Journal of Ambient Intelligence and Humanized Computing

https://doi.org/10.1007/s12652-020-02282-7

ORIGINAL RESEARCH

MHO: meta heuristic optimization applied task scheduling with load

balancing technique for cloud infrastructure services
S. Peer Mohamed Ziyath1 · S. Senthilkumar2

Received: 4 April 2020 / Accepted: 27 June 2020

© Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
The cloud computing provides on demand access to shared resources over internet in a cloud platform powerfully adaptable
and metered way. Cloud computing empowers the user get to wherever to a shared pool of configurable resources and gives
different administrations to the resource assignment like scientific operations, services computing through virtualization. To
give guaranteed productive execution to clients, tasks ought to be proficiently mapped to accessible resources. In this man-
ner, Task Scheduling is noteworthy issue in the cloud infrastructure administrations. The essential target of task execution
planning includes reserving the infrastructure assets and limiting the goal of the execution plan. In this research work, we
proposed metaheuristic optimization technique with load balancing to enhance the cloud infrastructure service provider’s
performance there by depleting the scheduling issues. The proposed technique is pertinent for static and dynamic task
condition, where static methods VM parameters are fixed, dynamic means parameters are chosen runtime. The proposed
algorithm consists of two phases MHOS-S and MHO-D for dealing with static and dynamic properties of the task submitted.
The result analysis by comparing with few traditional metaheuristic algorithms proves that the proposed technique performs
better in complex environments.

Keywords Task scheduling · Metaheurisitc · Optimization · Load balancing · Cloud infrastructure

1 Introduction the feedback of the users about the services offered by the
cloud service providers. Companies could make the applica-
Cloud computing is an up and coming developing paradigm tions execute quicker while IT infrastructure are overseen by
in distributed computing that provides programming appli- the cloud specialist organizations. These SLA incorporate
cations, development platform, and computing facilities as service accessibility, failure recovery, throughput, reaction
services. In cloud computing facility, the types of cloud time and so on. In this way, so as to ensure the SLA, cloud
service providers is partitioned into computing facility pro- providing organization desires to guarantee that current
viders and specialist web based service providers. Cloud framework provisions provider’s responsibilities. Among
specialist organization gives the administration dependent all the significant components, mainly the execution time
on the adjusted Cloud providers. According to the selected must be guaranteed. In customary processing condition, if
service level agreements (SLA), the users will receive the CPU and its associated assets for processing is allocated
the service which is requested by them to the cloud ser- leads to the fixation of execution time. Be that as it may, in
vice provider (CSP). Quality of Service (QoS) comprises distributed computing condition the process execution time
is reliant on which environment the execution is happened.
* S. Peer Mohamed Ziyath The time variance in execution is because of the territory
[email protected] of the information to executing device. The cloud comput-
S. Senthilkumar ing gives computational assets, for example, the Virtual
[email protected] Machines (VM) to cloud client’s on-request premise (Li
et al. 2012; Abrishami et al. 2013; Tsai et al. 2014). In this
1
Department of CSE, M.E.T Engineering College, Nagercoil, manner, the processing machine provided through infra-
Tamilnadu, India
structure framework is not a physical machine, instead it is
2
Department of CSE, University College of Engineering virtual one. Those virtual machines were conveyed through
Pattukkottai, Marudangavayal, Tamilnadu, India

13
Vol.:(0123456789)
 S. P. M. Ziyath, S. Senthilkumar

numerous physicals equipment utilizing proprietary hyper- et al. 2020). There are two types of scheduling techniques
visor. The infrastructure cloud framework comprises of for task: dependent scheduling and independent scheduling
different rack servers which might be placed in numerous (Masdari et al. 2017; Nagadevi et al. 2013). In independ-
geological areas. Every rack server contains various physi- ent scheduling jobs are autonomous of one another and
cal servers, and more than one virtual machine would be can be planned for any succession, anyway in dependent
running on one physical server. Servers on a similar rack scheduling, Jobs are assigned to Workflow Scheduling works
are straightforwardly associated with a similar switch yet according to the concept of directed acyclic graph (DAG).
those on various racks are associated through any switch The combination of edges and vertices are the components
or might through switches relying upon the open area of of a DAG. The vertex denotes the task and the edge denotes
the physical servers. The network used to set up communi- the purpose of connection among the vertexes (Singh et al.
cation incorporates copper, Fiber or SAS. The information 2015). Task scheduling is further divided into static and
required by a processing device (i.e., virtual machine) will dynamic task scheduling. While scheduling the task in static
be accessible through direct attached storage (DAS) asso- mode, each and every task or VMs are said to be from the
ciated with the physical server, network attached storage earlier to scheduling. Along these lines, it forces less runt-
(NAS) connected through Fiber/Ethernet switch, or storage ime overhead. Then again, in dynamic planning, the data
area network (SAN) open through fiber/Ethernet switch. The about the tasks is obscure ahead of time. Thus, Execution
physical remoteness between the executing and information time of task and information about VM’s, are not known
devices assumes a fundamental role in the real execution until it comes to scheduling, in order to execute the task
time required to execute any assignments (i.e., execution faster and assign the task to VM’s, scheduling is introduced
lessens with the reduction out yonder among processing and (Nagadevi et al. 2013; Mathew et al. 2014; Mohammed
information devices and furthermore relies upon the hidden et al. 2020). It has stayed a theme of research in different
communication and storage innovation). fields for quite a long time, may it schedule of procedures
Task scheduling enhancement is always a research region or threads in a working framework, work shop, stream shop
in infrastructure based cloud due to its property of NP-hard. or open shop planning for product production, imprinting
By the way, the independent quality and the infrastruc- circuit load up scheduling or scheduling of jobs in dispersed
ture differentiation And the different types of resources processing frameworks, for example grid computing, clus-
in infrastructure and the virtual machine placement needs ter computing etc. The popular metaheuristic enhancement
different techniques for the optimal task scheduling so that solution League Championship Algorithm (LCA) proposed
the resources in infrastructure based cloud services could by Kashan (2009). To frame an engineered championship
achieve minimum makespan time. This makespan is directly setting, the researcher expresses some glorified principles to
proportional to the cost for the execution in this condition pursue, at that point presents the promising computational
(Achary et al. 2015; Alkhanak et al. 2015; Srikanth et al. knowledge algorithm that is displayed dependent on various
2015). Expecting the calculation intensity of the each virtual interesting outcomes comparative with sports championship
machine is same and there is no waiting time for the assign- round robin timetable. The tasks are generally submitted and
ment than the execution time of the job will be subject to the placed in VM machines by taking the history of the tasks
region of the information. Consequently, the execution time supplied by the user in the cloud service provider. But this
diminishes concerning separation of the territory between would make issue if the tasks are highly dynamic in nature
data and process. The innovation used to have the informa- and user can’t provide the minimum required properties.
tion and the availability additionally assumes an imperative Hence it would be better if a solution is developed which
job in task consummation. Scheduling permits ideal portion could take the dynamic properties of the tasks submitted
of assets among given tasks in a limited time to accomplish into account for better VM placement as well as achieving
expected job administration. Officially, scheduling issue minimum completion time of the tasks.
includes jobs that must be booked on infrastructure subject In this paper, metaheuristic optimization applied task
to certain limitations to advance objectives. The point is scheduling techniques with load balancing are implemented
to fabricate a schedule that determines when and on which and compared to the traditional heuristic methods to solve
machine each task will be executed (Karger et al. 2010). To the independent static task scheduling problem in cloud
give guaranteed capable execution to clients, it is vital that infrastructure environment. We have employed two differ-
tasks ought to be mapped productively to accessible infra- ent environments: symmetric and asymmetric environments.
structure machines. The fundamental highlights of skilled In symmetric environment; the specifications of the VMs are
schedule planning for cloud computing condition are reduc- fixed. However, in the asymmetric environment, the VMs are
ing data flow time, total throughput of the system, level of decided randomly according to various specifications such
irregularity, makespan which results in increased utilization as RAM, Bandwidth and MIPS. Symmetric scheduling is
of resources (Widmer et al. 2008; Tsai et al. 2014; Malar unrealistic, because it does not take full advantage of the

13
MHO: meta heuristic optimization applied task scheduling with load balancing technique for…

asymmetric nature of VMs. The following section describes that the hybridized system introduced is progressively fit
various traditional metaheuristic task scheduling techniques for improving the exhibition of datacenter, increment the
and their drawbacks. Section 3 describes the developed opti- reaction speed and accuracy. Other ACO assignments sched-
mization technique in detail. Section 4 explains about the uling plans in Cloud are exhibited in (Achary et al. 2015;
simulation environment and various tests run based on the Sidhu 2015; Tawfeek et al. 2013; Tsai and Rodrigues 2014),
proposed technique and its comparison with few traditional while load balancing consciousness can be accomplished
metaheuristics techniques. Section 5 concludes the paper through planning procedures utilizing ACO as displayed in
and the further research enhancements that could adapted (Jin et al. 2015; Li et al. 2011). Hosen et al. (2015) use
in the proposed technique. neural frameworks to create and improve the idea of gauge
between times, through displaying a weighted averaging part
and utilizing reenacted fortifying calculation and hereditary
2 Related works figuring. Precup et al. (2017) and Vrkalovic et al (2018)
applied the grey wolf optimization (GWO) calculation to
As of now cloud based service computing is still evolving. tackle the advancement issue. Authors in Javad et al (2017)
It is still not yet reached its stability as new development applied concordance search estimation to plan Echo State
and delivery methods are evolving now and then. Essentially Networks. In addition, the HS-ESN model had the speediest
cloud computing are given under three layers, for example, movement speed and the least time ate up. The above dis-
software as service, infrastructure as service, platform as cussed methods were able to solve the issue through various
service. Since all the information technology based compa- techniques, still lack of adapting the dynamic requirements
nies are spending more amount of money on hardware, it is of the tasks submitted and also for fixed requirements, the
their intention to use infrastructure as a service more than make span time increases as these methods are dealing the
other two types of cloud services. Hence it becomes neces- problem as multi constraints based one.
sary for cloud service provides for concentrating more on Task scheduling optimization, this investigation is one
infrastructure to adopt new challenges then to develop new of the exemplary issues of improvement from the begin-
solutions. As of late, various metaheuristics based essential ning. The Task execution planning for versatile distributed
search methods were developed in taking care of schedul- computing primarily centers around limiting absolute fin-
ing issue in infrastructure based cloud service request pro- ishing time, asset usage, cost, vitality utilization and addi-
cessing. Metaheuristic able to order as populace based, for tions incredible consideration since the idea of portable
example, genetics based (Kolodziej et al. 2011), ant colony distributed computing is advanced. Authors in Fernando
based (Achary et al. 2015) and PSO (Zhong and He 2010); et al. (2012) and (Rahimi et al 2014) make an expansive and
and the course depends upon some of the techniques such broad audit of flexible distributed computing applications,
as simulated annealing (Geng et al. 2015), genetic algorithm structure and a logical characterization issue, the techniques
(GA) was utilized by the traditional approaches to attain where these issues have been taken care of, challenges and
the optimal solution for the issues regarding to the schedul- openings, insurance and security, future direction. Authors
ing the tasks in distributed environments such as grid and in Kulkarni and Annappa (2019) proposes a context aware
cloud computing (Beegom and Rajasree 2015). According optimized virtual machine placement technique. This
to Darwinian’s theory Genetic algorithm is an optimization method is developed based on heuristics based approach
approach which comes under model of metaheuristic (Dutta for optimization in heterogeneous cloud infrastructure pro-
and Joshi 2011; Yousif et al. 2012). Further a multi objective viders. This method uses the characteristics of the physi-
parallel virtual machines scheduling strategy utilizing genet- cal machines and the historical load log of the machines to
ics algorithm to expand adaptation to non-critical failure optimize the VM placement. This technique also proposed to
versatility in the Cloud service condition. The methodology limit the power utilization along with improved performance
gives a solitary ideal arrangement, yet a lot of results that of heterogeneous cloud service infrastructure providers. In
are not enslaved by each other. Be that as it may, the aim Chaudhary and Kumar (2019) a Hybrid Genetic-Gravita-
of this study is to determine which scheduling algorithm tional Search Algorithm (HG-GSA) proposed for reducing
provides best solution out of many outputs or scheduling. the total cost of computation. Here the cost for transfer of
Hu and Zhou et al. (2015) introduced task scheduling in data and execution cost are taken into account for calculating
Infra structure as service (IAAS) cloud developed (ACO) the total computational cost. This work proposed a gravita-
ant colony optimization and (DTP) dynamic trend predic- tional search algorithm to find the best suitable position of
tion are being used as planning strategy for infrastructure the particle using hybrid crossover technique. In the research
based cloud services. The plan uses DTP to anticipate store work Kaur and Kaur (2019), optimized usage of virtual
modification of cloud datacenter, a short time later shows machines technique for execution of workflows is proposed.
the physical equality through rule. Recreation results show The proposed techniques Ant Colony optimization (ACO),

13
 S. P. M. Ziyath, S. Senthilkumar

hybrid heterogeneous early finish time (HEFT). These tech- phases of algorithm why because the user could tell whether
niques implemented load balanced resource provision using the given task is constraints based or not. To reduce the
a hybrid approach. This technique utilizes to achieve its time for make span, these two phases would decide accord-
optimal performance in respect of make span and cost. The ing to the information given by the user. The task should
above discussed works could not handle the dynamic nature be maintained the dependency while sorting the task. The
of the tasks supplied for execution in the cloud environment. process is repeated until position update process is com-
Hence this research work takes the static and dynamic nature pleted. The best scheduling is done to produce an optimal
of job properties and proposed a metaheuristics optimization solution and improve the efficiency of the task and it’s the
applied technique with load balancing sensitivity. output of the parent node the child node will not get the input
for its execution performance. In a distribution of particles,
it depend on the behavior of the particle and many micro
3 Metaheuristic optimization with load particle having amazing echolocation. It helps to solve the
balancing multi-objective is used to solve the task with its sum of the
weighted task. It even helps to improve the efficiency based
The client gives a job work and the activity is planned to on the finding prey and avoid obstacles. It is very flexible
give an ideal task execution arrangement by utilizing meta- to solve the wide range of the task in NP-hard levels. The
heuristic methods dependent on optimization technique. The number of schedule will be handled with greater probabili-
significant part of any meta-heuristic calculation to come ties on different machine are scheduled in the appropriate
to an ideal arrangement is the utilization of seed arrange- machine. According to this, it will detect of all scheduled
ments. Seed arrangements are starting plausible answers task to produce an effective outcome. They use a random
for the issue which are utilized by an optimized calcula- phase to get an optimized result. Along these lines, locate
tion to assist the way toward finding an ideal arrangement. the most ideal answer for the cloudlets we use particles and
These arrangements are one of the significant elements for schedule them productively. This calculation gives the best
fast combination of any improvement optimized solution. position to the cloudlets for executions on VMs. The well-
Numerous methods have been utilized by analysts to produce ness work assesses the wellness estimation of the particles in
seed arrangements, contingent upon the idea of the issue. the pursuit space. The principal molecule is instated arbitrar-
These consolidate picking an unpredictable game plan, ily in the pursuit space pursued by choice of the following
logical course of action reliant on some restricted model of molecule dependent on better wellness esteem. It relies upon
issue, or possibly a heuristic based game plan. All of these data transfer capacity, MIPS, flow time, reaction time, asset
frameworks has its very own positive and negative points of use, throughput time, awkwardness degree estimations of
view. In any case, without an appropriate heuristic, uneven acquired from verifiable log of the cloud framework.
seed arrangement is every now and again used to create a
seed arrangement. 3.1 MHO algorithms (MHO‑S & MHO‑D)
The MHO Algorithm is used to minimize the execution
time and deadline. The task should maintain their depend- To execute or process a given set of tasks or work
encies even they change or be same from user to user. The W = {W1,W2,W3….,Wn) of a job J to a given set of pro-
input of the task is represented by the Directed Acyclic cessor P = {P1,P2,P3,….Pn} of a virtual machine cluster,
Graph (DAG) and it has the number of task with it and it subject to a predefined set of measurements as defined above
is denoted as T. The data transmission time between the is the optimized solution. The scheduling issue on cloud can
tasks is represented along the edges. They have an entry–exit be defined for solution S as follows
relationship. The child node depends on parent node without
t,m
MHO Algorithm is applied to schedule the task to achieve ∑
the optimistic schedule. It is the population based algo-
F(S) = min Ct (1)
i=1,j=1
rithm on the behaviour of tasks through its properties. The
solution to the problem is represented by the task position. where F(S) is fitness function of the solutions and Ct is com-
MHO algorithm is based on the behavior of swarm-intelli- pletion time of all the tasks and t is total number of task
gent algorithm and meta- heuristic techniques. It provide provided by the user and m is the total number of machines
an optimum solution to complex data and it is compared available. Here the fitness function value depends on the
with other algorithm. MHO algorithm consists of two phases machine parameters which is based on the type of jobs
(1) scheduling job with the static constraints, (2) schedul- provided. Here the type of job could be dynamic or static
ing jobs with dynamic constraints. In Scheduling job with kind. Static means the job properties are predefined like
execution time and deadline to obtain different scheduling total amount of data, flow of data in the system, time limita-
by changing its position of the task in the schedule. The two tion etc. Dynamic jobs represent jobs with undefined job

13
MHO: meta heuristic optimization applied task scheduling with load balancing technique for…

properties like data bursting, undefined data type etc. While where Pi defines the property of the job and CO(jo) defines
submitting the data, user should provide the job property the cost for executing the function and m defines the number
as static or dynamic. In static type the predefined proper- of jobs and n defines the number of properties of the job.
ties of the job will be fixed and then scheduling happens. Based on the cost of the job, the available VM cluster fitness
In dynamic type jobs, user would able to give the mini- values are calculated and the best fit function is applied to
mized property values of the job and the cloud scheduler select the optimal VM cluster for the execution of the jobs.
will decide allocation of the machine based on the runtime The fitness function is calculated as mentioned in Eq. (3).
operations. The machine properties are predefined by the n,m,p
cloud job manager like, RAM size, SSD size, processing
∑
Bestfit(F(S)) = min (P1, P2, … Px) + Tj + Mk (3)
speed, machine instructions per second, bandwidth, etc. The i=1,j=1,k=1
system architecture of the proposed technique is given below
as Fig. 1. where F(S) is the fitness value of corresponding VM cluster,
In static job scheduling, the properties of job is given n defines number of job, x value defines the value of corre-
by the user, where the property types are predefined by the sponding property P of the job, T defines the corresponding
cloud service provider which is ­(P1, ­P2, …..Pn). Based on the task, m defines the number of tasks, M defines the machine
values provided by the user, the current usage of the infra- and p defines the number of machines of the correspond-
structure is defined and the required VM cluster is defined. ing VM cluster. The static scheduling of tasks is defined in
Different types of VM clusters and their fitness value is MHO-S algorithm as given in Table 1.
calculated, by using the best fit function, the suitable VM From the MHO-S algorithm, the bestfit value obtained
cluster will be selected and thus load balancing is achieved VM cluster for static property task is selected for task
in this case. The cost of executing the jobs is calculated execution.
through property values defined by the user through the Eq. In Dynamic job scheduling, the properties of job could
(2). not be defined, so the service provider could define the mini-
mum number of properties and their corresponding prop-
n,m
∑ erty values as expected. These values are received from the
CO(jo) = Pi (2) user while submitting the initial job and then the required
i=1,j=1
machines are allocated for initial job with pool of machines

Fig. 1  System architecture

13
 S. P. M. Ziyath, S. Senthilkumar

Table 1  MHO-S scheduling algorithm where Pi defines the property max value and n defines the
Start; no of properties of the job, m defines the number of tasks for
Input: Job with set of Tasks J(T1, T2, ….. Tj); the job. During the runtime the maximum property values so
Set of task properties (P1, P2, ……Pn); far obtained by the tasks of the job is recorded and the mean
Set of VM cluster properties (M1, M2, …… Mk); value of the property is taken into account for allocation of
Output: bestfit VM cluster F(S) new VM cluster for the incoming tasks with dynamic prop-
erty values and defined by the Eqs. (5) and (6).
For I = 1 to n & J = 1 to m
Each task Ti with Property P
( n )
∑ /
Calculate Pi(T) = Vi n (5)
i=1
( ) ( )
(( n,m,p
) n,m
)
For all F(S)
∑ ∑
Bestfit(F(S)) = min (P1, P2, … Px) + Tj + Mk < max Pi
Find min F(S); i=1,j=1,k=1 i=1,j=1

(6)
Return Bestfit F(S); where in Eq. (5), Pi(T) defines the mean property value of
End; task, Vi defines the value of property, n defines the number
of property values obtained so far from the operation. In
equation VI, F(S) is the fitness value of corresponding VM
available to increase or decrease the number of machines cluster, n defines number of job, x value defines the value
depending upon the load of the job during runtime. The cost of corresponding property P of the job, T defines the cor-
of executing the job is calculated as maximum value that responding task, m defines the number of tasks, M defines
could be accepted by the user for initial startup and calcu- the machine and p defines the number of machines of the
lated as given in Eq. (4). corresponding VM cluster. The calculated value should be
the minimum of all fitness values which are less than the
n,m
∑ maximum value accepted by the user for the execution of
CO(jo) = max Pi (4) tasks. The dynamic scheduling of tasks is defined in MHO-D
i=1,j=1
algorithm as given in Table 2.

Table 2  MHO-D scheduling Start;

algorithm
Input: Job with set of Tasks J(T1, T2, ….. Tj);
Set of task properties (P1, P2, ……Pn);
Set of VM cluster properties (M1, M2, …… Mk);
Set of Task property values V;
Output: bestfit VM cluster F(S)

For each task T

Calculate ( ) ∑
For I = 1 to n & J = 1 to m
Each task Ti with Property P
Calculate

( ) ( )

For all F(S)

Find min F(S);

Return Bestfit F(S);

End;

13
MHO: meta heuristic optimization applied task scheduling with load balancing technique for…

From the MHO-D algorithm, the bestfit value obtained CA-O (Kulkarni and Annappa 2019), COH (Chaudhary
VM cluster is selected for task execution. and Kumar 2019), and LBOH (Kaur and Kaur 2019).The
performance parameters taken into account are makespan
time, response time, execution time and ratio of VM used.
4 Performance evaluation Makespan provides the value of the finishing time of the last
provided task, response time defines the time taken for the
For evaluating the performance of the proposed technique, scheduler to schedule the given tasks, execution time pro-
CloudSim (Calheiros et al 2011) simulator is used. The vides the time take for the tasks to execute completely in the
cloud platform parts like VMs, cloudlets, Datacenter, VMs, virtual machines and ratio of VM used defines the number
Client, cloudlets (undertakings) and cloud clients determi- of virtual machines used against the total number of virtual
nation are introduced in Table 3. Those values are defined machines available under the cloud infrastructure service
as fixed values for static simulation and minimum value for provider. Figure 2 depicts the comparison of proposed tech-
dynamic simulation where those values could be increased nique with existing methods under makespan metric.
during runtime. From Fig. 2 it is evident that, the proposed MHO algo-
The proposed MHO Algorithm is compared with three rithm provides optimized VM placement even though in
other methods described in related work section namely increased number of tasks than the existing systems. As the
algorithm deals with tasks by means of taking the properties
into consideration, which make it to provide optimal VM
Table 3  Simulation parameters placement. Figure 3 provides the comparison of response
Component Property Value time of the proposed algorithm with other existing systems.
Figure 3 proves that, the proposed MHO algorithm con-
Data Center Total no 2
verge faster and returns the placement of VM scheduling for
Virtual machine No of VM
the tasks given when compared with other existing solutions.
RAM 2 GB
The MHO algorithm can place the VM quickly as it takes
SSD 100 GB
the static property values first to deal with simpler tasks
Processor 2.4 GHz
which extends the capability of the algorithm to deal with
MIPS 1000
dynamic property based tasks strongly. Figure 4 provides the
OS Linux
comparison between MHO algorithms with other methods
Host Total No 3
in terms of execution time.
RAM 2 GB
Figure 4 defends that, the proposed MHO algorithm
Storage 100 GB
supports both dynamic and static property based tasks and
Bandwidth 10 GB/s
provide better execution time than other existing systems
Cloudlet Total no 500
through optimized VM placement scheduling. As the sim-
Length 50,000
pler tasks are placed in the VM quickly by MHO than
User No of users 1
any other methods, MHO achieves lower execution time.

Fig. 2  Makespan comparison 25000

20000
Makespan Time

15000

10000

5000

0
250 500 750 1000 1500 2000
Number of Tasks
MHO CA-O COH LBOH

13
 S. P. M. Ziyath, S. Senthilkumar

Fig. 3  Response time 120

100

80

Response Time
60

40

20

0
250 500 750 1000 1500 2000
No of Tasks

MHO CA-O COH LBOH

Fig. 4  Execution time 200000

180000
160000
140000
Execuon Time

120000
100000
80000
60000
40000
20000
0
250 500 750 1000 1500 2000
No of Tasks
MHO CA-O COH LBOH

Fig. 5  Number of virtual 14

machines
12

10
No of Virtual Machines

0
250 500 750 1000 1500 2000
No of Tasks

MHO CA-O COH LBOH

13
MHO: meta heuristic optimization applied task scheduling with load balancing technique for…

Figure 5 provides the comparison between proposed MHO it deals with tasks based in the properties defined by the
algorithms with other existing systems in terms of number user. The future work will look for extend this research in
of virtual machines used. This comparison is important incorporating the virtual network function provisioning
as service providers intended to provide service with less along with task scheduling.
number of virtual machines so that they could accommo-
date more user tasks.
Figure 5 defends that, MHO algorithm scheduling the References
tasks in an optimal way (i.e. using less number of virtual
machines) than the other existing systems. As the static Abrishami S, Naghibzadeh M, Epema DH (2013) Deadline-con-
strained workflow scheduling algorithms for infrastructure as a
property based tasks require minimum number of VM, service clouds. Future Gener Comput Syst 29:158–169
MHO achieves the lower number of VM used than other Achary R, Vityanathan V, Raj P, Nagarajan S (2015) Dynamic job
methods. By analyzing the performance comparison under scheduling using ant colony optimization for mobile cloud com-
different parameters, it is evident that MHO algorithm is puting. Intelligent distributed computing. Springer, Berlin, pp
71–82
providing an optimal virtual machine scheduling and also Alkhanak EN, Lee SP, Khan SUR (2015) Cost-aware challenges for
achieving load balancing, so that the cloud service pro- workflow scheduling approaches in cloud computing environ-
vider could accommodate more no of tasks from different ments: taxonomy and opportunities. Future Gener Comput Syst
users in runtime. The negative fact of the proposed work 50:3–21
Beegom AA, Rajasree M (2015) Genetic algorithm framework for
is that, it could utilize more VM placements incase more bi-objective task scheduling in cloud computing systems. Dis-
number of tasks with dynamic properties are submitted. tributed computing and internet technology. Springer, Berlin,
In this case, it is hard for the system to predict the future pp 356–359
use of the VM and could utilize a new VM for placing a Calheiros RN, Ranjan R, Beloglazov A, De Rose CAF, Buyya R
(2011) CloudSim: a toolkit for modeling and simulation of
static property based task. cloud computing environments and evaluation of resource
provisioning algorithms. Softw Pract Exp (SPE) 41(1):23–50
(ISSN: 0038-0644)
Chaudhary D, Kumar B (2019) Cost optimized hybrid genetic-grav-
itational search algorithm for load scheduling in cloud comput-
5 Conclusion ing. Appl Soft Comput 83:1–18
Dutta D, Joshi RC (2011) A genetic: algorithm approach to cost-
Virtual machine scheduling and load balancing is always based multi-QoS job scheduling in cloud computing environ-
an open issue for infrastructure based service providers as ment. In: Proceedings of the international conference & work-
shop on emerging trends in technology. ACM, pp 422–427
they tend to serve more users while utilizing less number Fernando N, Loke SW, Rahayu W (2012) Mobile cloud computing:
of virtual machines. There were more research solutions a survey. Future Generat Comput Syst 29(1):84–106
proposed for scheduling and load balancing, while they Geng J, Huang M-L, Li M-W, Hong W-C (2015) Hybridization of
have their own drawbacks like not considering runtime seasonal chaotic cloud simulated annealing algorithm in a SVR-
based load forecasting model. Neurocomputing 151:1362–1373
variation of task properties as well as virtual machine Hosen MA, Khosravi A, Nahavandi S, Creighton D (2015) Improving
properties. Hence to overcome this issue, this research the quality of prediction intervals through optimal aggregation.
work proposed a metahuristic optimization applied task IEEE Trans Ind Electron 62(7):4420–4429
scheduling with load balancing technique. This tech- Javad S, Payman M, Hamidreza K (2017) Training echo estate neu-
ral network using harmony search algorithm. Int J Artif Intell
nique is composed of two sub algorithms, MHO-S and 15(1):163–179
MHO-D, where MHO-S deals with tasks of confirmed Jin G, Liu L, Zhang P, Yu M (2015) Cost constrain load balanced
and static property values and virtual machine properties. ant colony scheduling of cloud environment. J Inf Comput Sci
MHO-D take care of scheduling the tasks which could 12:1045–1054
Karger D, Stein C, Wein J (2010) Scheduling algorithms. Algorithms
not provide the confirmed property values and submit a and theory of computation handbook: special topics and tech-
minimum property they will maintain. The user will also niques. Chapman & Hall/CRC, London
provide his maximum acceptable values for completion Kashan AH (2009) League championship algorithm: a new algo-
of tasks. Based on those values MHO-D schedules the rithm for numerical function optimization. In: 2009 interna-
tional conference of soft computing and pattern recognition.
tasks in an optimized way. The performance evaluation IEEE, pp 43–48
of the proposed MHO algorithm with other few exist- Kaur A, Kaur B (2019) Load balancing optimization based on hybrid
ing solutions proves that, MHO algorithm provides low heuristic-metaheuristic techniques in cloud environment. J King
execution time of tasks while utilizing minimum number Saud Univ Comput Inf Sci. https​: //doi.org/10.1016/j.jksuc​
i.2019.02.010
of virtual machines in the VM cluster of infrastructure Kolodziej J, Khan SU, Xhafa F (2011) Genetic algorithms for energy-
service provider. The proposed technique could provide aware scheduling in computational grids. In: 2011 international
the make span faster and keep the execution time lower as conference on P2P, parallel, grid, cloud and internet computing.
IEEE, pp 17–24

13
 S. P. M. Ziyath, S. Senthilkumar

Kulkarni AK, Annappa B (2019) Context aware VM placement opti- Sidhu J (2015) Ant colony optimization algorithm for independent task
mization technique for heterogeneous IaaS cloud. IEEE Access scheduling in cloud computing. Int J Appl Eng Res 10(1):535–544
7:89702–89713 Singh K, Alam M, Sharma SK (2015) A survey of static scheduling
Li K, Xu G, Zhao G, Dong Y, Wang D (2011) Cloud task scheduling algorithm for distributed computing system. Int J Comput Appl
based on load balancing ant colony optimization. In: 2011 sixth 129:25–30
annual ChinaGrid conference. IEEE, pp 3–9 Srikanth GU, Maheswari VU, Shanthi A, Siromoney A (2015) Task
Li J, Qiu M, Ming Z, Quan G, Qin X, Gu Z (2012) Online optimiza- scheduling model. Indian J Sci Technol 8:33–42
tion for scheduling preemptable tasks on IaaS cloud systems. J Tawfeek MA, El-Sisi A, Keshk AE, Torkey FA (2013) Cloud task
Parallel Distrib Comput 72:666–677 scheduling based on ant colony optimization. In: 2013 8th inter-
Malar ACJ, Kowsigan M, Krishnamoorthy N, Karthick S, Prabhu E, national conference on computer engineering & systems (ICCES).
Venkatachalam K (2020) Multi constraints applied energy effi- IEEE, pp 64–69
cient routing technique based on ant colony optimization used Tsai CW, Rodrigues JJ (2014) Metaheuristic scheduling for cloud: a
for disaster resilient location detection in mobile ad-hoc net- survey. IEEE Syst J 8:279–291
work. J Ambient Intell Human Comput. https​://doi.org/10.1007/ Tsai C-W, Huang W-C, Chiang M-H, Chiang M-C, Yang C-S (2014)
s1265​2-020-01767​-9 A hyper-heuristic scheduling algorithm for cloud. IEEE Trans
Masdari M, Salehi F, Jalali M, Bidaki M (2017) A survey of PSO-based Cloud Comput 2:236–250
scheduling algorithms in cloudcomputing. J Netw Syst Manag Vrkalovic S, Lunca E-C, Borlea I-D (2018) Model-free sliding mode
25:122–158 and fuzzy controllers for reverse osmosis desalination plants. Int
Mathew T, Sekaran KC, Jose J (2014) Study and analysis of various J Artif Intell 16(2):208–222
task scheduling algorithms in the cloud computing environment. Widmer M, Hertz A, Costa D (2008) Metaheuristics and scheduling.
In: Proceedings of the international conference on advances in In: Lopez CP, Roubellat F (eds) Production scheduling. Wiley,
computing, communications and informatics, Sept 24–27. IEEE Hoboken
Xplore Press, New Delhi, India, pp 658–664 Yousif A, Abdullah AH, Nor SM, Bashir MB (2012) Optimizing job
Mohammed AS, Balaji BS, Basha SMS, Asha PN, Venkatachalam scheduling for computational grid based on firefly algorithm. In:
K (2020) FCO—fuzzy constraints applied cluster optimiza- 2012 IEEE conference on sustainable utilization and development
tion technique for wireless adhoc networks. Comput Commun in engineering and technology (STUDENT). IEEE, pp 97–101
154:501–508 Zhong SB, He ZS (2010) The scheduling algorithm of grid task based
Nagadevi S, Satyapriya K, Malathy D (2013) A survey on economic on PSO and cloud model. Key Eng Mater 439–440:1487–1492
cloud schedulers for optimized task scheduling. Int J Adv Eng
Technol 4:58–62 Publisher’s Note Springer Nature remains neutral with regard to
Precup RE, David RC, Szedlak-Stinean AL, Petriu EM, Dragan F jurisdictional claims in published maps and institutional affiliations.
(2017) An easily understandable grey wolf optimizer and its appli-
cation to fuzzy controller tuning. Algorithms 10:68
Rahimi M, Ren J, Liu C, Vasilakos A, Venkatasubramanian N (2014)
Mobile cloud computing: a survey state of art and future direc-
tions. Mobile Netw Appl 19(2):133–143

13