electronics

Article
Cost-Effective and Energy-Aware Resource Allocation in Cloud
Data Centers
Abadhan Saumya Sabyasachi * and Jogesh K. Muppala

Department of Computer Science and Engineering, The Hong Kong University of Science and Technology,
Clear Water Bay, Kowloon, Hong Kong 999077, China
* Correspondence: [email protected]

Abstract: Cloud computing supports the fast expansion of data and computer centers; therefore,
energy and load balancing are vital concerns. The growing popularity of cloud computing has
raised power usage and network costs. Frequent calls for computational resources may cause
system instability; further, load balancing in the host requires migrating virtual machines (VM)
from overloaded to underloaded hosts, which affects energy usage. The proposed cost-efficient
whale optimization algorithm for virtual machine (CEWOAVM) technique helps to more effectively
place migrating virtual machines. CEWOAVM optimizes system resources such as CPU, storage,
and memory. This study proposes energy-aware virtual machine migration with the use of the
WOA algorithm for dynamic, cost-effective cloud data centers in order to solve this problem. The
experimental results showed that the proposed algorithm saved 18.6%, 27.08%, and 36.3% energy
when compared with the PSOCM, RAPSO-VMP, and DTH-MF algorithms, respectively. It also
showed 12.68%, 18.7%, and 27.9% improvements for the number of virtual machine migrations and
14.4%, 17.8%, and 23.8% reduction in SLA violation, respectively.

Keywords: virtual machine; SLA; WOA; migration; cloud computing

Citation: Sabyasachi, A.S.; Muppala,

J.K. Cost-Effective and Energy-Aware
Resource Allocation in Cloud Data
1. Introduction
Centers. Electronics 2022, 11, 3639.
https://doi.org/10.3390/ New computing paradigms, such as cloud computing, have emerged due to changes
electronics11213639 in the computer landscape over the last several years. Cloud computing systems have
become a popular computing paradigm as they can host huge computer systems, provide
Academic Editor: Fernando De la
various services for users to virtually access a host of resources through the internet, as well
Prieta
as have users only pay for the resources they use [1,2]. Virtualization technology is one of
Received: 4 October 2022 the most important features of modern data centers; further, it is particularly useful for
Accepted: 3 November 2022 infrastructure-based services that are housed on the cloud. Power usage in environmentally
Published: 7 November 2022 friendly data centers has skyrocketed as customers’ computing demands have, similarly,
Publisher’s Note: MDPI stays neutral
skyrocketed. With the explosion in both the number and size of cloud data centers has
with regard to jurisdictional claims in come a corresponding rise in the problem of energy consumption, which in turn drives up
published maps and institutional affil- the price of energy for service providers and is a major contributor to the release of harmful
iations. greenhouse gases [3,4].
As a result, several strategies for reducing energy use have been proposed in published
works. Researchers have, lately, shown an interest in this issue with the hope that virtual
machine (VM) migration may reduce the need for requiring many physical computers. In
Copyright: © 2022 by the authors. addition, moving virtual machines may significantly impact saving power [5]. Keeping
Licensee MDPI, Basel, Switzerland. up with the service level agreement (SLA) is a huge challenge when providing support for
This article is an open access article cloud customers, which is why distributed load balancing on physical computers is crucial.
distributed under the terms and A load balancing method is employed in the cloud in order to distribute work equitably
conditions of the Creative Commons
amongst available computer resources [6–8]. It is crucial in the cloud environment as it
Attribution (CC BY) license (https://
maximizes resource utilization, speeds up responses, and shortens the time it takes to do
creativecommons.org/licenses/by/
tasks. According to studies, an uneven distribution of resources may negatively impact
4.0/).

Electronics 2022, 11, 3639. https://doi.org/10.3390/electronics11213639 https://www.mdpi.com/journal/electronics

Electronics 2022, 11, 3639 2 of 15

Electronics 2022, 11, x FOR PEER REVIEW 2 of 17

the performance of a cloud data center. Thus, load balancing is crucial in order to ensure
the continued success of the cloud. On the other hand, in a cloud setting, the data center
negatively impact the performance of a cloud data center. Thus, load balancing is crucial
must regularly host the to
in order cloud
ensureservice. As a result,
the continued success cloud
of thedata
cloud.centers
On theuseothera great
hand, inamount
a cloud setting,
of power, addingthe to data
theircenter
operational expenses and, thus, leaving a carbon
must regularly host the cloud service. As a result, cloud data impact oncenters
the use a
environment [9–11].great amount of power, adding to their operational expenses and, thus, leaving a carbon
Metaheuristicimpact on the environment
approaches have been [9–11].
shown to be more effective in resolving the VM
allocation problem (i.e.,Metaheuristic
the NP-hard approaches
issue). Inhave been shown
particular, to be more
the whale effective in
optimization resolving the VM
algorithm
(WOA) and otherallocation problemsuch
metaheuristics, (i.e.,as
theparticle
NP-hard issue).optimization
swarm In particular,(PSO)the whale optimization
and grey wolf algo-
rithm (WOA) and other metaheuristics, such as particle
optimization (GWO, GA, etc.) methods have been utilized in order to optimize resource swarm optimization (PSO) and
usage and decrease power consumption. Ali et al. [12] developed an improved chaoticoptimize
grey wolf optimization (GWO, GA, etc.) methods have been utilized in order to
binary GWO method resource usage to
in order and decreasethe
improve power
VM consumption.
allocation and Alitoetoptimize
al. [12] developed
resource an use,improved
chaotic binary GWO method in order to improve the VM allocation and to optimize re-
balance multidimensional resources, and reduce communication traffic [13]. In order to
source use, balance multidimensional resources, and reduce communication traffic [13].
identify an ideal assignment issue homogeneously and to simplify the VM with the often
In order to identify an ideal assignment issue homogeneously and to simplify the VM
heterogeneous servers
with theinoften
cloud data centers,servers
heterogeneous Sasaninetcloud
al. [14]
datasuggested using
centers, Sasan et aal.chaotic
[14] suggested
hybrid optimizationusingalgorithm in order
a chaotic hybrid to anticipate
optimization and decrease
algorithm in order to energy usage
anticipate andindecrease
cloud energy
computing [15,16]. usage in cloud computing [15,16].
In this work, we In this work,
propose we propose
CEWOAVM CEWOAVM Whale
(Cost-Effective (Cost-Effective
OptimizationWhaleAlgorithm),
Optimization Algo-
a VM placementrithm),
method a VMthatplacement
is based method
on the that
whale is based on the whale
optimization optimization
algorithm. The algorithm.
WOA The
method has a highWOA method has
convergence ratea and
highisconvergence rate and
straightforward dueistostraightforward due to control
its resilience toward its resilience to-
ward control
parameters. Furthermore, in parameters. Furthermore,
order to maximize certainin order to maximize
aspects, certain aspects,
such as energy efficiency, such as en-
power consumption, ergyand
efficiency, power consumption,
total resource and total resource
usage, the suggested strategyusage, the suggested
considers a wide range strategy con-
siders
of resources to utilize a wideusing
beyond rangeonly
of resources
the centralto utilize beyond
processing unitusing
(CPU) only the central
[17–21]. processing unit
The system
(CPU) [17–21]. The system architecture of VM allocation is represented in Figure 1.
architecture of VM allocation is represented in Figure 1.

Figure 1. System architecture of VM allocations.

Electronics 2022, 11, 3639 3 of 15

The significant contributions of this paper are as follows:

1. Virtual machine (VM) placement is modeled in order to maximize energy efficiency
as an optimization issue.
2. The CEWOAVM technique reduces data center power usage via a cost-efficient VM
allocation without breaching SLAs.
3. CEWOAVM reduces energy consumption, VM migrations, and host shutdowns,
according to experimental results.
4. The fitness function is adjusted as time-varying in order to prevent the system from
settling into a local optimum.
The rest of the paper is structured as follows. In Section 2, we will go over the relevant
research; Section 3 will go over the suggested technique in detail; Sections 4 and 5 will
show the simulation results and the conclusions.

2. Related Work
This section will provide an overview of the VM consolidation algorithms that were
chosen as well as outline the essential characteristics of each. An approach to virtual
machine (VM) consolidation, which is based on the WOA and hybridized with a new
bandwidth allocation policy, was proposed by Abdel-baset et al. [22]. The team have been
concentrating on solving the VM consolidation issue, which has been phrased as a variable-
sized bin packing problem, considering the available frequency. They have executed
their idea utilizing the CloudSim toolkit with 25 distinct datasets and various bandwidths
implemented at random in order to verify its efficacy. The acquired findings verify that,
when compared to alternative metaheuristic techniques, the mechanism reduces the number
of PMs [23]. However, their approach has solely focused on improving bandwidth while
ignoring other critical resources, such as memory and processing power. Furthermore, the
issue of optimizing power use has not been considered.
Al-Moalmi et al. [24] proposed that the overhead caused by the dynamic placement of
VMs may be caused from the time that is spent migrating. As a result, they have thought of
using a static approach in order to distribute virtual machines. Their primary objective is
to unify the container-to-virtual-machine placement and the virtual-machine placement-
to-physical-machine placement into a single optimization challenge. The time needed to
resolve the positioning challenges may be drastically cut down by using this method. To
cut down on the time and energy spent on VM creation, WOA has been used to optimize
this issue. However, they must double check crucial metrics, such as migration time and
SLA breach [25].
A new multi-objective strategy, based on double thresholds and the ACO algorithm,
was presented by Xiao et al. [26]. It uses two levels of CPU usage as cut-offs in order to
determine whether or not the PM is overloaded. When a host becomes overburdened or
underutilized, virtual machine consolidation is initiated. In consolidating VMs, the ACO
algorithm uses several selection strategies that consider the loads of the PMs to choose
VMs from both the overloaded as well as the destination PMs. In addition to minimizing
migrations, performance degradation, service level agreement (SLA) violations, and overall
energy consumption, the suggested approach effectively uses available computing and
storage resources. However, the difficulty of calculating continues to be a significant
issue [27–29].
Fatima et al. [30] have proposed a new hybrid algorithm, LMOGWO. The suggested
method took design cues from grey wolves, modeling itself after the animals’ hunting and
pack-leading techniques. It also came with a storage archive for secondary options. The top
three answers are the alpha wolves who rally the pack to pounce on their victim. Alpha,
beta, and delta wolves represent the pack’s top three leaders, while omega wolves represent
the remaining members who have found a solution. The suggested method determines
the wolf step size based on the levy flight. Finally, the suggested method is put through its
paces using nine industry-standard benchmark functions.
Electronics 2022, 11, 3639 4 of 15

Dahsti et al. [31] developed a solution to address the requirements provided by both
service providers and users of these technologies. As a result of the research, a one-of-a-
kind PaaS service for organizing client errands was developed. In the cloud, excessive
energy usage and an energy performance tradeoff may occur if the specifications of the
physical machine and the user’s expectations are incompatible, resulting in lower provider
profitability. Using the PSO, energy efficiency is increased without compromising service
quality. The final aim of these strategies was to reallocate the relocated virtual machines in-
side the whole host. The results of the CloudSim simulations showed that the circumstances
inside the simulation were very comparable to those seen in the real world [32].
Therefore, we suggest a VM placement method based on the whale optimization
algorithm in order to lower data center power usage without SLA violation. The suggested
method concentrates on server CPU consumption while looking for the near ideal location
for VMs. The whale optimization algorithm (WOA) method, whose conventional form is
appropriate for continuous problems, is used in a discrete form in this system. In particular,
a suitable CEWOAVM method is employed in order to address the VM placement issue.

3. Problem Formulation
Virtual machines (VMs) may move from one host computer to another in the cloud
throughout a process. A VM has the option of moving to one of numerous host computers.
Simultaneously, VM migration may affect how much power a system uses. As a result,
it is essential to position correctly. Good VM organization thus organizes VMs on host
computers and powers down those not used in order to maximize efficiency and reduce
energy usage. In this scenario, n number of VMs are expected to run on m physical servers.
The challenge is developing a paradigm for the relocation of VMs on host machines. This
architecture should also facilitate the transfer of a more significant number of VMs while
simultaneously lowering energy usage.
The second kind of virtual machine placement happens when virtual machines are
moved as part of a consolidation effort. Let List of VMs = {VM1 , VM1 . . . . . . VMn } and
List of Hosts = {PM1 , PM2 . . . . . . PMm }.
Next, the aim is to influence the direction of each PM to host several VMs and offer
resources in order to accomplish this. Further, hypervisors are to maintain VMs for each
PM. The VM migration should consolidate VMs into the fewest active hosts, without
breaking SLA, for the purposes of power management. Local managers should continually
check server resource use. A global manager at the primary node should communicate
with local managers in order to monitor resource use. As demonstrated in Figure 1, we
can see what may determine VM migrations to reduce data center power usage. Each
server’s management would need to frequently check VM resource use. Local managers
would choose overcrowded and underloaded servers. Further, the local managers choose
the best VMs to move from overcrowded hosts. Local managers report resource use to
the global manager. The global manager should optimize migrating the VM placement
using power-aware VM placement based on the whale optimization algorithm. The global
manager then should instruct hypervisors to migrate VMs and shut down idle servers.
The objective is to place the migrated VMs into the respective physical machines
according to Equations (1)–(3).
k−1
∑i=0 CPUi + CPUmig < CPUj (1)

k−1
∑i=0 RAMi + RAMmig < RAMj (2)
k−1
∑i=0 BWi + BWmig < BWj (3)
where CPU, RAM, and BW are considered necessary resources for the virtual machine
(VM) to be migrated.
Electronics 2022, 11, 3639 5 of 15

In addition, a VM cannot be hosted by more than one PM at once. Hence, a VMi

can only be mapped to one server. Therefore, the suggested method represents mapping
moved VMs as in Equation (4).

1, VM and PM belong to the List of VM and Host
Vmap = (4)
0, otherwise

3.1. Energy Consumption Model

The energy needed to run a system is proportional to how often the CPU and RAM
are being used. The CPU mostly drives the energy needs of these devices. Therefore,
the percentage of time a computer’s CPU is actively used determines how much of the
machine’s resources are being used. A VM energy consumption model may be calculated
using Equation (5), which considers the processor’s utilization concerning the machine’s
current workload.
n
CPUUi = ∑j=1 CPUUi,j (5)

HostPi = f × Pmax + (1 − f) × Pmax × CPUUi (6)

Z t

E = HostPi CPUUi (t) dt (7)
t−1
where Pmax is the power spent at full utilization of the physical machine, f is the power
consumed by the machine while it is idle, and CPUUi is the processor utilization rate
(CPUUi [0, 1]) that is calculated from the workload caused by the execution of the cloud
service on the physical machine (HostPi ), which is represented by Equation (6).
In Equation (7), E is the energy used by a machine throughout the interval [t − 1, t],
and CPUUi (t) is the power drawn at time t, as determined.

3.2. Fitness Function

A fitness function assigns a value to a solution depending on the parameters effective
in the solution’s quality, allowing one to judge whether or not the solution can deliver a
timely response. When applied to all solutions generated by the suggested algorithm, this
function determines the value of each solution. The solution with the best value, either
the maximum or the lowest depending on the parameter placement strategy, is the most
practical. Thus, Equation (8) provides the fitness function that may be used in order to
evaluate the quality of each response.

1
F= Rt (8)
∑n1 t−1 HostPi


CPUUi (t) dt

3.3. Proposed Method

The cloud data center will have homogenous and heterogeneous VMs. VMs require
hardware from real machines in order to process services, and the cloud data centers host
virtual computers on physical hardware. Hardware resources are required as cloud services
demand rises. Hardware resources raise the expenses for these centers. Thus, deploying
VMs in the cloud in order to optimize resource use will save expenses and allow real
computers to host more VMs. Thus, effective VM placement maximizes physical machine
processing power, preventing the loss of hardware resources in the cloud data center. The
aim is to minimize physical machine resource utilization for a high-performance cloud data
center. Cloud data center VM utilization decreases migrations and bandwidth use. Details
of the CEWOAVM algorithm suggested in this research to schedule dependent activities in
order to reduce energy usage and improve load balancing, are presented in this section.
The reduced power consumption, CPU utilization, VM migration, and SLA violation in
cloud data centers are due to the CEWOAVM algorithm.
Electronics 2022, 11, 3639 6 of 15

WOA algorithms operate in a continuous cloud environment. Cloud data centers

map virtual computers to actual equipment separately. The suggested technique for VM
placement on physical machines in cloud data centers uses new operators in VMs in order
to discretely address the deployment issue.

Whale Optimization Algorithm (WOA)

We would propose to use a WOA-based mechanism for assigning resources to physical
hosts. In addition to the algorithm’s time complexity, load congestion reduces data center
energy usage. Furthermore, WOA is ideal for VM allocation to physical hosts (which, as
a function, is an NP-hard issue). Moreover, WOA is the finest metaheuristic for tackling
these situations. This method has accelerated and delivered a more efficient solution than
prior solutions.
Each whale conducts the exploration by randomly updating its location in the search
→ →
space or using the optimal search agent, which is determined by vector A. When A > 1, the
→
whales move randomly, but when A < 1 they prefer to undertake a local search. We may
represent the discovery stage using arithmetic as shown in Equations (9) and (10).
→ → → →
D = C × Wrand − X (9)

→ → → →
X (t+1) = Wrand − A × D (10)
→
where Wrand is a randomly picked whale.
The location of the prey is employed as a mathematical representation of the best
response available at the simulation time, which is the exploitation phase. Each second
that passes pushes the relative locations of the whales closer to the center of the circle,
representing the approaching prey. The predator’s behavior may resemble a spiral or
an encirclement of the victim. The encircling is mathematically defined by the following
Equation (11).
→ → → →
P (i+1) = X b ( i ) − A × D (11)
→
where at each iteration t, the agent’s location is represented by the vector P (i) , and the
→ →
vector X b indicate the best agent (i). In Equation (12), A stands for the coefficient vector,
and D stands for the distance from the best agent, as illustrated in Equation (13).
→ →
A = 2× a ×r − a (12)
→ → → →
D = C × X b (t) − X (t) (13)

→
C = 2×r (14)
→ →
where r ranges from [0, 1] is a random integer, a is a decreasing vector from 2 to 0, and C
is an adjustment factor by which search agents capture the local regions.
Algorithm 1 presents the pseudocode of the proposed CEWOAVM algorithm.

Algorithm 1 CEWOAVM Algorithm

Input: List of Hosts = {PM1 , PM2 . . . . . . PMm } and Migrated VM
Output: Best_VMmig
1:while (t < Itrmax ) do
2: for each Pi do
Electronics 2022, 11, 3639 7 of 15

4: if (q < 0.5) then

5: if |A| < 1 then
6: Update the Equation (12)
7: else if |A| ≥ 1 then
8: Initialize (Xrand )
9: Update the Equation (13)
10: end if
11: else if (q ≥ 0.5) then
12: Update the Equation (14)
13: end if
14: end for
15: for t ∈ max
16: for VM ∈ List of VMmig
1
17: Evaluate F = n R t
∑1 t−1 HostPi (CPUUi (t))dt
18: end for
19: end for
20: Count t = t + 1
21: end while
22: Return Best_VMmig

CEWOAVM cycles through various stages to get a VM position closer to the optimal
route. First, each particle’s velocity is updated with each iteration, allowing for a recalcu-
lation of its location. Next, the whale’s current location is compared to both the optimal
whale position and the optimal whale position in order to determine the appropriate speed
adjustment. Then, in order to obtain an index in the list of possible hosts, we round the
new location results and new speed calculations to the closest integer. Finally, the fitness
function for each whale is recalculated based on the updated locations—the results of these
computations aid particles in their quest to discover optimal solutions. Therefore, the
Electronics 2022, 11, x FOR PEER REVIEWswarm may use this to arrive at the best possible 8 of 17result. The whole working process is

shown in Figure 2.

Figure 2. Workflow diagram of the CEWOAVM method.

Figure 2. Workflow diagram of the CEWOAVM method.
4. Experimental Setup
The simulation test bench and analysis of outcomes from running the proposed
method in a simulated environment are described in this section. In the experiments
shown in Table 1, VM and PM sizes are varied, and random workloads are applied to
each. Each experiment is repeated ten times in order to guarantee the accuracy of the find-
ings. An accurate result may be given by comparing these findings to the widely used VM
placement method PABFD. Table 2 lists the experimentally used parameter values. Due
to the significant number of PMs and VMs used in our tests, scaling them up to evaluate
the suggested approach in a real-world setting would be impractical and expensive. As
an alternative, the suggested VM placement method is tested in a simulated setting. We
Electronics 2022, 11, 3639 8 of 15

4. Experimental Setup
The simulation test bench and analysis of outcomes from running the proposed
method in a simulated environment are described in this section. In the experiments
shown in Table 1, VM and PM sizes are varied, and random workloads are applied to
each. Each experiment is repeated ten times in order to guarantee the accuracy of the
findings. An accurate result may be given by comparing these findings to the widely used
VM placement method PABFD. Table 2 lists the experimentally used parameter values. Due
to the significant number of PMs and VMs used in our tests, scaling them up to evaluate
the suggested approach in a real-world setting would be impractical and expensive. As
an alternative, the suggested VM placement method is tested in a simulated setting. We
specifically chose the simulator platform CloudSim toolkit for the experiment. Further,
50–150 PMs and 50–200 VMs were employed for this study. Workload W1, W2, and W3
data centers contain 50 VMs and 50 PMs; 100 VMs and 100 PMs; and 150 VMs and 150 PMs,
respectively; further, they are utilized in order to assess the suggested method. The trials
are carried out using the suggested method in order to reduce VM migrations and PM
shutdowns, as well as energy consumption and SLA violations in data centers.

Table 1. Experimental cases.

Workload
W1 W2 W3
Number of VM 50 100 150
Number of PM 50 100 150

Table 2. Proposed CEWOAVM parameters.

HP ProLiant ML 110G4 (S1) 1860 (IMPS), Memory 6 GB

HP ProLiant ML 110G4 (S2) 2660 (IMPS), Memory 6 GB
High CPU VM 2200 IMPS
Small VM 1000 IMPS
Micro VM 500 IMPS
Np 30
It 100
IWmin 0.5
IWmax 1
C1 2.0
C2 2.0

Experiment-specific parameter values are listed in Table 1. Due to the enormous

number of PMs and VMs used in our studies, it would be impractical and expensive to
replicate them in a real-world setting in order to evaluate the suggested method. Instead,
the suggested method for VM placement is put through its paces in a simulated setting.
Experiments imitate two different kinds of servers and four different virtual machines,
the details of which are listed in Tables 2 and 3, respectively. To determine which hosts
are overburdened and which virtual machines should be moved, a combination of Lo-
cal Regression is used. As measured by various performance indicators, CEWOAVM
outperformed PSOCM, RAPSO-VMP, and DTH-MF.

Table 3. Energy utilization at S1 and S2.

Type of Server 0% 20% 40% 60% 80% 100%

S1 85.8 93.1 98.5 105.6 113 116.5
S2 94 100.8 111.2 120.61 128.5 136
Electronics 2022, 11, 3639 9 of 15

4.1. Simulation Results and Analysis

The CEWOAVM method is compared to existing state-of-the-art methods, including
PSOCM [28], RAPSO-VMP [29], and DTH-MF [30].

4.1.1. Energy Consumption Analysis

CEWOAVM’s objective is to reduce the number of hosts; therefore, it merges the
migrated virtual machines (VMs) into as few of them as possible. Thus, it boosts CPU use
in active servers while allowing other hosts to go to sleep. The results of this analysis are
shown in Figure 3, which shows how overall energy usage may be reduced. Compared
to PSOCM, RAPSO-VMP, and DTH-MF, the suggested method is shown to lower power
consumption by an average of 18.6%, 27.08%, and 36.3%. The suggested method10may
Electronics 2022, 11, x FOR PEER REVIEW of 17
reduce power consumption by a sufficient amount, provided that it is bound by the need
to avoid SLA violations.

80
70 CEWOAVM
Energy Consumptin (KWh)

PSOCM
60
RAPSO-VMP
50
DTH-MF
40
30
20
10
0
W1 W2 W3
Workload

Figure3.3.Comparison
Figure Comparisonanalysis
analysisof
ofenergy
energyconsumption.
consumption.

4.1.2. VM
4.1.2. VM Migration
Migration
Thereisisaarisk
There riskthat
thatlive
liveVM
VMmigrations
migrationsmay maycausecausesystem
systemperformance
performancedegradation
degradation
and, therefore,
and, therefore, breach
breach the
theSLA.
SLA.InInother
otherwords,
words,performance
performance degrades
degrades as as the
thenumber
number of of
migratedVMs
migrated VMsacross
acrosshosts
hostsincreases.
increases.CEWOAVM
CEWOAVM uses
uses a method
a method to to reduce
reduce thethe number
number of
of busy
busy servers
servers andand the percentage
the percentage of overworked
of overworked hosts.hosts. Consequently,
Consequently, increasing
increasing the CPUthe
CPU usage
usage of hosts of by
hosts by decreasing
decreasing the number
the number of activeof active
serversservers
resultsresults in an optimally
in an optimally small
small number
number of underloaded
of underloaded servers.servers. Moreover,
Moreover, it decreases
it decreases the number
the number of stressed
of stressed serv-
servers, as
ers, as can be seen in Figure 4; in addition, this attention to the two causes of migration
can be seen in Figure 4; in addition, this attention to the two causes of migration results
results
in fewerinVM fewer VM migrations.
migrations. Compared Compared
to PSOCM, to PSOCM,
RAPSO-VMP, RAPSO-VMP,
and DTH-MF,and DTH-MF,
the number the
number
of of VM migrations,
VM migrations, using the CEWOAVM
using the CEWOAVM method, may method, may be
be reduced by reduced by 12.68%,
12.68%, 18.7%, and
18.7%, respectively.
27.9%, and 27.9%, respectively.
CEWOAVM CEWOAVM
may accomplishmay accomplish
this goal withthis goal with a drastically
a drastically reduced
reduced number of VM migrations, in contrast to VM consolidation strategies, which on
number of VM migrations, in contrast to VM consolidation strategies, which focus fo-
performing several VM
cus on performing severalmigration operations
VM migration across servers
operations to lessentothe
across servers overall
lessen the power
overall
consumption
power consumptionin data centers.
in data As a result
centers. Asofa this reduction
result in the number
of this reduction in the of number
VM migration
of VM
operations, the newly offered services will be of a much higher quality.
migration operations, the newly offered services will be of a much higher quality.
Electronics 2022, 11, x FOR PEER REVIEW 11 of 17

Electronics 2022, 11, 3639 10 of 15

Figure4.4.Comparison
Figure Comparisonanalysis
analysisof
ofVM
VMmigration.
migration.

4.1.3. Average
4.1.3. AverageCPU CPUUtilization
Utilization
InFigure
In Figure5,5,we
wecan
cansee
seethe
thetypical
typicalCPUCPUload
load
ofof the
the assigned
assigned S1S1andand
S2 S2 kinds
kinds of serv-
of servers.
ers. FFD’s
FFD’s average
average server
server utilization
utilization andaverage
and the the average
usage usage
of allofactive
all active servers
servers are both
are both low.
low. FFD cannot effectively balance diverse resources as there is a large discrepancy the
FFD cannot effectively balance diverse resources as there is a large discrepancy between be-
tween
CPU the CPU consumption
consumption of types S1 andof types
S2. It isS1worth
and S2. It isthat
noting worth
typenoting that
S1 active type S1
servers haveactive
the
servers have
maximum CPUtheusage
maximum CPU usage whereas
for CEWOAVM, for CEWOAVM, whereas
type S2 servers type
are lessS2 servers
taxed thanare less
those
taxed than those for PSOCM, RAPSO-VMP, and DTH-MF. Compared to DTH-MF, which
for PSOCM, RAPSO-VMP, and DTH-MF. Compared to DTH-MF, which uses low-profile
uses low-profile
servers,
Electronics 2022, 11, x FOR PEER REVIEW
CEWOAVM servers,
favorsCEWOAVM favors configurations,
those with higher those with higher such configurations,
that the VMs can such that
work
12 of 17
together more seamlessly. Moreover, the maximum degree of consolidation
the VMs can work together more seamlessly. Moreover, the maximum degree of consoli- is limited by
memory capacity. CEWOAVM has the most significant average
dation is limited by memory capacity. CEWOAVM has the most significant average memory consumption,
close
memoryto 100%, demonstrating
consumption, closeits
toability
100%, to achieve the maximum
demonstrating its abilityconsolidation
to achieve the of maximum
VMs with
high resource use.
consolidation of VMs with high resource use.

Figure 5.
Figure 5. Comparison
Comparison analysis
analysis of
of average
average CPU
CPU utilization.
utilization.

4.1.4. SLA Violation

As a result, in an IaaS setting, evaluating the SLA offered to a VM’s user necessitates
the introduction of a statistic unrelated to workloads. Two metrics are established for the
purposes of monitoring the severity of SLA violations: the first measure is when the time
servers are at 100% utilization and the percentage of service degradation that occurs due
to migrations. Performance decline during migrations is the other measure (PDM). As a
Electronics 2022, 11, 3639 11 of 15

4.1.4. SLA Violation

As a result, in an IaaS setting, evaluating the SLA offered to a VM’s user necessitates
the introduction of a statistic unrelated to workloads. Two metrics are established for
the purposes of monitoring the severity of SLA violations: the first measure is when
the time servers are at 100% utilization and the percentage of service degradation that
occurs due to migrations. Performance decline during migrations is the other measure
(PDM). As a result, we present a composite measure, SLA violation (SLAV), which is
derived from the two individual metrics. In contrast to the PSOCM, RAPSO-VMP, and
DTH-MF methods, experimental findings demonstrated that the CEWOAVM does not
violate SLA (see Figure 6). Moreover, the CEWOAVM prevents an increase in SLA violation.
However, it may minimize SLA violations as the number of VMs and PMs grows. The
primary motivation for this effort is to reduce energy usage by consolidating migrating
virtual machines onto as few hosts as possible. However, this work may also prevent SLA
violations by avoiding their two significant causes. Compared to PSOCM, RAPSO-VMP,
Electronics 2022, 11, x FOR PEER REVIEW 13 of 17
and DTH-MF, respectively, CEWOAVM reduces the amount of SLA violations by 14.4%,
17.8%, and 23.8%, respectively, on average.

Figure 6. Comparison analysis of SLA violations.

Figure 6. Comparison analysis of SLA violations.

4.1.5.The
4.1.5. TheNumber
Numberof
ofHost
HostShutdowns
Shutdowns
Thefrequency
The frequencywith
withwhich
whichhosts
hostshave
haveshut
shutdown
downalso alsosignificantly
significantlyimpacts
impactsQoS.QoS.VMs
VMs
mustbe
must bemoved
movedoff
offaaserver
serverwhen
whenCPUCPUconsumption
consumptiondrops dropsbelow
belowaacertain
certainthreshold.
threshold.In In
thefuture,
the future,the
theserver
servermay
mayhost
hostvirtual
virtualmachines
machines(VMs)(VMs)that
thatwere
werepreviously
previouslymoved,
moved,and and
ititcould
couldthen
thenrelocate
relocatethose
thoseVMsVMsagain
againififthey
theyare
areunderutilized.
underutilized.As Asaaresult,
result,its
itscondition
condition
may be switched back to the low power mode to save energy. This situation harms
may be switched back to the low power mode to save energy. This situation harms not
not only
only energy
energy usage,usage,
but alsobut also
user user experience
experience becausebecause of the frequent
of the frequent VM migrations.
VM migrations. Repeatedly Re-
peatedlya running
running host witha low
hostloads
with is
low
notloads is not cost-effective.
cost-effective. If CEWOAVM If CEWOAVM
is configuredis to
configured
prevent
to prevent and
overloaded overloaded and cases,
underloaded underloaded
as shown cases, as shown
in Figure in host
7, fewer Figure 7, fewer will
shutdowns hostoccur.
shut-
downs willtooccur.
Compared PSOCM, Compared
RAPSO-VMP, to PSOCM, RAPSO-VMP,
and DTH-MF, and DTH-MF,
the suggested methodthe suggested
reduces the
method reduces the average number of host shutdowns by 45.39%, 52.94%, and 66.46%,
average number of host shutdowns by 45.39%, 52.94%, and 66.46%, respectively.
respectively.
 Electronics 2022, 11, x FOR PEER REVIEW 14 of 17

Electronics 2022, 11, 3639 12 of 15

Electronics 2022, 11, x FOR PEER REVIEW 15 of 17

Figure7.7.Comparison
Figure Comparisonanalysis
analysisofofhost
hostshutdowns.
shutdowns.

4.1.6.Error
4.1.6. ErrorAnalysis
Analysis
Figure 88shows
Figure showsthe
theerror
errorbar
barchart
chartwith
withananinterpolation
interpolationline
linecreated
createdto
toshow
showhow
howthethe
algorithmperformed
algorithm performeddifferently.
differently.Figure
Figure88demonstrates
demonstratesthat
thatthe
theCEWOAVM
CEWOAVMperformed
performed
betterthan
better thanthe
theRAPSO-VMP.
RAPSO-VMP. It It
waswas noted
noted that
that thethe PSOCM
PSOCM andand DTH-MF
DTH-MF algorithms
algorithms ei-
either
ther performed
performed averagely
averagely or poorly
or poorly in comparison.
in comparison.

Figure8.8.Error
Figure Erroranalysis
analysisbar
barplot.
plot.

4.2.Discussion
4.2. Discussion
Themethodology
The methodologyknownknownas asCEWOAVM
CEWOAVMhas hasbeen
beenassessed
assessedusing
usingthree
threedifferent
differenttest
test
caseswhile
cases whilebeing
beingsubjected
subjectedto tovarious
variousworkloads,
workloads,as asmentioned
mentionedininthe theprior
priorsection.
section.ItIt
does this
does thisby
byutilizing
utilizing the
thecapabilities
capabilities ofof the
the WOA
WOA in in order
order to tolook
lookfor
forthe
theoptimal
optimalVM VM
placement option both locally and globally. Its purpose is to reduce the amountof
placement option both locally and globally. Its purpose is to reduce the amount ofpower
power
wasted
wastedin indata
datacenters
centersby byimplementing
implementingan anefficient
efficientpower-aware
power-awaremethodmethodthat thatcan
canroute
route
migrating
migrating virtual machines to the servers most suited to host them. In particular,ititworks
virtual machines to the servers most suited to host them. In particular, works
to
tomaintain
maintain servers
servers in
inthe
thenormal
normalmode modeby byminimizing
minimizing the the number
number of of servers
servers that
that are
are
either overloaded or underloaded relative to their capacity. Overloaded servers
either overloaded or underloaded relative to their capacity. Overloaded servers harm the harm the
quality of the services that are provided to users. On the other hand, underloaded servers
cause a waste of resources and increase the data center power consumption that is needed
to perform virtual machine migrations, which violates the service level agreement (SLA).
The results of experiments, when compared to DTH-MF, have shown that CEWOAVM
Electronics 2022, 11, 3639 13 of 15

quality of the services that are provided to users. On the other hand, underloaded servers
cause a waste of resources and increase the data center power consumption that is needed
to perform virtual machine migrations, which violates the service level agreement (SLA).
The results of experiments, when compared to DTH-MF, have shown that CEWOAVM
can prevent SLA violations by preventing servers from entering an overload or underload
condition. Consolidating a large number of virtual machines onto a smaller number of
physical servers is another way CEWOAVM may cut down on the number of power servers
used. According to the findings of simulations, CEWOAVM has the potential to reduce an
average of roughly 36.3% of the power usage, 27.9% of the number of VM migrations, and
66.46% of host shutdowns.
In conclusion, the effectiveness of CEWOAVM is brought into focus by the SLA viola-
tion. CEWOAVM’s primary purpose is to reduce the power used while simultaneously
protecting SLAs from being violated. In order to demonstrate how effective CEWOAVM
is, we will proceed with an alternative scenario predicated on aggressive consolidation
utilizing WOA. The experimental results showed that the proposed algorithm saved 18.6%,
27.08%, and 36.3% of energy compared with the PSOCM, RAPSO-VMP, and DTH-MF algo-
rithms, respectively. In addition, it also showed 12.68%, 18.7%, and 27.9% improvements
for the number of virtual machine migrations and 14.4%, 17.8%, and 23,8% reduction in
SLA violation, respectively. The identical method is used in this circumstance, except that
the fitness function does not consider the total number of overloaded hosts. It is only
concerned with maximizing usage while simultaneously reducing the number of active
hosts. As a result, it can achieve a more significant decrease in the power utilized when
compared to CEWOAVM. However, this decrease comes at the price of the performance,
which entirely breaches the service level agreement (SLA).

5. Conclusions
Cloud computing and its related services are prevalent; hence, data center building has
risen. Data center power usage needs to be managed. Moreover, consolidating VMs onto
the fewest possible servers reduces power usage. Further, cloud computing migrates VMs
from underloaded servers to other hosts, such that their original hosts may go into sleep
mode. Migrating VMs from overloaded servers helps prevent SLA violations. Moreover,
finding hosts for migrating VMs is crucial. CEWOAVM, a cost-efficient VM placement
strategy based on the WOA algorithm, reduces power usage without breaching SLA. WOA
uses decimal encoding for ongoing issues, such as VM placement. A fitness function
reduced active and overburdened servers. In addition, the suggested approach was applied
in CloudSim, and simulation results proved its efficiency in terms of used energy, host
shutdowns, VM migrations, and CPU utilization. CEWOAVM’s efficiency may be validated
via real-world deployment. Moreover, memory, bandwidth, and network parameters may
also be optimized further.

Author Contributions: Conceptualization, A.S.S. and J.K.M.; methodology, A.S.S.; software, A.S.S.;
validation, A.S.S.; formal analysis, A.S.S.; investigation, A.S.S.; resources, A.S.S.; data curation,
A.S.S.; writing—original draft preparation, A.S.S.; writing—review and editing, A.S.S. and J.K.M.;
visualization, A.S.S.; supervision, J.K.M. All authors have read and agreed to the published version
of the manuscript.
Funding: This research received no external funding.
Informed Consent Statement: Not applicable.
Data Availability Statement: All data generated or analyzed during this study are included in this paper.
Conflicts of Interest: The authors declare no conflict of interest.
Electronics 2022, 11, 3639 14 of 15

References
1. Garg, H. A hybrid PSO-GA algorithm for constrained optimization problems. Appl. Math. Comput. 2016, 274, 292–305. [CrossRef]
2. Zhou, Z.; Chang, J.; Hu, Z.; Yu, J.; Li, F. A modified PSO algorithm for task scheduling optimization in cloud computing. Concurr.
Comput. Pract. Exp. 2018, 30, e4970. [CrossRef]
3. Patwal, R.S.; Narang, N.; Garg, H. A novel TVAC-PSO based mutation strategies algorithm for generation scheduling of pumped
storage hydrothermal system incorporating solar units. Energy 2018, 142, 822–837. [CrossRef]
4. Kumar, D.; Raza, Z. A PSO-based VM resource scheduling model for cloud computing. In Proceedings of the 2015 IEEE
International Conference on Computational Intelligence & Communication Technology, Ghaziabad, India, 13–14 February 2015;
pp. 213–219.
5. Xu, M.; Tian, W.; Buyya, R. A survey on load balancing algorithms for virtual machines placement in cloud computing. Concurr.
Comput. Pract. Exp. 2017, 29, e4123. [CrossRef]
6. Mondal, S.K.; Sabyasachi, A.S.; Muppala, J.K. On Dependability, Cost and Security Trade-Off in Cloud Data Centers. In
Proceedings of the 2017 IEEE 22nd Pacific Rim International Symposium on Dependable Computing (PRDC), Christchurch, New
Zealand, 22–25 January 2017; pp. 11–19.
7. Maciel, O.; Cuevas, E.; Navarro, M.A.; Zaldıvar, D.; Hinojosa, S. Side-blotched lizard algorithm: A polymorphic population
approach. Appl. Soft Comput. 2020, 88, 106039. [CrossRef]
8. Cuevas, E.; Fausto, F.; Gonzalez, A. The locust swarm optimization algorithm. In New advancements in Swarm Algorithms: Operators
and Applications; Springer: Cham, Switzerland, 2020; pp. 139–159.
9. Galvez, J.; Cuevas, E.; Becerra, H.; Avalos, O. A hybrid optimization approach based on clustering and chaotic sequences. Int. J.
Mach. Learn. Cybern. 2020, 11, 359–401. [CrossRef]
10. Arora, S.; Anand, P. Chaotic grasshopper optimization algorithm for global optimization. Neural Comput. Appl. 2019, 31,
4385–4405. [CrossRef]
11. Tharwat, A.; Elhoseny, M.; Hassanien, A.E.; Gabel, T.; Kumar, A. Intelligent Bezier curve-based path planning model using
Chaotic Particle Swarm Optimization algorithm. Clust. Comput. 2019, 22, 4745–4766. [CrossRef]
12. Ali, M.; Masdari, M.; Gharehchopogh, F.S.; Jafarian, A. Improved chaotic binary grey wolf optimization algorithm for workflow
scheduling in green cloud computing. Evol. Intell. 2021, 14, 1997–2025.
13. Ali, M.; Masdari, M.; Gharehchopogh, F.S.; Jafarian, A. A hybrid multi-objective metaheuristic optimization algorithm for
scientific workflow scheduling. Clust. Comput. 2021, 24, 1479–1503.
14. Sasan, G.; Masdari, M.; Jafarian, A. Power efficient virtual machine placement in cloud data centers with a discrete and chaotic
hybrid optimization algorithm. Clust. Comput. 2021, 24, 1293–1315.
15. Sasan, G.; Masdari, M.; Jafarian, A. Virtual machine placement in cloud data centers using a hybrid multi-verse optimization
algorithm. Artif. Intell. Rev. 2021, 54, 2221–2257.
16. Sasan, G.; Masdari, M.; Jafarian, A. The placement of virtual machines under optimal conditions in cloud datacenter. Inf. Technol.
Control 2019, 48, 545–546.
17. Hekimoglu, B. Optimal tuning of fractional order PID controller for DC motor speed control via chaotic atom search optimization
algorithm. IEEE Access 2019, 7, 38100–38114. [CrossRef]
18. Sayed, G.I.; Hassanien, A.E.; Azar, A.T. Feature selection via a novel chaotic crow search algorithm. Neural Comput. Appl. 2019, 31,
171–188. [CrossRef]
19. Qureshi, B. Profile-based power-aware workflow scheduling framework for energy-efficient data centers. Future Gener. Comput.
Syst. 2019, 94, 453–467. [CrossRef]
20. Li, X.; Yu, W.; Ruiz, R.; Zhu, J. Energy-aware cloud workflow applications scheduling with geo-distributed data. IEEE Trans. Serv.
Comput. 2020, 15, 891–903. [CrossRef]
21. Xu, P.; He, G.; Li, Z.; Zhang, Z. An efficient load balancing algorithm for virtual machine allocation based on ant colony
optimization. Int. J. Distrib. Sens. Netw. 2018, 14, 1550147718793799. [CrossRef]
22. Abdel-Basset, M.; Abdle-Fatah, L.; Sangaiah, A.K. An improved Le’vy based whale optimization algorithm for bandwidth-efficient
virtual machine placement in cloud computing environment. Clust. Comput. 2019, 22, 8319–8334. [CrossRef]
23. Yadav, R.; Zhang, W.; Kaiwartya, O.; Singh, P.R.; Elgendy, I.A.; Tian, Y. Adaptive energy-aware algorithms for minimizing energy
consumption and sla violation in cloud computing. IEEE Access 2018, 6, 55923–55936. [CrossRef]
24. Al-Moalmi, A.; Luo, J.; Salah, A.; Li, K.; Yin, L. A whale optimization system for energy efficient container placement in data
centers. Expert Syst. Appl. 2021, 164, 113719. [CrossRef]
25. Hsieh, S.Y.; Liu, C.-S.; Buyya, R.; Zomaya, A.Y. Utilizationprediction-aware virtual machine consolidation approach for energy
efficient cloud data centers. J. Parallel Distrib. Comput. 2020, 139, 99–109. [CrossRef]
26. Xiao, H.; Hu, Z.; Li, K. Multi-objective VM consolidation based on thresholds and ant colony system in cloud computing. IEEE
Access 2019, 7, 53441–53453. [CrossRef]
27. Gomathi, B.; Balaji, B.S.; Kumar, V.K.; Abouhawwash, M.; Aljahdali, S.; Masud, M.; Kuchuk, N. Multi-Objective Optimization of
Energy Aware Virtual Machine Placement in Cloud Data Center. Intell. Autom. Soft Comput. 2022, 33, 1771–1785. [CrossRef]
28. Bhagyalakshmi, M.; Malhotra, D. Resource-Efficient VM Placement in the Cloud Environment Using Improved Particle Swarm
Optimization. Int. J. Appl. Metaheuristic Comput. 2022, 13, 1–32.
Electronics 2022, 11, 3639 15 of 15

29. Fard, Z.; Yahya, S.; Reza Ahmadi, M.; Adabi, S. A dynamic VM consolidation technique for QoS and energy consumption in
cloud environment. J. Supercomput. 2017, 73, 4347–4368.
30. Fatima, A.; Javaid, N.; Anjum Butt, A.; Sultana, T.; Hussain, W.; Bilal, M.; Hashmi, M.; Akbar, M.; Ilahi, M. An enhanced
multi-objective gray wolf optimization for virtual machine placement in cloud data centers. Electronics 2019, 8, 218. [CrossRef]
31. Dashti, S.E.; Rahmani, A.M. Dynamic VMs placement for energy efficiency by PSO in cloud computing. J. Exp. Theor. Artif. Intell.
2016, 28, 97–112. [CrossRef]
32. Kim, M.; Hong, J.; Kim, W. An efficient representation using harmony search for solving the virtual machine consolidation.
Sustainability 2019, 11, 6030. [CrossRef]