Hybrid Scheduling Algorithm for Efficient Load Balancing In Cloud Computing

Int. J. Advanced Networking and Applications
Volume: 08 Issue: 05 Pages: 3181-3187 (2017) ISSN: 0975-0290 3181
Hybrid Scheduling Algorithm forEfficient
Load Balancing In Cloud Computing
Navpreet Singh
M.tech Scholar, CSE & IT Deptt., BBSB Engineering College, Fatehgarh Sahib,
Punjab, India
(IKG - Punjab Technical University, Jalandhar) navpreetsaini26@gmail.com
Dr. Kanwalvir Singh Dhindsa
Professor, CSE & IT Deptt.,
BBSB Engineering College, Fatehgarh Sahib,
Punjab, India
(IKG - Punjab Technical University, Jalandhar) kanwalvir.singh@bbsbec.ac.in
----------------------------------------------------------------------ABSTRACT-----------------------------------------------------------
In cloud computing environment, various users send requests for the transmission of data for different demands.
The access to different number of users increase load on the cloud servers. Due to this, the cloud server does not
provide best efficiency. To provide best efficiency, load has to be balanced. The highlight of this work is the
division of different jobs into tasks. The job dependency checking is done on the basis of directed acyclic graph.
The dependency checking the make span has to be created on the basis of first come first serve and priority based
scheduling algorithms. In this paper, each scheduling algorithm has been implemented sequentially and the
hybrid algorithm (round robin and priority based) has also been compared with other scheduling algorithms.
Keywords: Closest data center, Optimized response time, Dynamic Load, Round-Robin scheduling, Priority based
scheduling.
--------------------------------------------------------------------------------------------------------------------------------------------------
Date of Submission: March 29, 2017 Date of Acceptance: April 11, 2017
--------------------------------------------------------------------------------------------------------------------------------------------------
1. INTRODUCTION
CLOUD computing is a technology that delivers
computing as service, rather than a product, where
shared resources, information and software are provided
over the network. The cloud is a pool of heterogeneous
resources [1]. Its providers deliver the applications over
the internet, which are accessible from any web browser
connected to internet. The quality of the service
provided has improved since the burden of managing the
resources and their implementations has been shifted to
the provider of the service. Hence cloud computing
model has been of a huge benefit to the end users, IT
buyers, software developers, system administrators and
other corporate clients since it features low operational
cost, ensures availability of pool of resources, free from
capital cost and security. Cloud computing resources are
provisioned to the end users in the form of services as
pay-per-view basis.
The service providers own and manage the Data
Centers at various locations, and these data centers may
be configured with different hardware depending on its
utilization. The hardware also keeps on changing with
time depending upon the user requirement.
Cloud load balancing refers to distributing client
requests across multiple application servers that are
running in a cloud environment [3]. Cloud service
providers maintain the service level agreement by
scheduling the available resources efficiently and the
time performance optimization is achieved by deploying
the application on proper virtual machines according to
the service level agreement. Efficient allocation of
virtual machines is provisioned in two different steps: (a)
static planning initially: group the set of virtual machines;
classify them, and deployment on physical host. (b)
provisioning of resources dynamically: depending upon
the workload; virtual machines are created and additional
resources are allocated dynamically.
Cloud load can be balanced by serving the requests
received to the nearest data center. This approach is called
Closest Data Center approach. It reduces the network
costs and has serving capability to a limited area of
request generation.
2. REVIEW OF LITERATURE
Many existing literature works have been reviewed as
described below:
Xiao et.al[2] proposed an architecture, in which, to
achieve the service level objective, all available
hardware resources are pooled in a common shared
space in cloud computing infrastructure, from which the
hosted applications can access the resources as per their
needs. Das et.al[3] proposed a utility function as a
general two tier architecture for dynamic and
autonomous resource allocation. The function consisted
of a local agent that was responsible for calculating the
utilities, for current or fore casted workload. The results
were then transferred to global arbiter, which computes
near-optimal configuration of resources.Segal et.al[4]
described an architecture for dynamic scaling of web
application. It consisted of front end load balancer, a
number of web application virtual machine. Provisioning
and DE-provisioning of virtual machines were controlled

by a dynamic scaling algorithm based on the relevant
threshold of web application.
Jiyin et.al[5] proposed an adaptive resource
allocation mechanism for cloud system with pre-
emptible task execution to increase cloud utilization.
However, this approach was not apt for time
optimization and cost optimization.Buyya et.al[6]
through a case study presented how Cloud Analyst tool
can be used to model and evaluate a real world problem.
The case study was of a social networking application
hosted on the cloud. It illustrated the work as how a
simulator can be used to effectively identify the usage
patterns which can affect the data centers hosting the
application.Soklic [7] studied comparisons between
various load balancing techniques and proposed that
static load balancing algorithms are more stable but at
the same time, dynamic distributed algos are always
considered better than the static balancing algorithms.
Kaur [8] discussed VM Load balancer
algorithm to find the suitable virtual machine in a short
period of time. Author proposed to count the max length
of the virtual machine to allocate a new request. If the
length of the current virtual machine is not sufficient,
then a new virtual machine would be added. Jieqing
et.al[9] proposed an algorithm for adding capacity to the
dynamic balance mechanism for the cloud. The
algorithm obtained better load balancing degree by
taking lesser time all loaded tasks.
3. CLASSIFICATION OF
LOAD BALANCING
ALGORITHMS
Load balancers implement type specific algorithms to
make load balancing decisions. The decision determines
to which remote server a new job is to be forwarded.
Few of the algorithms for load balancing are studied in
this section. Depending on system state, load balancing
algorithms can be divided into two types as static and
dynamic [10]. A static load balancing algorithm does not
take into account the previous state or behavior of a node
while distributing the load. On the other hand, a dynamic
load balancing algorithm checks the previous state of a
node while distributing the load, such as CPU load,
amount of memory used, delay or network load, and so
on.
Static Algorithm: Static algorithms are appropriate
for systems with low variations in load. In static
algorithm, the traffic is divided evenly among the
servers. This algorithm requires a prior knowledge of
system resources. The performance of the processors is
determined at the beginning of the execution.
Therefore, the decision of shifting of the load does not
depend on the current state of system. However, static
load balancing algorithms have a drawback. In that, the
tasks are assigned to the processor or machines only
after it is created and those tasks cannot be shifted
during its execution to any other machine for load
balancing. Advantages: - Performs better in terms of
complexity issue.
Disadvantage: - Compromises with the result as decision
is solely made on statically gathered data. Also, the
algorithms are non-preemptive.
Types of Static Algorithm: -
Round Robin scheduling: Load is distributed evenly
to all the nodes. Equal load is assigned to
each node in circular order without any priority and back
to first node when the last node is reached. It is easy to
implement, simple and starvation free.
Threshold algorithm: Load is assigned immediately
on creation of the node. Each node has a load limit.
When load state of a node exceeds its limit, a message is
sent to all remote nodes regarding a new load state.
Randomized algorithm: A node is selected randomly.
When the load exceeds the node’s limit, it is migrated to
a randomly selected new neighbor. It does not send any
load message to other remote nodes. But it causes much
communication overheads due to random selection of
nodes.
Dynamic Algorithm: In dynamic algorithm, the
server with the least load in the whole network or system
is searched and preferred for balancing a load. For this,
real time communication with network is needed which
can increase the traffic in the system. Here, current state
of the system is used to make decisions to manage the
load. Dynamic algorithms respond to the actual current
system state in making load transfer decisions. Since
current state of the system is used to make dynamic load
balancing decisions, processes are allowed to move from
an over utilized machine to an underutilized machine in
real time dynamically.
Advantages: - No single web server will be overloaded.
Also, better performance report as it considers current
load of system to choose next data center. The
algorithms are preemptive.
Disadvantage: - Higher run time complexity.
Communication overheads occur more and more as
number of processes increase.
Types of Dynamic Algorithm: -
Priority based scheduling: The priorities are
calculated during the execution of the system. Higher the
priority of the request earlier will be resource allocation to
the request. The goal of dynamic priority is to adapt to
dynamically changing progress.
Central Queue Algorithm: Any new and pending
activities are stored in a cyclic FIFO queue. This
algorithm needs high communication among needs.
Whenever a new request is received, first activity is
removed from the queue.
Least Connection Algorithm: It decides the load
distribution depending upon the present number of
connections on a node. A load balancer maintains a log of
number of connections on a node. Load increases with
every new activity or a request, whereas load decreases
when an activity finishes. Nodes with lesser connections
are selected first.

4. CRITERIA FOR EFFICIENT LOAD
BALANCING IN CLOUD COMPUTING
Cloud computing is process of execution of various tasks
over the network in such a manner that user does not
know any information about hardware components.
Load balancing at different datacenters must be
achieved in such a way so that minimum response time
has been achieved by the system. In the process of
execution of different tasks, datacenters allocate virtual
machines to different resources that utilize different
components of virtual machines so that minimum
datacenter cost and minimum response time has been
achieved by the system. In the process of cloud
computing, load balancing policy has been designed
using hybrid round robin scheduling with priority based
approach. In the proposed work, resources have been
provided different levels of priorities for allocation of
virtual machines. Datacenters have different numbers of
physical machines and these physical machines contain
different number of virtual machines. The datacenters that
have highest number of virtual machines can process
large number of resources for output.
To schedule jobs for execution, the algorithms are very
vital. In the cloud computing domain, one of the most
challenging problems is the job scheduling algorithms.
In the table below are discussed some of the existing
algorithms for job scheduling.
Scheduling is one of the most vital tasks in cloud
computing atmosphere. In the Table 1 below, various
programming algorithmic rules and various parameters
have been analyzed. The table above suggests an
algorithmic rule for improving resource availableness
and computing in cloud computing environment.
S.No. Features Round Robin scheduling Priority Based scheduling Hybrid scheduling
1.
Load
distribution
Load is distributed evenly
among all nodes.
Equal load is assigned to
each node in circular node
without any priority and
will be back to first node if
last node has been reached.
Each request is assigned a
priority and the one with
highest priority is served first.
Requests with same priority
are served in FCFS manner. t
is a non-preemptive
scheduling algorithm.
It is a hybrid scheduling
algorithm which is a
combination of RR and PB.
Requests are first given the
priority and are then
executed in round robin
fashion.
2. Implementation
It is easy to implement, is
simple and is starvation
free.
It is complex in nature.
It is more complex and is
starvation free.
3.
Interprocess
communication
It does not require
interprocess
communication.
It requires interprocess
communication.
Requires much interprocess
communication.
4. Disadvantage
Cannot give expected result
when the jobs are of
unequal processing time.
Processes with lower priority
are not given much
opportunity to execute.
Still a better scheduling than
RR and PB individually.
Table 1: Comparison of scheduling algorithms

5. PROPOSED ALGORITHM
The proposed algorithm can be formally described by
the pseudo code as follows:
1. New Process P arrives.
2. P enters the ready queue.
3. Update the Service Rate and Arrival Rate.
4. Process P is loaded from ready queue into
CPU to be executed.
5. IF (Ready Queue is Empty), BT(P)
6. Time Quotient update SR and AR
7. End IF
8. IF (Ready Queue is Not Empty), AVG (Sum
BT of processes in TQ ready queue)
9. Update SR and AR.
10. End IF.
11. CPU executes P by TQTime.
13. IF (P is not terminated)
14. Return P to the ready queue with its updated
Burst Time.
16. End IF.
(where BT – Burst Time, TQ – Time Quotient, SR –
Service Rate, AR – Arrival Rate)
6. EXPERIMENTAL SETUP
Cloud resource allocation has been done using proposed
load balancing policy, that uses priority constraints for
allocation of different VM on to resources.
Various parameters have been used for simulation of
cloud services using different load balancing policies.
Table 2: Cloud Analyst Configuration for simulation
*(UB - user base, GMT – Greenwich Mean Time, DC -
Data Centre, PB - Priority Based, RR - Round Robin, VM
- Virtual Machine, ESCEL - Equally Spread Current
Execution Algorithm, ms – millisecond, MB – MegaByte,
bps – bits per second, MIPS – million instructions per
second)
Table 2 above represents various parameters that have
been used for simulation of the proposed work. These
parameters have been used for simulation of different
cloud service broker policies with different load
balancing policies.
7. RESULTS AND DISCUSSION
Various scheduling algorithms have been
implemented to evaluate the efficiency of overall
response time and the service time in response to
various service requests received. The results of
these evaluations have been discussed here.
Table 3 below represents overall response time and
Data Centre processing time for resources using
Closest Data Centre service broker policy.
VM Image Size 10000MB
VM Memory 1024 MB
VM Bandwidth 1000 bps
DC 1 – No. of Physical
Machine
20
Machine
10
Machine
3
DC – Memory Per Machine 2048MB
DC – Storage Per Machine 40000MB
DC – Available BW Per
Machine
40000 bps
DC – No. of Processors Per
Machine
4
DC – Processor Speed 1000MIPS
DC – VM Policy Time Shared
User Grouping Factor 10000
Request Grouping Factor 1000
Executable Instruction Length 250
Load Balancing Policy
Round Robin,
PB, ESCEL and
“Hybrid RR and
Priority”
Parameter Value Used
UB Name UB1
Region 2
Requests Per User Per Hour 60
Data Size Per Request 100
Peak hour start(GMT) 3
Peak hour end (GMT) 9
Avg Peak Users 400000
Avg Off Peak Users 40000
DC 1 – No Of VM 80

Over all
response
time (ms)
300.98 300.95 300.88 300.02
DC Service
Time (ms)
3.80 3.68 3.50 3.40
Table 3: Overall response time and DC Processing
time using Closest Data center.
Hybrid algorithm has optimized overall response time
as well as the data center service time which is 300.62
ms and 2.77 ms respectively for Closest data center
service broker policy. It is because of the proposed
hybrid algorithm, that the requests are served in the
minimal duration of time. Hybrid algorithm is most
efficient in case of Closest Data center approach because
in this approach, every service request is first forwarded
to the closest data center, hence resulting in lesser
response time. Also, its priority is sethigh.
datacenter processing time for resources using optimized
response time datacenter service broker policy.
RR PB ESCEL
Proposed
Hybrid
Over all
response
time (ms)
301.09 300.95 300.86 299.15
DC
Service
Time (ms)
3.47 3.15 2.83 2.80
time using Optimized Response Time.
Hybrid algorithm using Optimized response time
broker policy achieve optimal response time and data
center service time for all the request received. As it is
clear from table 4 above, that the suggested Hybrid
algorithm has the lowest overall response time of
299.15 ms as well as the lowest time taken to service
the data centre request
i.e. 2.80 ms, hence making it the most suitable
algorithm. This is because, unlike Round Robin
scheduling, the request having higher priority is served
first irrespective of its completion time and its request
age.
datacenter processing time for resources using
dynamic load datacenter service broker policy.
RR PB ESCEL Hybrid
time using Dynamic Load.
Hybrid algorithm is better than other mentioned
scheduling algorithms because when load is divided
dynamically, it gives better response time as
compared to all other algorithms. The overall response
time taken and Data Center Service Time is most
efficient in case of hybrid algorithm because unlike
PB scheduling, it automatically increases the priority
for the old processes having low initial priority, hence
executing them eventually. The different load
balancing and service broker policies have been used
for resource scheduling using different datacenters
virtual machines.
Fig. 1 below depicts the graphical representation of
overall response time of the user base for resource
processing by using different load balancing and
service broker policies. The overall response time of
the Hybrid scheduling algorithm to serve the request
in case of various service broker policies is much less
as compared to other scheduling algorithms which
makes it an efficient load balancing technique.
Fig.1: Over all response time using different Load
balancing and service broker policy.
The Fig.1 above shows that the overall response time
to service a request with Closest Data Center
approach is
RR PB ESCEL
Proposed
Hybrid
Over all
response
time (ms)
300.864 300.78 300.63 300.62
DC
Service
Time (ms)
3.46 3.04 2.809 2.77

least in case of the Hybrid scheduling algorithm that is
300.02 ms, which is lesser as compared to other
scheduling algorithms. This is because, every service
request is first forwarded to the closest data center, hence
resulting in lesser response time. Also, its priority is set
high.
Also, in case of Optimized Response Time approach,
the service request with a higher priority is executed first,
hence taking the least response time of 299.15 ms, which
is much lesser as compared to other scheduling
algorithms.
In the Dynamic Load approach for service execution,
Hybrid scheduling algorithm takes overall response time
of 300.02 ms which is much lesser as compared to other
scheduling algorithms. This is because, in this approach,
the load is divided among data centers dynamically. If a
server is having no load or lesser load, it is assigned a
request to service it. Hence making the process starvation
free and efficient.
Fig.2 below depicts graphical representation of
datacenter service times of various datacenters for
resource processing by using different load balancing and
service broker policies.
Fig. 2: Datacenter response time using
different Load balancing and service broker
policy
The response time taken by the datacenter to serve the
request in case of the Hybrid scheduling algorithm in
various service broker policies is much less as compared
to other scheduling algorithms which makes it an
efficient load balancing technique.
From fig.2 above, it is clear that in the Closest data
center approach, the Data center response time is least is
case of Hybrid scheduling algorithm, which is 2.77ms,
lower as compared to other scheduling algorithms. This
is because, in this approach, any service request received
by the broker is assigned to the closest data center to be
processed. Hence it takes the least time to service it.
Using the Optimized Response Time approach, Hybrid
scheduling takes only 2.80 ms for a Data center to
respond to a request. It is lesser as compared to other
scheduling algorithms for the same approach because in
Hybrid scheduling, the priority of a process increases
automatically with the age of the request received.
Hence making it more efficient as compared to other
algorithms. The proposed Hybrid scheduling algorithm
is a dynamic scheduling algorithm. That means, the
requests are alloted to different servers dynamically for
execution. The requests are divided depending upon the
load of a server and its capacity to service a request.
Because of this approach, datacenter response time
taken for Hybrid algorithm is 3.40 ms which is least as
compared to other scheduling algorithms for the same
approach.
Here various types of scheduling have been discussed.
And also, the results of implementation of various types
of scheduling algorithms in CloudSim have been shown
in the Fig.1 and Fig.2 above. With this experimental
setup, the performance of scheduling algorithms have
been analyzed. The proposed Hybrid algorithm of
Round Robin and Scheduling combined has shown
impressive results as compared to other algorithms in
achieving maximum utilization of resources in minimal
duration of time. According to the results of hybrid
Round Robin and Priority Based Scheduling algorithm,
all the scheduling criterion is perfect as compared to
other scheduling algorithms.
Table 6: Comparison of load balancing scheduling
algorithms
S.No. Features RR PB Hybrid
1. Load distribution
This is a priority free
algorithm.
Each process is given a
priority.
Each process has a priority.
2. Request execution
Execution is done in a circular
order, one after another.
Higher priority requests
are served first.
Larger priority processes have large
time slices.
3. Improvement
It doesn’t have starvation but for
large number of processes, wait
time can often be too long.
Indefinite blocking for
low priority requests.
Requests are executed on basis of
their priority and the priority
increases gradually for aging
requests.

The proposed hybrid scheduling algorithm is a
combination of round robin and priority based
scheduling algorithm. So it features all the properties
of both algorithms and also satisfies any limitation in
any of the two algorithms. As it is clear from table 6
above that, the proposed hybrid algorithm provides
each process a priority and priority increases for the
aging requests, it makes this algorithm more efficient.
By analyzing performance evaluation parameters, it
can be concluded that proposed approach provides
much better results than previous approaches.
7. CONCLUSION
In cloud computing scenario, number of tasks has to
be assigned on various processes to handle load on the
cloud. These tasks have been divided into sets and the
dependency checking is done for prevention of dead
lock state or to prevent demand of various extra
resources for allocation. Make span has been
developed based on the allocation. Tasks must be
checked for dependency by using directed Acyclic
Graph. The tasks that are divided by following the
Hybrid scheduling are more efficient as compared to
all existing load scheduling algorithms. It also ensures
the efficient and fair distribution of all computing
resources.
REFERENCES
[1] K. Kishor, V. Thapar, “An efficient service broker
policy for Cloud computing environment”,
International Journal of Computer Science Trends and
Technology (IJCST) - Vol. 2, Issue 4, July-Aug 2014.
[2] Qiang Li, Qinfen Hao, Limin Xiao, “Adaptive
management of virtualized resources in cloud
[3] T.C. Chieu, A. Mohindra, A.A. Karve, A. Segal,
“Dynamic Scaling of Web Applications in a
Virtualized Cloud Computing Environment,” in IEEE
International Conference on e-Business Engineering,
pp. 281-286, Dec. 2009.
[4] L. Jiyin, Q. Meikang, J.W. Niu, Y. Chen, and
Ming, Adaptive Resource Allocation for
Preeemptable Jobs in Cloud Systems, IEEE
[8] Zhang Bo, Gao Ji, Ai Jieqing, “Cloud
Loading Balance algorithm”, Information
Science and Engineering, Second International
Conference, Vol.2, No.5, 4-6 Dec. 2010.
[9] Z. Xiao, W. Song, and Q. Chen, “Dynamic
resource allocation using virtual machines for
cloud computing environment,” IEEE
Transactions on Parallel and Distributed
Systems, Vol. 24, No. 6, pp. 1107-1117, 2013.
[10] Di Caro, G. Dorigo, M. Mobile, “Agents
for adaptive routing” in Proceedings of the
Thirty-First Hawaii International Conference on
System Sciences, Kohala Coast, HI, USA, Vol.
7, pp. 74–83, January 1998.
[11] Bo, Z., Ji, G., Jieqing, A., “Cloud Loading
Balance algorithm”, Proceedings of the 2010
2nd International Conference on Information
Science and Engineering, Hangzhou, China, pp.
5001–5004, December 2010.
[12] Wu Lee, Chan, Huang, “Dynamic load
balancing mechanism based on cloud storage”,
Computing, Communications and Applications
Conference, Hong Kong, China, pp. 102–106,
January 2012.
[13] A. Bhadani, Chaudhary, “Performance
evaluation of web servers using central load
balancing policy over virtual machines on
cloud”, Third Annual ACM Bangalore
Conference, Bangalore, India, pp. 16-19,
January 2010.
[14] K. Nishant, P. Sharma, V.Krishna, C.
Gupta, K.P. Singh, N. Nitin, R. Rastogi, “Lord
Balancing of Nodes in Cloud Using Ant Colony
Optimization”, 2012 UK Sim 14th
International
Conference on Computer Modelling and
Simulation, Cambridge, UK, pp.3-8, March
2012.
[15] A.P. Deshmukh, Prof. K.Pamu, “Applying
Load Balancing: A Dynamic Approach”,
(IJARCSSE), Vol.2, Issue 6, June 2012.
International Conference on Intelligent Systems
Design and Applications, pp. 31-36, 2010.
[5] B. Wickremasinghe, N. Rodrigo Calheiros, and
R. Buyya, Cloud Analyst: A CloudSim-based Visual
Modeller for Analyzing Cloud Computing
Environments andApplications, IEEE, January 2011.
[6] Milan E. Soklic “Simulation of Load balancing
algorithms” ACM-SIGCSE Bulletin, Vol.34, Issue 4,
December 2002.
[7] Jaspreet Kaur, “Comparison of Load balancing
algorithms in a Cloud”, International Journal of
Engineering Research and Applications”, Vol. 2, Issue
3, May-June 2012.

Hybrid Scheduling Algorithm for Efficient Load Balancing In Cloud Computing

Recommended

More Related Content

What's hot (19)

Similar to Hybrid Scheduling Algorithm for Efficient Load Balancing In Cloud Computing (20)

More from Eswar Publications (20)

Recently uploaded (20)

Hybrid Scheduling Algorithm for Efficient Load Balancing In Cloud Computing