Hindawi

Computational Intelligence and Neuroscience

Volume 2022, Article ID 7873131, 10 pages
https://doi.org/10.1155/2022/7873131

Research Article
Virtual Machine Resource Allocation Optimization in Cloud
Computing Based on Multiobjective Genetic Algorithm

Feng Shi1 and Jingna Lin 2

1
Department of Computer Science and Technology, Taiyuan University, Taiyuan 030032, China
2
Shanxi Vocational University of Engineering Science and Technology, Taiyuan 030031, China

Correspondence should be addressed to Jingna Lin; [email protected]

Received 26 November 2021; Revised 17 January 2022; Accepted 4 February 2022; Published 10 March 2022

Academic Editor: Qiangyi Li

Copyright © 2022 Feng Shi and Jingna Lin. This is an open access article distributed under the Creative Commons Attribution
License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.
Cloud computing is an important milestone in the development of distributed computing as a commercial implementation, and it
has good prospects. Infrastructure as a service (IaaS) is an important service mode in cloud computing. It combines massive
resources scattered in different spaces into a unified resource pool by means of virtualization technology, facilitating the unified
management and use of resources. In IaaS mode, all resources are provided in the form of virtual machines (VM). To achieve
efficient resource utilization, reduce users’ costs, and save users’ computing time, VM allocation must be optimized. This paper
proposes a new multiobjective optimization method of dynamic resource allocation for multivirtual machine distribution
stability. Combining the current state and future predicted data of each application load, the cost of virtual machine relocation and
the stability of new virtual machine placement state are considered comprehensively. A multiobjective optimization genetic
algorithm (MOGANS) was designed to solve the problem. The simulation results show that compared with the genetic algorithm
(GA-NN) for energy saving and multivirtual machine redistribution overhead, the virtual machine distribution method obtained
by MOGANS has a longer stability time. Aiming at this shortage, this paper proposes a multiobjective optimization dynamic
resource allocation method (MOGA-C) based on MOEA/D for virtual machine distribution. It is illustrated by experimental
simulation that moGA-D can converge faster and obtain similar multiobjective optimization results at the same calculation scale.

1. Introduction included in the national science and technology strategy of

many countries [5–7].
Cloud computing is an emerging technology in the field of In cloud computing applications, Infrastructure as a
computers. Although it has been applied, its development is Service (IaaS) is an important service model, through which
not perfect. Many technical standards have not yet formed, cloud providers can provide massive computing resources to
and are still in the period of growth. As long as we grasp the cloud users through the Internet. In addition, the price
rare opportunity, we can occupy a place in the future cloud fluctuation and the uncertainty of demand and other factors
computing field [1–3]. In this critical period of the formation make this problem more complicated. Therefore, in order to
and development of new technology, whoever holds the meet the resource usage requirements and minimize user
commanding heights of technology will hold the initiative of expenses, the VM allocation scheme must be optimized. This
future technology strategy [4]. In addition to commercial problem is called VM allocation optimization (shown as
applications, many excellent characteristics of cloud com- Figure 1) [8, 9].
puting make it an essential aspect in military education, The emergence of cloud computing platform virtuali-
meteorology, astronomy, and other fields. Therefore, the zation technology brings new opportunities to realize energy
development of cloud computing technology has been saving in the cloud computing environment. The use of
2 Computational Intelligence and Neuroscience

Demand SaaS
Cloud brokering

Computer Result Optimization SaaS PaaS Software layer

The user terminal algorithm Virtual machine
IaaS Software layer Platform layer

Infrastructure layer
Cloud providers Cloud providers
Cloud providers Virtual resource pool

Physical Physical Physical Physical

resources resources resources resources
Figure 1: VM resource allocation in cloud computing.
Figure 2: Framework diagram of the cloud computing service
model.
virtualization technology enables multiple users to share the
computing resources pool according to the actual demand. 2. Related Works
The emergence of this technology has greatly improved the
resources utilization. With the emergence of a shared re- With the development of cloud computing, people find that
source pool, the number of servers in the cloud environment although cloud computing brings convenience, its energy
has reduced, thereby increasing the income of the cloud consumption is also huge compared with the traditional service
service provider [2, 10, 11]. And, resource allocation mode. Now, some suppliers in order to meet user require-
problems ultimately boils down to multiple application ments, which can respond to dynamic changes in user needs.
mapping relations between the virtual machine and the However, in the actual environment, this will cause a great
server, through reasonable allocation, adjusting the distri- waste of resources and increase unnecessary energy con-
bution of application virtual machines on the physical nodes, sumption. With the change of user demand, the application
making full use of the service condition of the server idle load request resources will change accordingly. Therefore, how
resources, so as to reduce the amount of activation server, to reduce energy consumption while ensuring service quality is
and thereby achieve the purpose of reducing energy con- a problem to be solved. In a cloud computing environment, the
sumption [12]. At present, there are three basic service optimal VM resource distribution scheme can effectively save
modes in the application of cloud computing technology: energy by properly deploying VM resources on physical nodes
Software as a Service (SaaS), Platform as a service (PaaS), and [15, 16]. Virtual machine deployment strategy is mainly
Infrastructure as a service (IaaS), as shown in Figure 2. Each considering the distribution and use of computing resources in
service mode corresponds to certain system levels. data center, so as to select the best distribution scheme,
At present, many energy-saving resource scheduling according to the distribution plan to make sure the virtual
methods of cloud platform focus on how to reduce the number machine and the mapping relations between physical nodes. If
of active servers to achieve energy saving and green computing. the new deployment scheme satisfies the constraints of the
However, the above resource scheduling schemes often ignore collection, then optimal allocation can be achieved. However,
the dynamic change of application load caused by the change of the allocation of multiple virtual machines to multiple physical
user demand in the cloud environment, and they also fail to nodes proved to be a problem, namely NP problem. Therefore,
consider the stability of application load distribution on each which strategy is adopted to realize the placement of virtual
physical node. Although the distribution of application VMS machines to achieve the purpose of energy saving has always
among physical nodes in the current state makes fewer physical been a hot research topic in the field of cloud computing
nodes active, the change of user requirements will lead to [17, 18].
changes in the load of each application; in this case, the dy- At present, the research of cloud computing is still in the
namic change of the required load will lead to resource hot initial stage. In terms of the development direction of the
spots on these physical nodes in the future with the change of future mainstream cloud computing industry, there are also
the application load. The status of these physical nodes may two different camps among the big companies that master
reappear in the near future, which will directly trigger a new the core technology of cloud computing. Emerging com-
round of dynamic resource configuration requirements [13]. panies represented by Google and Amazon are pushing the
To solve the above problems and to put forward a new thinking Internet cloud first, while traditional IT enterprises repre-
of physical nodes on the stability of the energy-saving resource sented by IBM and Microsoft are pushing the IT cloud. In
scheduling method, how to find a virtual machine distribution addition, due to commercial interests, many high-end cloud
that activates the stable state of each physical node and how to computing technologies are not disclosed as business secrets
reduce the energy consumption of virtual machines from the of enterprises [19]. Google is recognized as the leader in
old state to the new state are urgent and necessary [14]. cloud computing technology, occupying a great
Computational Intelligence and Neuroscience 3

technological advantage, but many of its technologies are not of resources on a single virtual machine. It is used to improve
disclosed. At present, scholars mainly study cloud com- the utilization of vm physical resources and meet users’
puting in two ways: one is based on part of the macroscopic computing requirements by dividing physical computing
technical data disclosed by Google and combined with resources [26]. In addition, most algorithms are put forward
traditional parallel computing-based grid computing vir- in specific demand scenarios, without considering various
tualization technology [20]. The other is based on the open uncertainties in practical applications.
source project Hadoop for communication and application From the above analysis, we know that the above
of different ideas and different application environments, methods have studied virtual machine resource allocation
resulting in the development of many different branches of optimization in cloud computing to some extent, but some
technology. Therefore, related research on cloud computing problem still exists. On the other hand, no scholar has
is still in a stage of contention, and the technical system has applied multiobjective genetic algorithm to this field till now,
not yet been formed. so the research here still draws a blank, which has great
As mentioned above, the widespread interest in cloud theoretical research and practical application value.
computing stems from its successful commercial applica- The contributions of this paper are as follows:
tion, so the earliest cloud computing research institutes are
(1) The proposed evolutionary algorithm provides a new
mainly concentrated in several major Western IT technology
solution idea for solving NP problems, and the
companies. However, with the gradual increase of the in-
emergence of multiobjective optimization algorithm
fluence of cloud computing, its many excellent character-
brings a new solution idea for complex multi-
istics make people see its application in other fields such as
objective optimization practical problems.
military scientific research. Cloud computing technology
began to receive attention from various countries and has (2) Because of its good development prospects, the
become an important aspect for countries to compete for the multiobjective optimization problem has become a
commanding heights of future science and technology [2]. research hotspot in the academic world. In this
In China, in recent years, due to the strong support of the paper, MOGANS- and MOEA/D-based multi-
state, as well as the extensive promotion of business plans, objective optimization scheduling schemes are
cloud computing has gradually become a popular research proposed according to the solution idea of multi-
content, has achieved rapid development, and its influence is objective optimization.
also increasing. In 2010, Beijing successfully hosted the This paper consists of five parts. The first and second
world cloud computing conference. Lenovo, Sohu, Baidu, parts give the research status and background. The third part
and other domestic companies have also begun to develop is the MOEA/D-based virtual machine resource allocation in
business based on cloud computing. In addition, various cloud computing. The fourth part shows the experimental
domestic universities began to set up the relevant research results and analysis. The experimental results of this paper
direction and discipline curriculum. Cloud computing re- are introduced, compared, and analyzed with relevant
search revealed a strong momentum, and has great devel- comparison algorithms followed. Finally, the fifth part
opment potential. However, there is still a big gap between summarizes the full paper.
China and western countries, and more investment and
policy planning are needed [21].
Resource management and allocation is a key problem in 3. MOEA/D-Based Virtual Machine Resource
cloud computing. In the IaaS mode, resource allocation is a Allocation in Cloud Computing
virtual machine allocation problem. Virtual machine allo-
cation, like other cloud computing technologies, is in a state 3.1. Conventional Particle Swarm Optimization Algorithm.
of constant discovery. Literature [22] proposed an algorithm Conventional particle swarm optimization (PSO) is an in-
based on cloud agent, which aims to provide virtual machine telligent bionic method, which was jointly proposed by
allocation scheme for physical server. Literature [23] pro- social psychologists Kennedy and Dr. Eberhart in 1995. It
posed an MMS virtual machine allocation algorithm based has become the mainstream intelligent optimization pro-
on min-max and shares features provided by the hypervisor. cessing method at present and has been comprehensively
However, these algorithms do not take into account the applied in the academic and industrial circles. Its optimi-
uncertain factors of demand and the minimization condi- zation performance is mainly reflected in convergence
tions of price. A dynamic integration mechanism based on [27, 28].
constraint programming is proposed in literature [24]. In 2
χ � 􏽱���������������
√����� 2 , l � c1 + c2 , l > 4, (1)
literature [25], demand and price factors are considered, but 􏼐2 − l − l2 − 4l 􏼑
response time and other issues are not taken into account.
These algorithms are based on different application sce- where χ is the convergence factor of particle swarm. For the
narios, and have their own characteristics. The existing al- particle in each iteration, the particle changes its velocity and
gorithm, however, is mainly to study the internal allocation position according to the following formula:
4 Computational Intelligence and Neuroscience

vk+1 k k k
(2) means of aggregation method, and the common aggregation
id � vid + χ 􏽨c1 r1 􏼐Fid − Qid 􏼑 + c1 r2 􏼐Hgd − Qid 􏼑􏽩,
methods include Chebyshev method, weighted sum method,
and so on. In this paper, the aggregation function is
Qk+1 k k+1
id � Qid + vid , (3) implemented by Chebyshev:

Fid ⩾ Hgd n, (4) x 􏼌􏼌 􏼌􏼌

minimize gte 􏼠 , z∗ 􏼡 � max􏽮λi 􏼌􏼌fi (x) − z∗i 􏼌􏼌􏽯,
y 1≤i≤x
(5)
where d � i � 1, 2, . . . , m, k is the number of iterations, and
c1 , c2 are decay acceleration factors.
subject to x ∈ Ωn.
The (PSO) method here is determined by cloud user U
and cloud VM, that is, the utility function of cloud user U is
used to obtain the optimal position of particle search, and And 􏽐m ∗ ∗ ∗ T
i�1 λi � 1(∀i, λi ≥ 0), z∗ � (z1 , z1 , . . . zm ) are the
∗
VM is used to obtain the optimal position of overall particle reference points. And, z is the optimal value of each target
swarm search, that is, one particle represents one cloud user, value in all solutions.
The whole particle represents the virtual machine resources
in the cloud computing environment of the whole field,
which corresponds to the particle swarm optimization 3.3. Implementation of MOGA-D Resource Optimization
method. In this way, not only can each cloud user get the Scheme. The main characteristics of MOEA/D algorithm are
maximum benefit but also can make the whole cloud as follows:
computing environment VM get global optimization. (1) MOEA/D algorithm does not transform the multi-
At present, many scholars have put forward many ex- target values into a single-target progressive solution
cellent algorithms for solving multiobjective problems, through the traditional method but decomposes the
among which the representative ones are NSGA-II, PSO, etc. multitarget space, thus transforming the multitarget
However, these algorithms do not adopt decomposition solution problem into the optimization problem of
thought in solving the MOP problem. These algorithms solving the sub-target space solution, and the sub-
mainly deal with the MOP problem through some algo- target space solution can be transformed into a
rithms, and finally integrate the MOP into a whole. Then, it single-target solution problem according to the
is solved by single objective optimization problem. But the vector of the solution space.
downside of this is that it becomes very difficult to allocate
(2) An excellent core idea of MOEA/D is to update the
fitness. However, many of the target values are independent,
new solution by exchanging information of neigh-
so there are many different cases of how to judge the su-
periority of a solution, and there are many other cases of the borhood solutions. This strategy greatly speeds up
relationship between the solutions besides being dominant. the search of understanding space, and has lower
Therefore, it is difficult to specify suitable fitness for these time and space complexity compared with some
solutions to compare the advantages and disadvantages of other multiobjective optimization solutions
these solutions in order to perform selection operations. In Because the resource scheduling problem in cloud
view of this problem, the decomposition idea is put forward, computing environment can finally be transformed into
and the decomposition strategy idea is to transform multiple solving a multiobjective optimization problem through
objectives into solving the problem of each single objective, evolutionary algorithm, the resource scheduling scheme
mainly by solving the problem with multiple objectives. It is based on MOEA/D algorithm is proved by experiments to
decomposed into a single solution optimization problem. By reduce the time complexity of the algorithm compared with
optimizing the solution of each sub-objective space, the MOGANS, so as to solve the shortage of MOGANS in
solution of the whole solution set can be optimized to computing performance.
achieve the purpose of multiobjective optimization, and The algorithm mainly consists of four parts: target space
finally the global optimization solution based on Pareto is decomposition, population classification, crossover, and
obtained. Therefore, MOEA/D comes into being. mutation. The core of the multiobjective optimization
problem is to decompose the solution space into N sub-
target space, where OMEGA represents the target space, so
3.2. Decomposition Strategy of MOEA/D Algorithm. The omega can be decomposed into a sub-target space composed
decomposition idea of MOEA/D is that, unlike other of (Ω1 , Ω2 , . . . , ΩN ). Individuals in the population can be
multiobjective optimization algorithms, the multiobjective mapped into different subspaces according to the generated
problem is not processed as a whole through integration. sub-target space, so as to achieve the classification of the
Instead, it transforms the multiobjective solution into population. The vector composed of each target value of
solving the single-objective optimization problem in N individuals in the initial population is mapped into N sub-
subspace through decomposition of the solution space. target space to ensure that each sub-target space has a so-
Among them, the solution in the subspace transforms a lution. The division of target space and population classi-
multiobjective problem into a single-objective problem by fication can be realized by the following formula:
Computational Intelligence and Neuroscience 5

3.4. Optimization of VM Allocation Problems. In the opti-

pi � 􏼨x|x ∈ POP, Δ􏼐F(x), λi 􏼑 � max 􏽮Δ􏼐F(x), λj 􏼑􏽯􏼩, (6) mization process, two schemes are proposed for users to
1≤j≤N
choose. One is the scheme without considering the time
λi ∗ (F(x) − Z)T factor, which only optimizes the cost and minimizes it while
Δ􏼐F(x), λi 􏼑 � �� i �� , i � 1, . . . , m, (7) ignoring the time factor. It is suitable for users who do not
��λ �� ∗ ‖F(x) − Z‖
care about time. Another kind is considering two factors of
cost and time; the scheme firstly sets up a relatively small
default time threshold. If the user has higher requirements,
Ωi � 􏼨F(x)|x ∈ Ω, Δ􏼐F(x), λi 􏼑 � max 􏽮Δ􏼐F(x), λj 􏼑􏽯􏼩,
1≤j≤N processing time can be dedicated to provide users with a
(8) time threshold that can be set up, after the cost optimization.
The linear programming method is adopted, with the
where Z is a reference point and Δ(F(x), λi ) is the optimal minimum cost as the optimization objective, and the pro-
solution of each target vector. It is also the cosine of the cessing time of resource price, virtual machine quantity, and
angle. In this way, POP population is divided into N classes. supplier’s supply capacity as the optimization conditions.
So there is a solution in every subspace and you have a Formula (10) represents the optimization goal, namely,
variety of systems that help maintain the solution. Then, the the overall cost; formula (11) indicates that the number of
multiobjective solving problem can be transformed into a VMS obtained from the supplier in each usage phase does
single-objective optimization problem by the following not exceed the number of VMS reserved. Equations
aggregation function, which is in the form of the following: (12)–(15) indicate that the total amount of basic resources
􏼌􏼌 􏼌 obtained from each supplier should not exceed the upper
∗ 􏼌⎪
te ∗ ⎨ 􏼌􏼌􏼌fj (x) − zj 􏼌􏼌􏼌⎫
⎪
⎧ ⎬ limit of the maximum resources that the supplier can
minimize g x|y, Z 􏼁 � max ⎪ ⎪ . (9)
1≤j≤m⎩ λj ⎭ provide. Formula (15) indicates that the number of VMS in
each stage of optimization is a nonnegative integer:
And the letters in formula (9) are the same as (6)–(8).

Minimize:
􏽘 􏽘 c[e] [e] [u] [u] [a] [a] (10)
j,i Xj,i + 􏽘 􏽘 􏽘 p(D)􏼐cj,i Xj,i (D) + cj,i Xj,i (D)􏼑,
pj ∈P vi ∈V D∈D pj ∈P vi ∈V

Subject to:
(11)
X(u) (e)
j,i (D) ≤ Xj,i , vi ∈ V, pi ∈ P, D ∈ D,

􏽘 n(a) [u] [a] (a)

j,i 􏼐Xj,i (D) + Xj,i (D)􏼑 ≤ sj , D ∈ D, (12)
vi ∈V

􏽘 n(b) [u] [a] (b)

j,i 􏼐Xj,i (D) + Xj,i (D)􏼑 ≤ sj , D ∈ D, (13)
vi ∈V

􏽘 n(c) [u] [a] (c)

j,i 􏼐Xj,i (D) + Xj,i (D)􏼑 ≤ sj , D ∈ D, (14)
vi ∈V

X(e) [u] [a]

j,i , Xj,i (D), Xj,i (D) ∈ 􏼈int eger􏼉, vi ∈ V, pi ∈ P, D ∈ Dn. (15)

4. Experimental Results and Analysis the maintenance cost of the cloud server into account in the
virtual machine configuration stage, the cost quotation of
4.1. Introduction to Experimental Environment and Data Set. the cloud server is set according to the pricing of Amazon
It includes a control node and four computing nodes [15] EC2 virtual machine, as shown in Table 1.
but due to limited hardware resources, this paper cannot
provide a better cloud environment in the authentication
algorithm, so a data center composed of 17 physical units is 4.2. Experimental Results Analysis. Figure 3 shows the ex-
simulated. P1 serves as the controller node, and the other perimental results of the difference in total revenue between
four compute nodes p2-P5 provide resources for users the algorithm in this chapter and the utility optimization
through virtualization [25]. As the algorithm in this paper algorithm based on non-cooperative game and resource
optimizes the revenue of the cloud service provider and takes reserve calculation method in cloud service providers. As the
6 Computational Intelligence and Neuroscience

time period increases, the total revenue of single bit time is Table 1: Cloud data center parameters.
equal to that of the comparison algorithm. Node CPU (MIPS) CPU cores Memory (GB) Bandwidth (M)
The main reason is that the algorithm makes a reasonable
P1 3000 2 4 100
planning of resources in the virtual machine configuration P2 2000 1 2 100
stage in this chapter, so as to maximize the revenue of cloud P3 1800 2 1 100
services and optimize resource allocation. However, the P4 2700 1 1 100
utility optimization algorithm based on cooperative game P5 2300 2 2 100
cannot guarantee the optimal game. With the increase of
time cycle, both the algorithm in this article and the game-
based utility optimization algorithm maintain good returns 40
under the effect of utility function optimization, while the

Total revenue (Ten thousand yuan)

35
resource reservation allocation algorithm starts to allocate
30
on-demand resources with higher prices.
Through the simulation experiment, the algorithm in 25
this article is compared with the noncooperative game-based
20
utility optimization algorithm, resource reservation algo-
rithm, and user utility on the unit revenue of cloud data 15
center. 0.4 is selected as the interval, and the statistical utility
10
value is between [1, 5]. The experimental results are shown in
Figures 4 and 5, with user utility on the horizontal axis and 5
unit revenue on the vertical axis Revenue represents the
0
revenue generated per request processed. 0 10 20 30 40 50 60 70 80
The experimental results show that with the increase of
Time period (Day)
user utility, unit income also increases gradually, but the
increasing trend gradually becomes slow, indicating that the Our algorithm
utility function has a certain promoting effect on the increase
Game utility based optimization
of unit income, but this promoting effect will begin to fail
after reaching a certain peak. At the stage of stimulative Reservation allocation algorithm
effect, the increase of yield per unit area is manifested. After Figure 3: Total revenue comparison of cloud service providers.
reaching the peak, the stimulative failure is mainly related to
the resource itself. Therefore, resources will not increase the tasks. The reason is that the algorithm in this article classifies
unit infinite revenue of the non-cooperative game optimi- user tasks and physical machines, and computationally in-
zation algorithm with the increase of the utility function, the tensive users are in the utility function during resource al-
profit per unit increases steadily at first and then falls location. In this way, computing resources can be fully utilized
sharply, indicating that on-demand resources play a decisive to improve computing resource utilization.
role in the allocation process. The experimental results of memory resource utilization
In addition, the user utility difference between the al- performance of OpenStack random scheduling algorithm
gorithm in this article and the virtual machine instance and the algorithm in this article are compared as shown in
based on noncooperative game utility optimization algo- Figure 8. The results in the Figure show that with the in-
rithm and resource reservation algorithm of the same user crease of the number of user tasks, resource utilization will
request is also compared as shown in Figure 6. The ex- increase steadily. The algorithm has a higher utilization of
perimental results of actual satisfaction and maximum computing resources, but the difference between the two is
satisfaction ratio of users show that there is no significant not very large. The reason is that the host filtering in the
difference between the proposed algorithm and other al- Nova component of OpenStack gives priority to memory
gorithms according to different users. Although the utility resource allocation and takes memory resource as the
value differs greatly, the proposed algorithm is still effective measurement standard. Therefore, the fluctuation of
on the whole. Combined with the analysis of the previous memory resource in the allocation process is smaller than
experimental results, it is proved that the algorithm in this that of the computing resource. Memory resource utilization
article has higher revenue and resource utilization of cloud has been improved to some extent.
service providers and better overall performance without Under the same initial conditions (the same initial state
affecting user utility. of VM distribution and the same load change of each ap-
The experiment compares the performance of OpenStack plication VM), the four algorithms are run to obtain the new
random scheduling algorithm and the algorithm in this article distribution states of each VM, respectively. The number of
in computing resource CPU utilization, as shown in Figure 7. migration times in the stable time and the number of idle
The horizontal axis represents the number of random user physical nodes of these new distribution states are shown in
tasks, and the vertical axis represents the CPU resource Tables 2 and 3.
utilization. This kind of performance results from the large As can be seen in Tables 2 and 3, MOGANS makes a
amount of idle resources in the initial stage of resource al- good balance between the stability of virtual machine dis-
location, which are used by users with the increase of user tribution and the cost of virtual machine migration required
Computational Intelligence and Neuroscience 7

1.2 0.9

Normalized value of user utility function

Unit returns (Ten thousand yuan)

1.1 0.8
1
0.7
0.9
0.6
0.8
0.7 0.5
0.6 0.4
0.5 0.3
0.4
0.2
0.3
0.1
0.2
1 1.5 2 2.5 3 3.5 4 4.5 5 0

User 1

User 2

User 3

User 4

User 5

User 6

User 7

User 8

User 9
User utility

Our algorithm
Game utility based optimization
Game utility based optimization
Reservation allocation algorithm
Reservation allocation algorithm
Our algorithm
Figure 4: Comparison of unit revenue of different algorithms.
Figure 6: Comparison of utility values of different users under
different algorithms.
0.85
0.8 0.8
0.75
Vm instance usage (%)

0.7
0.7
Computing Resource Usage

0.65 0.6
0.6 0.5
0.55
0.4
0.5
0.45 0.3
0.4
0.2
0.35
1 1.5 2 2.5 3 3.5 4 4.5 5 0.1
User utility 0
30 60 90 120 150 180
Our algorithm
Number of user tasks
Game utility based optimization
Reservation allocation algorithm Random algorithm
Our algorithm
Figure 5: Comparison of VM instance usage.
Figure 7: Comparison of computing resource usage.

by the old and new states, ensuring that the new multivirtual
machine distribution state not only has a long stability time least active physical nodes, and the remaining number of free
but also requires fewer virtual machine migration times physical nodes is the largest, which is about 8 times the
from the current virtual machine distribution state to the number of free physical nodes of MOGANS. This indicates
new virtual machine distribution state. Although GA-N can that the new virtual machine distribution state obtained by
change to a new VM distribution only after a minimum GA-NN is more energy saving, but this energy saving state
number of VM migrations, the new VM distribution can cannot be maintained for a long time. Once the distribution
only maintain stability for a short time. Once the VM state becomes unstable, additional new dynamic resource
distribution becomes unstable, new dynamic resource allocation cost will be added, and the energy saving effect will
configuration requests will be triggered, resulting in a new be greatly reduced.
round of VM migration. Ga-NN takes green calculation into In order to compare the computational overhead be-
account, and the new virtual machine distribution state tween MOGANS algorithm and MOGAD algorithm, the
obtained uses the least active physical nodes, and the experimental data tested in this section are basically the same
remaining number of free physical nodes is the largest, as the parameters as the former experiment, including two
which is about 8 times the number of free physical nodes of variables: population size and population total algebra. The
MOGANS. Ga-NN takes green calculation into account, and values of population size and population total algebra are
the new virtual machine distribution state obtained uses the constantly changed to measure the running time of the
8 Computational Intelligence and Neuroscience

0.8

0.7

Memory resource usage

0.6

0.5

0.4

0.3

0.2

0.1

0
30 60 90 120 150 180
Number of user tasks

Random algorithm
Our algorithm
Figure 8: Comparison of memory resource usage.

Table 2: Algorithms’ performance (16 physical nodes, 42 virtual machines).

Algorithms Number of migration Stable time Number of idle physical nodes
MOGANS 21 15 1.6
GA-S 6.8 3.8 0.5
GA-N 9.2 1.2 3.7
GA-NN 22.5 0.9 5.6

Table 3: Algorithms’ performance (32 physical nodes, 84 virtual machines).

Algorithms Number of migration Stable time Number of idle physical nodes
MOGANS 68 15.2 1.4
GA-S 81.5 13.8 0.9
GA-N 19.2 1.2 5.7
GA-NN 72.5 0.8 15.6

2000
1500
Time (s)

1000

500
800
600 e
0 400 siz
3000 2500 t i on
2000 1500 200 ul a
1000 500 op
0 0 ep
Number of executions Th

MOGANS
MOGA-D
Figure 9: Comparison of running time cost between MOAGNS and MOGA-D.

algorithm, respectively. The experimental results are shown MOGANS and MOGA-D algorithm has little difference.
in Figure 9: When the amount of data increases sharply, the time cost of
It can be seen from Figure 9 results that when the MOGA-D algorithm increases little. Therefore, when the
amount of data is small, the execution time cost of amount of data is large, the calculation cost of MOGA-D
Computational Intelligence and Neuroscience 9

algorithm is obviously superior as MOGANS has greatly computing environment,” ACM Transactions on Multimedia
improved the computational performance of the former. Computing, Communications, and Applications, vol. 16, no. 3s,
pp. 1–19, 2020.
5. Conclusions [8] N. Rasouli, R. Razavi, and H. R. Faragardi, “EPBLA: energy-
efficient consolidation of virtual machines using learning
Aiming at the deficiency of some current resource sched- automata in cloud data centers,” Cluster Computing, vol. 23,
uling schemes of energy-saving resource scheduling in the no. 4, pp. 3013–3027, 2020.
current cloud computing environment, this paper proposes [9] V. Bogatyrev and A. Derkach, “Evaluation of a cyber-physical
a multiobjective optimization scheme MOGANS based on computing system with migration of virtual machines during
continuous computing,” Computers, vol. 9, no. 2, p. 42, 2020.
NSGA-II multiobjective optimization.
[10] C. Li, C. Wang, and Y. Luo, “An efficient scheduling opti-
However, MOGANS has insufficient computing per-
mization strategy for improving consistency maintenance in
formance, especially when the amount of data to be cal- edge cloud environment,” The Journal of Supercomputing,
culated increases, the performance of MOGANS is not pp. 1–28, 2020.
satisfactory. After the test of the same amount of compu- [11] M. M. Shahriar Maswood, M. R. Rahman, A. G. Alharbi, and
tation, the energy-saving resource scheduling scheme of the D. Medhi, “A novel strategy to achieve bandwidth cost re-
new opportunistic MOEA/D algorithm, MOGA-D, per- duction and load balancing in a cooperative three-layer fog-
forms well. Compared with MOGANS, the computational cloud computing environment,” IEEE Access, vol. 8,
performance is significantly improved with the same cal- pp. 113737–113750, 2020.
culation scale. The similar multiobjective optimization re- [12] S. Lou, J. He, H. Li et al., “Current knowledge and future
sults can also be obtained. directions for improving s quality and reducing energy
consumption in conservation fields,” Agriculture, vol. 11,
Data Availability no. 7, p. 575, 2021.
[13] S. Wilson and K. Platts, “The role of resource configuration on
The data used to support the findings of this study are mix flexibility requirements,” Production Planning & Control,
available from the corresponding author upon request. vol. 20, no. 8, pp. 769–784, 2009.
[14] A. Khudoyberdiev, S. Ahmad, I. Ullah, and D. Kim, “An
optimization scheme based on fuzzy logic control for efficient
Conflicts of Interest energy consumption in hydroponics environment,” Energies,
The authors declare that they have no conflicts of interest. vol. 13, no. 2, p. 289, 2020.
[15] B. Baron, M. Campista, P. Spathis et al., “Virtualizing ve-
hicular node resources: f,” Vehicular Communications, vol. 4,
Acknowledgments pp. 39–46, 2016.
This study was supported by Shanxi Province Educational [16] S. Rawas, “Energy, network, and application-aware virtual
machine placement model in SDN-enabled large scale cloud
Science “Thirteenth Five-Year Plan” Project (GH-19136) and
data centers,” Multimedia Tools and Applications, vol. 80,
Shanxi Province Higher Education Reform and Innovation
no. 10, pp. 15541–15562, 2021.
Project (J2020380). [17] S. Namasudra, R. Chakraborty, S. Kadry, G. Manogaran, and
B. S. Rawal, “FAST: fast accessing scheme for data trans-
References mission in cloud computing,” Peer-to-Peer Networking and
Applications, vol. 14, no. 4, pp. 2430–2442, 2021.
[1] L. Wang, G. Von Laszewski, A. Younge et al., “Cloud com-
[18] S. Sun, C. Luo, and J. Chen, “A review of natural language
puting: a perspective study,” New Generation Computing,
processing techniques for opinion mining systems,” Infor-
vol. 28, no. 2, pp. 137–146, 2010.
mation Fusion, vol. 36, pp. 10–25, 2017.
[2] H. Shukur, S. Zeebaree, R. Zebari, D. Zeebaree, O. Ahmed,
[19] L. Wieclaw, V. Pasichnyk, and N. Kunanets, “Cloud com-
and A. Salih, “Cloud computing virtualization of resources
puting technologies in “smart city” projects[C]//2017 9th
allocation for distributed systems,” Journal of Applied Science
and Technology Trends, vol. 1, no. 3, pp. 98–105, 2020. IEEE International Conference on intelligent data acquisition
[3] S. Namasudra, “Data access control in the cloud computing and advanced computing systems: technology and applica-
environment for bioinformatics,” International Journal of Ap- tions (IDAACS),” IEEE, vol. 1, pp. 339–342, 2017.
plied Research in Bioinformatics, vol. 11, no. 1, pp. 40–50, 2021. [20] H. Shukur, S. R. M. Zeebaree, and A. Jamil Ahmed, “A state of
[4] K. Gai, J. Guo, L. Zhu, and S. Yu, “Blockchain meets cloud art survey for concurrent computation and clustering of
computing: a survey,” IEEE Communications Surveys & Tu- parallel computing for distributed systems,” Journal of Ap-
torials, vol. 22, no. 3, pp. 2009–2030, 2020. plied Science and Technology Trends, vol. 1, no. 4, pp. 148–154,
[5] H. Hassan, A. I. El-Desouky, A. Ibrahim, E.-S. M. El-Kenawy, 2020.
and R. Arnous, “Enhanced QoS-based model for trust as- [21] A. Kaur, P. Singh, and A. Nayyar, Fog Computing: Building a
sessment in cloud computing environment,” IEEE Access, Road to IoT with Fog Analytics, Springer, Singapore, 2020.
vol. 8, pp. 43752–43763, 2020. [22] M. Masdari, S. S. Nabavi, and V. Ahmadi, “An overview of
[6] Z. S. Ageed, R. K. Ibrahim, and M. A. M. Sadeeq, “Unified virtual machine placement schemes in cloud computing,”
ontology implementation of cloud computing for distributed Journal of Network and Computer Applications, vol. 66,
systems,” Current Journal of Applied Science and Technology, pp. 106–127, 2016.
pp. 82–97, 2020. [23] R. Meier, A. Rigo, and T. R. Gross, “Virtual machine design
[7] S. Namasudra, R. Chakraborty, and A. Majumder, “Securing for parallel dynamic programming languages,” Proceedings of
multimedia by using DNA-based encryption in the cloud the ACM on Programming Languages, vol. 2, pp. 1–25, 2018.
10 Computational Intelligence and Neuroscience

[24] M. E. Karim, M. M. S. Maswood, S. Das, and A. G. Alharbi,

“BHyPreC: a novel Bi-LSTM based hybrid recurrent neural
network model to predict the CPU workload of cloud virtual
machine,” IEEE Access, vol. 9, pp. 131476–131495, 2021.
[25] J. Parmar, P. Das, and S. M. Dave, “Study on demand and
characteristics of parking system in urban areas: a review,”
Journal of Traffic and Transportation Engineering, vol. 7, no. 1,
pp. 111–124, 2020.
[26] D. B. Christie III, H. M. Nemec, A. M. Scott et al., “Early
outcomes with utilization of tissue plasminogen activator in
COVID-19-associated respiratory distress: a series of five
cases,” Journal of Trauma and Acute Care Surgery, vol. 89,
no. 3, pp. 448–452, 2020.
[27] A. P. Piotrowski, J. J. Napiorkowski, and A. E. Piotrowska,
“Population size in particle swarm optimization,” Swarm and
Evolutionary Computation, vol. 58, Article ID 100718, 2020.
[28] X. Zhang, H. Liu, and L. Tu, “A modified particle swarm
optimization for multimodal multi-objective optimization,”
Engineering Applications of Artificial Intelligence, vol. 95,
2020.