanuel a
aDepartment of Computer Science, Landmark University, Omu-Aran, Nigeria; bDepartment of Computer Science,
University of Ilorin, Ilorin, Nigeria; cDepartment of Electrical and Information Engineering, Covenant University, Ota,
Nigeria
ABSTRACT
One of cloud computing’s fundamental problems is the balancing of loads,
which is essential for evenly distributing the workload across all nodes.
This study proposes a new load balancing algorithm, which combines
maximum-minimum and round-robin (MMRR) algorithm so that tasks with
long execution time are allocated using maximum-minimum and tasks with
lowest execution task will be assigned using round-robin. Cloud analyst tool
was used to introduce the new load balancing techniques, and a compara-
tive analysis with the existing algorithm was conducted to optimize cloud
services to clients. The study findings indicate that ’MMRR has brought
significant changes to cloud services. MMRR performed better from the
algorithms tested based on the whole response time and cost-effectiveness
(89%). The study suggested that MMRR should be implemented for enhanc-
ing user satisfaction in the cloud service.
ARTICLE HISTORY
Received 27 August 2020
Accepted 13 November 2020
KEYWORDS
Grid computing; cloud
computing; scheduling
algorithm; distributed
computing
1. Introduction
Cloud computing [27] is an expansive research area that brings excellent value to the costs of busi-
nesses around the world. Cloud computing [11] is a service model of information technology, using
the world’s previous networks and distributed computing concepts that assist users getting access to
dierent resources and data in the cloud. The ability to process complex data quickly and eciently
has made several organizations to support it such as Google, Facebook etc. [13,29]. Cloud computing
is established on virtualization technology [28], through network services to provide users with the
necessary resources, application platform, software and other services. Cloud computing is changing
theITindustry,changingthewayweusesoftwareandhardware[17]. Major companies like Google,
HP, Amazon, Oracle and others are embracing Cloud computing as a route to computer ability to
do many things. Cloud clients nowadays can purchase resources of dierent natures from various
providers. They can lease infrastructure resources, platform resources or software resources, possibly
all three types simultaneously. In the cloud environment, we have four distribution models in the
cloud; they are private, public, hybrid and community. Proper management of cloud resources is
maintained by the procient and accessible features of cloud computing. Most of the resources in the
cloud are in an easily accessible form, which is another crucial cloud system characteristic. The Cloud
ServiceProvider(CSP)oersservicestousersonarentedbasis[19]. The job of CSP to provide the
CONTACT Sanjay Misra sanjay.misra@covenantuniversity.edu.ng
© 2020 Informa UK Limited, trading as Taylor & FrancisGroup
8 A. K. MOSES ET AL.
facilities to the customer is a unique one with the array of available virtual cloud resources. Conse-
quently, scholars have given more consideration towards the balancing of the load. The impact of load
balancing on system performance is enormous.
Balancing of load [27]isaprocessthatensuresthejobbetweenvariousnodesinthegivensetting
are not overwhelmed or is idle for any moment. A practical algorithm for balancing of loads will
ensure that all the system nodes do almost the same amount of jobs. The algorithm for load balancing
is controlling the mapping of the tasks that are assigned to cloud territory, to unused assets, so overall
responsetimeavailableisenhancedaswellastheusefulresourceusageisgiven.Harmonizingthe
load has become a key concern in cloud computing because the number of requests in the cloud
environment that are issued within a second is not foreseeable. The unpredictability is because of the
cloud’s ever-changing behaviour.
Static and dynamic algorithm are the standard ways of classifying resource scheduling algorithm.
The static algorithm anticipates prior information on the structure and diverse factors of the frame-
work,likerestrictionsonastoragedevice,memoryetc.Anexampleofastaticapproachisround-robin
(RR) algorithm. The algorithm that can change as they know about the framework is referred to
as Dynamic algorithms. The demand of the customer can be eortlessly accomplished with vibrant
procedures. However, algorithms that are dynamic works well when equated with static algorithms;
the challenging task is in designing and developing an algorithm for the cloud environment that is
dynamic [21].
The dynamic algorithm classies further based on oine mode and online mode. The task to
be performed are done at some pre-characterized times for oine mode, and the circumstances are
selected based on possible ending time of more signicant tasks. The approach is call batch method
because the planning is completed in groups. For online mode, the job is allocated when the tasks
begantoarrive.Thecriticalgoalofbalancingloadinthecloudsphereistoallotjobsdynamically
between nodes to please user necessities and oer optimum resource usage by classifying the whole
obtainable jobs to dierent nodes.
The remaining part of this paper includes Section 2 that reviews related works of literature; Section
3discussestheapproachandtheperformancemetricsadoptedforthestudyandalsopresentsthe
proposed hybridization of MMRR algorithm. Section 4 gives results and discussion of simulation
outcomes. Lastly, Section 5 recapitulates the study results and proposal for work in the future.
2. Literature review
Individuals, companies and organizations have gained signicant benets and transformation from
cloud computing [1,5], which oers secure personalized and cost-eective services across the internet
to all aspects of life daily. Cloud derives from the grid computing, utility computing and simulation
concept [31]. Resources of computers in the cloud setting provide the ability to access computing
power, bandwidth, storage, etc. it gives users a variety of IT services [28]. The rapid progress in devel-
oping cloud systems and applications into cloud computing environments is a very challenging task,
according to [7] and requires capable algorithms for ecient resources distribution. Round Robin is
another algorithm used in cloud computing to assign resources to the Server; for a specic time, slice,
the request is performed in an RR fashion based on the community sharing order [12].
There are several articles on the study of various algorithms and cloud simulation applications
that address the relative merits of various options [3,26]. Such research relates to investigative and
nature-inspired algorithms [22,30] and evaluating the advantages of cloud analyst for investigating
the various algorithms [23]. Contrasted the various variations of RR algorithm like modied round-
robin and time slice-based priority-based round-robin (TSPBRR). These algorithms are evaluated in
terms of various parameters such as performance, processing time, waiting time and response time.
The results ndings show that TSPBRR oers a more robust solution with high throughput and better
response time.
INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS: COMPUTER SYSTEMS THEORY 9
RR algorithms had the drawback that when the VMs are assigned to the application request of
auser,itspositionwouldnotbemaintained,whichallowsthealgorithmtobeperformedforthe
request of similar demand again. The authors in [17] have built an enhanced RR which manages the
state every time the server runs an application request. Unique data structures achieved it called hash
maps to hold VM information allocated to a particular demand for user applications and VM state list
that retains the VM status (busy or free). The simulation results showed that this algorithm increased
theresponsetimecomparedwiththenormalone.
The authors in [8] built an updated algorithm for load balancing called throttled. The algorithm
makes sure the best vacant VM is assigned according to the response and processing time to an appli-
cation request. Here the state of each VM is maintained by ‘throttle VM load balancer’. On request
to the ‘data centre manager’ for an application, it requests VM load balancer for throttle to select the
right VM based on the application specications. A technique for biologically inspired bee colony
optimization in distributing load in a cloud hosting web services is dened in [20].
Under this methodology, the allocation of web services varies accordingly as the amount of
requests for web services varies. The servers are here considered to be virtual servers that contain
alistofdemandsforservice.Theservercalculatesameasurecalled‘prot’whichgivesustheappli-
cation request service, founded on performance metrics such as CPU time, etc. The ‘advert board’
also shows the idle servers if they need any services. Depending on various metrics, the net income
of the associated network is measured. The servers will act the character of bees here, and lazy severs
willbelikethebeeswaitingforthenectartofetch.
The ant colony optimization (ACO) updated technique [20] is another biologically inspired load
balancing technique. The ACO methods are established on the ant movements in a specic direction
using pheromone. ACO is related to the balancing of cloud loads by seeing the system module as ants
and the virtual machines (VMs) as nodes. One node will initially be chosen, as the Regional Load
Balancing Node (RLBN), which performs as the head node. Ants will emerge from this RLBN and
will be able to travel towards the nodes in any direction, according to a load of nodes.
In ACO approach, a table of pheromone is preserved, which keeps the information about loads
on the nodes. The main goal of ant is to nd and assign the least loaded node to the client, thus
distributing the node in the network equally. Established on the standard of analytical hierarchy pro-
cess, the author in [10] provided a load balancing process called priority-based task scheduling in
cloud. The algorithm ow built with three prioritization stages, like resource, scheduling and job lev-
els. The downside of this work was the complexity of the system, the incoherence and the time of
consumption.
In [9], the author implemented a new updated HEFT (heterogeneous earliest nish time)
algorithm to resolve the HEFT drawbacks. It works on assigning the inbound job amongst dier-
ent resources in two steps, which reduces the makes-pan [15]. Introduced elastic cloud max–min
(ECMM) algorithm with signicant enhancements in the basic max–min algorithm. This algorithm
maintains a table of all tasks with status. It locates the current job for each server, and the estimated
timeforeachtasktobecompleted–thetabledetailsforspecictoolstobeusedwhenallocating
various tasks.
ECMM process is evaluated using cloudsim with other algorithms, and the result shows that it
increases the job pending time in comparison with max–min and RR algorithms. Still, the max–min
algorithmisfasterthanECMM.
NBST algorithm [25] proposed a method that attempts to strike a balance with the load by arrang-
ing the VM according to their processing power and the cloudlets according to their length. The list of
VM and cloudlets is then submitted to the broker for allocation. Broker assigned through mid-point
algorithm and divides the VM list and cloudlet list till we have at most one cloudlet or VM in the
inventory and then the allocation of resources is done.
Arulkumar and Bhalaji [4] present the evaluation result of the load balancing algorithm inspired by
nature in a cloud environment. The analysis concludes that water wave algorithm is at least 2.5% better
than particle swarm optimization; 4.5% better than ACO and 6.9% better than genetic algorithm.
10 A. K. MOSES ET AL.
2.1. Load balancing algorithms types used in cloud
The load balancing algorithms that are currently being employed in cloud computing is described
below, along with specic considerations:
(1) Random Algorithm
The nature of the algorithm is static [12], typically specied in system design or implementation.
It randomly selects the node by using a random number generator [16]foreachprocessor;the
procedures for the delivery guideline are retained. Although the process works well with similarly
loaded systems, when the loads are of diverse computational complexity, there may be some problem.
Another problem is that there is no deterministic approach to the algorithm.
(2) FCFS
FCFS algorithm [6] is mostly used for parallel processing, and it selects a task coming in and aimed
at resource with the shortest waiting queue time. The load balancing strategy that each load balancer
hasaworkqueueinwhichthejobiswaitingtobeexecutedforitsturn.FCFS’sadvantagesaredueto
its quick and comfortable design.
(3) Round Robin
Allocatesthenodesinawaythatitgivestimetoeachjobequallytosatisfytherequestsinthetime-
sharing process, i.e. based on the locally dened process distribution directive. RR gives the benet
of a quick response in case workload is distributed equally among the methods. The job cycle time
for various systems, however, is not the same. Many nodes will, therefore, be hugely overburdened,
whereassomeotherswillkeeponidle[12].
(4) Weighte d R R
Thealgorithmshareweightforeachmachineaccordingtoadeneways[16], the algorithm con-
siders prociencies of the VM resource. It gives a greater quantity of jobs to the higher capacity VM
according to the weightage given to each of the systems. This system works eciently, nevertheless
has an issue in the beginning due to the static interpretation of the weights. It also failed to consider
the length of the tasks to select the appropriate VM.
(5) Dynamic RR
The dierence of dynamic RR compared with weighted RR is that on a continuous cycle, the servers
are tracked, and the weights continue to change. That is a form of active load balancing. Dynamic
RR uses numerous facets of instantaneous server capabilities scrutiny, as the recent quantity of links
pernodeorthequickestreactiontimeforthelinkstobedistributed.Theonlyproblemwiththis
algorithm is that, because of its complicated existence, load balancing is rarely used [16].
(6) Least connections
The load balancing algorithm least connections modes for pool members, an algorithm that is
dynamic and allocates links to the pool member (node/server) that is presently handling the fewest
open links at the time when a new link request is received. The program transfers a unique link
to the server using this algorithm that has the minimum number of new connections [16]. The
algorithmiswellsuitedinenvironswhereidenticalcapabilitiesexistfortheserversorloadbal-
ancingofothermaterials.Theapplicationthatusesthisisscarcelypossibleinasimpleload
balancer.
INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS: COMPUTER SYSTEMS THEORY 11
(7) Observed
Although it is not common, the system calculates a vibrant ration value used to allocate connections
along with available pool members. It is the alignment of least connections and fastest algorithm.
Servers with a better combination between the smallest networks and quickest reaction time earn a
larger share of the systems. The observed mode balances the load of the networks by the ratio values
given to individual pool participant, favouring the pool participant with the highest ratio value [16].
(8) Equally spread current execution load (ESCE)
Communication occurs among the load balancer and the Datacentre Controller for bringing up-to-
date the index table leading to overhead in the ESCE algorithm. Besides, this overhead leads to a
delay in getting a response to the arrived requests. Continuous monitoring of jobs is required for
this algorithm [18] that is available for execution to schedule the tasks to various VM. The task is
arbitrarily issued by testing the size and then giving the job to a VM that is not loaded much or can
performthisfunctioneortlesslyandusesalesseramountoftime,thusprovidingfullthroughput.
(9) Throttledloadbalancing
Within this system [2],asuitableVMislookedfortoperformthecorrectoperationwhenauserdesire
is received. The algorithmic process begins withholding a list of all accessible VMs. The list is much
and to make it faster in the method of lookup; indexing is carried out. A user’s request is approved
when a suitable match is identied depending on the machine’s dimension and accessibility. The
Server will get a virtual machine assigned. On the other hand, the demand will be lined up by the
load balancer.
(10) Min-Min Algorithm
The algorithm [12] starts by looking for the shortest completion time for all jobs. Then the quickest
time in all the tasks are pick for all resources and scheduled based on the equivalent device. After-
wards, the time use of the allocated task is joined with the completion times of other jobs on that
system to update the nishing time on that machine. The task completed is then removed from the
machine’s list of tasks yet to be assigned. This process is repeated; however, again till every job are
dispersed on the resources. The biggest shortcoming of this technique is starvation [14].
(11) Max–Min Algorithm
The max–min algorithm [14] is like the min-min process, but the maximum value is chosen after
nding the total execution times. The highest completion time on any resources in the jobs, based
on what the job is the plan on the equivalent machine. Subsequently, the completing time of the
apportioned task is joined to the performance periods of new tasks on that machine to update the
performance period on that machine. The apportioned job is removed from the list of jobs yet to be
allotted to the devices. The process is followed again until every task is allocated to the resources.
(12) Token R o u ting
The algorithm was designed to minimize the cost of the system by shifting tokens around the device.
Because of communication bottleneck, agents cannot have enough workload distributing infor-
mation. With the support of token-based load balancing heuristic approach, the downside of this
algorithm can be eliminated, thus providing quick, ecient routing decisions. The proxies might
not have comprehensive information about the workload of their international state and neighbours;
however, they make their specic conclusions about places to transfer the token by producing their
particular knowledge base. Thus, the information base used is generated from the tokens received
before, and so no unnecessary cost is created.
12 A. K. MOSES ET AL.
3. Methodology
The new emerging technology is cloud computing in the academics and industry. In the proposed
work, cloud job scheduling and balancing of load strategy were used for allocation of the virtual
machine to dierent user bases requests. In this process, the user base deployed in diverse regions
of the world that transmit their request to dierent datacentres allocated by the third-party cloud
server. The cloud analyst tool was used for development and simulation of cloud environment
before actual deployment to the real-world application. In the processing of cloud analyst, numer-
ous provision broker policies and policies of load balancing have been used for response of dierent
users.
In the Max–Min Algorithm, the pseudocode is described below:
For i =1 to K // K represents task numbers to be scheduled.
For j =1 to N //N represents the virtual machine numbers.
TCOij =TETij +TRTj // CTij represents Task Completion Time
// TETij denotes Task Execution Time
// TRTj denotes Task Ready Time of job i on VM j
End For
End For
Do until all the unscheduled tasks are exhausted.
For each unscheduled task
Find the maximum completion time of the task and virtual machine that
obtains it.
End for
Locate the task tp with the maximum completion time
Allocate task tp to the virtual machine that gives the maximum completion time
Delete task tp from the pull of unscheduled tasks
Update the ready time of the machine that oers the maximum completion time
End for
3.1. Proposed algorithm
This paper advocates a new fusion method for load balancing using maximum-minimum and RR
algorithm to assign VM to dierent user base requests called MMRR Load balancing. The proposed
hybrid will overcome the problem of maximum and minimum that usually restrict job with least
completion time from being executed on time. In contrast, the job with the highest nishing time
will be accomplished rst. RR that assigns tasks without priority. The proposed ow of work of the
new hybrid algorithm is shown in Figure 1.
Figure 1showing the systematic approach to implement the new algorithm. The data centres and
user base are rst initializing, then the characteristics of the cloud environment and its bandwidth are
set.Thenewhybridalgorithmisthenapplied,andtheresultofthesimulationisnoted.Thedownside
of maximum and minimum procedure shows the implementation of jobs with the highest execution
period may rst escalate the overall response period of the system [24]. Eventually, it may cause a
slowing down in implementing tasks with smallest execution period, therefore, the reason to work
on the new maximum and minimum procedure to lessen the delay in implementing jobs with lowest
completion time.
3.2. Performance evaluation parameters
These parameters are used for performance evaluation of the proposed approach.
INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS: COMPUTER SYSTEMS THEORY 13
Figure 1. Flow step for the hybrid.
(i) Over All Response Time: provide information about time use by the CSP to respond to requests.
(ii) Datacentre Request servicing time: give information on a single data centre for serving requests
of dierent user bases
(iii) Cost Analysis: provide information about dierent load balancing approaches using data
transmission cost and total VM cost occurred.
The pseudocode for MMRR
Minimum →5
Calculate expected completion time, for all job in the job set
If the job set is empty then
end the program
else
Find a job with least execution time and job with high execution time
Calculate the dierence between high execution time and least execution time
End if
If dierence <=least then
Assignjobwithhighexecutiontimewithmax–min
Else
Assign job with least execution time accordingly
End if
Remove the job from the job set
Update the ready time for the selected resource and expected completion time for all tasks
End
ThepseudocodeshowstheimplementationmethodsofMMRR;withtheassignedtaskofthe
shortest execution time process rst; the job with the highest expected time will be allocated with RR
so that it does not starve the job with minimum execution time tasks. This approach enhances the
maximum-minimum algorithm while the RR algorithm will only be used on the task with maximum
execution task.
14 A. K. MOSES ET AL.
3.3. Conguration of simulation software
The implementation of the algorithms is achieved using Cloud Analysts, a graphical user simulator
for the cloud environment. The Eclipse Java EE IDE for Web Developers is used as a container for
it. The study adds a new algorithm MMRR to Cloud Analyst source code. In the proposed work,
sixuserbasesandtwodatacentreshavebeendenedinthecloudcomputingenvironment.They
have been allocated in dierent regions for processing. Dierent colour regions have identied region
boundaries.
Figure 2displays the distribution of the data centres and the user base for this simulation. We have
six user base spread across the continent and two data centres.
Table 1representstheuserbasecongurationthathasbeenusedforthesimulationofthecloud
computing environment.
Table 2represents a datacentre conguration that has been used for simulation of the cloud
computing environment.
Figure 2. User base and DC allocation in cloud analyst.
Tab le 1 . Configuration for the userbase.
Name Regions Request/h Size Peak/h Peak Users Normal users
UB1 0 600 100 1–3 1500 100
UB2 1 1600 1000 3–5 500 1000
UB3 2 2000 1000 5–7 500 1000
UB4 3 1000 1000 7–9 1600 600
UB5 4 1000 1000 9–11 2000 700
UB6 5 7000 300 11–12 2000 800
Tab le 2 . Configuration of the datacentre.
DC name RCost Vm/h Memory cost/s Data transfer/Gb Speed H/w units
DC-1 0 0.1 0.5 0.1 10,000 2
DC-2 4 0.1 0.8 0.1 10,000 3
INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS: COMPUTER SYSTEMS THEORY 15
Figure 3. The cloud analyst configuration page.
Tab le 3 . Extra simulation parameter.
Parameter Value
User grouping factor in user base 1000
Request grouping factor 10
Executable instruction length/request 100
Load balancing policy
Round Robin, Max-Min, Equally Spread
Current Execution, Throttle and MMRR
Simulation duration 60.0 min
VM image size 10,000
VM memory 512
VM bandwidth 1000
Data centre architecture X86
Data centre OS Linux
Data centre VMM Xen
Figure 3display the conguration page of the cloud analyst used. We dene specic parameters in
the main, data centre and advance conguration.
Table 3shows the extra parameters set for this simulation.
Figure 4isthepagewhereweselectadierentalgorithmtobetested.
Based on the simulation parameters, various ways of balancing load policies have been used for the
simulation of computing in the cloud. Based on the simulations, numerous parameters were evaluated
for performance assessment of the proposed approach.
4. Results and discussion
Theoutcomesgavethedierentalgorithmsofthedierentscenariosaftersimulationaredisplayed
in Tables 4 and 5.
Table 4shows the outcomes of the comparative analysis of the four algorithms show that the new
algorithm MMRR performed well in terms of overall response time. The Throttle seems better at data
centre processing time and the cost of data transfer from one place to the other, the new algorithm is
cheaper to use than any of the dierent algorithms compared.
16 A. K. MOSES ET AL.
Figure 4. The advanced configuration page.
Tab le 4 . Comparing all the algorithms.
Algorithms
Parameter Round Robin Max-Min ESCE Throttle MMRR
Overall response time (ms) 228.29 228.20 228.12 227.28 227.26
Processing time of data centre (ms) 1.48 1.40 1.39 0.50 0.57
Cost of processing ($) 630.20 630.20 630.20 630.20 222.32
Tab le 5 . Comparing all the algorithms using optimize response time policy.
Algorithms
Parameter Round Robin Max-Min ESCE Throttle MMRR
Total response time (ms) 228.29 228.20 228.12 227.28 227.26
Data centre processing time (ms) 1.48 1.43 1.39 0.50 0.57
DC1 request servicing time (ms) 1.45 1.40 1.36 0.49 0.56
DC2 request servicing time (ms) 1.76 1.68 1.61 0.61 0.63
DC1 transfer cost ($) 562.36 562.36 562.36 562.36 196.56
DC2 transfer cost ($) 67.05 67.05 67.05 67.05 23.53
Table 5shows the result when using the optimized time policy to rerun the simulation; the result
gotten is exciting. The new algorithm MMRR still performs better than any other algorithm. The
throttle algorithm is much better at data centre processing time, the request servicing time for Dat-
acentres 1 and 2 is fastest using the Throttle and followed by the new algorithm. When comparing
the cost of data transfer, for both data centres, MMRR algorithm is far cheaper to use than any other
algorithm compared.
The chart in Figure 5shows the result of the ve algorithms that were tested. The new algorithm
performed slightly well than the Throttle procedure looking at the overall response time. It also
displays that the new algorithm MMRR is better than ESCE, max–min and RR algorithm.
The chart in Figure 6shows that concerning the processing time for data centre, the Throttle
algorithmisbetterthanthenewalgorithmjustproposesslightly.
INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS: COMPUTER SYSTEMS THEORY 17
Figure 5. Overall response time of load balancing algorithms.
Figure 6. Load balancing algorithms processing time of data centre.
18 A. K. MOSES ET AL.
Figure 7. Total cost of data transfer of load balancing algorithms.
Figure 7chart shows that in terms of the cost of data transfer, the new algorithm MMRR fared
better than others.
The results show that the time of the new hybrid MMRR algorithm is better for two data centre
and six user bases as compared to ESCE, Round-Robin, Max–Min and Throttled Algorithms. The
time taken for processing by the data centre is worthy from both Throttled and MMRR, it was worst
for Round Robin. The cost of data transfer was similar for ESCE, Throttled, Max–Min and Round
Robin. Interestingly the improved MMRR recorded a low data transfer cost in both data centres,
which makes it very cost-eective and cheaper to transfer data across the globe. The overall response
time is 227.26 ms and 89.0% cost-eectiveness.
5. Conclusion and future work
In conclusion, the proposed hybrid approach of Round Robin with Maximum Minimum load bal-
ancing algorithm brought about signicant improvement in the cloud services. The processing time
taken by the data centre is good from both Throttled (0.50 ms) and MMRR (0.57ms), but it was worst
for Round Robin (1.48 ms). Out of the dierent algorithms that were evaluated, MMRR performed
well in terms of an overall response time of 227.26ms and 89% cost-eectiveness.
Thus, the study recommends that MMRR should be adopted in the cloud service to improve people
satisfaction. The response time is encouraging; the processing time of the data centre is above average,
and the cost of processing a job with the new algorithm is cheaper compare to others.
The role of various service factors produces noteworthy results when VMs are assigned to incom-
ing tasks in a cloud environment. This study would be expanded using dierent quality of service
constraints like cost, throughput and delay for various routing policies in future. More data centres,
user base and more algorithms will be used to nd out the best algorithm that will give optimum
result in the cloud environment.
INTERNATIONAL JOURNAL OF COMPUTER MATHEMATICS: COMPUTER SYSTEMS THEORY 19
Disclosure statement
No potential conict of interest was reported by the authors.
ORCID
Abiodun Kazeem Moses http://orcid.org/0000-0002-3049-1184
Awotunde Joseph Bamidele http://orcid.org/0000-0002-1020-4432
Ogundokun Roseline Oluwaseun http://orcid.org/0000-0002-2592-2824
Sanjay Misra http://orcid.org/0000-0002-3556-9331
Adeniy i Abidemi Emmanuel http://orcid.org/0000-0002-2728-0116
References
[1] R.L. Adelaar, Cloud computing technology in manufacturing rms-Identifying adoption factors, challenges, and
corresponding solutions, Master’s thesis, 2020.
[2] T.Ahmed,andY.Singh,Analytic Study of Load Balancing Techniques Using Tool Cloud Analyst,International
Journal of Engineering Research and Applications, New Delhi, 2012. Vol. 2. ISSN: 2248.
[3] D.F. Altayeb, and F.A. Mustafa, Analysis of load balancing algorithms implementationoncloudcomputingenviron-
ment,Int.J.Innov.Res.Adv.Eng.6(2)(2016), pp. 1–32.
[4] V. Arulkumar, and N. Bhalaji, Performanceanalysisofnatureinspiredloadbalancingalgorithmincloudenviron-
ment, J. Ambient. Intell. Humaniz. Comput. (2020), pp. 1–8.
[5] M.G. Avram, Advantages and challenges of adopting cloud computing from an enterprise perspective,Procedia
Technol. 1 2 ( 2014), pp. 529–534.
[6] K. Bala. A review of the load balancing techniques at cloud server. 1, International Journal of Advances in Com-
puter Science and Communication Engineering, Chandigarh, March 2014, International Journal of Advances in
Computer Science and Communication Engineering, Vol. 2. ISSN 2347-6788.
[7] R. Buyya, and K. Sukumar, Platforms for building and deploying applications for cloud computing. ArXiv Preprint
arXiv 1104 (2011), pp. 4379.
[8] S.G. Domanal, and G.R. Mohana Reddy, Load balancing in cloud computing using modied throttled algorithm,
Cloud Computing in Emerging Markets (CCEM), 2013 IEEE International Conference on. IEEE, 2013.
[9] K.Dubey,M.Kumar,andS.C.Sharma,Modied HEFT algorithm for task scheduling in cloud environment,
ProcediaComput.Sci.125(2018), pp. 725–732.
[10] S. Ghanbari, and M. Othman, A priority based job scheduling algorithm in cloud computing,ProcediaEng.50
(2012), pp. 778–785.
[11] H. Gibet Tani, C. El Amrani. Smarterroundrobinschedulingalgorithmforcloudcomputingandbigdata,J.Data
Min. Digital Humanities, Special Issue on Sci. Technol. Strategic Intell. jdmdh:3104 (2016).
[12] T.C. Hung, L.N. Hieu, P.T. Hy, and N.X. Phi, MMSIA: improved max-min schedulingalgorithm for load balancing on
cloud computing, In Proceedings of the 3rd International Conference on Machine Learning and Soft Computing
(pp. 60-64), 2019, January.
[13] B.T. Khiet, N.T. Nguyet Que, H.D. Hung, P.T. Vu, and T.C. Hung, A Fair VM Allocation for Cloud Computing
basedonGameTheory, Proceedings of the 10th National Conference on Fundamental and Applied Information
Technology Research (FAIR’10), Da Nang, Viet Nam, 2017.
[14] T. Kokilavani, and G. Amalarethinam, Load balanced Min-Min algorithm or static Meta-task scheduling in grid
computing, Int. J. Comput. Appl. 20(2) (2011), pp. 0975–8887.
[15] X.Li,Y.Mao,X.Xiao,andY.Zhuang,An improved max–min taskscheduling algorithm for elastic cloud, In: IEEE
international symposium on computer, consumer and control, pp. 340–343, 2014.
[16] D. MacVittie, Intro to Load Balancing for Developers – The Algorithms. DevCentral. [Online] 31
March 2009. https://devcentral.f5.com/articles/intro-to-load-balancing-for-developers-ndash-the-algorithms#.
U5k6Xnbm4oE.
[17] K. Mahajan, A. Makroo, and D. Dahiya, Round robin with server anity: a VM load balancing algorithm for cloud
based infrastructure, J. Inf. Process. Sys. 9(3) (2013), pp. 379–339.
[18] S.H.Mahalle,R.P.Kaveri,V.Chavan,Load balancing on cloud datacenters,Int.J.Adv.Res.Comput.Sci.Softw.
Eng., 3 (January 2013), pp. 1–4.
[19] S.K. Mishra, B. Sahoo, and P.P. Parida, Load balancing in cloud computing: a big picture,J.KingSaudUniversity-
Comput. Inf. Sci. 32(2) (2020), pp. 149–158.
[20] K. Nishant, P., Sharma, V., Krishna, C., Gupta, K. P., Singh, & R. Rastogi, Load balancing of nodes in cloud using ant
colony optimization, Computer Modelling and Simulation (UKSim), 2012 UKSim 14th International Conference
on. IEEE, 2012.
20 A. K. MOSES ET AL.
[21] S. Patel, H. Patel, and N. Patel, Dynamicloadbalancingtechniquesforimprovingperformanceincloudcomputing,
Int. J. Comput. Appl. 138(3) (2016), pp. 1–5.
[22] G.A.E.N.A. Said, Nature inspired algorithms in cloud computing: A survey,Int.J.Intell.Inf.Syst.5(5)(2016), pp.
60–64.
[23] P. Samal, and P. Mishra, Analysis of variants in round robin algorithms for load balancing in cloud computing,Int.
J.Comput.Sci.Inf.Technol.4(3)(2013), pp. 416–419.
[24] B.Santhosh,andD.Manjaiah,Animprovedtaskschedulingalgorithmbasedonmaxminforcloudcomputing,
2014.
[25] S. Sidana, N. Tiwari, A. Gupta, and I.S. Kushwaha, NBST algorithm: A load balancing algorithm in cloud computing,
In 2016 International Conference on Computing, Communication and Automation (ICCCA), IEEE, 2016, April,
pp. 1178–1181.
[26] S.P. Singh, A. Sharma, and R. Kumar, Analysis of load balancing algorithms using cloud analyst,Int.J.Grid.Distrib.
Comput. 9(9) (2016), pp. 11–24.
[27] A.N. Singh, M. Hemalatha, An approach on semi distributed load balancing algorithm for cloud computing systems,
Int. J. Comput. Appl. 56(12) (2012), pp. 5–10.
[28] A.Thomas,G.Krishnalal,andV.J.Raj,Credit based scheduling algorithm in cloud computing environment,Procedia
Comput.Sci.46(2015), pp. 913–920.
[29] Y.F. Wen and C.L. Chang, 2014.Load balancing job assignment for cluster-based cloud computing,SixthInterna-
tional Conference on Ubiquitous and Future Networks (ICUFN), Shanghai, pp. 199–204.
[30] B. Wickremasinghe, R.N. Calheiros, and R. Buyya, Cloudanalyst: A cloudsim-based visual modeller for analyz-
ing cloud computing environments and applications, In: IEEE international conference on advanced information
networking and applications, pp. 446–452, 2010.
[31] C. Xiong, L. Feng, and L. Chen, A new task scheduling algorithm based on an improved genetic algorithm in a cloud
computing environment, Adv. Inf. Sci. Serv. Sci. 5(3) (2013), pp. 32.
