Improving Cloud Performance through Performance Based Load Balancing Approach

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395 -0056
Volume: 04 Issue: 06 | June -2017 www.irjet.net p-ISSN: 2395-0072
© 2017, IRJET | Impact Factor value: 5.181 | ISO 9001:2008 Certified Journal | Page 523
Improving Cloud Performance through Performance Based Load
Balancing Approach
Er. Surbhi Sharma1, Er. Intiyaz Ahmad2, Er. Sourav Mirdha3
1, 2M.Tech. Student, Computer Science & Engineering, International Institute of Engineering & Technology,
Samani, Kurukshetra, Haryana, India
3Assistant Professor, Computer Science & Engineering, International Institute of Engineering & Technology,
Samani, Kurukshetra, Haryana, India
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - Among the major issues ofcloudcomputing, load
balancing is the critical issue. It can be achieved through task
scheduling, resource management, task resource mapping,
efficient virtualization and alsobyavoidingfaultandhandling
the situation of fault. Fault tolerance is also one of the critical
issues. Major work associated with fault tolerance is its
detection in advance followed by recovery from it. To tackle
this issue, different researchers have given different
methodologies. Quality of service provided by CSP can be
improved by providing desired resources well in time with
minimization of response, service time and failure.
In this paper, authors have tried to improve the cloud
performance through load balancing with fault tolerance.
Fault handler, redundancy and check pointing have beenused
to implement fault tolerance (reactive and proactive). This
removes the faulty node and does not make them availablefor
task assignment till its recovery. Also while distributing load
among nodes, success ratio and past load data is also
considered. This has improved the quality of service as task is
getting mapped with that node whosesuccessrateismoreand
present load is less.
Key Words: Cloud Computing, Load Balancing, Fault
Tolerance, Virtualization, Cloudsim
1. INTRODUCTION
Cloud computing has recently emerged as a new form of the
utility-based computing paradigmforhostinganddelivering
hardware and software “as services”. It provides its users
with the illusion of infinite computing and storageresources
which are potentially available on-demand from anywhere
and anytime. Cloud computing is attractive since it
eliminates the requirement for its users to plan ahead for
provisioning, by allowing IT enterprises to start from the
small and to increase resources only when there is a rise in
service demand. However, despite of this, the development
of techniques to make cloud computing effectiveiscurrently
at its infancy, with many issues still to be addressed [1].
“A Cloud is a type of parallel and distributed system
consisting of a collection of inter-connected and virtualized
computers that are dynamically provisioned and presented
as one or more unified computing resource(s) based on
service-level agreements established through negotiation
between the service provider and consumers” [2].
Essential characteristics of cloud computing are as follows:
 On-demand self-service: a consumer can
autonomously provision computing capabilities (e.g.,
computing power, storage space, network bandwidth),
that is without requiring human interaction with the
respective provider(s);
 Rapid elasticity: the above capabilities may be
dynamically resized in order to quickly scale up (to
potentially unlimited size) or down in according to the
specific needs of the consumer [3].
1.1 Architecture of Cloud System
A cloud system, that is a system which adopts the cloud
computing paradigm, canbecharacterizedbyitsarchitecture
and the services it offers. The architecture of a cloud
computing system is usually structured as a set of layers. A
typical architecture of a cloud system is shown in Figure 1
(from [4]). At the lowest level of the hierarchy there is the
hardware layer, which is responsible for managing the
physical resources of the cloud system, such as servers,
storage, network devices, power and cooling systems. Onthe
top of the hardware layer, resides the infrastructure layer,
which provides a pool of computingandstorageresourcesby
partitioning the physical resources of the hardware layer by
means of virtualization technologies. Built on top of the
infrastructure layer, the platform layer consists of operating
systems and application frameworks. The purpose of this
layer is to minimize the burden of deploying applications
directly onto infrastructure resources by providing support
for implementing storage, database and business logic of
cloud applications. Finally, at the highest level of the
hierarchy there is the application layer, which consists of
cloud applications.
For what regards services implemented on top of a cloud
computing system, they can be provided in three modality,
according to the abstraction level of the capability provided
and the service model of providers [2]:
 Infrastructure as a Service (IaaS), which
comprises servicestoallowitsconsumerstorequest
computational, storage and communication

resources on-demand, thus enabling the so called
“pay-per-use” paradigm whereby consumers can
pay forexactly the amount ofresource theyuse(like
for electricity or water). The consumers can use the
provided resources to deploy and run arbitrary
software; however, the management and control of
the underlying cloud infrastructure is possible only
by the provider. An example is Amazon EC2 [5].
Fig -1: The architecture of a cloud system (from [4])
 Platform as a Service (PaaS), which comprises
high-level services providing an independent
platform to manage softwareinfrastructures,where
consumers (i.e., developers) can build and deploy
particularclassesofapplicationsusingprogramming
languages, libraries, and tools supported by the
provider. Usually, consumers don’t manage or
control the underlying infrastructure (such as
servers, network, storage, or operating systems),
which can only be accessed by means of the high-
level services provided by the provider. An example
is Google App Engine [6].
 Software as a Service (SaaS), which comprises
specific end-user applications running on a cloud
infrastructure. Such applications are delivered to
consumer as a network service (accessible from
various client devices, ranging from desktop
computers to smart phones), thus eliminating the
need to install and run the application on the
consumer’s own computers and simplifying
maintenance and support. Consumersdon’tmanage
or control the underlying infrastructure and
application platform; only limited user-specific
application configurations are possible. An example
is Salesforce.com [7].
The traditional approach to deploy a cloud system is a public
computing system. However, other deployment models are
possible which differentiate each other’s by variations in
physical location and distribution.Forinstance,thefollowing
models are taken from NIST [2]:
 Publiccloud:thecloudinfrastructureisprovisioned
for open use by the general public and is made
available in a “pay-per-use” manner;
 Private cloud: the cloud infrastructure is
provisioned for exclusive use by a single
organization comprising multiple users;
 Community cloud: the cloud infrastructure is
provisioned for exclusive use by a specific
community of users from organizations that have
shared concerns (e.g., mission, security
requirements, policy, and compliance
considerations);
 Hybrid cloud: the cloud infrastructure is a
composition of two or more distinct cloud
infrastructures (private, community, or public) that
remain unique entities, but are bound together by
technology that enables data and application
portability. A typical example is when a private
cloud is temporarily supplemented with computing
capacity from public clouds, in order to manage
peaks in load (also known as “cloud-bursting”) [3].
2. LOAD BALANCING & FAULT TOLERANCE
Load balancing can be defined as the process of task
distribution among multiple computers, processes, disk, or
other resources in order to get optimal resource utilization
and to reduce the computation time. Load balancing is an
important means to achieve effective resource sharing and
utilization. In general, load balancing algorithms can be
divided into following three types[8]:
o Centralized approach: In thisapproach, a singlenodeis
responsible for managing the distribution within the
whole system.
o Distributed approach: In this approach, each node
independently builds its own load vector by collecting
the load information of other nodes. Decisions are made
locally using local load vectors. This approach is more
suitable for widely distributed systems such as cloud
computing.
o Mixed approach: A combination between the two
approaches to take advantage of each approach [9].
Fault tolerance is an approach where a system continues to
work properly even if there is a fault. There are number of
fault tolerant techniques areavailable butstillfaulttolerance
in cloud computing is a difficult task. Because of the wide
spread infrastructure of cloud and the increasing demand of
services, an efficient fault tolerant technique for cloud
computing is essential. But due to its virtualization and
internet based service providing behavior, fault tolerance in
cloud computing is still a major problem. The main fault
tolerance issues in cloud computing are detection and
recovery. Fault tolerance mechanism can be implemented at
task and work flow level. Fault tolerance mechanism can be
divided into two categories [10]:
 Proactive Fault Tolerance
 Reactive Fault Tolerance

Proactive Fault Tolerance:
We try to identify the components which may cause fault
and replace them in advance. Some of the commonly used
techniques based upon this theory are as follows:
 Preemptive Migration: It depends upon the
feedback mechanism where system is consistently
analyzed.
 Self Healing: This is automatically used to handle
the failure situation when many instances of the
same application are running.
 Software Rejuvenation: In this methodology
system reboots itself after certain period of time
with clean state [11].
Reactive Fault Tolerance:
This type of policies comes in action after occurrence of
failure and tries to minimize the effect of failure. Techniques
based upon this policy are as follows:
 Rescue workflow: In this technique, system will
keep on working until it becomes impossible to
move forward.
 Task Resubmission: This is the most commonly
used technique where failed task is resubmitted
from the beginning.
 Task Migration: After failure, pending task may be
migrated to other machines.
 Check Point: When a task fails, it is allowed to
restart from the last entry done for check point
purpose [12, 13].
3. PROPOSED WORK
The proposed load balancing model has used the logic of
reactive fault tolerance. Success ratio is assigned to all
virtual modes. In the beginning it is .5, while its maximum
value is 1. A virtual node becomes eligible for selection if
success ratio is lying in (0, 1]. If it is not lying in this interval
then that node is not eligible for selection. Diagrammatical
representation of proposed approach is shown in figure 1.
Fig -1: Proposed Approach
Steps of the proposed approach are as follows:
1. User interacts with the CSP through the provided
graphical interface.
2. CSP forwards the user request to cloud manager
(CM). It maintains the Performance record (PR)
table which stores the following entries:
a. Id of virtual node
b. Id of associated physical machines
c. Success ratio of virtual node
d. Task assignment counter
e. No of times node has given the successful
results (a)
f. No of times tasks has been assigned to
node (b)
3. CM forwards the request towards scheduler which
does the load balancing activity. Scheduler access
the PR table to assign the task to that VM whose SR
is good and present load is less.
4. Whenever a node get fails, fault handler comes in to
action. It updates the record of nodes performance
in PR table and either restart the server or calls
scheduler to transfer the pending task.
5. Execution results are transferred to decisionmaker
module (DM). Through status checker (SC), it gets
the information about the status of all virtual
machines. DM checks the deadlines of the tasks
through Task Deadline Component (TDC).
a. If both SC & TDC for a VM results in success
then its SR is incremented and PR table is
updated. SR=a++/b++;
b. If SC results in fail, then fault handler is
called to handle the situation. SR=a/b++;
c. If SC does not return in fail but TDC results
in fail, then its SR is decremented and PR
table is updated
d. DM maintains the list of all those VM who’s
SC & TDC results in success. Highest SR
value of VM is considered as checkpointfor
further executions.
Proposed work in the algorithmic form is as follows:
Algorithm LBFT ( )
{
Identify the different available virtual machines;
V= {V1, V2,……..Vn}
// Set of available virtual machines
For (i=1 to n)
{
SR(Vi)=.5
// Initially Success ratio of all virtual machine is same
Store the following info in the
performance record table for each VM;
i. Id of virtual node
ii. Id of associated physical machines
iii. Success ratio of virtual node
iv. Task assignment counter

v. No of times node has given the
successful results, its initial value is
1 (a)
vi. No of times tasks has been assigned
to node, its initial value is 2 (b)
}
While (Task is there in the data center)
{
 Calculate Priority=success
ratio/load for each virtual
machine in the performance
record table. If load is zero then
Priority=success ratio;
 Sort the performance record
table on the basis of Priority;
 Select the highest Priority virtual
machine from the priority table;
 Assign the task to the selected
VM;
 Update the performance record
table.
If (status checker of the machine is not
fail and task is completed before
deadline)
{
Value of SR is updated,
SR=a++/b++;
}
Else if (status checker returns in fail task
is completed before deadline)
{
Value of SR is updated,
SR=a/b++;
Fault handler is called to handle
fault situation;
}
Else
{
Value of SR is decremented;
}
}
Decision maker maintains the list of all those VM
whose status checker & task deadline controller
results in success. Highest SR value of VM is
considered as checkpoint for further executions.
}
Algorithm Fault_handler(id of virtual machine)
{
 Recalculate the success ratio of received virtual
machine;
 Transfer the pending task to other VM using the
same approach;
}
4. SIMULATOR AND RESULTS
We can analyze the performance of any load balancing
algorithm by actually testing it in cloud environment on
various parameters. But it is very costly and difficult to
manage the cloud environmentonlyforexperimentpurpose.
So there is a need of simulator to test the load balancing
algorithm in cloud environment.
We have used Cloudsim simulator which is free and open
source software available at
http://www.cloudbus.org/CloudSim/. It is a code library
based on Java. This library can be directly used by
integrating with the JDK to compile and execute the code.
For rapid applications development and testing, Cloudsimis
integrated with Java-based IDEs (Integrated Development
Environment) including Eclipse or NetBeans. Using Eclipse
or NetBeans IDE, the Cloudsim library can be accessed and
the cloud algorithm can be implemented [14].
To analyze the variation of success ratio for different virtual
machines, we have considered three different virtual
machines having initial success ratio .5. We have analyzed
the performance of three virtual machines 10 times. Same
set of tasks are assigned to all virtual machines. In chart 1, 2
and 3 analysis of success ratio for three virtual machines is
given. It has been found that success ratio of virtual machine
1 is almost continuously increasing while for virtual node 2
it is following a random walk pattern. For virtual machine 3
success ratio is decreasing for first half while in second half
it is increasing.
Chart -1: Success ratio analysis of virtual node 1

Table -1: Node Selection Procedure
Sr.
No.
Task
Deadline
Status Deadline
Finish
Time
Success
Ratio
Status Deadline
Finish
Time
Success
Ratio
Status Deadline
Finish
Time
Success
Ratio
Start - - - - 0.5 - - - 0.5 - - - 0.5 -
1 1700 Success Success 1600 0.667 Success Success 1602 0.667 Success Success 1610 0.667 1
2 1602 Success Success 1600 0.75 Success Success 1602 0.75 Success Fail 1610 0.5 2
3 1601 Success Success 1600..8 0.8 Success Fail 1602 0.6 Success Fail 1611 0.4 1
4 1605 Success Success 1601 0.833 Success Success 1603 0.667 Success Fail 1611 0.333 1
5 1600 Success Fail 1602 0.714 Success Fail 1603 0.571 Success Fail 1612 0.286 -
9 1700 Success Success 1603 0.818 Fail Fail - 0.636 Success Success 1613 0.545 1
VirtualMachine1 VirtualMachine2 VirtualMachine3
Selected
Node
Table 1 shows the execution of proposed algorithmforthree
virtual machines and same set of tasks for 10 times.
Arbitrarily task deadlines are assigned. Initially,all thethree
virtual machines have same success ratio .5. So during first
execution, any virtual machine can be selected randomly. In
the second execution, two machines have the same success
ratio, so among them any one can be selected randomly. In
this way, algorithm ensures that every time task will be
mapped with the best available virtual machine. In case
status of machine become fail, them immediately fault
handler is called. Pending tasks are transferred to other
virtual machines using the same strategy. Maximumsuccess
ratio is updated after execution of every cycle by the highest
success ratio of the virtual machine.
Chart 4 shows the comparison of three virtual machines on
the basis of number of average task completion time. It has
been found that VM1 shows the best performance and VM3
shows the worst performancewhileperformanceofVM2lies
between VM1 & VM3. Same kind of relationship has been
found when we have compared the performanceofthreeVM
on the basis of number of tasks completed before deadline
out of 10. VM1 has completed maximum number of tasks
while VM3 has completed the minimum number of tasks
while VM2 performance lies between VM1 & VM3. This
result has been shown graphically in chart 5.
So it can be concluded that priority calculated on the basisof
success ratio and load is good scale to select the appropriate
VM.
5. CONCLUSIONS
In this paper, author has presented a new load balancing
approach by impartingtheconceptoffaulttolerance,success
ratio and present load. Success ratio for every virtual
machine is calculated based upon its past performance.
Based upon the success ratio and present load, priority of
each virtual machine is calculated which becomes the
deciding factor for selection of virtual machine. Also, on
failure of any virtual machine, pending tasks are transferred
to other machines. As we are considering the past
performance of virtual machine, while mapping task with it,
so this makes our approach faulttolerant.Aslessperforming
virtual machines have low priority and less chance of
selection. So in this way we have added proactive and
reactive fault tolerance feature with the proposed load
balancing approach.
To improve the proposed approach, we can embed the
mechanism of load transferfromoverloadedvirtual machine
to under loaded virtual machine. Also we can embed the
resource allotment logic with the proposed load balancing
approach.

REFERENCES
[1] A. Jain and R. Kumar, “A TaxonomyofCloudComputing,”
International Journal of Scientific and Research
Publications. vol. 4(7), Jul. 2014, pp. 1-5.
[2] P. Mell and T. Grance, “The NIST definition of Cloud
computing: Recommendations of the national institute
of standards and technology,” Special Publication 800-
145, NIST, Sep 2011.
[3] R. Buyya, C. S. Yeo, S. Venugopal, J. Broberg, and I.
Brandic, “Cloud computing and emerging IT platforms:
Vision, hype, and reality for delivering computing asthe
5th utility,”. Future Generation Computer Systems, vol.
25(6), 2009, pp. 599–616.
[4] Q. Zhang, L. Cheng, and R. Boutaba, “Cloud computing:
state-of-the-art and research challenges,” Journal of
Internet Services and Applications, vol. 1(1), 2010, pp.
7–18.
[5] Amazon elastic compute cloud (EC2). Available:
http://aws.amazon.com/ec2.
[6] Google AppEngine: Run your web apps on Google’s
infrastructure.Available:http://code.google.com/appeng
ine/
[7] Salesforce.com. Available: http://www.salesforce.com.
[8] A. Jain and R. Kumar, “A Multi Stage Load Balancing
Technique for Cloud Environment.” International
Conference on Information Communication and
Embedded Systems (ICICES), Feb 2016, pp. 1-7.
[9] A. Khiyaita, M. Zbakh, H. El Bakkali and D. El Kettani,
"Load balancing cloud computing: State of art," in
Proceedings of 2012 National Days of Network Security
and Systems (JNS2), 20-21 April 2012, pp. 106-109.
[10] A. Bala and I. Chana, “Fault Tolerance- Challenges,
Techniques and Implementation in Cloud Computing,"
IJCSI International Journal of Computer Science Issues,
vol. 9(1),, January 2012.
[11] R. Jhawar, V. Piuri, and M. D. Santambrogio, "Fault
Tolerance Management in Cloud Computing: A System
Level Perspective," IEEE InternationalSystems Journal,
vol. 7(2), June 2013, pp. 288-297.
[12] I. P. Egwutuoha., S. Chen, D. Levy, and B. Selic, “A Fault
Tolerance Framework for HighPerformanceComputing
in Cloud,” IEEE/ACM International Symposium on
Cluster, Cloud and Grid Computing (CCGrid), IEEE, May
2012, pp. 709-710.
[13] D. Poola, K. Ramamohanarao, and R. Buyya, “Fault
Tolerant Workflow Scheduling Using Spot Instances on
Clouds,” 14th ELSEVIER International Conference on
Computational Science (ICCS), 2014, pp. 523-533.
[14] RN. Calheiros, R. Ranjan, A. Beloglazov, CA. De Rose, R.
Buyya, “CloudSim: a toolkit for modeling andsimulation
of cloud computing environments and evaluation of
resource provisioning algorithms,” Software: Practice
and experience. vol. 41(1), Jan 2011, pp.23-50.

Improving Cloud Performance through Performance Based Load Balancing Approach

Recommended

More Related Content

What's hot (20)

Similar to Improving Cloud Performance through Performance Based Load Balancing Approach (20)

More from IRJET Journal (20)

Recently uploaded (20)

Improving Cloud Performance through Performance Based Load Balancing Approach