You are on page 1of 3

2018 International Conference on Virtual Reality and Intelligent Systems

Cloud Computing Task Scheduling Policy Based on Improved Particle Swarm


Optimization

Wu.DaQin
Department of Information Engineering, Jilin Business and Technology College, Changchun, China
liulan2003@sina.com

Abstract—The task scheduling policy is the important Particle Swarm Optimization (PSO) is a swarm
factors for achieving efficient calculation in a cloud intelligence algorithm based on group collaboration
computing environment. This article put forwards a developed by simulating bird foraging behavior. It is simple
task scheduling method based on improved particle and easy to understand and has been widely used. The basic
swarm algorithm against the present inefficiency. algorithm proposed and improved by Eberhart. However,
Particle Swarm Optimization (PSO) algorithm is used to PSO is an optimization algorithm based on swarm
solve task scheduling optimization by introducing the intelligence. With the advantages of few set parameters,
iterative selection operator. Improved particle swarm simple calculation and fast convergence speed, it has great
optimization algorithm (IPSO) can improve the ability application value. However, Standard PSO (SPSO) easily
of the optimization, as much as possible avoiding falling reduces the diversity of population in the evolution resulting
into a local optimum. The convergence effect is so better in algorithm premature convergence. Therefore, Zhang et al.
that the task scheduling time costs can be reduced. By maintained the diversity of population by reinitializing
simulation on a CloudSim simulation platform, the particle velocity. Riget et al. measured according to the
experimental results show that the algorithm has the characteristics of population diversity and alternately
advantages of improving optimization and taking less performed attraction and repulsion of particle swarm in the
time. So it can be used to research and practice about evolution. To perform task scheduling in cloud computing
cloud computing problem for complex scheduling environments, this article alternately performs the iterations
optimization. of optimal particle based on multiple iterative methods,
maintaining the diversity of particle swarm and decreasing
Keywords- improved Particle Swarm Optimization; the possibility of local optimization while ensuring the
cloud computing system; task scheduling; selection computing speed of the algorithm. The experimental results
operators show that the improved algorithm proposed in this article
can improve the time efficiency while maintaining the
I. INTRODUCTION
optimization scheduling of original resources to further
Cloud computing is one of the fastest growing IT optimize the space-time process of task scheduling in cloud
technology now, even it is becoming more and more get the environments.
favour of people. In order to adapt to the large data
processing requirements, cloud computing as a new parallel
computing technology is the extension and development for II. CLOUD COMPUTING TASK SCHEDULING
distributed computing, grid computing and parallel Task scheduling algorithm in cloud computing is
computing. Cloud computing consists of infrastructure layer, performed with the scheduling strategy built in virtual
platform layer and application layer. The function of each machine, but its scheduling strategy is limited to local
layer offered by the form of services. performance and cannot effectively meet the requirements
Cloud computing task scheduling refers to the process in for global performance with high time costs. Resource
which resources are allocated and adjusted between different scheduling in cloud environment aims to assign x tasks to W
users according to resource usage strategies of the computing resources in a cloud environment. To achieve this
computing in specific cloud configuration environment. In goal, this article introduces the improved PSO algorithm to
the process, specific task scheduling is performed at the calculate the average minimum time of performing tasks in
infrastructure layer (IaaS). each resource to control the global and local search so as to
How can task scheduling be performed in cloud achieve resource optimization of system. The time for
computing environments to meet the maximum needs of determining resource optimization is expressed in the
users? Currently, most of known strategies are used to equation (1) shows:
perform virtual machine scheduling and attempt to perform ∑
minF(xi)= , i=1,2,…S (1)
scheduling for their internal applications combined with
certain scheduling strategies, but they are not accurate and As the equation (1), the count of tasks is X in the cloud
efficient unable to meet the increasing needs of cloud environment within the time given. Then ∑ =X, in this
computing development. indicates tasks are allocated to the No. i calculation
resource. If the max value of calculation resources is Maxi,

978-1-5386-8031-5/18/$31.00 ©2018 IEEE 99


DOI 10.1109/ICVRIS.2018.00032
then 0< <Maxi, indicates the No. i calculation 3) Sequentially perform the following operations from
resource will cost time for processing the tasks. Therefore, the second particle to the last particle:
F(xi) is minimum, that is the cloud schedule optimum value. a) If the distance from the current particle to all
designated initial particles is greater than , the particle is
III. PARTICLE SWARM OPTIMIZATION
designated as a new network center and added into the
A. Standard Particle Swarm Optimization (PSO) memory group as a memory particle in the same way.
Particle Swarm Optimization (PSO) proposed by b) If the distance from the current particle to a
Eberhart and Kennedy in 1995 is based on the simulation of designated initial particle is less than , the particle is
bird flocks or fish schools. The information queue of best eliminated.
known positions of particle and entire swarm can be Increase the local search ability of the algorithm while
memorized in the PSO algorithm based on group ensuring the global optimization can be searched to control
collaboration. Each particle in the population represents a global search and avoid premature convergence of original
possible solution of the object to be optimized. Let the algorithm which would easily get into local optimization.
population X composed of m particles be X={xl, x2, x3, …, With this algorithm, task scheduling can be better performed
xn} in a n-dimensional search space. Where, the position of in cloud enviroments to reduce the overall time of task
No. i particle is xi={ xi1, xi2, xi3,…, xim}T , where 1≤ i≤ m, implementation. The similarity equation in this phase is as
the velocity is vi ={vi1, vi2, vi3,…, vim}T, the current follows:
optimal position of particle i itself at time t; is the global simility( , )= ∑ (4)
1
optimal position of particle swarm at time t; c1 and c2 are
the learning factors; r1 and r are the random numbers evenly Where, 1≤ i≤ M, M is the numbers of particle swarm;
distributed in (0,1). , the current k-dimensional position of the particle i, j;
Then the equations of speed and position update of the upper, low indicates the value range of particle X={xl, x2,
particle x are as follows: x3, …, xn}, represents the inhibition factor of threshold.
1 The flow of improved PSO algorithm shows as Figure 1.
= + c1r1( )+ c2r2( ) (2)
1 1
= + (3)
At time t, m is the number of particles in the particle
swarm; represents current position of particle i,
represents the current velocity of particle i. If the whole
particle swarm is within the neighborhood of particles, the
above algorithm indicates the global-PSO (PSO in global
model), or the local-PSO (PSO in local model).
B. Improved PSO algorithm
In the standard PSO, PSO algorithm has fast
convergence speed in the early stage of search, but its
convergence speed slows down when the optimal solution of
particles approaches gradually decreasing the diversity of the
population. Once the velocity of particles in the population is
zero, the algorithm gets into premature if the optimal
position of particle swarm is not the optimal value of the
objective function at this time. This article proposes that
iterative selection operators should be added into the PSO
algorithm. The initial values of population and compared
values of results whose are gained by iterations should be
first recorded. In the inhibition phase, the particles with low
similarity and high adaptability to current particles should be
Figure 1. The flow chart of improved PSO algorithm
maintained so as to ensure that the particles with best
positions in the set range where threshold value is inhibited IV. THE EXPERIMENT RESULTS ANALYSIS
can enter the next iteration after high-adaptability particles
are compared with initial particles in terms of adaptability. The C1oudSim simulation platform is selected for
The steps for network inhibition are as follows: simulation,meanwhile the experiments are expanded using
1) Arrange particles according to their adaptability programming tools Eclipse. Then rewrite DataCenterBroker
after each iteration in descending order. class, and test simulation again.
2) Designate the first particle as the first initial particle
and add it into the memory group as a memory particle.

100
“Applied Technical Talents Training Based on Diversified
School-running Mode”(2015).
REFERENCES
[1] Gao Y.A M ulti—obiective Ant Colony System Algorithm
for Virtual Machine Placement in Cloud Computing [J].
Journal of Computer and System Sciences, 2013, 79(8):l230-
l242.
[2] Litvinski O, Gherbi A. Experimental Evaluation of Open
Sack Compute Scheduler[J]. Procedia Computer Science,
2013, 19(1):116—123
[3] Ghanbari H.Exploring Alternative Approaches to Implement
an Elasticity Policy[C]//IEEE 4th International Conference on
Figure 2. The contrast chart of execution time CloudComputing.Washington,DC,USA:IEEE,2011,716-723.
[4] Padhy R P.SLAs in Cloud Systems:The
In this experiment, we respectively adopted C1oudSim
BusinessPerspective[J].International Journal of Computer
existing Round Robin (Round Robin, RR) scheduling Science and Technology(IJCST), 2012,3(1):481-488.
algorithm, the PSO algorithm and the IPSO algorithm for
[5] Zissis D . Lekkas D . Addressing Cloud Computing
task scheduling. The results of using the algorithms for
Security Issues J 1 . Future Generation Computer
calculating stability are compared.
The experiment parameters are set as follows: the Systems,2012,28(3):583—592.
learning factors cl=c2=2; the number of initial particle swarm
is equal to the number of jobs; the number of iterations
Gmax from 50 to 350 times, every 50 times as a gradient;
the inhibition factor delta = 0. 01. Compared in the following:
Comparing with two cases, it shows that : Set the number of
virtual machines is 6 in the experiments, and the number of
tasks is 15。The result which is gained by the RR algorithm
is 1, 162.57 seconds. If they running 50 times independently,
average total tasks execution time by two kinds of algorithm
time are contrasted in the curve graph shown as Figure 2.
V. CONCLUSION
To remedy the defects of PSO algorithm in addressing
the issue of large-scale optimization, this article analyzes the
time efficiency of task scheduling in cloud environments and
adds iterative selection inhibition operators into the PSO
algorithm, which can increase the convergence speed of PSO
algorithm, avoid local optimization and efficiently complete
the search and distribution of computing resources in cloud
computing environments with higher efficiency than general
resource scheduling strategies, giving consideration to the
fairness of resource allocation with better usability and
scalability. However, the strategy still needs to be improved.
The parameters of inertia weight and learning factors will be
improved and the issue of load balancing will be solved in
the future.
This work was financially supported by Jilin Province
Educational Science topics " The Research and Realization
of Self-help Building Website Platform Based on Saas"
(2013399) and "The Tutorial Expert System For Computer
Rank Examination Based on AI Technology"(2013542).
VI. ACKNOWLEDGEMENTS
This work was financially supported by Jilin Province
Higher Education Teaching Reform Research Topics

101

You might also like