Cloud partitioning with load balancing a new load balancing technique for public cloud

1. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 5, Issue 8, August (2014), pp. 86-94 © IAEME
INTERNATIONAL JOURNAL OF ADVANCED RESEARCH
IN ENGINEERING AND TECHNOLOGY (IJARET)
ISSN 0976 - 6480 (Print)
ISSN 0976 - 6499 (Online)
Volume 5, Issue 8, August (2014), pp. 86-94
© IAEME: http://www.iaeme.com/IJARET.asp
Journal Impact Factor (2014): 7.8273 (Calculated by GISI)
IJARET
© I A E M E
CLOUD PARTITIONING WITH LOAD BALANCING: A NEW LOAD
BALANCING TECHNIQUE FOR PUBLIC CLOUD
Shamna S S1, Manu J Pillai2
1(Mtech Student, Computer Science and Engineering, TKM College of Engineering, Kollam, India)
2(Assistant Professor, Computer Science and Engineering, TKM College of Engineering,
Kollam, India)
86
ABSTRACT
Cloud Computing provides lot of services which can be easily consumed over the internet as
on needed basis. One of the main challenges faced by cloud network is load balancing. It is needed in
a distributed computing like cloud computing to distribute the dynamic workload across multiple
nodes to ensure that no single node is overwhelmed. Since job arrival pattern is not predictable and
the capacities of each node in the cloud differ, a dynamic load balancing technique is proposed here,
mainly for public cloud because public cloud has numerous nodes which are distributed across
different geographical locations. Cloud partitioning concept is also included in this proposed
technique. Thus the cloud has a cloud controller that chooses the suitable partitions for arriving jobs
while the balancer for each partition chooses the best node for job allocation.
Keywords: Cloud Computing, Cloud Partitioning, Distributed Computing, Load Balancing, Public
Cloud.
1. INTRODUCTION
Cloud computing is the use of computing resources that are delivered as a service over
a network [1]. Cloud computing entrusts remote services with a user's data, software and computation.
Many industry experts dispute the validity of the four deployment models. For them only public
clouds are true clouds, but when the user experience and functional capabilities are the same, and
there is the possibility of moving seamlessly across cloud boundaries the distinctions become, well,
cloudy. The cloud deployment models are public cloud, private cloud, community cloud and hybrid
cloud [2].
Cloud computing is efficient and scalable but maintaining the stability of processing so many
jobs in the cloud computing environment is a very complex problem with load balancing receiving

2. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 5, Issue 8, August (2014), pp. 86-94 © IAEME
much attention for researchers. Load balancing is one of the central issues in cloud computing. It is a
mechanism that distributes the dynamic local workload evenly across all the nodes in the whole
cloud to avoid a situation where some nodes are heavily loaded while others are idle or doing little
work. It helps to achieve a high user satisfaction and resource utilization ratio, hence improving the
overall performance and resource utility of the system. Load balancing schemes depending on
whether the system dynamics are important can be either static or dynamic. Since the job arrival
pattern is not predictable and the capacities of each node in the cloud differ, for load balancing
problem, workload control is crucial to improve system performance and maintain stability. Static
schemes do not use the system information and are less complex while dynamic schemes will bring
additional costs for the system but can change as the system status changes. So a dynamic load
balancing method is needed.
The load balancing model mentioned in this paper is aimed at the public cloud which has
numerous nodes with distributed computing resources in many different geographic locations. Thus,
this model divides the public cloud into several cloud partitions. When the environment is very large
and complex, these divisions simplify the load balancing. The cloud has a cloud controller that
chooses the suitable partitions for arriving jobs while the partition balancer for each cloud partition
chooses the best load balancing strategy.
The article covers the concepts and implementation of the proposed work. Section 2 contains
a brief review of related works. Following Section 6 gives the proposed system in detail. Next
section describes detailed result and discussion. Final section shows the conclusions and scopes for
the future enhancement.
87
2. RELATED WORKS
This paper [3] presents and analyses the most important issues which need to be considered
in the development of an effective load balancing algorithm: load estimation, load levels comparison,
performance indices, stability, amount of information exchanged among nodes, job resource
requirements estimation, jobs selection for transfer, remote nodes selection, and more. This paper
also describes various components of dynamic load balancing algorithms such as Information
Strategy, Transfer Strategy, and Location Strategy. The paper [4] review various different load
balancing algorithms for the grid based distributed network, identify several comparison metrics for
the load balancing algorithms and to carry out the comparison based on these identified metrics
between them. The paper discusses the policies of dynamic load balancing for heterogeneous
resources such as Transfer, Location, Selection, Information policy. The different metrics for
comparing the load balancing algorithms are identified: Load balancing time, Makespan, Scalability,
Fault tolerance, Reliability, Stability and so on. Different load balancing algorithms and comparison
between them have been discussed.
H. Mehta et al. [5] proposed a new content aware load balancing policy named as workload
and client aware policy (WCAP). It uses unique and special property (USP) to specify the unique and
special property of the request as well as computing nodes. Y. Lua et al. [6] proposed a Join- Idle-
Queue load balancing algorithm for dynamically scalable web services. This algorithm provides
large-scale load balancing with distributed dispatchers by, first load balancing idle processors across
dispatchers for the availability of idle processors at each dispatcher and then, assigning jobs to
processors to reduce average queue length at each processor. Z. Zhang et al. [7] proposed a load
balancing mechanism based on ant colony and complex network theory in an open cloud computing
federation. It uses small-world and scale-free characteristics of a complex network to achieve better
load balancing. S.C. Wang et al. [8] proposed a two- phase scheduling algorithm that combines OLB
(Opportunistic Load Balancing) and LBMM (Load Balance Min-Min) scheduling algorithms to
utilize better executing efficiency and maintain the load balancing of the system. M. Randles et al.

3. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 5, Issue 8, August (2014), pp. 86-94 © IAEME
[9] investigated a distributed and scalable load balancing approach that uses random sampling of the
system domain to achieve self organization thus balancing the load across all nodes of the system.
88
3. PROPOSED SYSTEM
There are so many cloud computing categories such as private cloud, public cloud, hybrid
cloud etc. But the model proposed here is mainly for public cloud. The public cloud is based on
standard cloud computing model. A public cloud has numerous nodes with distributed computing
resources in many different geographic locations [4]. And also so many challenges are faced by
public cloud such as security, fault tolerance, load balancing etc. Here a technique is proposed for
efficiently balancing the load in the public cloud.
A large public cloud will include many nodes and the nodes in different geographical
locations. Here to manage this large cloud, cloud partitioning is used. Thus this model divides the
public cloud into different cloud partitions. When the environment is very large and complex these
partitions simplify the load balancing.
3.1 Cloud Partitioning
Fig 1: Typical Cloud Partitioning
The load balancing strategy is based on cloud partitioning concept. The architecture is shown
in fig 1. After creating the cloud partitions load balancing procedure starts: When a job arrive at the
cloud system, cloud controller deciding which cloud partition should receive the job. Then the
partition load balancer will decide to which node, job should be given for execution. The load status
of each cloud partition is checked when a job arrives. If the partition is normal then only job is given
to that partition, otherwise choose another partition for job allocation.
Cloud Controller Partition Balancer
The load balance solution is done by the cloud controller and the partition balancers. The
cloud controller first assigns jobs to the suitable cloud partition and then communicates with the
balancers in each partition to refresh this status information. Since the cloud controller deals with
information for each cloud partition, smaller data sets will lead to the higher processing rates. The
partition balancers in each partition gather the status information from every node and then choose
the right strategy to distribute the jobs. The relationship between the balancers and the main
controller is shown in fig 2.

4. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 5, Issue 8, August (2014), pp. 86-94 © IAEME
Fig 2: Relationships between cloud controllers, Partition Balancers and the Nodes
89
Algorithm 1: Cloud Partitioning Algorithm
1. First select a random node(take it as a centre node)
2. Select another center node which is far apart from selected center node
a. Selection is based on geographical location of each node(latitude and longitude)
3. Thus select ‘n’ center node if we want ‘n’ partitions
4. After that, partition other nodes based on the minimum distance from centre node
3.2 Proposed Algorithm: Load Based Allocation
When a job arrives at the public cloud, the first step is to choose right partition. The cloud
controller will choose the best partition and provide the job to that partition. The best partition is
selected based on partition status of each partition. If the standard deviation of the workload of the
partition is less than the threshold value, then only that partition is selected otherwise choose another
partition and check the status.
When the job reaches at the best partition’s balancer, the partition balancer will choose the
best node for job allocation. The partition balancers choose the best node based on load on each
node. Balancers first calculate load on all nodes and choose the node with minimum load as best
node. The job assignment strategy of proposed algorithm, Load based Allocation is given in fig 3.

5. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 5, Issue 8, August (2014), pp. 86-94 © IAEME
Fig 3: Job Assignment Strategy
90
3.2.1 Best Partition Selection
When the jobs arrive at the cloud controller, job will be received by the best partition selected
by the cloud controller. Cloud controller select suitable partition based on overall status of the
partition. Partition status is checked based on standard workload deviation
Calculation of workload and standard deviation
• Capacity of a Node(N)
Cj = PEnumj * PEmipsj + Nbwj (1)
where processing element, PEnumj is the number of processors in Nj ,PEmipsj is million
instructions per second of all processors in Nj and Nbwj is the communication bandwidth ability of Nj
[10].
• Capacity of all Nodes
(2)
c ci
Summation of capacity of all nodes is the capacity of partition.
• Load on a node
Total length of tasks that are assigned to a node is called load
Lni,t = total length of tasks that are assigned to a node and not yet completed at
time t
=
=
m
i
1

6. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 5, Issue 8, August (2014), pp. 86-94 © IAEME
• Load of all nodes in a partition is calculated as
a
jPj
91
(3)
m
=
=
L LNi
i
1
• Processing time of a Node:
PTi = LNi/Ci (4)
• Processing time of all nodes:
PT = L/C (5)
• Standard deviation of load
(6)
m
2 s 1/
( )
=
= −
m PTi PT
i
1
Algorithm 2: Best Partition Selection Algorithm
I f standard deviat ion of workload () Ts (Threshold)
Select the part i t ion
Else
Choose another par t i t ion
3.2.2 Best Node Selection
Jobs arrive at the cloud controller will be received by the suitable partition selected by the
cloud controller. When the job arrives at the partition balancer, a best node in the partition will be
selected for job allocation by the balancer. The partition balancer will select the best node based on
load on each node in the partition. The node with minimum load is selected as the best node for job
allocation.
Calculation of load
Define a load parameter set: P = {P1, P2…….Pm) with each Pi (1 i m, Pi € [0, 1]), where
m represents the total no. of parameters [5].
Service rate of node Ni,
m
S (Ni, t) = (7)
=
j
1
m
( =1) are weights that we assign for each parameter.
ai
=
i
a
i
1
Load on a node Ni at time t,
LNi,t = N(T,t)/S(Ni, t) (8)
where N(T,t) is number of tasks at time t on service queue of Ni and S(Ni, t) service rate of
node at time t.

7. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 5, Issue 8, August (2014), pp. 86-94 © IAEME
92
Algorithm 3: Best Node Selection Algorithm
1. For each node in the par t i t ion
Calculate current load of the node
2. Node wi th minimum load is taken as best node
4. RESULTS AND DISCUSSION
The classes of CloudSim simulator have extended to simulate this algorithm. A public cloud
has numerous nodes with distributed computing resources in many different geographical locations.
A public cloud needs a load balancing method that can complete jobs within reasonable response
time. It is because services rendered over the network by the public cloud that is open for public use.
Each user who accesses the public cloud wants his jobs completed in shortest time. So to analyze the
proposed algorithm, the metrics [11] such as response time, waiting time and completion time of jobs
are considered.
• Average Waiting Time
Waiting Time is the sum of the periods spent waiting in the ready queue i.e.it is the time
between the submission of the request and initiation of the response. Here we compare the average
waiting time of proposed algorithm with existing load balancing technique Round Robin. From the
Fig 4 it is understood that our proposed technique is much better than Round Robin because it have
reduced waiting time than Round Robin.
Fig 4: Comparison between algorithms Load based Allocation Round Robin based on Average
Waiting Time
• Average Response Time
Response Time is the time between the submission of a request and the completion of the
response. Because the response time equals service time plus wait time, we can increase performance
in the area by reducing wait time or reducing service time. Service time is the time between the
initiation and completion of the response to a request. From the above figure we understood that wait
time is reduced. In the fig 5 average response time of both load based allocation and round robin is
compared. The response time of load based allocation algorithm has better response than Round
Robin.

8. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 5, Issue 8, August (2014), pp. 86-94 © IAEME
Fig 5: Comparison between algorithms Load based Allocation Round Robin based on Average
Response Time
93
• Completion Time
Completion time is the time taken to complete particular number of jobs. Here we compared
the completion time of different number of jobs in Load based Allocation algorithm and Round
Robin. It is illustrated in the fig 6.From the figure we can uunderstood that Completion time also
improved in Load based Allocation algorithm than Round Robin algorithm
Fig 6: Comparison between algorithms Load based Allocation Round Robin based on Completion
Time
5. CONCLUSION
Load balancing is needed to improve the performance of a parallel and distributed system
through a redistribution of load among the processors. But static load balancing technique does not
use the system information so a dynamic based load balancing technique is proposed here. The Load
based allocation algorithm proposed here is mainly for public cloud which is open for public. This
algorithm not only balances the load but also reduces the amount of time a task or job has to wait and
thus reduces the response time. In future we can improve the algorithm by including more
techniques.

9. International Journal of Advanced Research in Engineering and Technology (IJARET), ISSN 0976 –
6480(Print), ISSN 0976 – 6499(Online) Volume 5, Issue 8, August (2014), pp. 86-94 © IAEME
94
REFERENCES
[1] http://en.wikipedia.org/wiki/Cloud_computing.
[2] National institute of standards and Technology computer security Resource Center –
www.CSRC.nist.gov.
[3] Ali M. Alakeel (2010) “A Guide to Dynamic Load Balancing in Distributed Computer
Systems”, College of Computing and Information Technology, University of Tabuk, Tabuk,
Saudi Arabia, IJCSNS International Journal of Computer Science and Network Security,
VOL.10 No.6.
[4] Gaochao Xu, Junjie Pang, and Xiaodong Fu” A Load Balancing Model Based on Cloud
Partitioning for the Public Cloud” TSINGHUA SCIENCE AND TECHNOLOGY ISSNl
l1007-0214l l04/12l lpp34-39 Volume 18, Number 1, February 2013
[4] Sachin Kumar, Niraj Singhal, “A Study on the Assessment of Load Balancing Algorithms in
Grid Based Network”, published in the International Journal of Soft Computing and
Engineering (IJSCE) ISSN: 2231-2307, Volume-2, Issue-1, March 2012.
[5] H. Mehta, P. Kanungo, and M. Chandwani (2011) “Decentralized content aware load
balancing algorithm for distributed computing environments”, Proceedings of the
International Conference Workshop on Emerging Trends in Technology (ICWET),
pages 370-375.
[6] Y. Lua, Q. Xiea, G. Kliotb, A. Gellerb, J. R. Larusb, and A. Greenber (2011) “Join- Idle-
Queue: A novel load balancing algorithm for dynamically scalable web services”, An
international Journal on Performance evaluation, In Press, Accepted Manuscript, Available
online.
[7] Z. Zhang, and X. Zhang (2010) “A Load Balancing Mechanism Based on Ant Colony and
Complex Network Theory in Open Cloud Computing Federation”, Proceedings of 2nd
International Conference on Industrial Mechatronics and Automation (ICIMA), Wuhan,
China, pages 240-243.
[8] S. Wang, K. Yan, W. Liao, and S. Wang (2010) “Towards a Load Balancing in a Three-level
Cloud Computing Network”, Proceedings of the 3rd IEEE International Conference on
Computer Science and Information Technology (ICCSIT), Chengdu, China, pages 108-113.
[9] M. Randles, D. Lamb, and A. Taleb-Bendiab (2010) “A Comparative Study into Distributed
Load Balancing Algorithms for Cloud Computing”, Proceedings of 24th IEEE International
Conference on Advanced Information Networking and Applications Workshops, Perth,
Australia, pages 551-556.
[10] Dhinesh Babu L.D, P. Venkata Krishna” Honey bee behavior inspired load balancing of tasks
in cloud computing environments” Applied Soft Computing13 (2013) 2292–2303.
[11] Sandeep Sharma, Sarabjit Singh, and Meenakshi Sharma, “Performance Analysis of load
Balancing Algorithms”, published in the proceedings of World Academy of Science,
Engineering and Technology Volume, 28 April 2008, ISSN 2070-3740.
[12] D.Asir, Shamila Ebenezer and Daniel.D, “Adaptive Load Balancing Techniques in Global
Scale Grid Environment”, International Journal of Computer Engineering Technology
(IJCET), Volume 1, Issue 2, 2010, pp. 85 - 96, ISSN Print: 0976 – 6367, ISSN Online:
0976 – 6375.
[13] M. A. Mahajan and G. T. Chavan, “Use of Multiple Ant Colony Optimization for Load Balancing
in Peer to Peer Networks”, International Journal of Computer Engineering Technology
(IJCET), Volume 4, Issue 3, 2013, pp. 419 - 425, ISSN Print: 0976 – 6367, ISSN Online:
0976 – 6375.