A Host Selection Algorithm for Dynamic Container Consolidation in Cloud Data Centres using Particle Swarm Optimization

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 09 Issue: 03 | Mar 2022 www.irjet.net p-ISSN: 2395-0072
© 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page 1
A Host Selection Algorithm for Dynamic Container
Consolidation in Cloud Data Centres using Particle Swarm
Optimization
Prof.Radwan Saleh Dandah1, Dr.Kasem Kabalan2, Eng.Hayyan Rajab3
1 Professor, Department of Systems and Computer Networks, Faculty of Information Engineering, Tishreen
University, Lattakia, Syria
2 Assistant professor, Department of Systems and Computer Networks, Faculty of Information Engineering,
Tishreen University, Lattakia, Syria.
3 Postgraduate student (Phd), Department of Systems and Computer Networks, Faculty of Information
Engineering, Tishreen University, Lattakia, Syria.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - One of the leading causes of excessive
energy consumption in cloud data centers is inefficient
resource use. To address the issue,researchersproposed the
Dynamic Container consolidation approach, which aims to
consolidate containers into the fewest number of VMs and
hosts. In this research, we introduce a novel host selection
policy for container consolidation calledtheEnergyEfficient
Particle Swarm Optimization (EE-PSO) Algorithm to reduce
energy consumption while maintaining the required
performance levels in a cloud data center. We performed
experimental evaluations using the ContainerCloudSim
toolkit to validate the proposed algorithm's effectiveness
with real-world workloads. The simulation results show
that our proposed algorithm outperforms existing works in
terms of energy consumption, QoS guarantees, number of
newly created VMs, and number of container migrations.
Key Words: Cloud Computing, Container as a Service
(CaaS), Energy Efficiency, Dynamic Container
Consolidation, Particle Swarm Optimization.
1.INTRODUCTION
As a result of the rise of web-based applications such as
micro-services and server-less architectures, containers
have become more popular for creating an isolated, low
overhead environment for deploying applications [1].
Container is an operating system-level virtualization that
offers various advantages over virtual machines, including
lightweight, mobility, low start-up time, and low resource
usage [2]. Thus, containers might be seen as a new
revolution in the cloud era and have been adopted from
many cloud providers. Container as a service (CaaS) is the
new service model that has been introduced in addition to
traditional cloud services, including software as a service
(SaaS), platform as a service (PaaS), and infrastructure as a
service (IaaS).
Containers can be deployed either on physical machines
or on virtual machines. Although deploying containers
without the overhead of the hypervisor achieves high
performance levels, there are some limitations for usingthis
model such as: the dependency between containers and
operating system type, suffering fromsecuritythreatsdue to
the fact that containers do not provide the same level of
isolation as VMs [3]. Consequently, many cloud providers
use a two-level virtualization architectureas showninFig -1.
Host OS
Hardware
Hypervisor
Container Engine
Guest OS2
VM2
Container Engine
App
C3
Bins/Libs
App
C4
Bins/Libs
Guest OS1
VM1
App
C1
Bins/Libs
App
C2
Bins/Libs
Fig -1: Two-level virtualization architecture [4]
In order to meet the increasing demand for cloud
services, there has been a significant expansion in building
data centers around the world, and statistics indicate that
there are approximately 8.2 million data centers [5]. A

recently published study showed that data centers
consumed about 205 TWh during 2018, which represents
about 1% of global electricity consumption [6], and
expectations indicate that this number could reach about
353 TWh by 2030 [7].
Servers are responsible for a large part of the power
consumed in a cloud data center. Consequently,reducingthe
number of active servers, through dynamic consolidation of
containers (or virtual machines), can significantly reduce
power consumption while maintaining quality of service.
The following issues should be able to be addressed by any
container consolidation framework: [8]:
 When the host is detected as being overloaded and
unable to provide the required resources for
containers and virtual machines running on this
host?
 Which containers should be selected to migrate
from an overloaded host?
 When the host is identifiedasbeingunderloaded?Is
it possible to migrate all hosted containers andshut
down this host?
 How to choose a destination (host/VM) for
migrated containers?
According to the above questions, there are four sub-
problems in dynamic container consolidation, in this paper
we will focus on the sub-problem of destination host
selection.
2. Related Work:
In contrast to the substantial study on energy efficiency
of computing, for virtualized cloud data centers, only a few
studies explored the challenge of energy efficient container
management.
In [7], the researchers proposed a framework for
container dynamic consolidation on virtual machines. They
used static thresholds to determine the status of the hosts,
Maximum Usage (MU) and Most Correlated (MCor) to select
containers from overloaded hosts, First Fit (FF), least full
(LF), and Correlation threshold (CorHs) to select a
destination host for a migrated container, and finally, they
used First Fit to place the migrated containers ona VMatthe
selected destination.
In [9], the researchers evaluated some of the container
placement algorithms, including First Fit, Least Full, Most
Full, and Random. These algorithms are used to select a
running virtual machine on the host to be allocated to the
container. The results showed that First Fit outperforms all
the other algorithms in terms of energy consumption and
number of migrations.
In [10], the authors showed the relationship between
container and host selection policy. ACO, Least Full, Most
Full, Most Correlated, Least Correlated, Max Variance, and
Min Variance were considered for host selection, while for
container selection, they considered max-correlated, max-
variance, max-usage, and min-variance. They achieve a
superior result using the maximum-variance host selection
strategy and the greatest utilization container selection
policy.
In [11], the researchers proposed MESF,ormostefficient
server first, which is a greedy containerplacementtechnique
that assigns containers to the most energy efficient
computers first. The suggested MESF method may greatly
improve energy usagewhen comparedtotheLeastAllocated
Server First (LASF) and random scheduling schemes,
according to simulation findings utilizing an actual set of
Google cluster data as task input and machine set.
In [12], the researchers designed a fitness function to
evaluate the resource wastage of VMs and Hosts, then they
used Best Fit algorithm to create VMs in the hosts, and ACO
to place containers on these VMs.
3. Background:
3.1 Data Center Power Model:
The power consumption of a data center at time t
( ) can be calculated as the sumofpowerconsumption
of its servers at time t ( ) as shown in equation 1.
(1)
where is the number of servers. According to [13]
there is a linear relationship between the CPU utilization
and power consumption of the server. This relation can be
formulated as follows:
(2)
where is the CPU utilization of server i at time t,
and represent theconsumed powerwhentheserveris
idle, or fully utilized respectively.
3.2 SLAV Metric:
Meeting QoS requirements is a very critical issueincloud
computing environment. These requirements can be
formulated using several metrics such as: minimum
throughput or maximum response time, but due to the fact
that we do not have any prior knowledge about the behavior
of the application running inside the container, it is
important to identify a metric which does not depend on the
workload. Researchers in [14] showed that the SLA can be
violated if the virtual machine cannot get the required CPU
which has been requested. Equation 3 shows how to
calculate the SLAV metric:
(3)

where , are the number of VMs, the number of SLA
Violations respectively, and ,
are CPU amount requested and allocated
by VM j on server i at the time at which the violation p
happened.
3.3 System Model:
In order to consolidate containers on the minimum
number of virtual machines and hosts, we will use a similar
model to the proposed model in [8], which consists of two
modules: Host Status Module and Consolidation Module.
The Host Status Module exists in each active host, and it
has three main components:
 Host Over-load/ Under-Load Detector: This
component is responsible of deciding if the host is
detected as being either over-loaded or under-loaded.
In this research, the resource utilization is checked
every 5 minutes using static thresholds to identify the
status of the host.
 Container Selector: Thiscomponentdetermineswhich
containers should be selected to migrate from an
overloaded host.
 Container Migration List: This list storesall containers
selected by the Container Selector component.
The Consolidation Module runs on a separate host. This
module is responsible for choosing a destination (host/VM)
for migrated containers, and consists of the following
components:
 Over-loaded Host List
 Over-loaded Destination Selector: This component
uses a host selection policy to select a destinationfora
migrated container from an over-loaded host, and a
container placement policy to choose one of the
running VMs on that host to be allocated to the
container. If there is no running vm to host a migrated
container, the Vm Creator component is called.
 VM Creator: this component creates the largest
possible vm on the host and assign the container to it.
 Destination List: This component stores the migration
map decided by the over-loaded Destination Selector.
 Under-loaded Host List: This listcontainsall hoststhat
are identified as under-loaded after finding a
destination for all migrated containers from over-
loaded hosts.
 Under-loaded Destination Selector: This component
tries to find appropriate destinations for all hosted
containers by an under-loaded host. If this mission
could be achieved, it sends the hostIDtounder-loaded
host deactivator component.
 VM-Host Migration Manager: This componenttriggers
the migration process after selecting all destinations.
 Under-loaded Host Deactivator: after migrating all
containers of an under-loaded host, this component
turns it off.
3.4 Standard Particle Swarm Optimization
PSO is a meta-heuristic algorithm that wasintroducedby
Eberhart and Kennedy in 1995[15]. It mimics the social
behavior of a flock of birds searching for food. Each bird
inside the algorithm is called a particle, andeachparticlehas
a position vector that represents one of the possible
solutions to the problem, as well as a velocity vector that
describes its movement within the solution space. First, the
position vectors are generated randomly. Second, at each
iteration of the algorithm, each particle moves to a new
position which depends on the values of the following: the
particle's velocity in the previous iteration, the best position
found by the particle (pbest); and the best position found in
the entire swarm (gbest). The positions are evaluated using
a fitness function. The following equations are used to
calculate the new position and velocity vectors. [16]:
(5)
where:
: velocity vector of particle i at iteration t
: velocity vector of particle i at iteration t+1
: current position of particle at iteration t
: new position of particle i at iteration t+1
: personal best position of particle i
: global best position
w: inertia weight
c1,c2: cognitive and social learning parameters
respectively
rand1,rand2 : random values between 0 and 1
Fig-2 shows the flowchart of the algorithm.
(4)

Start
Swarm initialization
Particle fitness evaluating
Calculating the individual
historical optimal position
Calculating the swarm
historical optimal position
Updating particle velocity and
position according to the velocity
and position updating equation
Satisfying the
ending condition?
End
Yes
No
Fig-2: Flowchart of PSO [17]
4. Proposed Method:
4.1 Initialize Particles:
In the standard PSO, initial particles are generated
randomly. However, this randomness reduces the
algorithm’s likelihood of converging to the optimal solution.
As a result, effective initialization solutions can vastly
increase its performance [18]. In our proposed method, we
choose a random container from the migrated container list
and assign it to an available host using the algorithm
described in Fig-3. This approach ensures thattheinitialized
solutions are good and feasible. The proposed host selection
method is considered a modified version of the well-known
bin-packing algorithm, First Fit. The hosts are arranged in
descending order according to their power efficiencies and
CPU utilization, respectively. The power efficiency of a host
is a ratio between its CPU capacity and its maximum
consumed power [19]. The algorithm examines hosts and
their running VMs one by one. If there is a suitable vm and
the host would not become overloadedaftertheallocationof
the container, the selected host along with the selected vm
are returned as the destination of the migrated container. If
there is no running vm, the algorithm tries to create the
largest possible VM on that host and assigns the container to
it. It is important to note that creating VMs is only done in
the process of finding destinations for migrated containers
from overloaded hosts. Theotherdifferencebetweenfinding
destinations for migrated containers from overloaded and
underloaded hosts is the excluded hosts list, which includes
the overloaded hosts in the first phase, and overloaded,
switched off, and hosts which have been selected as
destinations in the second phase.
Algorithm1: Destination Selection Process
Input: hostList,excludedHosts,container
Output: Destination(allocatedHost, allocatedVM)
1: allocatedHost NULL;
2: allocatedVM NULL;
3: hostList sort(hostList,PE,U,’Des’);
4: foreach host in hostList do
5: if excludedHosts.contains(host) then
6: continue;
7: end if
8: vmList host.getVmList();
9: foreach vm in vmList do
10: if (vm.isSuitable(container) and
host.isNotOverLoadedAfterAllocation(vm,contain
er)) then
11: allocatedHost host;
12: allocatedVM vm;
13: break;
14: end if
15: end foreach
16: if (allocatedVM == NULL) then
17: newVM = host.createLargestPossibleVM();
18: if (newVM != NULL) then
19: if (newVM.isSuitable(container) and
host.isNotOverLoadedAfterAllocation(newV
M, container)) then
20: allocatedHost host;
21: allocatedVM vm;
22: break;
23: end if
24: end if
25: end if
26: end foreach
27: return Destination;
Fig-3: Destination Selection Process

4.2 Fitness Function:
To formulate the fitness function of our proposed
algorithm, we use a multi-criteria algorithm called TOPSIS
[20]. According to this method, the best solution is the one
which has the greatest distance from the negative-ideal
solution and the smallest distance from the positive-ideal
solution. There are four criteria depicted in table 1 used to
rank the particles.
Table-1: Considered Criteria in Fitness Function
No criteria Description Cost/benefit
1
Energy
Consumption
The new power
consumption of the
data center after
container
migrations
Cost
2
Number of
Successfully
migrated
containers
The higher this
number, the higher
the probability that
the overloaded
host will return to
normal state, and
the under-loaded
host will be shut
down
Benefit
3
The Sum of
Energy
Efficiency
Factors of
Selected Hosts
Selecting the most
energy-efficient
hosts has a great
impact on energy
consumption and
performance
Benefit
4
Number Of
Newley Created
VMs
The lower this
number, the lower
overhead of
launching a new
operating system
Cost
First, we calculate the value of each parameter for every
particle in the swarm. Then, these values are normalized by
dividing them by the maximum value of each parameter
found in the swarm using the equation:
(6)
In the next step, the and are
determined according to the following equations:
Then, the Euclidean distances from the positive ideal
solution and the negative ideal solution
are calculated for each particle using the equation 9 and
equation 10, respectively.
(9)
(10)
Finally, the score of the particle i is calculated using
Equation 11:
(11)
This score is regarded as the fitness function value that
the algorithm seeks to maximize.
4.3 PSO Parameters:
The value of inertia weight w and its changes during the
iterations of the algorithm are very critical to performance.
In our proposed algorithm, we will use the approach
introduced in [21], which leadstogoodresults.,asitdepends
on a linear reduction of w value for 70% of the iterations,
and in the last stages, it gives w a random value within the
range (0.4,0.7), which means greater valuesofwtoallowthe
algorithm to jump outside the local optimal solution. The
equation for calculating w is given as follows:
Where t is the current iteration, T is the total number of
iterations.
When a particle updates its position, the proposed
algorithm checks the feasibility of the new solution. For
simplicity, it examines the statusoftheselectedhostforeach
migrated container one by one, and if there is no suitable vm
(running or newly created), the value of the corresponded
element of the unallocated container at thepositionvectoris
set to null.
The rest of the algorithm parameters are shown in the
Table -2.
(7)
(8)
(12)

Table-2: PSO Parameters
100
Number of particles
1.4
Initial inertia weight w
0.4
Minimum Value of w
2
Learning factors c1,c2
100
Number of iterations
5. Performance Evaluation:
5.1 Simulation Setup:
To evaluate the performance of our proposed policy, we
will use the ContainerCloudSim toolkit [22]. In our
experiments, the cloud datacenter consists of 100 PMs, 200
VMs, and more than 1000 containers. The characteristics of
their configurations are shown in Table-3. PMs, VMs, and
containers each have 1 TB, 2.5 GB, and 0.1GBofdisk storage.
The network bandwidth of PMs, VMs, and containers is
1GB/s, 10MB/s, and 250KB/s, respectively. Because the
startup delay of each container and VM creation directly
affects the SLA measurements, these startup delays are
significant and are set to 0.4 seconds for containers [8] and
100 seconds for VMs [23].
The CPU utilization of eachcontainerisassignedtooneof
PlanetLab workload traces [19]. These traces consist of 10
days of workload gathered every 5 minutes between March
and April 2011 [9] as shown in Table-4.
We use the First Fit algorithm as a container placement
policy, and the Maximum Usage (MU) algorithm as a
container selection policy.
The performance of our proposed algorithm (EE-PSO) is
compared with the following algorithms: Correlation
Threshold Host Selection (CorHS), First Fit Host Selection
(FFHS), Least Full Host Selection (LFHS), and Most Full Host
Selection (MFHS). This comparison will be done in terms of
energy consumption, SLAV, total number of migrations, and
number of newly created VMs.
Table-4: PlanetLab Workload Traces
Date
Number of
Containers
Mean(%) St.dev(%)
2011/03/03 1052 12.31% 17.09%
2011/03/06 898 11.44% 16.83%
2011/03/09 1061 10.70% 15.57%
2011/03/22 1516 9.26% 12.78%
2011/03/25 1078 10.56% 14.14%
2011/04/03 1463 12.39% 16.55%
2011/04/09 1358 11.12% 15.09%
2011/04/11 1233 11.56% 15.07%
2011/04/12 1054 11.54% 15.15%
2011/04/20 1033 10.43% 15.21%
5.2 Experiment Results:
5.2.1 Scenario 1:
In this set of experiments, the upper and lower
thresholds are set at 80% ,70%, respectively. Because there
are 10 days' workload data, each performance metric might
Table-3: Configuration of PMs, VMs, and containers
PM Configurations and power models
PM type # CPU [3GHz] (mapped on 37274 MIPS Per core)
Memory
(GB)
Pidle(Watt) Pmax(Watt)
# 1 4 cores 64 86 117
# 2 8 cores 128 93 135
# 3 16 cores 256 66 247
Container and VM Types
Container
type #
CPU MIPS (1 core) Memory (MB) VM type #
CPU [1.5 GHz]
(mapped on 18636
MIPS Per core)
Memory
(GB)
# 1 4658 128 # 1 1 core 1
# 2 9320 256 # 2 2 cores 2
# 3 18636 512
# 3 4 cores 4
# 4 8 cores 8

yield ten results. The average of these results is used as the
final result of the algorithm on the metric. The results
(Charts [1-4]) show that our proposed method outperforms
all the other algorithms for all metricsbecauseouralgorithm
aims to minimize the power consumption in its fitness
function. EE-PSO also maximizes the number of successfully
migrated containers, which means a lower number of
overloaded hosts and SLA violations, anda higher numberof
underloaded hosts that will beshutdown.Ontheotherhand,
selecting the most energy efficient hosts means not only
minimizing the energy but also means higher capacities so
the container will get its required resources and thenumber
of migrations and created VMs will decrease.
Chart -1: Energy consumption in scenario 1
Chart -2: Total number of container migrations in
scenario 1
Chart -3: SLAV in scenario 1
Chart -4: Total number of newly created VMs in
scenario 1
5.2.2 Scenario 2:
In this set of experiments, we will study the effect of
varying the upper-threshold value while keeping the lower
threshold fixed at 70%. The number of containers and their
workload traces are set according to day 1 of Table-4. The
results (Charts [5-8]) show that increasing the upper-
threshold value increases the powerconsumptionmetric for
all algorithms because of the linear relationship between
CPU utilization and energy consumed by the server. Also,
when the upper-threshold is increased,theprobabilitythata
host would be identified as overloaded is decreased, which
in turn decreases the number of containers selected to
migrate, which results in a lower number of VM creations.
On the other hand, the higher the upper-threshold, themore
likely the host will not have adequate resources to adapt to
fluctuations in container resourcerequirements,resultingin
more SLA violations. At all upper-threshold values, our
proposed algorithm performs better than all other
algorithms.

scenario 2
scenario 2
5.2.3 Scenario 3:
In this set of experiments, the upper threshold is set at
80%, and the lower threshold will be varied. Also, the
number of containers and their workload traces are set
according to day 1 of Table-4. As shown in Charts [9-12],
decreasing the lower-threshold increases the energy
consumption metric since more hosts will be active, and the
number of migrations will decrease becausefewerhosts will
be identified as underloaded. A lower container migration
rate results in fewer VMsbeingcreated.Althoughthesmaller
number of created VMs when we decrease the lower-
threshold, the size of these VMs will be bigger since we have
more resources to allocate and the possibilitythatthese VMs
will not get their required resources in the future is high,
which results in high SLA violations. The findingsreveal that
the performance of our proposed method outperforms that
of competing algorithms across all metrics and lower-
threshold values.

scenario 3
scenario 3
6. CONCLUSION AND FUTURE WORK
Despite the growing popularity of Container as a Service
(CaaS), the energy efficiency of resource management
algorithms in this service paradigm has received little
attention.
In this paper, we introduced A novel host selection
algorithm for container consolidationbytakingadvantageof
particle swarm optimization and energy efficiency of hosts.
Three sets of simulation tests were performed to compare
the performance of our approach with existing algorithms.
Results show that our proposed method outperforms all
competitive algorithms regardingenergyconsumption,total
number of migrations, SLAV, and number of created VMs.
As for future work, we will improve our algorithm to
make it aware of operating system type as a new constraint
for the problem, and will extend our work to solve other
container consolidation sub-problems.
REFERENCES
[1] B. Tan, H. Ma, Y. Mei, and M. J. I. T. o. C. C. Zhang, "A
cooperative coevolution genetic programming hyper-
heuristic approach for on-line resource allocation in
container-based clouds," 2020.
[2] F. Paraiso, S. Challita, Y. Al-Dhuraibi, and P. Merle,
"Model-driven management of docker containers," in 2016
IEEE 9th International Conference on cloud Computing
(CLOUD), 2016, pp. 718-725: IEEE.
[3] O. Smimite and K. J. a. p. a. Afdel, "Container placement
and migration on cloud system," 2020.
[4] C. Zhang, Y. Wang, H. Wu, and H. J. I. A. Guo, "An energy-
aware host resource management framework for two-tier
virtualized cloud data centers," vol. 9, pp. 3526-3544, 2020.
[5] B. K. Sovacool, C. G. Monyei, P. J. R. Upham, and S. E.
Reviews, "Making the internet globally sustainable:
Technical and policy options for improved energy
management, governance and community acceptance of
Nordic datacenters," vol. 154, p. 111793, 2022.
[6] E. Masanet, A. Shehabi, N. Lei, S. Smith, and J. J. S.Koomey,
"Recalibrating global data center energy-use estimates,"vol.
367, no. 6481, pp. 984-986, 2020.
[7] M. Koot and F. J. A. E. Wijnhoven, "Usage impact on data
center electricity needs: A system dynamic forecasting
model," vol. 291, p. 116798, 2021.
[8] S. F. Piraghaj, A. V. Dastjerdi, R. N. Calheiros, and R.
Buyya, "A framework and algorithm for energy efficient
container consolidation in cloud data centers," in 2015 IEEE
International Conference on Data Science and Data Intensive
Systems, 2015, pp. 368-375: IEEE.
[9] P. Mishra, S. Bhatnagar, and A. Katal, "Cloud Container
Placement Policies: A Study and Comparison," in

International Conference on Computer Networks and
Inventive Communication Technologies, 2019, pp. 513-524:
Springer
[10] W. A. Hanafy, A. E. Mohamed, and S. A. Salem, "Novel
selection policies for container-based cloud deployment
models," in 2017 13th International Computer Engineering
Conference (ICENCO), 2017, pp. 237-242: IEEE.
[11] Z. Dong, W. Zhuang, and R. Rojas-Cessa, "Energy-aware
scheduling schemes for cloud data centers on google trace
data," in 2014 IEEE Online Conference on Green
Communications (OnlineGreencomm), 2014, pp. 1-6: IEEE.
[12] M. K. Hussein, M. H. Mousa, and M. A. J. J. o. C. C. Alqarni,
"A placement architecture for a container as a service(CaaS)
in a cloud environment," vol. 8, no. 1, pp. 1-15, 2019.
[13] D. Kusic, J. O. Kephart, J. E. Hanson, N. Kandasamy, and
G. J. C. c. Jiang, "Power and performance management of
virtualized computing environmentsvia lookaheadcontrol,"
vol. 12, no. 1, pp. 1-15, 2009.
[14] A. Beloglazov, R. J. C. Buyya, C. Practice, and Experience,
"Optimal online deterministic algorithms and adaptive
heuristics for energy and performance efficient dynamic
consolidation of virtual machines in cloud data centers,"vol.
24, no. 13, pp. 1397-1420, 2012.
[15] R. Eberhart and J. Kennedy, "A new optimizer using
particle swarm theory," in MHS'95. Proceedings of the sixth
international symposium on micro machine and human
science, 1995, pp. 39-43: Ieee.
[16] F. Ebadifard, S. M. J. C. Babamir, C. Practice, and
Experience, "A PSO‐based task scheduling algorithm
improved using a load‐balancing technique for the cloud
computing environment," vol. 30, no. 12, p. e4368, 2018.
[17] D. Wang, D. Tan, and L. J. S. C. Liu, "Particle swarm
optimization algorithm: an overview," vol. 22, no. 2,pp.387-
408, 2018.
[18] S. Alsaidy, A. Abbood, and M. Sahib, "Heuristic
initialization of PSO task scheduling algorithm in cloud
computing," Journal of King Saud University-Computer and
Information Sciences, 2020.
[19] F. F. Moges and S. L. J. J. o. C. C. Abebe, "Energy-aware
VM placement algorithms for the OpenStack Neat
consolidation framework," vol. 8, no. 1, pp. 1-14, 2019.
[20] M. Behzadian, S. K. Otaghsara, M. Yazdani, and J. J. E. S.
w. a. Ignatius, "A state-of the-art survey of TOPSIS
applications," vol. 39, no. 17, pp. 13051-13069, 2012.
[21] F. Luo, Y. Yuan, W. Ding, and H. Lu, "An improved
particle swarm optimization algorithm based on adaptive
weight for task scheduling in cloud computing," in
Proceedings of the 2nd International ConferenceonComputer
Science and Application Engineering, 2018, pp. 1-5.
[22] S. F. Piraghaj, A. V. Dastjerdi, R. N. Calheiros, R. J. S. P.
Buyya, and Experience, "ContainerCloudSim: An
environment for modeling and simulation of containers in
cloud data centers," vol. 47, no. 4, pp. 505-521, 2017.
[23] M. Mao and M. Humphrey, "A performance study on the
vm startup time in the cloud," in 2012 IEEE Fifth
International Conference on Cloud Computing,2012,pp.423-
430: IEEE.

A Host Selection Algorithm for Dynamic Container Consolidation in Cloud Data Centres using Particle Swarm Optimization

More Related Content

Similar to A Host Selection Algorithm for Dynamic Container Consolidation in Cloud Data Centres using Particle Swarm Optimization (20)

More from IRJET Journal (20)

Recently uploaded (20)

A Host Selection Algorithm for Dynamic Container Consolidation in Cloud Data Centres using Particle Swarm Optimization