ISSN (Online) 2394-2320

International Journal of Engineering Research in Computer Science and Engineering

(IJERCSE)
Vol 5, Issue 2, February 2018

A Resource Allocation Strategy using PSO in

Heterogeneous Cloud
[1]
Neeraj Kumar Pandey, [2] N.K.Joshi
[1]
GLA University, Mathura, Uttar Pradesh, [2] Vice Chancellor, Uttaranchal University Dehradun,
Uttarakhand
Abstract - The most important problem in the cloud service provider is to maintain the elastic property of the cloud in such a way
that user will pretend the cloud as limitless. So the challenge is how to make the limited sources unlimited. Every task must be
granted what it requires by any mean otherwise it will degrade the performance of cloud. So resource allocation has a lot of
solution. Resource allocation is an NP-hard problem so no particular solution can perform well always. But these kinds of problems
are solved by nature in many ways such that such as ant colony optimization (ACO) algorithm, particle swarm optimization (PSO)
algorithm and firefly algorithm. In this paper, a particle swarm optimization technique has been used to resolve the most critical
problem of the cloud service provider at cloud data centre. This technique is basically taken from the collective and collaborative
nature of the nature swarm. This technique can be used to allocate the resource to the task request by minimizing the makes span,
flow time and task execution cost. The simulation and test results show the better efficiency than the other similar existing
technique.

the proposed algorithmic solution of the resource

1. INTRODUCTION allocation problem. Section 5 will describe simulation
result and section 6 will conclude the results.
Cloud computing is a long term result of the „utility
computing‟ or on demand computing. It simplifies the 2. RELATED WORK
goal of achieving high scale computing without the
maintenance pitfalls and IT expertise. It provides the In paper [1] new job scheduler presented by using particle
model of pay as per use like other utilities used in day swarm optimization (PSO) by minimizing execution cost
today life. So performance is always a challenge for the by meeting the timestamp required for the execution. Its
cloud service provider (CSP) which includes availability shows better simulation results as compared to different
of the demanded resource at that instant of time. So CSP genetic algorithms. In paper[2] proposed approach gives
must have a strategy to making resources available as per random generation of the particles which represents the
demand at any point of time. no of resource which lies between 1 to M. random
selection gives equal chance of selection of every
A CSP must provide pool of heterogeneous, virtualized, resource and every job will get the sufficient amount of
highly available, dynamically-scalable, and configurable the resource to get executed. Paper [3] gives the
and reconfigurable computing resources (e.g. storage, comparative solution of the task scheduling by embedding
applications, networks, computing units, data) can be the PSO with small position value (SPV). PSO algorithm
rapidly provisioned and released with minimal embeds in crossover and mutation (CM-PSO), PSO
management effort in the data centres [1] so that a QoS algorithm embeds in local search (L-PSO) to compare
must be maintained in the cloud. A scheduling system with the PSO metric [3]. Another solution is proposed in
required efficient allocation of task, security, reliability the paper [4] in which set of task-service pair is
and fault tolerance. In cloud computing different user represented as a candidate set. Each particle will learn
have various demands like real time assurance, low cost , from each feasible pair of different dimension. The
strong stability. So there is different parameter to satisfy position building technique guarantees every position is
the different needs of users. In this paper a resource reasonable. This scheme significantly minimizes the
scheduling algorithm is proposed based on PSO. search space and improves the algorithm performance.
The new algorithm produces outstanding performance on
Rest of the paper is organised as follows: section 2 job scheduling–resource allocation schemes in cloud
describe the related research done by the researchers in environment.
the task scheduling using PSO. Section 3 formulates the
job-resource scheduling problem. Section 4 will discuss

All Rights Reserved © 2018 IJERCSE 423

 ISSN (Online) 2394-2320

International Journal of Engineering Research in Computer Science and Engineering

(IJERCSE)
Vol 5, Issue 2, February 2018

implement this strategy smallest task must be allocated to

3. JOB-RESOURCE SCHEDULING MODEL the lowest capacity processor (Resource).

The task scheduling is represented by the Directed 4. RESOURCE ALLOCATION ALGORITHM

Acyclic Graph (DAG) where G:{V,E}. The set of nodes USING PSO
are called vertices which are a set of task V= {V1, V2,
V3…..Vn} represents in workflow application. The edge Kennedy and Eberhart proposed particle swarm
which is connecting the task node represents the exchange optimization algorithm in 1995 [6].PSO is used to
of information between the tasks. The swarm can be optimize the problem. It is inspired by the self-guided
represented as Xi=Xi1,Xi2,………..XiD . every iteration nature of the flocks of birds. In this strategy each particle
update the particle position and draw it to the its best is considered as path of solution in an evolutionary
position which helps the every particle to provide the best algorithm. In the first distribution the particle are
solution of the task scheduling. Figure 1 depicts the edge positioned randomly. Each particle has a position vector
weight Eij denote the edge between node i and j which (Xi) which is updated on the regular interval. Based on
share the information. The weight is the simply work proposed algorithm by Kennedy and Eberhart [9] the
capacity of node and edge. formula that exists for updating position vector is:

Xik+1=Xik+Vik+1 (1)

Figure 2 simply describe the PSO algorithm which states

that that each particle contributes to find the optimal
solution of the problem. Initially random matrix is taken
as position of particle so that every particle must be
selected. Then the current position and velocity matrix is
updated.

Figure 1. A DAG example on Heterogeneous Cloud

Paper [9] describes the other QoS parameter like

minimizing the execution by applying the multi-objective
task scheduling algorithm. More than one objective of
task scheduling have been discussed and implemented
successfully. The same concept of multi objective task
scheduling framework is implemented in paper [10] using
PSO algorithm. This strategy works on the different work
flow modules of the cloud like network, image, pricing
and certification module.

The paper consider that data centre has different capacity

of processor (Resource) and different types of tasks
demanding for resource having heterogeneous constraints
like execution time, cost, priority and resource
reservation. So in this variety of the task this paper
focuses on the minimization of execution cost so. To
Figure 2: flowchart of PSO algorithm

All Rights Reserved © 2018 IJERCSE 424

 ISSN (Online) 2394-2320

International Journal of Engineering Research in Computer Science and Engineering

(IJERCSE)
Vol 5, Issue 2, February 2018

Following algorithm merged the extended particle swarm

optimization with the shortest job to smallest processor. There are 4 VM and 30 GB of hard disk and 1 up to 4
The vectors simply determine the current position of the cores of processor. Before examining the proposed
particle which directly participate the solution of the algorithm the slandered PSO and Genetic Algorithm (GA)
scheduling of the task. So that every particle can guide it have been tested and results are collected for comparison
successive particle and can affect the result. with the proposed one. In the table 1 and figure 3 there
EPSO Based Task-Resource Scheduling Algorithm are three types of algorithms have been analysed GA,
1. Begin traditional PSO and hybrid (EPSO+SJSP). Three range of
2. Initialize the particle based on SJSP Algorithm task have been taken 100, 200 and 400 and 6 resources.
3. Apply fitness function on each particle. This scenario is implemented on the cloudsim and criteria
4. Update position and velocity matrix which is taken for consideration is makespan, and cost of
5. Evaluate termination criteria execution (number of the task executed per unit time)
6. Calculate fitness value
7. If (Fitness value ==Null) goto step 3
8. Terminate algorithm

Firstly existing PSO is optimized into extended PSO then

it is merged with SJSP so that the cost of smallest job
must be minimum because of assigning of smallest
processor (processor having minimum processing power).
By this strategy total cost of processing the task can be
reduced.

5. EXPERIMENTS AND RESULT

Cloudsim 3.0.3 is an open source simulator which has

been developed by Gridbus project team and the grid
Laboratory of the University of Melbourne in Australia.
The Cloudsim can run on Linux and Windows systems
[7] [8]. It creates suitable simulation environment for the Figure 3 Results and Analysis of Proposed Algorithm
implementation and analysis of the modified task
scheduling algorithm. A cloudsim 3.0.3 was taken with As it can be observed that as we increase the number of
core i5 2.4GHz processor and Eclipse Jee Oxygens used task on the fix amount of resource (VM) it is producing
to run the cloudsim. It is used to implement extended PSO an optimum result than a traditional PSO and GA.
and SJSP algorithm and a comparative study is also been
done to evaluate the performance proposed algorithm. 6. CONCLUSION AND FUTURE SCOPE

Task- The evaluation of the PSO is done with the genetic

Resource No. of No.of Task/Ti Improveme algorithm and it is observed that the proposed
Scheduling Task Resource me nt combinations of two strategies are performing better
Algorithm together. As we increase the number of resource and the
GA 6.23 available VM remains the same still it is giving the better
PSO 100 6 5.97 2.06 results. The parameter was taken was makespan and total
EPSO+SJSP 5.36 cost of execution. In future the other parameter like load
GA 13.03 balancing, VM allocation strategy and optimizing energy
PSO 200 6 12.86 9.13 efficiency. There may be other major objective which can
EPSO+SJSP 12.16 be targeted using this hybrid PSO algorithm.
GA 26.8
PSO 400 6 26.31 3.27
EPSO+SJSP 24.94
Table: 1 Result Analysis of Proposed Algorithm

All Rights Reserved © 2018 IJERCSE 425

 ISSN (Online) 2394-2320

International Journal of Engineering Research in Computer Science and Engineering

(IJERCSE)
Vol 5, Issue 2, February 2018

REFERENCES Computing Throughput Optimization” International

Conference on Intelligent Computing, Communication &
[1] Jing Liu, Xingguo Luo, Xingming Zhang, and Fan Convergence (2015) pp.107-113.
Zhang “Job scheduling algorithm for cloud computing
based on particle swarm optimization” Advanced [10] R K Jena “Multi objective Task Scheduling in Cloud
Materials Research Online, Trans Tech Publications, Environment Using Nested PSO Framework” 3rd
Switzerland ISSN: 1662-8985, Vol. 662, (2013) pp 957- International Conference on Recent Trends in Computing
960. 2015 (ICRTC-2015) pp. 1219-1227.

[2] Vahid Asadzadeh Chalack,Seyed Naser Razavi,Sajjad

Jahanbakhsk Gudakahriz “Resource Allocation in Cloud
Environment Using Approaches Based Particle Swarm
Optimization” International Journal of Computer
Applications Technology and Research Volume 6–Issue
2, 87-90 ISSN:-2319–8656 (2017) pp. 87-90.

[3] Lizheng Guo, Shuguang Zhao, Shigen Shen1,

Changyuan Jiang “Task Scheduling Optimization in
Cloud Computing Based on Heuristic Algorithm” journal
of networks, journal of networks, vol. 7, no. 3, march
(2012) pp.547-553.

[4] Zhangjun Wu, Zhiwei Ni, Lichuan Gu, Xiao Liu “A

Revised Discrete Particle Swarm Optimization for Cloud
Workflow Scheduling” IEEE International Conference on
Computational Intelligence and Security (2010) pp 184-
188.

[5] Solmaz Abdi, Seyyed Ahmad Motamedi, and Saeed

Sharifian “Task Scheduling using Modified PSO
Algorithm in Cloud Computing Environment”
International Conference on Machine Learning, Electrical
and Mechanical Engineering (ICMLEME'2014) Jan. 8-9,
Dubai (UAE) (2014) pp.37-41.

[6] J. Kennedy, R.C. Eberhart, “Particle swarm

optimization”, Proc, IEEE Conf. Neural Netw., vol. IV,
IEEE, Piscataway, NJ, 1995,pp.1942-1948.

[7] Abraham, R. Buyya, and B. Nath.” Nature‟s heuristics

for scheduling jobs on computational Grids”, 8th IEEE
International Conference on Advanced Computing and
Communications (ADCOM 2000), India, 2000.

[8] Ali Al-maamari and Fatma A. Omara “Task

Scheduling Using PSO Algorithm in Cloud Computing
Environments” International Journal of Grid Distribution
Computing Vol. 8, No.5, (2015), pp.245-256.

[9] Atul Vikas Lakra, Dharmendra Kumar Yadav “Multi-

Objective Tasks Scheduling Algorithm for Cloud

All Rights Reserved © 2018 IJERCSE 426