Multi-objective-Oriented Cuckoo Search Optimization-Based Resource Scheduling Algorithm for Clouds

Abstract

Scheduling problems in cloud computing environment are mostly influenced by multi-objective optimization but frequently deal with using single-objective algorithms. The algorithms need to resolve multi-objective problems which are significantly different from the procedure or techniques used for single-objective optimizations. For this purpose, meta-heuristic algorithms always show their strength to deal with multi-objective optimization problems. In this research article, we present an innovative Multi-objective Cuckoo Search Optimization (MOCSO) algorithm for dealing with the resource scheduling problem in cloud computing. The main objective of resource scheduling problem is to reduce the cloud user cost and enhance the performance by minimizing makespan time, which helps to increase the revenue or profit for cloud providers with maximum resource utilization. Therefore, the proposed MOCSO algorithm is a new method for solving multi-objective resource scheduling problems in IaaS cloud computing environment. Moreover, the effects of the proposed algorithm are analyzed and evaluated by comparison with state-of-the-art multi-objective resource scheduling algorithms using simulation framework. Results obtained from simulation show that the proposed MOSCO algorithm performs better than MOACO, MOGA, MOMM and MOPSO, and balance multiple objectives in terms of expected time to completion and expected cost to completion matrices for resource scheduling in IaaS cloud computing environment.

Full text

Arabian Journal for Science and Engineering
https://doi.org/10.1007/s13369-018-3602-7
RESEARCH ARTICLE - COMPUTER ENGINEERING AND COMPUTER SCIENCE
Multi-objective-Oriented Cuckoo Search Optimization-Based Resource
Scheduling Algorithm for Clouds
Syed Hamid Hussain Madni1·Muhammad Shafie Abd Latiff1·Javed Ali2·Shafi’i Muhammad Abdulhamid3
Received: 11 July 2018 / Accepted: 10 October 2018
© King Fahd University of Petroleum & Minerals 2018
Abstract
Scheduling problems in cloud computing environment are mostly influenced by multi-objective optimization but frequently
deal with using single-objective algorithms. The algorithms need to resolve multi-objective problems which are significantly
different from the procedure or techniques used for single-objective optimizations. For this purpose, meta-heuristic algorithms
always show their strength to deal with multi-objective optimization problems. In this research article, we present an innovative
Multi-objective Cuckoo Search Optimization (MOCSO) algorithm for dealing with the resource scheduling problem in cloud
computing. The main objective of resource scheduling problem is to reduce the cloud user cost and enhance the performance
by minimizing makespan time, which helps to increase the revenue or profit for cloud providers with maximum resource
utilization. Therefore, the proposed MOCSO algorithm is a new method for solving multi-objective resource scheduling
problems in IaaS cloud computing environment. Moreover, the effects of the proposed algorithm are analyzed and evaluated
by comparison with state-of-the-art multi-objective resource scheduling algorithms using simulation framework. Results
obtained from simulation show that the proposed MOSCO algorithm performs better than MOACO, MOGA, MOMM and
MOPSO, and balance multiple objectives in terms of expected time to completion and expected cost to completion matrices
for resource scheduling in IaaS cloud computing environment.
Keywords Cloud computing ·Cuckoo search ·Meta-heuristic algorithm ·Multi-objective optimization ·Resource scheduling
1 Introduction
The term cloud computing has become one of the most
demanding topics in academic research and IT industries
since 2006 [1]. It is a new paradigm that offers the soft-
ware, database, platform, infrastructure, hardware, comput-
ing, storage and security as services through the Internet. In
cloud computing environment, computing entities are vir-
tualized, dynamically configured and driven by economic
scale [2]. It is progressively adopted by individuals, small-
and medium-scale enterprises (SMEs) and government orga-
BSyed Hamid Hussain Madni
madni4all@yahoo.com
1School of Computing, Faculty of Engineering, Universiti
Teknologi Malaysia, 81310 Skudai, Johor, Malaysia
2College of Computing Informatics, Saudi Electronic
University, Madinah Munawarah, Kingdom of Saudi Arabia
3Federal University of Technology Minna, Minna, Niger State,
Nigeria
nizations to run their common and critical applications. The
reason for this adoption is the special characteristics of cloud
including the administration, agility, availability (anytime
and anywhere access), benefits of pay-per-use, delegation
of maintenance, disaster recovery, elasticity, economic, on
demand service, operational expenditure, performance, reli-
ability, scalability, security and trust [3,4].
Resource management is the primary problematic issue
raised in cloud computing. It includes the resource allocation,
resource scheduling, resource discovery, resource monitor-
ing, resource availability, resource pricing, etc [5]. From
cloud providers prospective, large-scale virtual resources
need to be allocated to millions of cloud users, dynamically,
correctly and profitably. From cloud users point of view,users
are economically driven entities when they choose to use
cloud services. For sufficient resources, cloud users always
compare the prices between various cloud providers [6].
Resource scheduling chooses the best virtualized resources
from the appropriate physical resources. In fact, it classifies
the physical resource where the virtual machines (VMs) are
to be generated to allocate or assign the resources from cloud
123

Arabian Journal for Science and Engineering
infrastructure [7]. Therefore, resource scheduling is consid-
ered to be a multi-objective or multi-criteria problem under
the Infrastructure as a Service (IaaS) in cloud computing
environment as shown in Fig. 1. For this purpose, an algo-
rithm is required to solve multi-objective problems, which is
expressively opposite or not the same as the single-objective
problems.
In cloud computing, optimization problems are mostly
considered to be single objective, but in reality, these are
typically multi-objective in nature. For example, every cloud
user wishes to get cloud services with minimum cost and
high performance. On the other hand, cloud providers want
to provide cloud services with maximum utilization of
resources, high revenue and profit. For this purpose, the
algorithms need to resolve the multi-objective problematic
issues in an expressively different way than algorithms for
single-objective optimizations. Meta-heuristic optimization
algorithms always show their advantages and benefits in deal-
ing with multiple-objective functions.
Optimization problems deal with the research on those
types of problematic issues to maximize or minimize one or
more objectives that are functions of certain real or integer
variables. These are implemented in an organized manner to
select the specific values of real or integer variables from a
given set. The main objectives of optimization are to achieve
the optimum values from various objective functions. It
encompasses both minimum and maximum of the objective
function and defined as follows:
Given an objective function f:Rn→Rfrom some set
Rnto the set of real numbers that aim to determine a member
of x,yin Rsuch that:
f(x)< f(y), ∀(x,y)∈Rn(for minimization)
or
f(x)> f(y), ∀(x,y)∈Rn(for maximization)
Here, Rrepresents a subset of the Euclidean space Rn,
which is a group of objects including the constraints, equali-
ties or inequalities. The function fis known as an objective
function, which consists of various parameters like cost, time,
energy. A realistic solution which is used to optimize the
objective function is known as an optimal solution.
Multi-objective optimizations are either multi-criteria or
multi-attribute optimization, which deal with the task of
concurrently optimizing two or more conflicting objectives
regarding a group of definite limits. In optimization, if one
objective indicates to the other objective optimization auto-
matically, then it cannot be considered a multi-objective
problem. However, in cloud computing environment, we
come across problems where the challenges are to improve
one objective that sometimes leads to degradation of the
other. Such problems fit in the set of multi-objective problems
and seem in several fields related to cloud computing includ-
ing resource management [5], resource allocation [8] and
scheduling [8], green computing (energy and heat consump-
tion) [9], big data (storage) [10–12], fog computing [13],
Internet of Things (IoT) [14].
In recent times, an innovative optimized meta-heuristic
Cuckoo Search (CS) algorithm is created by “Yang and
Deb” [15]. Existing studies confirm that it is reliable and per-
forms better than current meta-heuristic algorithms includ-
ing the Ant Colony Optimization (ACO), Genetic Algo-
rithm (GA), Particle Swarm Optimization (PSO), Simulated
Annealing (SA) in multi-discipline areas [16–20]. In this
research article, we formulate a new and innovated Multi-
objective Cuckoo Search Optimization (MOCSO) algorithm
for multi-objective optimizations of resource scheduling for
enhancing the quality of service, to assign a task to a spe-
cific VM with lowest execution time and cost to fulfill the
cloud user’ s demand and enhance the resource utilization for
cloud providers with less delay. Thus, MOCSO algorithm is
proposed for resource scheduling in IaaS cloud computing,
which helps to reduce the makespan and cost and enhance
the resource utilization.
Our major contributions of this research article are as fol-
lows:
– An innovative Multi-objective Cuckoo Search Optimiza-
tion (MOCSO) algorithm is generated for optimum
resource scheduling in IaaS cloud computing.
– Formulate the makespan, cost and resource utiliza-
tion through mathematical models for optimal resource
scheduling as the objective functions.
– Design the MOCSO algorithm to address the proposed
scheduling models.
– Comparative and statistical analysis of the MOCSO algo-
rithm with existing heuristic and meta-heuristic multi-
objective algorithms is presented for resource scheduling
in IaaS cloud computing environment.
– Performance evaluation of the existing meta-heuristic
algorithms with MOCSO optimization algorithm by
considering the matrices of makespan, cost, resource uti-
lization and performance improvement rate (%).
The remaining sections of this research article are as fol-
lows: In Sect. 2, we review the current comparison studies
and related works for multi-objective resource scheduling in
the field of IaaS cloud computing. Problem formulation is
discussed in Sect. 3. We provide a brief introduction of CS
algorithm with a detailed description of MOCSO algorithm
in Sect. 4. Section 5discusses the parameters for evalua-
tion, while Sect. 6outlines the simulation setup. Results
and discussion demonstrate the performance evaluation with
the help of experimental simulation and statistical analysis
in Sect. 7. The last section is Sect. 8, which comprises the
specifics of the conclusion and recommendations.
123

Arabian Journal for Science and Engineering
Fig. 1 Resource scheduling problem in IaaS cloud computing environment
2 Related Works
In this section, we review the existing studies of multi-
objective resource scheduling algorithm for IaaS cloud com-
puting environment and analyze the CS algorithm to solve
the scheduling problems in various multi-discipline areas of
research. Most of the studies are focused on heuristics, meta-
heuristics and hybrid algorithms for multi-objective resource
scheduling by considering cost, energy, makespan, resource
utilization, time, etc.
Chen et al. [21] combine GA algorithm with knapsack
problem for scheduling the resources with multiple objec-
tives considering the utilization of resources and energy
consumption in cloud computing. Similarly, Sindhu and
Mukherjee [22] use dual-objective GA algorithm for enhanc-
ing the makespan and utilization of resources. For this
purpose, GA algorithm is combined with different heuristic
algorithms consisting of Longest Cloudlet to Fastest Proces-
sor (LCFP), Minimum Completion Time (MCT) and Shortest
Cloudlet to Fastest Processor (SCFP) algorithms for improv-
ing the convergence rate. However, Zhang et al. [23] present
a hierarchal resource scheduling strategy based on PSO and
weight inertia for reducing the communication cost and bal-
ancing the load. For enhancing the cloud providers’ profit
and minimizing the consumption of energy, a job schedul-
ing model is established based on PSO algorithm in cloud
environment [24]. Zhang et al. [25] enhance the Vector-
ized Ordinal Optimization (VOO) approach from single to
multi-objective for multiple tasks scheduling for reducing
scheduling overhead time and performance. In the same way,
Tsai et al. [26] propose Improved Differential Evaluation
Algorithm (IDEA) by merging the Taguchi method and DEA
for optimal resource scheduling. Improved DEA shows the
ideal results for multiple objectives including reducing cost
and makespan.
123

Arabian Journal for Science and Engineering
To enhance the task scheduling, Ramezani et al. [27]
develop a Multi-objective PSO (MOPSO) algorithm for
reducing the execution cost, execution time and transferring
time, and extend the Jswarm package to the Multi-objective
Jswarm package for determining the task arrangement among
the VMs regarding to MOPSO algorithm. Similarly, Alka-
yal et al. [28] apply MOPSO algorithm for selecting suitable
VM based on three conflict objectives including the expected
completion time, execution cost and VM processing. More-
over, He et al. [29] adjust adaptive acceleration coefficient
for maintaining the particle diversity of PSO algorithm as
an Adaptive Multi-objective Task Scheduling (AMTS) Strat-
egy for task scheduling. It enhances the cost, execution time
and resource utilization and reduces the energy consump-
tion. Similarly, Tchernykh et al. [30] emphasize bi-objective
scheduling strategies in IaaS cloud computing in terms of
cost, energy and speed, whereas Zuo et al. [31] propose a
multi-objective scheduling based on ACO regarding the per-
formance matrices consisting of cost, dead line violation,
makespan and utilization of resources.
A Dynamic Objective GA (DOGA) algorithm is devel-
oped by [32], for solving the resource scheduling problem
under the minimization of cost and time model in cloud
computing. Further, Shojafar et al. [33] propose a hybrid
algorithm based on Fuzzy theory and GA (FUGA) algo-
rithm for assigning the jobs to the optimal resources. For
achieving the multiple objectives, fuzzy theory and GA mod-
ify with diverse fuzzy-based stable state GA to minimize
the makespan, execution cost, execution cost and average
degree of imbalance. Moreover, Jena [34] presents a Task
Scheduling PSO (TSPSO) algorithm for optimizing the mul-
tiple objectives, including the energy and execution time,
to effectively utilize the cloud resources in cloud comput-
ing environment. Likewise, Lakra and Yadav [35] present a
multi-objective task scheduling algorithm for assigning tasks
to the VMs in terms of minimizing the cost and enhancing
the throughput without SLA violation in cloud computing
environment.
A normalized Multi-Objective Min-Max Schedulingline-
break (NMOMXS) algorithm is based on Min-min and
Max-min algorithms for task scheduling in multi-cloud
computing by considering the makespan and resource uti-
lization [36]. Similarly, Panda and Jana [37] propose a
Multi-Objective Task Scheduling Algorithm for scheduling
the resources in multi-cloud environment. The technique is
used for three performance metrics including the makespan,
cost and resource utilization. Furthermore, Choudhary et
al. [38] propose a Hybrid Gravitational Search Algorithm
(HGSA) for bi-objective scheduling workflows, by reducing
the makespan as well as the cost of execution. The sim-
ulation results show the effectiveness of the HGSA over
standard Gravitational Search Algorithm (GSA), Hybrid
Genetic Algorithm (HGA) and the Heterogeneous Earliest
Finish Time (HEFT).
Nguyen and Vo [39] recommend a Modified CS (MCS)
algorithm for solving the short-term hydrothermal schedul-
ing problem. The hydrothermal scheduling problem focuses
on minimizing the total cost of thermal generators regard-
ing the loading effects, satisfying power balance constraint,
water availability and generator operating limits. Also,
Bibiks et al. [40] present a Discrete CS (DCS) algorithm
for scheduling the resource-constrained project problem. The
proposed DCS algorithm is evaluated on a set of bench-
mark instances and compared with existing meta-heuristic
algorithms. The numerical results show that the proposed
algorithm is very efficient and outperforms in terms of quality
of the solutions and execution time. Similarly, Marichel-
vam et al. [41] suggest the Improved CS (ICS) algorithm for
minimizing the makespan for the hybrid flow shop schedul-
ing problems. The proposed ICS algorithm is implemented
with real-world application data from a furniture manufac-
turing company. Furthermore, Marichelvam and Geetha [42]
present a hybrid algorithm based on CS algorithm to solve
the single machine total weighted tardiness scheduling prob-
lems. Three different dispatching rules are combined of the
CS algorithm for improving the quality of the solution. Com-
putational results show that proposed algorithms are very
competitive and outperform than other meta-heuristic com-
parison algorithms.
From the analysis of existing studies, the following obser-
vations are obtained. Multi-Objective Min-Max (MOMM)
algorithm uses the heuristic approach, which produces the
constant values at each run, yet is insufficient for the global
search optimization. MOACO algorithm is also considered
inadequate for the local search since it always gets trapped at
the local optima. MOGA algorithm is considered better for
both local and global search, but its convergence rate is very
slow. MOPSO algorithm is also considered superior for both
local and global search, but it gets trapped at the local optima.
Hence, MOCSO algorithm based on CS algorithm uses ran-
dom walk for the local search that enhances the search speed
and the major advantage is to neglect the repeated selection.
The global search is calculated by the Levy flight procedure,
which has infinite mean and variance that explores the search
more efficiently than the other algorithms. Due to these rea-
sons, CS algorithm is comprehensively rewarded attention
by researchers to get the optimal results in multi-discipline
research areas [43–47].
Therefore, the current available resources need to be
proper scheduled with respect to the multi-objective per-
spective regarding the cloud users and providers in cloud
computing. The existing related works show that the cur-
rent resource scheduling techniques neither consider multi-
objective parameters together nor provide efficient near-
optimal solutions for resource scheduling at different levels.
123

Arabian Journal for Science and Engineering
3 Problem Formulation
Scheduling comprises the consignment of starting and com-
pletion times for the several operations to be executed. Like
other scheduling problems, resource scheduling in cloud
computing is a procedure that applies to the distribution
of valuable cloud resources, generally processors, networks,
storage and VMs to accomplish the demands of cloud users
by the cloud providers. It is applied for optimal utilization of
resources to ensure the equal distribution of resources accord-
ing to the demand, and to give some prioritization according
to set rules. It also makes sure that cloud computing is able
to serve all the cloud users’ requests, with a certain quality
of service by the cloud providers. Resource Scheduling (RS)
problem can be clarified with the help of Eq. (1) as depicted
in Fig.1.
RS =
m,n

x=1
(Rx+Sx+··· + Nx)×Tx→Uz
x(1)
where it maps m required numbers of cloudlets/task T=
(T1,T2,T3,...,Tm)onto navailable physical resources to
virtual resources in cloud data centers R=(R1,R2,R3,
...,Rn), S=(S1,S2,S3,...,Sn)and so on, to the cloud
users U=(U1,U2,U3,...,Un)such that the fitness of z
particular objectives F=(F1,F2,F3,..., Fz)is enhanced.
Cloud computing consists of various data centers, and
all data centers are interrelated with VMs with different
specification. Suppose there is a set of cloudlet/task Ti=
(T1,T2,T3,...,Tn)that is originated from the cloud users
as their required demands. Cloud broker is responsible for
mapping the cloudlet/task to requisite virtual resources Vj=
(V1,V2,V3,...,Vm)as virtual resources with minimum
completion time and cost. The Expected Time to Comple-
tion (ETC), described as the expected execution time of all
cloudlets/tasks on a definite virtual resource acquired by
using ETC matrix, is presented in Eq. (2). Total number of
cloudlets/tasks multiplied by the total number of resources
gives the ETC matrix’ s dimension, and their elements are
characterized as an ETC(Ti,Vj).
ETC(Ti,Vj)=
⎡
⎢
⎢
⎣
T1V1T1V2T1V3... ... T1Vm
T2V1T2V2T2V3... ... T2Vm
T3V1T3V2T3V3... ... T3Vm
TnV1TnV2TnV3... ... TnVm
⎤
⎥
⎥
⎦
(2)
Suppose there is a set of cloudlet/task cost Ci=
(C1,C2,C3,...,Cn)that is calculated from the cloud users
as their required demands. Cloud broker is responsible
for mapping the cloudlet/task to requisite virtual resources
Vj=(V1,V2,V3,...,Vm)as virtual resources with min-
imum completion time and cost. The Expected Cost to
Completion (ECC), described as the expected cost of all
Table 1 Options for multi-objective resource scheduling problem
Options Cost ($) Time (min)
→A 10 570
B 9 700
→C 8 600
D 7 850
→E 6 820
cloudlets/tasks, is calculated on a definite virtual resource
acquired by using ECC matrix and is presented in Eq. (3).
Total number of cloudlets/tasks cost multiplied by the total
number of resources gives the ECC matrix’ s dimension, and
their elements are characterized as an ECC(Ci,Vj).
ECC(Ci,Vj)=
⎡
⎢
⎢
⎣
C1V1C1V2C1V3... ... C1Vm
C2V1C2V2C2V3... ... C2Vm
C3V1C3V2C3V3... ... C3Vm
CnV1CnV2CnV3... ... CnVm
⎤
⎥
⎥
⎦
(3)
Most of the time, IaaS cloud resources are often under-
utilized due to the poor scheduling and make a reason of
SLA violation in cloud computing environment, whereas the
foremost influence of cloud computing is to optimally utilize
the all available cloud resources. The problem of determin-
ing the right settlement between utilization of resources with
minimum cost and makespan time of the superiority con-
straints is the multi-objective optimization problem.
In multi-objective resource scheduling problem, let us take
a simple example to illustrate the complexity of the situa-
tion with the help of Table 1. Suppose cloud user wants the
demands of the IaaS cloud resources, which is having the
cheapest cost and least execution time, which is impossible
for the cloud provider to fulfill exactly. Table 1has options
for the cloud providers that are the best suitable combinations
for the cloud user demands.
Table 1shows the decision for the least time and cost at
the same time. It means that objective functions minimize the
time and cost together. In reality, it is very difficult to attain
both objectives at the optimal level because the cost as well
as the time is low if the data are very clear. From options A
and B, A has less cost, but time is high and B has bit high cost
but less time than A. That is why, it is unable to select the
option from A and B. From options B and C, C has less time
and cost than B, so B is dominated by C. Again, the problem
is about the selection of options A and C because in one case
the time is high, and another the cost is high, so both are non-
dominated. From options D and E, option E is better because
E has less time and cost, so D is dominated by E. The problem
is which option is best between A, C and E because all three
are equally acceptable. In Fig. 2, options A, C and E lie in a
curve line. If there are more data, then all acceptable options
123

Arabian Journal for Science and Engineering
Fig. 2 Non-dominated curved achieved by Table 1for multi-objective
resource scheduling problem
lie on that curve. It means that the options are not invariant
to each other, all lie in the curve but the options which are
dominated by some other options lay in feasible space and do
not lie in the curve. In multi-objective scheduling problem,
the main goal is to find the non-dominated curve or solutions.
Therefore, the core objective of this research article is to
propose MOCSO algorithm for mapping the cloudlets/tasks
on virtual resources with minimum ETC and ECC matrices
in order to minimize makespan time and cost for enhancing
the utilization of resources.
3.1 Mathematical Models for the Multi-objective
Resource Scheduling Problem
This research article considers the makespan time, cost
and utilization as the multi-objective functions for optimal
resource scheduling in IaaS cloud computing environment.
Therefore, the fitness values of a MOCSO algorithm can
be calculated for makespan, cost and utilization by using
Eqs. (4), (5) and (6), respectively.
3.1.1 Makespan Model
f(x)=max
m

i=1
Ti,∀i∈N,i=1,2,3,...,m(4)
where Tiindicates the completion time of specific
cloudlet/task.
3.1.2 Cost Model
f(y)=
m

i=1
resourcei(Ci∗Ti), ∀i∈N,i=1,2,3,...,m
(5)
where Cidenotes the cost of resource ith per unit time and
Tisignifies the time of utilization of resourcei.
3.1.3 Utilization Model
f(z)=m
i=1resourcei(Ti)
max m
i=1Ti
,∀i∈N,i=1,2,3,...,m
(6)
where Tiindicates the completion time of specific
cloudlet/task.
The lowest makespan and cost while maximizing utiliza-
tion of resources show the better efficiency of the proposed
MOCSO algorithm. In this research article, the main objec-
tive is to reduce the ETC and ECC matrices of specific
cloudlets/tasks on all VMs during the resource scheduling,
and then we achieve the minimum makespan, cost and max-
imum utilization of resources.
4 Methodology
In this section, we extend the Cuckoo Search (CS) algo-
rithm from single optimization to resolve the multi-objective
optimization for resource scheduling problematic issue in
IaaS cloud computing environment. Briefly, the introduc-
tion of cuckoo behavior, the effectiveness of Levy flight
and detailed description of MOCSO algorithm are discussed
below.
4.1 Basic Terminologies
4.1.1 Cuckoo
Cuckoo is considered to be a fascinating bird, not only for
its beautiful voice but also for its aggressive reproduction
strategy. It lays and dumps the eggs in a random nest of other
species. The egg is either hatched to the next generation or
abandoned by the host bird.
123

Arabian Journal for Science and Engineering
Fig. 3 Solutions in the search space by 3-option moves in cloud computing
4.1.2 Cuckoo Behavior
The strange offspring generation method of CS algorithm
efficiently enhances its problem solving aptitude. Some of
the cuckoo species, “Ani” and “Guira” lay their eggs in some
other communal nest and throw the host eggs out of the nest to
increase the possibility of their own eggs. Similarly, “Tapera”
female cuckoo is very expert in mimicking the color and
shape of the eggs of the selected host bird. This behavior
decreases the possibility of their eggs being excluded and
increases their hatching reproduction.
4.1.3 The Egg
It is assumed that a cuckoo lays one egg in a nest so that each
egg in a nest represents a solution. A new solution is also
represented by an egg that reserved virtual resources or VMs
for mapping task in cloud environment.
4.1.4 The Nest
There is fixed quantity of nest in CS algorithm, which repre-
sents the size of the population. A nest is an individual of the
population, and its rejection is to change in the population
with a different one.
4.1.5 Objective Function
In the search space, every solution is interrelated with the
numerical value so that quality of the solution is relational to
the value of the objective function. In CS algorithm, an egg
in the nest leads to the new generation, which shows that the
quality of an egg effects on the birth of new cuckoo directly.
4.1.6 Search Space
In the case of three dimensions, the search space expresses
the locations of the prospective nest in terms of cloud virtual
resources as shown in Fig. 3. These locations are (x,y,z)∈
R×R×R. To modify the location of nest, there is need
to change the coordinates. It occurs in most continuous opti-
mization problems.
4.1.7 Movement in Search Space
The coordinates of virtual cloud resources are fixed; the
movement is based on the demanded virtual cloud resources
to fulfill the cloud users’ demands. The move from one solu-
tion to another in the search space is prepared with the help
of small steps in the small region nearby the existing solu-
tion. Consequently, various steps move toward the further
solutions.
4.1.8 The Step
The step is distance between two solutions, which depends
upon space topology and neighbors. The step length is rela-
tional to the sum of successful 3-option moves of solutions.
Big step always represents the double bridge move.
4.1.9 Levy Flights
Levy flights have the ability of thorough search close to the
solution, tracked by big step in the long run. It is mentioned
that most of the optimization problems efficiently search for
a new best solution with the help of Levy flights [48]. For
improving the solutions, the step length is related to values
generated by Levy flights.
Moving from one solution to another is defined by step
unit, which is assumed to be a 3-opt move. It is a choice
among a small step, kth steps and a big step for moving in
search space. These are managed through Levy flights and
related to the interval between 0 and 1, which helps to select
the suitable step length. The value of Levy is expressed as
follows:
123

Arabian Journal for Science and Engineering
(0,i,j)with one step 3-opt moves.
(k−1)∗i∗j,k∗i,k∗jmoves from kth steps.
(k∗i∗j,1)big step by double bridge move.
The value of i=1/(n+1),j=1/(n+2)where n is the
maximum numbers of steps and k=(3,4,...,n). Assume
that n=5, so i=0.16, j=0.14 and interval is divided into
six steps.
Step 1: Levy in (0, i, j) (0, 0.16, 0.14) by 3-opt move.
Step 2:Levyin((2−1)∗i∗j,2∗i,2∗j)(0.0224,0.32,0.28)
by 3-opt move.
Step 3:Levyin((3−1)∗i∗j,3∗i,3∗j)(0.0448,0.48,0.42)
by 3-opt move.
Step 4:Levyin((4−1)∗i∗j,4∗i,4∗j)(0.0672,0.64,0.56)
by 3-opt move.
Step 5:Levyin((5−1)∗i∗j,5∗i,5∗j)(0.0896,0.80,0.70)
by 3-opt move.
Step 6:Levyin((6−1)∗i∗j,1)(0.112,1)by double bridge
move.
4.1.10 Dominate Solution
If all objective functions use minimization, then a realistic
solution Xis supposed to be dominated by alternative real-
istic solution Yexpressed as (X≺Y),if(fi(x)≤fi(y))
for (i=1,2,...,K)and (fi(x)< fi(y)) for at least one
objective function.
4.1.11 Pareto Optimal Solution
If a solution is not dominated by another solution in search
space, then it is said to be Pareto optimal. It cannot be
enhanced without making at least one of the other objectives
worse.
4.2 Multi-objective Cuckoo Search Optimization
(MOCSO) Algorithm
The original CS algorithm deals with single optimization
function and uses the three idealize rules:
– Each cuckoo lays one egg at a time and dumps it in a
randomly chosen nest.
egg =nest +solution
– The best nest with the highest quality of eggs (solution)
will carry over to the next generation.
best solution =min f(x)or max f(x)
where f(x)is an objective function
– The number of accessible host nest is static, and a host
can realize an unfamiliar egg with probability (0 or 1). In
this case, the host can throw the egg or abandon the nest
to build a new nest in a completely new location.
For multi-objective optimization with kth various objec-
tives, we adjust only first and third rules into integrate the
requirements of multi-objective:
– Each cuckoo lays keggs at a time and dumps these eggs
in a randomly chosen nest. Egg i,i∈(1,2,3,...,k),
corresponds to the solution for kth objective.
– The best nest with the highest quality of eggs (solution)
will carry over to the next generation.
best solution =min fi(x)or max fi(x)
where fi(x)is representing one objective function
– Each nest can be abandoned with probability Pαand new
nest with kth egg is built, depending on resemblance or
dissimilarity of eggs. Some random mixing is applied for
generating diversity.
Mathematically, the first rule is transformed into a ran-
domization procedure, so that a new solution is arbitrarily
originated with the help of either a random walk or Levy
flight. Besides this, a local random permutation occurs over
solutions. For each nest, there are the kth solutions by Eq. (7).
In theory, the second rule is still the same so that the best solu-
tions are moved toward the next generation by checking the
quality, which ensures the algorithm converges appropriately.
Finally, the third rule is considered for the transformation pro-
cess; in this way, the poorest solutions are rejected with the
help possibility and new solutions are created with respect
to the resemblance of the solutions to the supplementary
solutions. It indicates that the transformation is a vectored
function performed through the difference of quality solu-
tion with the help of Levy flight. These unique functions
guarantee the efficiency of the MOCSO algorithm.
Three parameters are used in MOCSO algorithm
–Pα∈[0,1]The probability of worse nest to be aban-
doned.
–α>0 step size, which should be associated with the
scale of the problem of attention. In most cases, α>1.
–λrandom step length.
Levy flight is a random walk in which the steps are defined
in terms of the step length, which are distributed according
to a heavy-tailed probability distribution with the direction
of steps being isotropic and random.
xt+1
i=xt
i+α⊕Levy(λ) (7)
where
123

Arabian Journal for Science and Engineering
xt
i—current solution
x(
it+1)—new solution
α⊕Levy(λ)—transaction probability
For easiness, the most recent supposition can be estimated
by a fraction Pαof the n nests being substituted by new nests
with a new random solution at the new location. For the objec-
tives’ minimization or maximization, the fitness and quality
of the solutions are simply proportional to every replace-
ment and non-dominated solutions are found, respectively.
Furthermore, we enhanced the original algorithm to deal with
multi-objective resource scheduling in IaaS cloud computing
as follows:
– A Pareto function is applied for keeping all non-
dominated solutions sets encountered during the search.
– The worst nests are found through the Pareto dominance.
In the single optimization problems, it is easy to rank the
nests depending upon the quality of the solution, where one
egg exists in a nest in terms of the value of the objective
function. However, in multi-objective problems that consid-
ered more than one egg in each nest, the nests’ ranking is not
an insignificant process. Contrasting the solutions is based
on the values of objective functions, which is unreliable and
delivered the incorrect ranking. It will occur when objec-
tives are conflicted with each other. For this purpose, nests
are distributed into two groups with the help of Pareto domi-
nance. The first group is focused on non-dominated nests with
Pareto optimal solution, and the second group is focused on
dominated nests with dominated solutions that are not Pareto
optimal. All the nests in the first group are non-dominated
measured being the best, while all the nests in the other
group are measured being the worst. Both groups are cre-
ated by examining and comparing the nests to one another.
A fraction of nests with probability Pαfrom all denominated
groups is rejected as the worst nests, and the new nests are
generated.
Based on two modified above rules and their extensions,
the basic steps of MOCSO algorithm are exactly as shown
in the pseudo-code displayed in Fig. 4.InFig.5, an under-
standing of the flow of MOCSO algorithm is presented.
In this research article, k=2 is used for considering
the two objectives to resolve the multi-objective resource
scheduling problem for IaaS in cloud computing environ-
ment.
There are two goals of multi-objective optimization as
detailed below.
– Obtain the solution that is near to the true Pareto front.
– Generate the solutions as possible in the non-dominated
front.
5 Performance Metrics
Resource scheduling parameters are used for computing
results and significance analysis of the scheduling algo-
rithms in IaaS cloud computing environment. These param-
eters are based on computing, network, node and storage
including the availability, bandwidth/speed, cost, degree
of imbalance, energy, execution time, makespan, memory,
performance, priority, reliability, response time, SLA, tem-
perature, throughput, time and utilization [7,8,49]. This study
considers the makespan, cost, utilization and performance
improvement rate for multi-objective resource scheduling in
IaaS cloud computing environment. These performance met-
rics are discussed below.
5.1 Makespan
Makespan is used to determine the maximum completion
time for execution, by calculating the finishing time of all
tasks, when they are all scheduled. If the makespan of specific
cloudlet or task is more than the demands of cloud users, these
demands will not be fulfilled on time [50,51]. This research
article reduces the makespan of all task mapping on VMs as
defined in Eq. (4).
5.2 Cost
Cost indicates the overall amount produced against the usage
or utilization of resources, which is paid by the cloud users
to the cloud providers. The real intention is to reduce the
expenses for cloud user while enhancing the growth of profit
and revenue for cloud providers with competent utilization
[8,52]. Assume that the cost of a VM fluctuates from alter-
native based on substantial time and VMs’ description as
identified by the cloud providers, so that Eq.(5)isusedfor
the calculating cost of completing the task of a specific VM.
5.3 Utilization
Utilization is the whole quantity of resources that are actually
consumed in the data centers. The main goal is to utilize
the resources efficiently by competent mapping [53,54]as
defined in Eq. (6). It helps to enhance the cloud providers’
profit and revenue while fulfilling the cloud users’ demands
successfully.
5.4 Performance Improvement Rate
Performance improvement rate (PIR) is used to estimate the
percentage of performance improvement for the ith algo-
rithm with regard to the kth algorithm as presented in Eq. (8).
It is always represented in percentage (%) [55].
123

Arabian Journal for Science and Engineering
Fig. 4 Flowchart of Multi-objective Cuckoo Search Optimization (MOCSO) algorithm
PIR(%)=kth PM −ith PM
ith PM ∗100 (8)
6 Simulation Setup
The simulation setup is described in this section for the com-
putational results and significance analysis obtained after
the implementation of MOCSO algorithm and other existing
meta-heuristic algorithms for multi-objective optimization
of resource scheduling in IaaS cloud computing environ-
ment. All algorithms are implemented in CloudSim simulator
[56,57] by using three workload traces as S01, S02 and S03,
respectively.
S01, S02 and S03 are generated from the “Parallel Work-
load Archives” that consists of HPC2N (High-Performance
123

Arabian Journal for Science and Engineering
Fig. 5 Pseudo-code of
Multi-objective Cuckoo Search
Optimization (MOCSO)
algorithm
Computing Center North) [58], NASA Ames iPCS/860
[59] and SDSC (San Diego Supercomputer Center) [60].
These workload archives are offered by “Ake Sandgren, Bill
Nitzberg and Victor Hazlewood”, in the standard workload
format (swf) recognized by the CloudSim tool. HPC2N com-
prises the statistics of 527,371 tasks, NASA comprises the
statistics of 14,794 tasks, and SDSC comprises the statistics
of 73,496 tasks. In cloud computing environment, mostly
these workloads are applied for evaluating the performance
of algorithms [30,61–65].
The specification of users, cloudlets, hosts, VMs and data
centers are shown in Table 3for the computation results
and significance analysis based on [55,62,66,67]. The larger
cloudlets/tasks will improve the observation in scalability
of the performance of the algorithms with the large prob-
lem sizes and little user demand. The parameter values of
MOACO based on [66,68], those of MOGA based on [32,66],
those of MOPSO based on [69,70], and those of MOCSO
based on [41,71] are shown in Table 2. MOACO, MOGA,
MOMM and MOPSO algorithms are used for the compari-
son with MOCSO algorithm, on a set of parameters including
makespan, cost, utilization and performance improvement
rate for resource scheduling in IaaS cloud computing envi-
ronment. In the statistical analysis, mean, standard deviation
and the best value are mentioned after 25 runs of the sim-
Table 2 Parameters setting of meta-heuristic algorithms in cloud com-
puting environment for resource scheduling
Algorithms Parameters Values
MOACO Number of ants 10
Vaporization factor 0.4
Pheromone tracking weight, α0.3
Heuristic information weight, β1
Pheromone updating constant, Q100
MOCSO Population size 20
Abandon probability Pα0.25
Max iteration 1000
Step size λ0.01, 1
MOGA Population size 1000
Max iteration 1000
Crossover rate 0.5
Mutation rate 0.1
MOPSO Particle size 100
Self-recognition coefficients, c1,c22
Uniform random number, R10,1
Max iteration 1000
Inertia weight, W0.9–0.4
123

Arabian Journal for Science and Engineering
Table 3 Simulation parameters setting of CloudSim in cloud comput-
ing environment for multi-objective resource scheduling
S. no Entities Parameters Values
1 User Number of users 100
Number of brokers 10
2 Cloudlet Number of cloudlets 200–2000
Length 800,000
File size 600
3 Host RAM 2048 MB
Storage 1,000,000
Bandwidth 10,000
4 VM Number of VMs 50
Type of policy Time shared
RAM 512 MB
Bandwidth 10,000
MIPS 1000
Size 10,000
VMM Xen
Operating system Linux
Number of CPUs 1 on each
5 Data center Number of data centers 2
ulation. A comparison of performance metrics is shown in
Tables 4,5and 6. The comparison of MOCSO algorithm’
s PIR% over existing algorithms is presented in Tables 7,8
and 9. To calculate the cost for resource scheduling based
on VMs’ descriptions are considered. Hence, the cost of
resources is as follows: $0.52, $0.48, $0.17, $0.13 and $0.12
per hour [26,31,72].
7 Results and Discussion
This section identifies and discusses the computational
outcomes of the simulation formulated with statistical sig-
nificance analysis, to assess the performance of the proposed
MOCSO algorithm for multi-objective resource scheduling
in IaaS cloud computing environment. The makespan time,
cost and utilization for multi-objective resource scheduling
are illustrated in Figs. 6,7and 8.
Figure 6a–c shows the makespan time of MOCSO algo-
rithm as computed with the help of HPC2N, NASA and
SDCS workloads. Thus, simulation has been performed
with various numbers of cloudlets/tasks from the range
of 200–2000. The four cloud computing resource schedul-
ing algorithms (MOACO, MOGA, MOMM and MOPSO)
are used for the comparison with proposed MOCSO algo-
rithm. Figure 6a–c identifies that makespan time of resource
scheduling algorithms is increased with increasing the num-
ber of cloudlets/tasks. The unit of makespan time is seconds.
The makespan time as calculated by the MOCSO algorithm
is lower than the other four algorithms, especially as the
cloudlet/tasks increases. The consequence of this outcome
is that the proposed MOCSO algorithm would support the
cloud users to save more money while using the cloud com-
puting environment.
Table 4 Statistical significance of multi-objective algorithms for makespan after 25 runs
Algorithms Workloads S01 S02 S03
Statistical dispersion/
number of cloudlets
¯
XσBest ¯
XσBest ¯
XσBest
MOACO 200 2175.31 9.15 2150.12 117.74 14.25 93.64 182.82 22.02 150.58
1200 4557.74 13.96 4502.44 740.585 29.07 701.26 1626.01 40.41 1559.25
2000 5846.90 21.23 5781.03 1349.20 32.72 1307.80 2874.93 63.30 2753.00
MOGA 200 2297.73 11.27 2260.32 115.40 11.50 98.53 182.58 13.69 154.59
1200 4554.12 15.16 4505.80 733.25 22.48 691.45 1672.09 24.90 1592.10
2000 5865.64 24.90 5759.90 1292.34 31.35 1239.34 2880.69 37.33 2760.80
MOMM 200 2491.25 0.00 2491.25 136.24 0.00 136.24 249.25 0.00 249.25
1200 4766.25 0.00 4766.25 816.75 0.00 816.75 1667.25 0.00 1667.25
2000 6324.02 0.00 6324.02 1704.00 0.00 1704.00 3304.02 0.00 3304.02
MOPSO 200 2052.88 12.34 2015.50 120.08 20.17 89.39 183.05 19.50 156.52
1200 4561.36 17.56 4508.14 747.92 26.62 720.92 1579.92 31.41 1479.03
2000 5828.15 22.02 5761.02 1406.06 31.52 1357.46 2869.16 40.84 2780.24
MOCSO 200 2056.80 6.84 2041.06 63.40 2.42 58.48 168.20 6.16 149.54
1200 3957.93 11.27 3920.37 657.15 9.40 642.84 1314.30 9.79 1289.42
2000 5687.50 14.51 5641.20 1069.00 17.15 1013.9 2694.18 18.39 2635.09
S01: HPC2N, S02: NASA, S03: SDSC
123

Arabian Journal for Science and Engineering
Table 5 Statistical significance of multi-objective algorithms for cost after 25 runs
Algorithms Workloads S01 S02 S03
Statistical dispersion/
number of cloudlets
¯
XσBest ¯
XσBest ¯
XσBest
MOACO 200 99.66 2.02 95.30 1.55 0.24 1.13 0.97 0.53 0.51
1200 194.66 3.40 185.31 66.47 9.73 54.65 21.59 0.22 18.22
2000 271.01 7.80 249.65 132.88 14.93 101.70 59.38 0.79 56.81
MOGA 200 42.76 0.98 40.47 0.55 0.30 0.29 0.34 0.19 0.22
1200 76.53 1.76 70.66 23.59 5.47 18.97 7.64 0.26 7.32
2000 118.90 4.08 107.03 47.09 8.91 36.39 18.90 1.39 14.43
MOMM 200 161.04 0.00 161.04 2.16 0.00 2.16 1.12 0.00 1.11
1200 267.45 0.00 267.45 75.10 0.00 75.10 28.08 0.00 28.08
2000 433.92 0.00 433.92 163.59 0.00 163.59 64.48 0.00 64.48
MOPSO 200 28.53 0.94 27.44 0.42 0.26 0.32 0.26 0.20 0.16
1200 61.25 1.95 55.19 17.96 4.28 12.89 5.85 0.25 5.59
2000 95.71 3.28 85.15 35.97 7.61 29.06 16.04 0.99 13.23
MOCSO 200 22.51 0.38 21.43 0.31 0.14 0.18 0.24 0.11 0.17
1200 40.47 1.02 27.61 10.02 1.10 9.74 3.26 0.17 3.02
2000 62.55 2.60 54.68 26.97 3.10 23.60 7.16 0.59 7.03
S01: HPC2N, S02: NASA, S03: SDSC
Table 6 Statistical significance of multi-objective algorithms for utilization after 25 runs
Algorithms Workloads S01 S02 S03
Statistical dispersion/
number of cloudlets
¯
XσBest ¯
XσBest ¯
XσBest
MOACO 200 10.48 1.03 13.60 58.29 8.39 62.53 56.69 9.83 63.63
1200 11.24 1.11 13.83 16.47 4.75 19.84 11.74 2.95 13.84
2000 10.36 1.20 12.95 10.91 2.43 13.65 10.39 1.54 12.56
MOGA 200 10.21 0.87 11.86 57.13 7.96 62.76 56.53 8.86 61.65
1200 11.31 0.80 12.80 16.32 3.52 19.54 12.05 3.52 14.54
2000 10.36 1.02 13.58 11.63 1.99 14.31 10.40 1.74 12.90
MOMM 200 8.04 0.00 8.04 63.37 0.00 63.37 60.06 0.00 60.06
1200 9.27 0.00 9.27 15.13 0.00 15.13 11.54 0.00 11.54
2000 7.58 0.00 7.58 13.74 0.00 13.74 10.50 0.00 10.50
MOPSO 200 10.79 0.90 11.84 59.45 9.65 67.53 56.85 8.55 65.53
1200 11.17 0.78 13.79 16.61 4.22 21.04 11.43 4.08 14.33
2000 10.35 0.93 12.44 11.39 2.10 14.54 10.37 1.84 12.54
MOCSO 200 10.96 0.53 12.58 32.35 5.30 38.94 65.45 6.02 75.10
1200 11.74 0.56 13.88 24.18 2.47 28.67 15.74 1.28 18.67
2000 10.91 0.59 13.80 17.92 1.63 19.53 11.99 0.74 12.45
S01: HPC2N, S02: NASA, S03: SDSC
Table 7 MOCSO performance
improvement rate (%) on
Makespan for resource
scheduling
MOACO MOGA MOMM MOPSO MOCSO
Total Makespan 61802.97 62616.03 69179.40 60989.91 55933.44
PIR over MOACO – – 1.32 – 11.94 1.32 9.50
PIR over MOGA – – – 10.48 2.60 10.67
PIR over MOMM – – – 11.84 19.15
PIR over MOPSO – – – – 8.29
123

Arabian Journal for Science and Engineering
Table 8 MOCSO performance
improvement rate (%) on cost
for resource scheduling
MOACO MOGA MOMM MOPSO MOCSO
Total cost 2660.46 1043.64 3840.55 795.14 546.89
PIR over MOACO – – 60.77 – 44.36 70.11 79.44
PIR over MOGA – – – 68.00 23.81 47.60
PIR over MOMM – – – 79.30 85.76
PIR over MOPSO – – – – 31.22
Table 9 MOCSO performance
improvement rate (%) on
utilization for resource
scheduling
MOACO MOGA MOMM MOPSO MOCSO
Total utilization 536.13 539.04 524.62 534.48 599.50
PIR over MOACO – 0.54 – 2.15 – 0 .31 11.82
PIR over MOGA – – – 2.68 – 0.85 14.22
PIR over MOMM – – – 1.88 14.27
PIR over MOPSO – – – – 12.17
Similarly, Fig. 7a–c presents the cost of MOCSO algo-
rithm as computed with the help of HPC2N, NASA and
SDCS workloads. Thus, simulation has been performed
with various numbers of cloudlets/tasks from the range
of 200–2000. The proposed MOCSO algorithm is com-
pared with the four cloud computing resource scheduling
algorithms (MOACO, MOGA, MOMM and MOPSO). Fig-
ure 7a–c shows that cost of resource scheduling algorithms is
increased with increasing the number of cloudlets/tasks. The
unit of cost is $/h. The cost as calculated by the MOCSO
algorithm is lower than the other four algorithms, specif-
ically as the cloudlet/tasks increases. The consequence of
this outcome is that the proposed MOCSO algorithm would
support the cloud users to reduce the expenses and cloud
providers to increase the revenue and profit while using the
cloud computing environment.
In the same way, Fig.8a–c presents the utilization of
resources for MOCSO algorithm as computed with the help
of HPC2N, NASA and SDCS workloads. Thus, simulation
has been performed with various numbers of cloudlets/tasks
from the range of 200–2000. The four cloud computing
resource scheduling algorithms (MOACO, MOGA, MOMM
and MOPSO) are used for the comparison with MOCSO
algorithm. Figure 8a–c shows that resource utilization for
resource scheduling algorithms is decreased with increasing
the number of cloudlets/tasks. The utilization of resources
as calculated by the MOCSO algorithm is higher than by
the other four algorithms, especially as the cloudlets/tasks
increase. The consequence of this outcome is that the pro-
posed MOCSO algorithm would support the cloud providers
to increase the revenue and profit while using the cloud com-
puting environment.
The statistical analysis of the makespan, cost and resource
utilization data achieved after 25 runs is shown in Tables 4,5
and 6, respectively. The significance test of the data generated
is to check the robustness of the proposed MOCSO algorithm.
The mean, standard deviation and best values of MOCSO
algorithm and all four comparison algorithms are computed
with help of all three workloads. This significance analysis
shows that the results obtained from MOCSO algorithm fol-
low the normal distribution and robustness as compared to
other algorithms.
Performance improvement rate (%) based on makespan,
cost and utilization of the MOCSO algorithm as it relates to
the MOACO, MOGA, MOMM and MOPSO algorithms are
presented in Tables 7,8and 9, respectively. For the makespan,
MOCSO algorithm shows 9.5%, 10.67%, 19.15% and 8.29%
makespan time improvements on the MOACO, MOGA,
MOMM and MOPSO algorithms. In case of cost, MOCSO
algorithm produces 79.44%, 47.60%, 85.76% and 31.22%
cost improvements on the MOACO, MOGA, MOMM and
MOPSO algorithms. In the same way, MOCSO algorithm
generates 11.82%, 11.22%, 14.27% and 12.17% of uti-
lization improvements on the MOACO, MOGA, MOMM
and MOPSO algorithms. This identifies that the proposed
MOCSO algorithm achieves the better performance in terms
of minimizing the makespan and cost while maximizing the
utilization for resource scheduling in IaaS cloud computing
environment.
Pareto efficiency is not always equal to the fairness and
equity. A resource scheduling is inefficient if there does not
exist another solution in which at least one objective is bet-
ter off, and none is worse off. Figure9a–c shows the Pareto
front lines are obtained by MOSCO algorithm using HPC2N,
NASA and SDCS workload for multi-objective resource
scheduling for IaaS in cloud computing environment. These
Pareto front lines lie on the non-dominated curves at the 1000,
1200 and 1400 cloudlets/tasks according to Fig. 2.Itisclearly
shown that MOCSO algorithm tries to achieve the best solu-
123

Arabian Journal for Science and Engineering
(a) HPC2N
0
1000
2000
3000
4000
5000
6000
7000
200 400 600 800 1000 1200 1400 1600 1800 2000
Makespan Time
Number of Cloudlets
MOACO
MOCSO
MOGA
MOMM
MOPSO
(b) NASA
(c) SDCS
0
200
400
600
800
1000
1200
1400
1600
1800
200 400 600 800 1000 1200 1400 1600 1800 2000
Makespan Time
Number of Cloudlets
MOACO
MOCSO
MOGA
MOMM
MOPSO
0
500
1000
1500
2000
2500
3000
3500
200 400 600 800 1000 1200 1400 1600 1800 2000
Makespan Time
Number of Cloudlets
MOACO
MOCSO
MOGA
MOMM
MOPSO
Fig. 6 Makespan time for resource scheduling using aHPC2N, b
NASA and cSDCS in IaaS cloud computing environment
tions for both ETC and ECC matrices without making the
better utilization of the resources worse for any one.
The results of the simulation and significance analysis
show that the MOCSO algorithm returns the multi-objective
resource scheduling as measured in terms of makespan time,
cost, utilization and performance improvement rate parame-
ters. The computational results and statistical analysis show
that our proposed MOCSO algorithm provides better-quality
(a) HPC2N
0
50
100
150
200
250
300
350
400
450
200 400 600 800 1000 1200 1400 1600 1800 2000
Cost
Number of Cloudlets
MOACO
MOCSO
MOGA
MOMM
MOPSO
(b) NASA
(c) SDCS
0
10
20
30
40
50
60
70
200 400 600 800 1000 1200 1400 1600 1800 2000
Cost
Number of Cloudlets
MOACO
MOCSO
MOGA
MOMM
MOPSO
0
20
40
60
80
100
120
140
160
180
200 400 600 800 1000 1200 1400 1600 1800 2000
Cost
Number of Cloudlets
MOACO
MOCSO
MOGA
MOMM
MOPSO
Fig. 7 Cost for resource scheduling using aHPC2N, bNASA and c
SDCS in IaaS cloud computing environment
results than MOACO, MOGA, MOMM and MOPSO algo-
rithms. It further clarifies that MOCSO algorithm is very
appropriate for the ETC and ECC matrixes in IaaS cloud
computing environment. From the performance evaluation,
it can be suggested that the MOCSO algorithm is potentially
a powerful search and optimization technique sufficient for
resolving the multi-objective issue of resource scheduling in
IaaS cloud computing environment.
123

Arabian Journal for Science and Engineering
(a) HPC2N
0
2
4
6
8
10
12
200 400 600 800 1000 1200 1400 1600 1800 2000
Utilization
Number of Cloudlets
MOACO
MOCSO
MOGA
MOMM
MOPSO
(c) NASA
(c) SDCS
0
10
20
30
40
50
60
70
200 400 600 800 1000 1200 1400 1600 1800 2000
Utilization
Number of Cloudlets
MOACO
MOCSO
MOGA
MOMM
MOPSO
0
10
20
30
40
50
60
70
200 400 600 800 1000 1200 1400 1600 1800 2000
Utilization
Number of Cloudlets
MOACO
MOCSO
MOGA
MOMM
MOPSO
Fig. 8 Utilization of resource for resource scheduling using aHPC2N,
bNASA and cSDCS in IaaS cloud computing environment
8 Conclusion and Recommendations
Resource scheduling in IaaS cloud computing environ-
ment is considered to be an NP-hard problematic issue
that continues to attract attention from the scientific world.
The motivation of this research article is to discuss the
multi-objective resource scheduling problem by providing
maximum utilization of resources in IaaS cloud comput-
ing environment. Therefore, Multi-objective Cuckoo Search
Optimization (MOCSO) algorithm is suitable for cloud
Fig. 9 Pareto front is obtained by MOCSO algorithm using aHPC2N,
bNASA and cSDCS in IaaS cloud computing environment
computing environment due to its effective utilization of
resources through the minimization of cost and makespan
time. The computational results and significance analysis
illustrated that MOCSO algorithm outperformed MOACO,
MOGA, MOMM and MOPSO algorithms. This indicates
that the MOCSO algorithm is more precise and suitable for
multi-objective resource scheduling in IaaS cloud comput-
ing. The proposed algorithm can be used for solving other
optimization issues in cloud computing environment and
other discrete optimization issues in different domain. In the
123

Arabian Journal for Science and Engineering
future, hybridization of CS algorithm with other optimized
heuristic and meta-heuristic algorithms may also prove to be
fruitful for cloud computing environment.
References
1. Lu, G.; Zeng, W.H.: Cloud computing survey. Appl. Mech. Mater.
530, 650–661 (2014)
2. Foster, I.; Zhao, Y.; Raicu, I.; Lu, S.: Cloud computing and grid
computing 360-degree compared. In: Grid Computing Environ-
ments Workshop, 2008. GCE’08 pp. 1–10. IEEE (2008)
3. Rimal, B.P.; Choi, E.; Lumb, I.: A taxonomy, survey, and issues
of cloud computing ecosystems. In: Antonopoulos, N., Gillam,
L. (eds.) Cloud Computing. Computer Communications and Net-
works, pp. 21–46. Springer, London (2010)
4. Mell, P.; Grance, T.: The NIST definition of cloud computing. Natl.
Inst. Stand. Technol. 53(6), 50 (2009)
5. Manvi, S.S.; Krishna Shyam, G.: Resource management for Infras-
tructure as a Service (IaaS) in cloud computing: a survey. J. Netw.
Comput. Appl. 41, 424–440 (2014)
6. Jennings, B.; Stadler, R.: Resource management in clouds: survey
and research challenges. J. Netw. Syst. Manag. 23(3), 567–619
(2015)
7. Madni, S.H.H.; Latiff, M.S.A.; Coulibaly, Y.: Resource scheduling
for infrastructure as a service (IaaS) in cloud computing: challenges
and opportunities. J. Netw. Comput. Appl. 68, 173–200 (2016)
8. Madni, S.H.H.; Latiff, M.S.A.; Coulibaly, Y.; Abdulhamid, S.I.M.:
Recent advancements in resource allocation techniques for cloud
computing environment: a systematic review. Clust. Comput.
20(3), 2489–2533 (2017)
9. Patel, Y.S.; Mehrotra, N.; Soner, S.: Green cloud computing: a
review on Green IT areas for cloud computing environment. In:
2015 International Conference on Futuristic Trends on Compu-
tational Analysis and Knowledge Management (ABLAZE), pp.
327–332. IEEE (2015)
10. Yang, Z.; Hoseinzadeh, M.; Andrews, A.; Mayers, C.; Evans, D.T.;
Bolt, R.T.; Bhimani, J.; Mi, N.; Swanson, S.: AutoTiering: auto-
matic data placement manager in multi-tier all-flash datacenter. In:
2017 IEEE 36th International Performance Computing and Com-
munications Conference (IPCCC), pp. 1–8. IEEE (2017)
11. Yang, Z.; Awasthi, M.; Ghosh, M.; Mi, N.: A fresh perspective on
total cost of ownership models for flash storage in datacenters. In:
2016 IEEE International Conference on Cloud Computing Tech-
nology and Science (CloudCom), pp. 245–252. IEEE (2016)
12. Yang, Z.; Jia, D.; Ioannidis, S.; Mi, N.; Sheng, B.: Intermediate
Data Caching Optimization for Multi-Stage and Parallel Big Data
Frameworks. arXiv:1804.10563 (2018)
13. Bonomi, F.; Milito, R.; Zhu, J.; Addepalli, S.: Fog computing and
its role in the internet of things. In: Proceedings of the First Edition
of the MCC Workshop on Mobile cloud computing, pp. 13–16.
ACM (2012)
14. Gubbi, J.; Buyya, R.; Marusic, S.; Palaniswami, M.: Internet of
things (IoT): a vision, architectural elements, and future directions.
Future Gener. Comput. Syst. 29(7), 1645–1660 (2013)
15. Yang, X.-S.; Deb, S.: Cuckoo search via Lvy flights. In: World
Congress on Nature and Biologically Inspired Computing, 2009.
NaBIC 2009, pp. 210–214. IEEE (2009)
16. Burnwal, S.; Deb, S.: Scheduling optimization of flexible manu-
facturing system using cuckoo search-based approach. Int. J. Adv.
Manuf. Technol. 64(5–8), 951–959 (2013)
17. Gunavathi, C.; Premalatha, K.: Cuckoo search optimisation for fea-
ture selection in cancer classification: a new approach. Int. J. Data
Min. Bioinform. 13(3), 248–265 (2015)
18. Majumder, A.; Laha, D.: A new cuckoo search algorithm for 2-
machine robotic cell scheduling problem with sequence-dependent
setup times. Swarm Evolut. Comput. 28, 131–143 (2016)
19. Wang, H.; Wang, W.; Sun, H.; Cui, Z.; Rahnamayan, S.; Zeng, S.: A
new cuckoo search algorithm with hybrid strategies for flow shop
scheduling problems. Soft Comput. 21(15), 4297–4307 (2017)
20. Zendaoui, Z.; Layeb, A.: Adaptive cuckoo search algorithm for
the bin packing problem. In: Chikhi, S., Amine, A., Chaoui, A.,
Kholladi, M., Saidouni, D. (eds.) Modelling and Implementation
of Complex Systems. Lecture Notes in Networks and Systems, vol.
1, pp. 107–120. Springer, Cham
21. Chen, S.; Wu, J.; Lu, Z.: A cloud computing resource scheduling
policy based on genetic algorithm with multiple fitness. In: 2012
IEEE 12th International Conference on Computer and Information
Technology (CIT), pp. 177–184. IEEE (2012)
22. Sindhu, S.; Mukherjee, S.: A genetic algorithm based scheduler
for cloud environment. In: 2013 4th International Conference on
Computer and Communication Technology (ICCCT), pp. 23–27.
IEEE (2013)
23. Zhang, H.; Li, P.; Zhou, Z.; Yu, X.: A PSO-based hierarchical
resource scheduling strategy on cloud computing. In: Trustworthy
Computing and Services, pp. 325–332. Springer (2013)
24. Liu, J.; Luo, X.G.; Zhang, X.M.; Zhang, F.: Job scheduling algo-
rithm for cloud computing based on particle swarm optimization.
Adv. Mater. Res. 662, 957–960 (2013)
25. Zhang, F.; Cao, J.; Li, K.; Khan, S.U.; Hwang, K.: Multi-objective
scheduling of many tasks in cloud platforms. Future Gener. Com-
put. Syst. 37, 309–320 (2014)
26. Tsai, J.-T.; Fang, J.-C.; Chou, J.-H.: Optimized task scheduling
and resource allocation on cloud computing environment using
improved differential evolution algorithm. Comput. Oper. Res.
40(12), 3045–3055 (2013)
27. Ramezani, F.; Lu, J.; Hussain, F.: Task scheduling optimization
in cloud computing applying multi-objective particle swarm opti-
mization. In: Service-Oriented Computing, pp. 237–251. Springer
(2013)
28. Alkayal, E.S.; Jennings, N.R.; Abulkhair, M.F.: Efficient task
scheduling multi-objective particle swarm optimization in cloud
computing. In: 2016 IEEE 41st Conference on Local Computer
Networks Workshops (LCN Workshops), pp. 17–24. IEEE (2016)
29. He, H.; Xu, G.; Pang, S.; Zhao, Z.: AMTS: adaptive multi-objective
task scheduling strategy in cloud computing. China Commun.
13(4), 162–171 (2016)
30. Tchernykh, A.; Lozano, L.; Schwiegelshohn, U.; Bouvry, P.;
Pecero, J.E.; Nesmachnow, S.; Drozdov, A.Y.: Online bi-objective
scheduling for IaaS clouds ensuring quality of service. J. Grid Com-
put. 14(1), 5–22 (2016)
31. Zuo, L.; Shu, L.; Dong, S.; Zhu, C.; Hara, T.: A multi-objective
optimization scheduling method based on the ant colony algorithm
in cloud computing. Access IEEE 3, 2687–2699 (2015)
32. Chen, Z.-G.; Du, K.-J.; Zhan, Z.-H.; Zhang, J.: Deadline con-
strained cloud computing resources scheduling for cost optimiza-
tion based on dynamic objective genetic algorithm. In: 2015 IEEE
Congress on Evolutionary Computation (CEC), pp. 708–714. IEEE
(2015)
33. Shojafar, M.; Javanmardi, S.; Abolfazli, S.; Cordeschi, N.: FUGE:
a joint meta-heuristic approach to cloud job scheduling algorithm
using fuzzy theory and a genetic method. Clust. Comput. 18(2),
829–844 (2015)
34. Jena, R.: Multi objective task scheduling in cloud environment
using nested PSO framework. Procedia Comput. Sci. 57, 1219–
1227 (2015)
35. Lakra, A.V.; Yadav, D.K.: Multi-objective tasks scheduling algo-
rithm for cloud computing throughput optimization. Procedia
Comput. Sci. 48, 107–113 (2015)
123

Arabian Journal for Science and Engineering
36. Gajera, V.; Gupta, R.; Jana, P.K.: An effective multi-objective task
scheduling algorithm using min-max normalization in cloud com-
puting. In: 2016 2nd International Conference on Applied and The-
oretical Computing and Communication Technology (iCATccT),
pp. 812–816. IEEE (2016)
37. Panda, S.K.; Jana, P.K.: A multi-objective task scheduling algo-
rithm for heterogeneous multi-cloud environment. In: 2015 Inter-
national Conference on Electronic Design, Computer Networks
and Automated Verification (EDCAV), pp. 82–87. IEEE (2015)
38. Choudhary, A.; Gupta, I.; Singh, V.; Jana, P.K.:A GSA based hybrid
algorithm for bi-objective workflow scheduling in cloud comput-
ing. Future Gener. Comput. Syst. 83, 14–26 (2018)
39. Nguyen, T.T.; Vo, D.N.: Modified cuckoo search algorithm for
multiobjective short-term hydrothermal scheduling. Swarm Evo-
lut. Comput. 37, 73–89 (2017)
40. Bibiks, K.; Hu, F.; Li, J.-P.; Smith, A.: Discrete cuckoo search for
resource constrained project scheduling problem. In: 2015 IEEE
18th International Conference on Computational Science and Engi-
neering (CSE), pp. 240–245. IEEE (2015)
41. Marichelvam, M.; Prabaharan, T.; Yang, X.-S.: Improved cuckoo
search algorithm for hybrid flow shop scheduling problems to min-
imize makespan. Appl. Soft Comput. 19, 93–101 (2014)
42. Marichelvam, M.; Geetha, M.: A hybrid cuckoo search metaheuris-
tic algorithm for solving single machine total weighted tardiness
scheduling problems with sequence dependent setup times. Int. J.
Comput. Complex. Intell. Algorithms 1(1), 23–34 (2016)
43. Civicioglu, P.; Besdok, E.: A conceptual comparison of the Cuckoo-
search, particle swarm optimization, differential evolution and
artificial bee colony algorithms. Artif. Intell. Rev. 39(4), 315–346
(2013)
44. Civicioglu, P.; Besdok, E.: Comparative analysis of the cuckoo
search algorithm. In: Yang, X.S. (ed.) Cuckoo Search and Firefly
Algorithm. Studies in Computational Intelligence, vol. 516, pp.
85–113. Springer, Cham (2014)
45. Gandomi, A.H.; Yang, X.-S.; Alavi, A.H.: Cuckoo search algo-
rithm: a metaheuristic approach to solve structural optimization
problems. Eng. Comput. 29(1), 17–35 (2013)
46. Gandomi, A.H.; Yang, X.-S.; Talatahari, S.; Deb, S.: Coupled
eagle strategy and differential evolution for unconstrained and con-
strained global optimization. Comput. Math. Appl. 63(1), 191–200
(2012)
47. Marichelvam, M.; Tosun, : Performance comparison of cuckoo
search algorithm to solve the hybrid flow shop scheduling bench-
mark problems with makespan criterion. Int. J. Swarm Intell. Res.
7(2), 1–14 (2016)
48. Kamaruzaman, A.F.; Zain, A.M.; Yusuf, S.M.; Udin, A.: Levy flight
algorithm for optimization problems–a literature review. Appl.
Mech. Mater. 421, 496–501 (2013)
49. Mustafa, S.; Nazir, B.; Hayat, A.; Madani, S.A.: Resource man-
agement in cloud computing: taxonomy, prospects, and challenges.
Comput. Electr. Eng. 47, 186–203 (2015)
50. Abdulhamid, S.M.; Latiff, M.S.A.; Idris, I.: Tasks scheduling
technique using league championship algorithm for makespan min-
imization in IaaS cloud. ARPN J. Eng. Appl. Sci. 9(12), 2528–2533
(2015)
51. Madni, S.H.H.; Latiff, M.S.A.; Abdulhamid, S.M.: Optimal
resource scheduling for IaaS cloud computing using cuckoo search
algorithm. Sains Humanika 9(1–3), 71–76 (2017)
52. Li, X.; Li, Y.; Liu, T.; Qiu, J.; Wang, F.: The method and tool of cost
analysis for cloud computing. In: IEEE International Conference
on Cloud Computing, 2009. CLOUD’09, pp. 93–100. IEEE (2009)
53. Zhang, Q.; Cheng, L.; Boutaba, R.: Cloud computing: state-of-
the-art and research challenges. J. Internet Serv. Appl. 1(1), 7–18
(2010)
54. Madni, S.H.H.; Latiff, M.S.A.; Coulibaly, Y.; Abdulhamid, S.M.:
An appraisal of meta-heuristic resource allocation techniques for
IaaS cloud. Indian J. Sci. Technol. 9(4), 1–14 (2016)
55. Abdulhamid, S.I.M.; Latiff, M.S.A.; Madni, S.H.H.; Abdullahi, M.:
Fault tolerance aware scheduling technique for cloud computing
environment using dynamic clustering algorithm. Neural Comput.
Appl. 29(1), 279–293 (2018)
56. Calheiros, R.N.; Ranjan, R.; Beloglazov, A.; De Rose, C.A.; Buyya,
R.: CloudSim: a toolkit for modeling and simulation of cloud
computing environments and evaluation of resource provisioning
algorithms. Softw. Pract. Exp. 41(1), 23–50 (2011)
57. Buyya, R.; Ranjan, R.; Calheiros, R.N.: Modeling and simulation of
scalable Cloud computing environments and the CloudSim toolkit:
challenges and opportunities. In: International Conference on High
Performance Computing and Simulation, 2009. HPCS’09, pp. 1–
11. IEEE (2009)
58. HPC2N: The HPC2N Seth log. http://www.cs.huji.ac.il/labs/
parallel/workload/l_hpc2n/ (2016)
59. NASA: The NASA Ames iPCS/860 log. http://www.cs.huji.ac.il/
labs/parallel/workload/l_nasa_ipsc/ (2016)
60. SDSC: The SDSC SP2 log. http://www.cs.huji.ac.il/labs/parallel/
workload/l_sdsc_sp2/ (2016)
61. Abdullahi, M.; Ngadi, M.A.; Abdulhamid, S.M.: Symbiotic organ-
ism search optimization based task scheduling in cloud computing
environment. Future Gener. Comput. Syst. 56, 640–650 (2016)
62. Abdullahi, M.; Ngadi, M.A.: Hybrid symbiotic organisms search
optimization algorithm for scheduling of tasks on cloud computing
environment. PLoS ONE 11(6), e0158229 (2016)
63. Barquet, A.L.; Tchernykh, A.; Yahyapour, R.: Performance eval-
uation of infrastructure as service clouds with SLA constraints.
Computacin y Sistemas 17(3), 401–411 (2013)
64. Zhan, J.; Wang, L.; Li, X.; Shi, W.; Weng, C.; Zhang, W.; Zang, X.:
Cost-aware cooperative resource provisioning for heterogeneous
workloads in data centers. IEEE Trans. Comput. 62(11), 2155–
2168 (2013)
65. Mehrotra, P.; Djomehri, J.; Heistand, S.; Hood, R.; Jin, H.;
Lazanoff, A.; Saini, S.; Biswas, R.: Performance evaluation of
Amazon Elastic Compute Cloud for NASA high-performance com-
puting applications. Concurr. Comput. Pract. Exp. 28, 1041–1055
(2013)
66. Abdulhamid, S.M.; Latiff, M.S.A.; Abdul-Salaam, G.; Madni,
S.H.H.: Secure scientific applications scheduling technique for
cloud computing environment using global league championship
algorithm. PLoS ONE 11(7), e0158102 (2016)
67. Abdullahi, M.; Ngadi, M.A.: Symbiotic Organism Search opti-
mization based task scheduling in cloud computing environment.
Future Gener. Compu. Syst. 56, 640–650 (2016)
68. Li, K.; Xu, G.; Zhao, G.; Dong, Y.; Wang,D.: Cloud task scheduling
based on load balancing ant colony optimization. In: 2011 Sixth
Annual Chinagrid Conference (ChinaGrid), pp. 3–9. IEEE (2011)
69. Eberhart, R.C.; Shi, Y.: Comparing inertia weights and constriction
factors in particle swarm optimization. In: Proceedings of the 2000
Congress on Evolutionary Computation, 2000, pp. 84–88. IEEE
(2000)
70. Gabi, D.; Ismail, A.S.; Zainal, A.; Zakaria, Z.; Abraham, A.:
Orthogonal Taguchi-based cat algorithm for solving task schedul-
ing problem in cloud computing. Neural Comput. Appl. 30(6),
1845–1863 (2018)
71. Ouaarab, A.; Ahiod, B.; Yang, X.-S.: Discrete cuckoo search algo-
rithm for the travelling salesman problem. Neural Comput. Appl.
24(7–8), 1659–1669 (2014)
72. Madni, S.H.H.; Latiff, M.S.A.; Abdullahi, M.; Usman, M.J.: Per-
formance comparison of heuristic algorithms for task scheduling in
IaaS cloud computing environment. PLoS ONE 12(5), e0176321
(2017)
123