See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/377255456

An efficient reducing mechanism for energy consumption in data center using

hybrid consolidation techniques

Article in Ife Journal of Science · January 2024

DOI: 10.4314/ijs.v25i3.8

CITATIONS READS

0 17

2 authors:

Ezekiel Oyekanmi Olasupo Modupe Adegoke

Achievers University Joseph Ayo Babalola University
23 PUBLICATIONS 19 CITATIONS 1 PUBLICATION 0 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Ezekiel Oyekanmi on 09 January 2024.

The user has requested enhancement of the downloaded file.

https://dx.doi.org/10.4314/ijs.v25i3.8
Ife Journal of Science vol. 25, no. 3 (2023) 427
AN EFFICIENT REDUCING MECHANISM FOR ENERGY CONSUMPTION IN
DATA CENTER USING HYBRID CONSOLIDATION TECHNIQUES

Oyekanmi, E.O.1,* and Adegoke, O.M.2

1
Department of Computer Science, Achievers University, Owo, Nigeria.
2
Department of Computer Science, Joseph Ayo Babalola University, Ikeji-Arakeji, Nigeria.
*
Corresponding Author's Email: [email protected]
(Received: 19th may, 2023; Accepted: 24th October, 2023)

ABSTRACT

The rise in internet user demand is a major factor in the expansion of infrastructure and the upsurge in energy
use in cloud, colocation, and some business data centres. The advent of 5G has compounded the situation, as it
substantially gives room for many new types of digital services, resulting in a need for richer media formats and
higher resolution content which consume a lot of energy when no scaling method is applied. Services such as
email, data storage and retrieval and other cloud services also require a lot of high energy consumption which
eventually result into carbon(IV) oxide (CO2) emissions to the environment. This research therefore, focuses on
lowering the energy usage of a data centre with heterogeneous power awareness either in an idle server state or
high-performance state using a novel hybridized algorithm called “DyVoFesLoReMu”, comprising Dynamic
Voltage Frequency Scaling (Dvfs) and a modified Local Regression Minimum Utilization (LrMu). A real dataset
(workload) obtained online from PlanetLab consisting of hosts and Virtual Machines (VM) was simulated on a
data center in CloudSim 3.0.3. Tool kit with preset parameters consisting of VM Allocation Policy and VM
Selection Policy was used. The tool kit was utilised to create cloud infrastructure and simulate the essential
features of a cloud environment. The Cloudsim was installed on Eclipse Integrated Development Environment
(IDE) 2019 version on Windows 10 operating system. The hybridized algorithm was compared with other five
(5) existing energy reducing algorithms and it was found to be more efficient with a range of 41-90% reduction
in energy usage from the ten days workload traces and in comparison with the existing algorithms used for the
simulation.

Keywords: Energy Consumption, High performance, Data center, Cloud.

INTRODUCTION (Spengler and Wilmsmeier, 2019). The main

There had been a tremendous increase in the reason for this extreme high energy consumption
usage of internet resources during the pandemic is not just the use of computing resources in large-
era (COVID-19). Numerous meetings and scale and the energy inefficiency of hardware, but
services are carried out using Whatsapp, Zoom, rather lies in the inefficient usage of these
Facebook, and other social media platforms. resources (Yadav et al., 2019).
Business, trainings, and conferences are handled
virtually. All these social media activities are World widely, datacenters are estimated to spend
controlled and coordinated from a data center in a $27 billion annually on energy (Jay and Chuck,
cloud environment, thus increasing technology 2021),and to meet up with this increasing demand
drive and creating new business opportunities of cloud services, new datacenters may need to
revolution for the sustenance of economic be constructed, else the capabilities or existing
development and new technology drive. datacenters may be expanded. However,
Consequently, revolution in information expanding datacenter capability is not always
technology (IT) and advancement in data storage feasible due to economical and physical
and retrieval are having a major negative impact on constraints such as fixed capacity of power
the data center when the cloud is loaded with more generating facilities and limited achievable server
tasks on data storage and retrieval, email services, density (Leverich, 2014). Building new data
and other cloud services, thus leading to high centres may be extremely expensive, greatly
energy consumption and CO2 emission. Recent raising the total cost of ownership in terms of
research has shown that energy consumption will both the capital expense needed to build a new
likely increase in the future due to growing facility and the operational expense needed to
equipment stocks to support the Internet manage the datacentre ( Leverich, 2014 ).
infrastructure and use of modern IT equipment
428 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption

Investigating and locating inefficiencies in order simulated to aid energy reduction in datacenter.
to enhance the capabilities of current facilities is
an alternative to constructing new data centres or In order to increase energy efficiency and lower
expanding existing ones. energy consumption of the facilities used in data
centres, this research developed a novel strategy
Data growth trends will continue to increase as the called DyVoFesLoReMu.
world consumes more and more data especially
now that pandemic is ravaging the world and there RELATED WORK
was a paradigm shift to the usage of cloud Many authors have contributed to knowledge in
services. Survey conducted by Bashroush (2020) the area of reducing energy consumption and
suggests that users of the internet, particularly their works span through placement of servers
those who use social media and cryptocurrencies and adjustment of power state of servers. It was
run by datacenters, require large-scale data centers suggested in (Abdelsamea et al., 2017) to improve
that need enormous amount of electric power to virtual machine consolidation using hybrid
provide good quality of service (Yadav and Garg, regression algorithms. This technique, called a
2019). According to US figures, data centres Hybrid Local Regression Host Overload
quadruple every five years and use around 3% of Detection algorithm (HLRHOD), was created to
the world's electrical generation (Rallo, 2014; reduce energy usage while assuring a high level of
Shaw, 2007). adherence to Service Level Agreements (SLA).
Instead of using just one utilisation parameter
One of the ways to ensure effective cloud (CPU usage), three were used: CPU, Memory, and
computing technology is to avoid wastage or loss Bandwidth. VM behaviour is rarely a function of
of data stored in the cloud. This can only be done one variable; rather, it should be a function of
when power is conserved. Hence, a novel several factors since the results from the
technique is presented to continue in the combined factors are better than the results from
conservation of energy usage on datacenter as a the single-factor methods. The outcome was
way of reducing energy consumption by its IT tested on a single PlanetLab workload with 800
infrastructure. hosts and 1033 virtual machines, however the
percentage reduction was between 20 and 24. This
Energy conservation methods have been study has motivated us to develop a hybrid
employed in data centres. These methods provide method that can improve energy efficiency in the
servers with fundamental energy management. cloud.
The methods include putting servers in sleep
mode, turning them on and off, using dynamic Zahedi et al. (2017) accomplished energy
voltage/frequency scaling (DVFS) to change the reduction, Quality of Service and temperature
power states of servers, and putting virtual balancing in the cloud data centre using a unique
migration strategy into practise. Despite all the virtual machine consolidation technique. The
methods for conserving energy in data centres, the authors used CloudSim tool to simulate the
intricate operations of cloud data centres suggested technique. The outcome of the
necessitate an upgrade to a high efficiency simulation certified that physical machine
management system so that servers can operate at temperature, SLA, and migration technique can
their peak efficiency. The energized processing together control the energy consumption and
power of data centers during the distribution, QoS in a cloud data center.
processing and storage of data also necessitate the
need for the establishment of an effective, As an alternative to Dynamic Voltage Frequency
efficient, reliable way that can withstand or Scaling and VM consolidation, a new algorithm
prevent any loss, wastage or reduction in energy dubbed Brownout was also proposed by Buyya
consumption during the idle or busy state of and Gill (2018). The combined brownout
servers. Leveraging on the earlier discussed issues, approach reduces energy consumption by
in this research, a hybridized scheme comprising selectively and dynamically deactivating
Dvfs and LrMu algorithms was designed and application optional components, and it can also
 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption 429

be used with self-contained microservices. The Zhou et al. (2021) addresses the issue of load
findings demonstrate that their method can variation during VM allocation and placement, an
reduce energy use by more than 20%, albeit there energy-efficient VM allocation and deployment
are trade-offs between energy savings and user technique based on an adaptive energy-aware
discounts. It was discovered to be better than architecture for internet of thing (IOT)
DVFS at preventing overloading. In comparison applications was given and the experimental
to the outcome of our established hybrid analysis was carried out using PlanetLab virtual
approach, the reduction in percentage is still machines running a real-world workload with
modest. various safety parameter ranges between 0.5 and
3.0.
Ahmad et al. (2018) proposed a solution for the
energy saving problem by enabling dynamic A dynamic VM selection algorithm known as
voltage and frequency scaling technique for Minimum Size Utilisation (MSU) was devised by
gaming data centers. The dynamic voltage and Yadav et al (2021) to choose the VMs from an
frequency scaling technique was compared against overloaded host for VM consolidation. The goal
non-power aware and static threshold detection of this study was to improve cloud data centres'
techniques. This helps service providers to meet energy efficiency. In comparison to the other
the quality of service and quality of experience baseline schemes, the suggested algorithm was
constraints by meeting service level agreements. able to reduce energy usage and SLA violation by
Game traces were used as a workload for testing an average of 23% and 27.5%, respectively.
the technique. Selection of better techniques
helped gaming servers to save energy cost and The performance of energy consumption
maintain a better quality of service for users reduction using our suggested strategy was found
placed globally. This work provides an to be more effective than the state-of the-art
opportunity to investigate which technique based on our findings.
behaves better — dynamic, static or non-power
aware. The results demonstrate that less energy is SYSTEM MODEL ARCHITECTURE
consumed by implementing a dynamic voltage One way to reduce energy consumption is by way
and frequency approach in comparison with static of dynamically readjusting the VM. With this
threshold consolidation or non-power aware approach of consolidation, virtual machines are
technique. However, the work failed to carried out placed on all host as a form of initialization based
enhancement for energy saving technique in Big on the requirement of the host. This is referred to
Data (such as the one used in the experimentation as default placement of VMs. However, as the
of our study) and detailed analysis were not needs of the VMs evolve over time, readjustment
provided. (consolidation) of the VM happens dynamically.
As a result of this dynamic shift, periodic
LrMu algorithm was analysed and compared with provisioning of the underlying resources is
Non Power aware, Dvfs, ThRs, IqrMc,Madmmt required to adequately support the VMs while
algorithms (Nagpal et al., 2018). But, the work was minimising energy consumption and Service
carried out using a random workload and less VM Level Ag reement (SLA) violations. A
and host. complementary approach to VM consolidation
using DyVoFesLoReMU while finding suitable
Yavari et al. (2019) also improve the consolidation host for a VM is designed as shown in Figure 1.
problem of VMs by combining two novel During checking of Overloaded Host, a Mean
techniques which are heuristic energy and Size Utilization Scheme was also implemented
temperature aware based virtual machine towards reducing the energy consumption in a
consolidation (HET-VC) and Firefly energy and datacenter.
temperature aware based virtual machine
consolidation (FET-VC).
430 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption

Figure 1: DyVoFesLoReMU Architecture.

There are 5 components to the design depicted in Figure 1. They are discussed as follows.

Initial VM Placement accordance with the migration map. The

The first step is to set up the virtual machines virtualization technology allows one to build many
(VMs) on the hosts (physical machines), taking virtual machines (VM1, VM2,..., VMn) on a real
into account the resources that each machine machine, hence reducing the amount of hardware
needs. However, while the machines are operating, required and increasing resource utilisation. This
the demand could alter, necessitating the use of technology can be used to access the resources of
overloading or underloading detection cloud infrastructure.
mechanisms to condense them.
Many algorithms had been used in power
Detecting Overloaded Hosts consumption reduction in order to consolidate
The hosts are all examined by the overload the virtual machine by cloud computing data
detection algorithm for overload. The VMs must center as discussed in the related works. In this
be moved away from the hosts if any of them are research, Median Absolute Deviation (MAD)
overcrowded in order to lessen the CPU burden algorithm was used as the host overload detection
and avoid SLA violations. algorithm whose VM selection algorithm is
Minimum Utilization. This scheme was used
Detecting Underloaded Hosts purposely to solve the problem of outliers. Some
The underload detection method evaluates each of the benchmark algorithms in the same family in
host for underload so that all of the VMs can be CloudSim environment were compared and
moved to other hosts and the hosts can be evaluated in this research. Energy consumption
switched to a power-saving state. value, SLATAH (SLA Violation Time per Active
Host), number of hosts shut down, and number
Selecting the VMs for Migration from the of migrations are among the comparison metrics
Hosts used.
The combined migration map for the overloaded
and underloaded hosts is returned by the VM Workload Characterization
selection algorithm and shows where to put the Simulation and evaluation employed workload
selected VMs for migration. traces from a real system. Workload information
was imported from the CoMon project's
VM Placement PlanetLab component (Tang et al., 2007). In a
The VM placement is then completed in random 10-day period from March to April 2011,
 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption 431

the workload traces were gathered. You may find characteristics of the workload are shown in Table
more information about these traces at 1. More than 500 servers around the world's
https://github.com/beloglazov/planetlab- locations were used to gather data on CPU usage.
workload-traces. According to workload traces, The task involved thousands of VMs, and
the CPU utilization used during data collection measurements of CPU utilisation showed that it
was less than 50%, and during the simulation, the took 5 minutes (Beloglazov and Buyya, 2012).
VM assignment had been random. The
Table 1: CPU utilization of workload data (Beloglazov and Buyya, 2012).
Date Number Number Mean (SD) % Quartile1 (%) Median (%) Quartile3 (%)
of VMs of Host
03/03/2011 1052 800 12.31(17.09) 2 6 15
06/03/2011 898 800 11.44(16.83) 2 5 13
09/03/2011 1061 800 10.70(15.57) 2 4 13
22/03/2011 1516 800 9.26(12.78) 2 5 12
25/03/2011 1078 800 10.56(14.14) 2 6 14
03/04/2011 1463 800 12.39(16.55) 2 6 17
09/04/2011 1358 800 11.12(15.09) 2 6 15
11/04/2011 1233 800 11.56(15.07) 2 6 16
12/04/2011 1054 800 11.54(15.15) 2 6 16
20/04/2011 1033 800 10.43(15.21) 2 4 12

RESEARCH MIGRATION POLICY In CloudSim 3.0.3, a standard median deviation

SCHEME VM power policy allocation file named
In statistics, outliers are known to be extreme “PowerVmAllocationPolicyMigrationMedianAbs
values in a data distribution which are likely to be oluteDeviation” was used for migration of virtual
erroneous. Finding a method to recognise and machines during the simulation. However, in the
remove such data from the distribution is one overloading module of the file, a new scheme was
technique to address this extreme situation. Since implemented which is discussed in the next
the number of host is eight hundred (800) while section.
number of VM is varied with a range of 898 and
1516, there is possibility of having extremely high VM Migration Scheme for Overloaded Host
or extremely low values and non-normality during One of the main causes of an increase in energy
simulation. Median Absolute Deviation (MAD) is consumption is thought to be an overworked host.
used because of its various advantages. It In order to solve this problem, the cost of
enhances robust measure for data which are not migration and service level agreement violation
normally distributed, serves as an outlier-tuner are needed to be reduced by using a novel scheme
technique which uses standard deviation from the called Mean Size Utilization scheme (MSU). This
mean and sometimes its uses deviation from the method was used to choose the VMs when the
median, which is less susceptible to distortion hosts were found to be overloaded, as stated in
caused by outlying values. In this research equation 2. The scheme reduces the idleness of
therefore, host, maximize its capacity utilization and prevent
it from being overloaded. The algorithm for MSU
scheme is shown in Algorithm 1
Where I is a set of utilization history and Ui is the
ith utilization value of VM measured in MIPS, (2)
then
MAD = median (|median(Ui) - Ui|) (1) where U(Hi) is the cpu utilization of host i and |H|
where 0 £ I < |utilization history| is the total number of all the host in the datacenter.
One (1) was added to both the numerator and
MAD is the returned utilization mean in MIPS as denominator to prevent null value and division
shown in equation 1. error especially when the length of host is zero.
Although this kind of scenario is actually not
common.
432 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption

Algorithm for MSU Migration Scheme DyVoFesLoReMU procedure

1: Input: vmTomigrateList 1: Input: HostList, VM
2: Output: Selected VM 2: Output: Select Suitable Host
3: vmTotalUtilization ← 0 3: FOR each host in HostList DO
3: For each VM in vmTomigrateList do 4: IF (getUtilizationOfCpuMips(host)!=0)
4: vmTotalUtilization ← vmTotalUtilization + 5: //Begin LrMu implementation
vmUtilization(VM) 6: IF((getUtilizationOfCpuMips(host)>
5: EndFor vmUtilization(VM)) andand
(getUtilizationOfCpuMips(host)<=vm_utilizati
6: onmetric))
7: allocate (host, VM)
7: For each VM in vmTomigrateList do 8: suitableHost.add(host)
9: ELSE
8: IF vm_utilizationmetric < vmUtilization (VM) 10: IF (getUtilizationOfCpuMips(host)<
then vmUtilization(VM))
9: vmToMigrate.add (VM) 11: Switchoff(host) // to conserve power and
10: EndIf reduce energy consumption
11: EndFor 12: ELSE IF(getUtilizationOfCpuMips(host) >
12: return vmToMigrate vm_utilizationmetric))
13: FOR each vm in host DO
DyVoFesLoReMU Algorithm 14: vmToMigrate.add (vm)
This algorithm combined VM migration scheme 15: IF(g etUtilizationOfCpuMips(host)<
with host suitability. While finding suitable host, vm_utilizationmetric)
the utilization of the energy consumption of the 16: Break;
host is calculated at the end of each time frame 17: ENDIF
before starting another simulation. 18: ENDFOR
DyVoFesLoReMU, which is a combination of 19: ENDIF
Dynamic Voltage/Frequency Scaling (DyVoFes) 20: ENDIF
and Local Regression VM allocation policy with 21: //End of LrMu implementation
Mean Utilization VM selection policy (LoReMU) 22: ELSE
was implemented, such that at each time frame the 23: Switchoff (host)//Applying DvFS
utilization value equals 0%, Dynamic Voltage 24: ENDIF
Frequency Scaling power aware policy is invoked 25: ENDFOR
and any host of such value is shutdown (turned
26: RETURN suitableHost
off) to save the energy. LoRe VM allocation policy
determines whether a host is overcrowded or not
Hardware and Software Configuration
by using Local Regression (LR) to forecast host
The simulation for this study was conducted on a
utilisation (load). The VM with the Mean
server with a 64-bit Core i3 operating system (OS)
(Average) CPU Utilisation (MU) is the one that is
and four logical processors, which divide a server's
chosen for migration under the MU VM Selection
computing capacity into separate halves to allow
Policy. If the required MIPS at the end of each
for parallel processing. Each logical processor has
time frame computation is higher than the total
the capacity to carry out its own stream of
MIPS allotted to the host, the host is regarded as
instructions concurrently, and the OS can assign
being under allocated for the VM. If a virtual
concurrent independent units of work to each
machine's utilisation is average, migration is
logical processor as indicated in Table 2. As a
considered. MSU has been discussed in Algorithm
framework for modelling and simulating cloud
1 and implemented for VM migration in LoReMU
computing infrastructures and services with
policy file. The algorithm is shown in Algorithm 2
limited computing system capability, CloudSim is
as follows:
one of the software requirements. It may replicate
cloud-based systems on any computer system
 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption 433

with a dual-core processor, 2 GB RAM, and 1 GB Java Eclipse. For example, regression calculations
storage using a Java Runtime Environment (JRE). performed in Local Regression Mean Utilisation
The CloudSim toolkit 3.0.3 was installed and set are addressed by this library of lightweight, self-
up for efficient operation using Java Eclipse as the contained mathematics and statistics components.
development environment, as indicated in Table 3. The Java Runtime Environment (JRE) was able to
The Apache Commons Math library, which execute the CloudSim source since once it has
addresses the most prevalent practical issues not been converted by the Java Development Kit
immediately addressed in the Java programming (JDK).
language or commons-lang, was imported into the

Table 2: The configuration of hardware used for the simulation.

Operating System Processor Installed RAM Hard Disk
Windows (10) 64 -bit operating Intel (R) Core(TM) 4.00 GB 320GB
system i3 Processor
Table 3. The configuration of software used for the simulation.
CloudSim Eclipse Java Version Apache
Cloud Sim Tool kit Java Eclipse IDE Java Development Kit Apache commons -
3.0.3
(2019) (JDK 1.8) maths3-3.6.1

RESULTS AND DISCUSSION Simulation result without adjusting safety

The simulation was done in two ways. The first parameter
aspect of the results was based on the efficiency The hybrid method was simulated within a total
of the proposed method on ten days workload simulation in general based on the workload traces
traces from PlanetLab without adjusting the safety from PlanetLab for ten days. Table 4 shows energy
parameter. The second result was focused on the reduction value in kilowatt per hour for both the
workload trace of 03/03/2011 labelled as proposed method and compared methods for ten
“20110303” file. The safety parameter on this file days workload traces from PlanetLab. Figure 2
was adjusted and the result of the proposed shows the bar chart of the comparison with
method was compared with other method in the DyVoFesLoReMU having lower energ y
same policy selection category. consumption value compare to the other five
methods.
Table 4: Energy Consumption result of cloud methods and DyVoFesLoReMU
Energy Consumption (KWh)
Workload
DyVoFesLoReMU Dvfs lrmu thr mad iqrm
20110303 102.79 803.91 174.24 206.73 275.84 204.22
20110306 76.4 623.77 132.33 157.41 151.49 153.02
20110309 83.18 708.68 151.74 182.62 177.7 180.17
20110322 104.34 1014.21 188.31 220.64 214.02 219.49
20110325 89.09 785.49 162.39 189.41 183.73 188.44
20110403 133.84 1071.9 240.02 279.8 271.67 275.84
20110409 112.48 928.59 190.87 223.5 217.06 222.36
20110411 108.54 903.08 187.12 221.31 212.9 218.16
20110412 89.98 766.75 163.6 193.24 186.8 191.27
20110420 77.06 688.63 139.98 166.19 164.81 170.5
434 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption

Figure 2: Energy Consumption for ten days workload traces.

Table 5 shows the number of VM migration that displays the workload trace graphed against the
occurs at every instance of workload. Zero (no number of migrated virtual machines during the
migration) was recorded for Dvfs, because it does experiment.
not involve any virtual migration scheme. Figure 3

Table 5: Simulation result of Number of Virtual machine Migration.

NUMBER OF VM MIGRATION
Workload
DyVoFesLoReMU dvfs lrmu thrmu madmu iqrmu
20110303 2458 0 29555 30188 30051 29901
20110306 2166 0 21915 23729 23566 23256
20110309 2657 0 25876 27194 26536 27177
20110322 3423 0 32710 33552 32035 33084
20110325 2536 0 27426 27939 27485 27754
20110403 3180 0 40277 40430 39815 39910
20110409 3041 0 32546 33047 32292 32502
20110411 2925 0 31007 32703 31336 32064
20110412 2528 0 27882 28813 27150 28026
20110420 2483 0 24998 26250 25890 26511
 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption 435

Figure 3: Graph of VM migration against workload traces.

Table 6 displays the number of hosts that were shut Figure 4 shows the graphical representation of the
down during the ten-day workload traces for the workload trace against the number of host shut down
proposed approach and the compared methods. during the simulation.

Table 6: Simulation result of number of host shutdown.

Number of Host Shutdown

Workload
DyVoFesLoReMU dvfs lrmu thrmu madmu iqrmu
20110303 774 457 5525 6474 6353 6420
20110306 781 545 4305 5146 5062 5017
20110309 786 582 4899 5827 5640 5786
20110322 774 385 5744 6878 6592 6775
20110325 775 467 5090 5928 5847 5978
20110403 762 358 7237 8360 8186 8260
20110409 771 412 5892 6897 6675 6814
20110411 772 419 5728 6836 6603 6685
20110412 778 476 476 6125 5908 6079
20110420 781 519 4670 5482 5442 5539

Figure 4: Graph of the number of host shut down against workload trace.
436 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption

Table 7 shows the execution total mean time for shown in Figure 5. The equation from the graph
both the proposed method for ten days workload shows that the performance of the method used is
traces from PlanetLab. The graph of the total 82% efficient.
mean time plotted against the number of VMS is
Table 7: Simulation result of total mean time using DyVoFesLoReMU.

Workload VM Execution total mean time(SD) (sec)

20110303 1052 0.14975 (0.404)
20110306 898 0.10039 (0.311)
20110309 1061 0.11941 (0.352)
20110322 1516 0.17758 (0.418)
20110325 1078 0.11207 (0.388)
20110403 1463 0.20926 (0.393)
20110409 1358 0.18378 (0.397)
20110411 1233 0.17152(0.348)
20110412 1054 0.12463 (0.332)
20110420 1033 0.11585 (0.296)

Figure 5: Graph of total mean time against number of VMS.

Simulation Result with Adjustment of Safety following the MSU scheme earlier discussed in
Parameter section III. The MSU scheme was induced on all
In CloudSim 3.0.3, the standard policy used for the methods in Cloudsim 3.0.3 that uses median
VM allocation is median deviation VM power absolute deviation scheme with minimum
policy which is implemented in a file labelled as utilization and the safety parameter of the
“PowerVmAllocationPolicyMigrationMedianAbs methods were also adjusted. This section's results
oluteDeviation”. This file uses a VM selection demonstrate the high effectiveness of the
policy to choose the VM with the lowest CPU suggested induced MSU on all the approaches.
Utilisation for migration. This is checked The performance was measured by observing the
especially for overloaded host. However, we percentage of reduction on average using the
modified this file and a mean utilization was used report of March 3, 2011 from PlanetLab which is
 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption 437

labelled as “20110303”. compared with when the scheme was changed to

mean utilization scheme. In this result, safety
The energy consumption in kilowatt per hour is parameter was not adjusted, and was allowed to
shown on Table 8 for four different methods that run only under the adjusted scheme for CPU
uses CPU minimum utilization for overloaded utilization. The result shows that with LRMU has
host. The normal value when minimum utilization better percentage reduction report compared with
was used for overloaded host was higher other methods with the same scheme.

Table 8: Energy consumption under CPU mean (induced) and minimum utilization.
LRMU_1.2 THRMU_0.8 MAD_2.5 IQRMU_1.5
(kw/h) (kw/h) (kw/h) (kw/h)
With MSU Induced 102.79 102.41 100.12 100.7
normal 174.24 206.73 275.84 204.22
Reduction rate (%) 59 50 36 49
The safety parameter was adjusted and simulated but yet with the variant safety parameter. The
under the proposed CPU utilization scheme output was recorded on average scale as shown on
induced in all the related methods. Another Tables 9-12 and the graphs are represented on
simulation was run without the proposed scheme Figures 6-9.

Table 9: Energy consumption reduction under Threshold Random selection method

Method
Induced MSU
ThrMu
Safety POLICY
Coefficient
0.5 226.02 133.21
1 190.04 88.49
1.5 95.28 63.56
2 93.33 52.8
2.5 91.57 43.91
3 91.57 36.78
AVERAGE 131.30 69.79
% Average
46.84
Reduction
Figure 6: Proposed MU policy on ThrMU.
Table 10: Energy consumption reduction under Median Absolute Deviation selection method
Method
Induced
MadMu MSU
Safety
POLICY
Coefficient
0.5 193.1 91.55
1 195.98 91.9
1.5 196.88 94.44
2 196.41 97.59
2.5 200.4 100.12
3 201.6 102.24
AVERAGE 197.40 96.31
% Average
51.21 Figure 7: Proposed MU policy on MadMu.
Reduction
438 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption

Table 11: Energy consumption reduction under Inter Quartile Range selection method
Method
Induced
IqrMu MSU
Safety
POLICY
Coefficient
0.5 194.29 92.1
1 198.46 98.14
1.5 204.22 100.7
2 210.16 105.92
2.5 214.46 108.97
3 222.78 113.34
AVERAGE 207.40 103.20
% Average
50.24
Reduction Figure 8: Proposed MU policy on IqrMU.
Table 12: Energy consumption reduction under local regression selection method
Method
Induced
LrMu MSU
Safety
POLICY
Coefficient
0.5 92.96 54.49
1 142.73 91.21
1.5 199.99 117.65
2 236.23 131.13
2.5 268.77 133.61
3 304.35 133.81
AVERAGE 207.51 110.32
% Average
46.84
Reduction
Figure 9: Proposed MU policy on LrMU.
The report shown on Tables 9-12 reflect that the selection method with 51.21% and 50.24%,
implementation of mean size utilization of the respectively. Both Threshold Random and Local
CPU on Median Absolute Deviation selection Regression selection methods have same value of
method, has the highest reduction value on 46.84% reduction on average. The report of the
average, followed by Inter Quartile Range SLA violation is shown in Table 13 and Figure 10.

Table 13: SLA Violation at default safety parameter.

LRMU_1.2 THRMU_0.8 MAD_2.5 IQRMU_1.5
(%) (%) (%) (%)
With MSU Induced 11.57 13.16 12.6 12.69
normal 9.71 10.09 10.28 10.17
 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption 439

Figure 10: SLA Violation of Proposed MSU.

CONCLUSION REFERENCES
This paper has shown improvement in energy Abdelsamea, A., El-Moursy, A. A., Hemayed, E.
consumption reduction by applying hybrid of E., and Eldeeb, H., 2017. Virtual machine
DvFs and LrMU methods called consolidation enhancement using hybrid
DyVoFesLoReMU in power Cloudsim. The reg ression alg orithms. Eg yptian
results show that under default safety parameter, Informatics Journal, 18(3): 161-170.
the proposed method achieved a range of 41-90% doi: 10.1016/j.eij.2016.12.002
reduction in energy usage from the ten (10) days Ahmad, B., McClean, S., Charles, D., and Parr, G.,
workload traces and in comparison with the 2018. Analysis of energy saving technique
existing algorithms. Also, it was established that in CloudSim using gaming workload.
the proposed mean size utilization CPU for Cloud Computing, 2018, 143.
overloaded host performed better than the default Bashroush, R., 2020. Data center energy use goes
minimum size CPU utilization under varying up and up and up. Uptime Inst.
safety parameter. Future work would examine Beloglazov, A., and Buyya, R., 2012. Optimal
improving on the reduction of SLA violation online deterministic algorithms and
value which is also a metric for high performance adaptive heuristics for energy and
in Cloudsim computation. perfor mance efficient dynamic
consolidation of virtual machines in cloud
CONFLICT OF INTEREST: d a t a c e n t e r s. C o n c u r r e n c y a n d
Authors declare no conflicts of interest on this Computation: Practice and Experience,
manuscript 24(13): 1397-1420.
doi: 10.1002/cpe.1867
AUTHOR'S CONTRIBUTION Buyya, R., and Gill, S. S., 2018. Sustainable cloud
O.E.O: Conceptualization, Methodology, computing: foundations and future
Software. O.E.O: Data curation, Writing- directions. arXiv preprint
Original draft preparation. A.O.M: Visualization, arXiv:1805.01765.
Investigation. O.E.O: Supervision. O.E.O: doi.org/10.48550/arXiv.1805.01765
Software, Validation. A.O.M: Writing- Reviewing
and Editing
 440 Oyekanmi and Adegoke: An Efficient Reducing Mechanism for Energy Consumption

Jay, K. and Chuck, R., 2021. Improving data center Yadav, A. K., Garg, M. L., and Ritika, 2019. The
efficiency through Intel technologies and Issues of Energy Efficiency in Cloud
high ambient temperature operation. Computing Based Data Centers.
Retrieved September 2 0 2 1 , f r o m Bioscience Biotechnology Research
https://www.intel.com/content/dam/d Communications, 12(2).
oc/technolog y-brief/efficient- doi: 10.21786/bbrc/12.2/35
datacenter-high-ambient-temperature- Yadav, R., Zhang, W., Li, K., Liu, C., and Laghari,
operation-brief.pdf, 2021. A. A., 2021. Managing overloaded hosts
Leverich, J. B., 2014. Future scaling of datacenter for energy-efficiency in cloud data centers.
power-efficiency. Stanford University. Cluster Computing, 1-15.
Nagpal, S., Neeraj, D. and Surjeet D., 2018. doi.:10.1007/s10586-020-03182-3
Analysis of LrMu Power Algorithm in the Yavari, M., Ghaffarpour Rahbar, A., and Fathi, M.
Cloud Computing Environment using H., 2019. Temperature and energy-aware
CloudSim Toolkit. IJRECE, 6(3): 1175- consolidation algorithms in cloud
1177. computing. Journal of Cloud Computing,
Rallo, A., 2014. Industry outlook: Data center 8(1), 1-16.
energy efficiency. Retrieved September, doi: 10.1186/s13677-019-0136-9
24, 2018. Zahedi Fard, S. Y., Ahmadi, M. R., and Adabi, S.,
Shaw, D., 2005. EPA's 2007 report on the 2017. A dynamic VM consolidation
environment: Science report (SAB review technique for QoS and energ y
draft). consumption in cloud environment. The
Spengler, T., and Wilmsmeier, G., 2019. Journal of Supercomputing, 73(10), 4347-
Sustainable performance and 4368.
benchmarking in container doi: 10.1007/s11227-017-2016-8
terminals—The energy dimension. In Zhou, Z., Shojafar, M., Alazab, M., Abawajy, J.,
Green ports (pp. 125-154). and Li, F., 2021. AFED-EF: An energy-
doi: 10.1016/B978-0-12-814054- efficient VM allocation algorithm for IoT
3.00007-4 applications in a cloud data center. IEEE
Tang, L., Chen, Y., Li, F., Zhang, H., and Li, J., Transactions on Green Communications
2007. Empirical study on the evolution of and Networking, 5(2), 658-669.
planetlab. In Sixth International doi.org/10.1109/TGCN.2021.3067309
Conference on Networking (ICN'07) (pp.
64-64).
doi:10.1109/ICN.2007.40

View publication stats