Deployment Method of Virtual Machine Based on PSO

Algorithm Modified by Gauss Strategy

Shibiao Mu, Zhongming Li

Yiwu Industrial & Commercial College, Yiwu, 322000 China

{Email: 164297164@qq.com}

Abstract. A Gaussian correction scheme is proposed and a G-PSO algorithm is

proposed. Through the introduction of Gaussian coefficients, we modify the

fitness function of PSO algorithm, the particle individual velocity and the

position updating strategy. Furthermore, two test functions are used to analyze

the performance of G-PSO algorithm to solve the global optimal solution.

Experimental results of virtual machine deployment show that G-PSO

algorithm is faster to deploy, has higher deployment precision, and has higher

resource utilization after deployment.

Keywords: PSO algorithm, Gaussian correction, virtual machine, resource

utilization

1 Introduction

Cloud computing is being widely concerned by academia and industry. It is a further

development of distributed computing, parallel processing and grid computing. The

goal of cloud computing is to provide users with secure, fast and convenient data

storage and network computing services with the Internet as the center. As a new

computing technology, cloud computing is designed to dynamically provide the

resources needed for computing and storage, and manage workloads to meet the needs

of a large number of applications [1]. The core idea of cloud computing is to use a

large number of distributed computers to achieve fast and efficient computing, rather

than in the local computer or a separate remote server [2].

Infrastructure-as-a-service is the most basic type of service in cloud computing.

When the user asks the IaaS service provider for service, IaaS service providers use

virtualization technology to encapsulate the CPU, memory, hard disk storage and

other resources in a virtual machine. And the virtual machine is provided to the user

[3].

Different from the traditional application model, in the cloud computing

environment, the user's application is deployed to the virtual resources rather than

physical resources. In recent years, driven by the rapid development of computer

hardware resources, the physical resources (including CPU, network, storage, I / O

devices, VMware as the representative) virtualization technology has made

considerable progress, and provides a solid foundation for the rapid development of

cloud computing [4].

Advanced Science and Technology Letters Vol.142 (GDC 2016), pp.34-39

http://dx.doi.org/10.14257/astl.2016.142.06

ISSN: 2287-1233 ASTL Copyright © 2016 SERSC

Virtualization technology is to make the entire operating system of the virtual

machine as a separate program, running on the original physical computer. Running

multiple virtual machines in it, and consolidating multiple virtual machines onto the

same physical platform to reduce the space and administrative costs of hardware

maintenance. Physical resources based on the same virtual platform provide a great

convenience for unified automated management, so that users can also solve the

heterogeneous problems of operating systems between the physical platforms [5].

Cloud data centers make extensive use of this virtualization technology, so that

multiple virtual machines can be deployed in a single physical machine. However,

due to the difference between the virtual machine and the physical machine

configuration, it is necessary to solve the problem of virtual machine deployment

when constructing the cloud data center [6].

Cloud computing is a business model that benefits from providing services and

delivers high-quality services through the cloud data center to meet customer needs.

Yu demonstrated that the quality of service can be improved by using a load-

balancing scheduling, and that the strategy’s quality determines the degree of

improvement [7]. Bayyapu found that this load-balancing scheduling problem can be

reduced to a subset-sum problem in a polynomial time [8] when only load balancing

scheduling of single CPU resources is considered. Beloglazov pointed out that, the

resources need to be considered is not only CPU in the real environment, and the

problem will become more complex [9]. Therefore, when deploying virtual machines,

how to balance the load of physical machine resources is an NP-Hard combinatorial

optimization problem. At present, most of the open source IaaS solution simply uses

greedy algorithms for virtual machine deployment. Since the greedy algorithm does

not consider the difference between virtual machine and physical machine, in order to

balance the physical load, in the actual use of these solutions, usually by the user to

write their own algorithms.

Compared with other intelligent optimization algorithms, Particle Swarm

Optimization (PSO) has the advantages of simple concept, easy implementation, less

parameters and no gradient information. Therefore, there are researchers using PSO

algorithm to solve the problem of virtual machine deployment. However, when

dealing with complex problems, PSO still has some disadvantages such as slow

convergence speed and easy to fall into local optimum. Therefore, this paper

introduces the Gaussian processing mechanism to modify the PSO algorithm to make

it better applied to the deployment of cloud computing virtual machine.

2 Gaussian Modification Scheme of PSO Algorithm

PSO algorithm has many advantages and has been widely used in engineering design,

artificial intelligence, multi-objective optimization and other fields. However, PSO

algorithm also has some shortcomings, mainly are:

(1) Although the PSO algorithm provides the possibility of global search, it

cannot guarantee convergence to the global optimal point. For a function with

multiple local extremum points, it is easy to fall into the local extremum, and the

correct result cannot be obtained. There are two reasons for this phenomenon, one is

Advanced Science and Technology Letters Vol.142 (GDC 2016)

Copyright © 2016 SERSC 35

the nature of fitness function itself, and the other is due to the diversity of particles in

PSO algorithm, making it disappear too fast, resulting in a premature convergence.

(2) PSO algorithm does not make full use of the information obtained in the

calculation process. It can be seen that in each iteration, the PSO algorithm uses only

the information of the group optimal and the individual optimal. Coupled with the

lack of precision of the algorithm itself, the results of the PSO algorithm cannot be

guaranteed that is always accurate.

(3) PSO algorithm is a heuristic bionic optimization algorithm. There is no strict

theoretical basis. It is only a simple simulation of a population search phenomenon.

Because the PSO algorithm is easy to fall into the local optimal solution, it is

considered that the particle may find a better position under the influence of the

current bestg . Therefore, this paper will improve the PSO algorithm, the idea is to

improve the Gaussian parameters of the PSO algorithm is optimized, the specific

optimization process is as follows: 2 is introduced to represent the Gaussian variance of the fitness of each particle

after each iteration, and if is the fitness value of the i particle. Let f̂ denote the

mean of the current fitness values of all particles, and introduce the f factor to limit

the size of 2 , where f satisfies two conditions:

(1) For all if , satisfy:

1ˆ ff i (1)

(2) f can evolve with the evolution of the particle, which is calculated as follows:

1ˆmax1

1ˆmaxˆmax

ff

fffff

i

ii

(2)

The relationship between Gaussian variance 2 and f is as follows:

2

1

2ˆ

n

i

i

f

ff

(3)

When the optimal solution is reached, all particles have the same fitness value, and

02 . If the population fitness variance is equal to zero and the optimal solution is

not the theoretic optimal solution or the desired optimal solution, then the PSO will

fall into the local optimum and the algorithm will converge prematurely when the

PSO runs.

Advanced Science and Technology Letters Vol.142 (GDC 2016)

36 Copyright © 2016 SERSC

Therefore, if we want to overcome the premature convergence problem, we must

provide a mechanism. When the algorithm converges prematurely, the algorithm can

jump out of the local optimum and continue to search the other regions of the solution

space until the global optimal solution is found.

The mutation operation is designed as a random operator. First, we need to

determine the convergence degree of all the particles. When 02 , the mutation

bestg satisfies the mutation condition, and it varies with a certain probability p.

Probability p is automatically adjusted, the formula is as follows:

bestbestd

bestbestd

fgf

fgftp

)(,0

)(,22

22

(4)

Here, )( bestgf is the currently obtained extreme value, bestf is the optimal

value of the function. t takes the value in the range of 0.1 to 0.4. 2

d takes the

value according to the actual situation and the value of different targeted, and

generally takes much less than the value of 2 .

4 Experimental Results and Analysis of Virtual Machine

Deployment

In this paper, we use CloudSim, an open-source cloud computing simulation platform

launched by Grid University of Melbourne, Australia and Gridbus project (refer to

Amazon EC2 (Elastic Compute Cloud)), and extend it on this basis. The physical

configuration of the resulting cloud data center is shown in Table 1, and the virtual

machine configuration is shown in Table 2.

Table 1. Physical Host Configuration

Physical Host

Class

Number of CPU

cores

Memory

GB

Hard

disk GB

Bandwidth

MB

1 8 40 3000 1000

2 16 50 5000 1500

3 24 60 8000 2000

Table 2. Virtual machine configuration

Physical Host

Class

Number of

CPU cores

Memory

GB

Hard

disk GB

Bandwidth

MB

Advanced Science and Technology Letters Vol.142 (GDC 2016)

Copyright © 2016 SERSC 37

1 1 1.70 150 50

2 1 1.70 300 50

3 1 2.00 300 100

4 2 1.70 600 100

5 2 2.00 600 200

6 4 1.70 800 200

7 4 2.00 800 400

8 8 1.70 1200 400

9 8 1.70 1200 600

10 8 2.00 1200 800

Between these 10 virtual machines, there are still some communication relationship.

In the following, we will give the realization of the optimal solution of the Gaussian

modified PSO algorithm in the algorithm. Taking into account the actual situation, we

use reverse solutions to solve the virtual machine placement program.

In order to verify the performance of the Gaussian modified PSO algorithm in the

implementation of virtual machine deployment, the following experimental study is

carried out. First, the G-PSO algorithm and the PSO algorithm are compared to obtain

the global optimal solution, as shown in Figure 1.

200017501500125010007505002500

Iteration times

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.50

Iter

ati

on

err

or

PSO Algorithm

G-PSO Algorithm

Fig.1. Comparison results of iterative performance of two algorithms

From the comparison of the curves in Fig. 1, it can be seen that the G-PSO algorithm

not only converges faster, but also has smaller iteration error than the PSO algorithm.

Advanced Science and Technology Letters Vol.142 (GDC 2016)

38 Copyright © 2016 SERSC

It can be seen that the PSO algorithm after Gaussian correction has better iterative

performance and can obtain the global optimal solution faster and more accurately.

5 Conclusions

In this paper, a Gaussian modified PSO algorithm is proposed to solve the problem of

virtual machine deployment in cloud computing. Experimental results show that

compared with the PSO algorithm, the iterative process of Gaussian modified PSO

has faster convergence speed, higher optimization precision and higher resource

utilization of virtual machine deployment.

References

1. Calheiros, R.N., Ranjan, R., Beloglazov, A., De Rose, C.A.F.: CloudSim: a toolkit for

modeling and simulation of cloud computing environments and evaluation of resource

provisioning algorithms[J]. Practice and Experience, 41(1): 23-50. (2011)

2. Faragardi, H.R., Rajabi, A., Shojaee, R., Nolte, T.: Towards energy-aware resource

scheduling to maximize reliability in cloud computing systems[C]. Proceedings IEEE

International Conference on High Performance Computing and Communications, 1469-

1479. (2014)

3. Abbadi, I.M., Ruan, A.B.: Towards trustworthy resource scheduling in clouds[J]. IEEE

Transactions on Information Forensics and Security, 8(6): 973-984. (2013)

4. Sadooghi, I., Palur, S.: Achieving efficient distributed scheduling with message queues in

the cloud for many-task computing and high-performance computing[C]. 14th IEEE/ACM

International Symposium on Cluster, Cloud, and Grid Computing, 404-413. (2014)

5. Quarati, A., Danovaro, E.: Scheduling strategies for enabling meteorological simulation on

hybrid clouds[J]. Journal of Computational and Applied Mathematics, 273: 438-451.

(2014)

6. Tordsson, J., Montero, R., Moreno-Vozmediano, R.: Cloud brokering mechanisms for

optimized placement of virtual machines across multiple providers[J]. Future Generation

Computer Systems, 28(2): 358-367. (2012)

7. Yu, J., Buyya, R.: Workflow scheduling algorithms for grid computing[J]. Metaheuristics

for Scheduling in Distributed Computing Environments, 146: 173-214. (2008)

8. Bayyapu Karunakar, R., Fischer, P.: Load scheduling in a cloud based massive video-

storage environment[C]. Proceedings-16th International Symposium on Symbolic and

Numeric Algorithms for Scientific Computing, 349-356. (2014)

9. Beloglazov, A., Abawajy, J., Buyya, R.: Energy-aware resource allocation heuristics for

efficient management of data centers for cloud computing[J]. Future Generation Computer

Systems, 28(5): 755-768. (2012)

Advanced Science and Technology Letters Vol.142 (GDC 2016)

Copyright © 2016 SERSC 39